The principal aim of this book is to develop and defend an analysis of the

concept of moral obligation. The analysis is neutral regarding competing

substantive theories of obligation, whether consequentialist or deontolog-
ical in character. It seeks to generate new solutions to a range of philo-
sophical problems concerning the concept of obligation and its application.
Among the topics treated are deontic paradoxes, the supersession of
obligation, conditional obligation, prima facie obligation, actualism and
possibilism, dilemmas, supererogation, and cooperation. By virtue of its
normative neutrality, the analysis provides a theoretical framework within
which competing substantive theories of obligation can be developed and
assessed.
CAMBRIDGE STUDIES IN PHILOSOPHY

The Concept of Moral Obligation

 CAMBRIDGE STUDIES IN PHILOSOPHY

General editor Ernest Sosa

Advisory editors
Jonathan Dancy University of Keele
Gilbert Harman Princeton University
Frank Jackson Australian National University
William G. Lycan University of North Carolina, Chapel Hill
Sidney Shoemaker Cornell University
Judith J. Thomson Massachusetts Institute of Technology

RECENT TITLES
Andrew Newman The Physical Basis of Prediction
David Owens Causes and Coincidences
Jaegwon Kim Supervenience and Mind
Warren Quinn Morality and Action
Michael Jubien Ontology, Modality, and the Fallacy of Reference
John W. Carroll Laws of Nature
Henry S. Richardson Practical Reasoning about Final Ends
Joshua Hoffman and Gary S. Rosencrantz Substance among Other Categories
M. J. Cresswell Language in the World
Noah Lemos Intrinsic Value
Lynne Rudder Baker Explaining Attitudes
Robert A. Wilson Cartesian Psychology and Physical Minds
Michael Devitt Coming to Our Senses
Barry Maund Colours
 The Concept of
Moral Obligation
Michael]. Zimmerman
University of North Carolina, Greensboro

CAMBRIDGE
UNIVERSITY PRESS
 CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, Sao Paulo

Cambridge University Press

The Edinburgh Building, Cambridge CB2 8RU, UK

Published in the United States of America by Cambridge University Press, New York

www.cambridge.org
Information on this title: www.cambridge.org/9780521497060

© Cambridge University Press 1996

This publication is in copyright. Subject to statutory exception

and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.

First published 1996

This digitally printed version 2007

A catalogue record for this publication is available from the British Library

Library of Congress Cataloguing in Publication data

Zimmerman, Michael J., 1951—
The concept of moral obligation / Michael J. Zimmerman,
p. cm. - (Cambridge studies in philosophy)
Includes bibliographical references and indexes.
ISBN0-521-49706-X(hc)
1. Duty. 2. Responsibility. I. Title. II. Series.
BJ1451.Z56 1996 95-19155
170-dc20 CIP

ISBN 978-0-521-49706-0 hardback

ISBN 978-0-521-03874-4 paperback
 Contents

Preface page ix
Acknowledgments xiii

1 Groundwork: Some distinctions 1

1.1 Moral and nonmoral obligation 1
1.2 Binding and nonbinding senses of "ought" 2
1.3 Overall and prima facie obligation 5
1.4 Objective and subjective obligation 10
1.5 Act-evaluation and other types of evaluation 20

2 Moral obligation: An analysis 21

2.1 Value maximization 21
2.1.1 Two problems 21
2.1.2 The analysis 25
2.1.3 Extension of the analysis 31
2.1.4 Perfect and imperfect duties 38
2.2 Underlying metaphysical and conceptual issues 40
2.2.1 "Can" and control 40
2.2.2 "Can" and accessibility 45
2.2.3 "Can" and knowledge 49
2.2.4 "Can" and control again 50
2.2.5 The ontological status of actions 53
2.2.6 Indefiniteness 57
2.3 Implications of the analysis 63
2.3.1 Certain deontic implications 63
2.3.2 Certain deontic paradoxes 72
2.3.3 Analysis and neutrality 75
3 The dynamics of obligation 79
3.1 "Ought" and "can" 79
3.1.1 An internal criticism 79
3.1.2 "Can" and "can avoid" 81
 3.1.3 An alleged redundancy 82
3.1.4 Past and present obligations 82
3.1.5 Alternate possibilities 85
3.1.6 Psychological incapacitation 89
3.1.7 The obligatory and the ideal 90
3.1.8 Excuses 93
3.1.9 Self-imposed impossibility 95
3.2 Immediate and remote obligation 96
3.2.1 Self-imposed impossibility 97
3.2.2 The time at which a wrong is done 100
3.2.3 The time at which an obligation is satisfied 105
3.2.4 The cancellation of obligation 107
3.2.5 The supersession of obligation 111
Conditional obligation 114
4.1 Varieties of conditions on obligation 114
4.2 The analysis 117
4.3 Certain deontic paradoxes 122
4.4 Subsidiary obligation 128
4.4.1 Factual detachment 128
4.4.2 Levels of obligation 131
4.4.3 Overriding 135
4.4.4 A rival account 136
4.4.5 Level of wrongdoing and seriousness of
wrongdoing 138
4.4.6 Level of wrongdoing and level of obligation 139
Prima facie obligation 141
5.1 A tale of two obligations 141
5.2 Defective analyses 143
5.3 The analysis 145
5.3.1 "All else being equal" 145
5.3.2 Three objections 154
5.3.3 Detachment 160
5.3.4 Overriding 162
5.4 Ross on prima facie obligation 163
5.4.1 Conditional and unconditional prima
facie obligation 163
5.4.2 The self-evidence of prima facie obligations 164
5.4.3 Prima facie obligation and overall obligation 169
5.4.4 Residual obligation 174
5.5 Rights 176
5.5.1 The correlativity of rights and obligations 176

VI
 5.5.2 Absolute and prima facie rights 181
5.5.3 Rights and overall obligation 187

6 Actualism and possibilism 189

6.1 The dispute 189
6.2 Objections to actualism 191
6.2.1 Certain theses denied 191
6.2.2 Sanctioning wrongdoing 193
6.2.3 A nondiachronic case 195
6.2.4 A slightly amended version of actualism 195
6.3 Objections to possibilism 196
6.3.1 Arbitrariness 197
6.3.2 Advice 198
6.3.3 Detachment 200
6.3.4 Act-individuation 203
6.4 What it is best to do 203

7 Dilemmas 207
7.1 The nature of moral dilemmas 207
7.2 The case against basic moral dilemmas 211
7.3 The case for basic moral dilemmas 217
7.4 Nonbasic moral dilemmas 225

8 Supererogation 232
8.1 The nature of supererogation 232
8.1.1 Going beyond prima facie obligation 232
8.1.2 Going beyond perfect obligation 233
8.1.3 Going beyond overall obligation 234
8.2 The possibility of supererogation 237
8.2.1 Supererogation and discretion 239
8.2.2 The division of values 244
8.2.3 Continuity 248

9 Cooperation 254
9.1 The problem 254
9.2 Nonsolutions 258
9.2.1 Unrequited cooperation 258
9.2.2 Procedural cooperation 259
9.2.3 Group obligation 260
9.2.4 Counterfactual cooperation 263
9.2.5 Mere openness 265
vii
 9.3 The solution: Unintrusive transigence 268
9.4 Interpersonal moral dilemmas 274

Postscript 277
Appendix: List of propositions 278
List of works cited 282
Index of names 293
Index of subjects 297

vin
 Preface

On October 18,1993, the News and Record of Greensboro, North Carolina,

carried the following item:
Dear Ann Landers: I am 37 years old and have never married. I dated a 35-year-
old divorced man for five years. "Jack" and I were very much in love. I wore
his engagement ring for a year, and we talked of marriage, but we bickered a
lot. I assumed this was just our style of communication.
We decided to get premarital counseling about our never-ending arguments.
One counselor said our values were too different. Another counselor said we
were made for each other. I postponed the wedding because Jack wouldn't
continue with the counseling. Also, he wanted to elope, and I wanted a church
wedding.
Last January, I went over to Jack's house and caught him with his secretary.
It turned out this 21-year-old gorgeous thing had moved in with him. Jack
insists that he doesn't love her, but she cooks and cleans for him, which I never
did.
Ann, we've been seeing each other behind the secretary's back. We agree
that we love each other, but he's afraid he hasn't the will-power or self-disci-
pline to remain true to me. I still want to marry him.
I went to a doctor who told me I was depressed. I've tried dating other men,
but I have no desire for them. What should I do?

Indianapolis Dilemma
I record this for two reasons. First, its comic value. Perhaps, as you were
reading it, a slight smile played over your lips. I hope so; the rest of this book
is dead serious and will afford very little opportunity for amusement.
The second reason is this. A preface should let the reader know what to
expect from a book, and I want you to know that you should not expect an
answer to Indianapolis's question, even if the "should" in her "What should
I do?" is a moral one. Nor should you expect answers to any questions
remotely like hers. Some people seem to think that the task of moral phi-
losophy is simply to provide answers to such questions; if you are among
ix
these, then this book will sorely disappoint you and I advise you to read no
further. Once, many years ago, I wrote a paper on the relation between
intrinsic value and the appropriateness of pleasure, and I submitted it
to a leading philosophy journal. One referee recommended rejection,
saying:
Actual cases of moral questions and issues are hardly mentioned; yet what else
is morality about except Joan wondering whether to leave her husband or have
an abortion, Albert wondering whether to tell his grandmother what her doc-
tor just told him, or whether a new bike would spoil his son? That Albert and
Joan would find this paper totally irrelevant to their dilemmas, even if they were
philosophers and understood it, seems to me to sit very differently than the idea
of the ordinary man who knows his coffee-cup is real not being interested in
The Theory of Knowledge or Science, Perception, and Reality... If I were advising
Joan, I would tell her to go and read The French Lieutenant's Woman, and not
this paper, and I consider that a wholly legitimate criticism of a piece of philo-
sophical ethics.
I hope you are as appalled by this as I still am. (If not, then, again, please
read no further.) I have found that this referee's attitude is by no means an
isolated one, but I find it quite baffling. Joan's situation, whatever it is, may
be a difficult one, and certainly moral philosophers may legitimately try
their hands at resolving it; but it had never been my intention in the paper
even to pretend to be concerned with the resolution of such matters, and
to claim that such resolution is the only legitimate task of moral philosophy
is utterly preposterous. At any rate, this book attempts something quite dif-
ferent and much less ambitious.
A great deal of contemporary moral philosophy is concerned with find-
ing answers to such general substantive questions as what it is that makes
right acts right, what the various virtues and their interrelations are, what
constitutes a proper excuse for vile behavior, and so on. These are all very
important questions, and moral philosophers (some of them, anyway) are
especially well equipped to handle them. But, again, it is not with such
ambitious questions as these that this book has to do.
No, the focus of this inquiry is a concept: the concept of moral obliga-
tion. The main question that is addressed is not what it is that Joan (or
Albert, or Indianapolis) ought to do in particular, nor what it is that makes
an act obligatory in general; it is, rather, what it is for an act to be obligato-
ry. The task that I have set myself is thus one of conceptual cartography, of
mapping the contours of the concept of moral obligation. The goal is sim-
ply that of understanding this concept better. This may be a modest goal,
but I have found its achievement difficult.
 What follows, then, is an unabashed exercise in metaethics. Although
you won't find in it an answer to Indianapolis's predicament, you will find
answers to these questions (and many more besides): What is the relation
between "ought" and "good"? If someone ought to do each of two things,
ought he to do both? If he ought to do both, ought he to do each? If some-
one ought to do something but cannot do it without doing something else,
ought he to do the something else? What are imperfect duties? Does
"ought" apply only to actions, or does it have broader scope? Ought one
to be perfect? Does "ought" imply "can"? If so, can someone divest him-
self of an obligation simply by rendering himself unable to fulfill it? Can
someone succeed in sloughing off an obligation through sheer laziness? Can
someone be obligated to do something he cannot avoid doing? Can some-
one have an excuse for doing something that it was not wrong for him to
do? Can something that it was obligatory to do become wrong? Can some-
thing that it was wrong to do become obligatory? Are there gradations of
obligation and wrongdoing? Can there be obligations without rights? Can
someone be obligated to do, or to cause himself to do, wrong? If one per-
son ought to advise another to do something, ought the latter to do that
thing? Can someone (Indianapolis, say) be in a genuine moral dilemma?
What is the relation between wrongdoing and guilt? Can one go beyond
the call of duty? Can it be right to prevent someone from doing what's
right? Can two wrongs make a right? If questions like these interest you,
please read on.
(Interlude on political correctness. In the last paragraph, I have used
"he" in what is intended as a gender-neutral manner. I am quite aware —
who these days could not be? - that some find this objectionable. In reply,
let me simply quote Judith Thomson, who has recently stated, with cus-
tomary eloquence, exactly what I would want to say:
On the other hand, "she" is no better. Indeed, it is in two ways worse. In the
first place, prose should be transparent, like a pane of glass through which one
sees the thoughts behind it; the use of "she" for these purposes is like a smudge
on the pane - it captures the attention. In the second place, those who now use
"she" in this way are making a moral point in doing so, a moral point I think
entirely right; but it is annoying to have that moral point introduced (with the
back of the hand, as it were) into matters it has no connection with — one feels
nagged. All the same, "he or she" very soon yields impossible clutter.1
I would merely add that "one" also very soon yields impossible clutter,
while "they" simply generates barbarisms.)

1 Thomson (1990), p. 3, n. 1.

xi
 It is customary for prefaces to include an outline of the text to follow,
but frankly I don't think I can say, in brief, much that is useful beyond what
a perusal of the Table of Contents will reveal, and so I will not try to do so.
Instead, let me just deal with one other preliminary matter, namely, the sys-
tem of numbering that is adopted in the text. Numbers will be used to refer
to chapters, sections, subsections, charts, figures, and tables. The practice
that I shall adopt is illustrated in the following examples: Chapter 1, Section
1.1, Subsection 2.1.1, Chart 1.1, Figure 2.1, Table 1.1.1 hope this practice
is self-explanatory. In no case will such numbers be used alone; in every
case it will be explicitly stated whether reference is being made to a chap-
ter, section, chart, and so on. I shall also use numbers to refer to proposi-
tions (and portions thereof). Here the numbers will stand alone, but they
will always appear within parentheses. (Thus, for example, "(1.1)" refers to
the first proposition so treated in Chapter 1.) When, in referring to propo-
sitions, I make use of a prime or primes, this indicates a variation on some
other proposition. (Thus (2.41') is a variation on (2.41).) When I make use
of a lower-case letter, this indicates a variation on part of some other propo-
sition. (Thus (2.41'a) is a variation on part of (2.41').) In addition I have
employed the capital Roman numerals (I) — (XVI) to refer to certain core
analyses of concepts having to do with obligation; (I) — (XII) are introduced
in Chapter 2, (XIII) - (XVI) in Chapter 4. Finally, in Chapter 1 I also use
the capital letters (A) - (E) to refer to certain portions of propositions that
I there call "act-categories." All of this is, I confess, rather cumbersome, but
I have sought in vain for a more wieldy method of reference.

xn
 Acknowledgments

It is my pleasant duty to acknowledge the help that I have received from

others in the writing of this book.
I thank the students in my graduate seminars on obligation given at
Brown University in 1984 and 1987. They helped me deal with the mate-
rial in its early stages of development.
I was awarded a research leave by the University of North Carolina at
Greensboro for the fall semester of 1993. This enabled me to complete a
first draft of the book. I believe I would still be engaged in writing the book
had I not received this leave, and I am very grateful for it.
Several anonymous referees, both for Cambridge University Press and
for certain journals, provided penetrating criticism and useful advice con-
cerning portions of the manuscript. I am very thankful for this help. I am
thankful, too, for permission to reproduce (often with modification) por-
tions of previously published works. These works are listed in the List of
Works Cited as: Zimmerman (1986) - copyright 1986 by D. Reidel
Publishing Company, reprinted by permission of Kluwer Academic
Publishers; (1987) - reproduced by permission of the editor of the American
Philosophical Quarterly; (1988b) - reproduced by permission of the editor
of Philosophical Papers; (1990b) - copyright 1990 Kluwer Academic
Publishers, reprinted by permission of Kluwer Academic Publishers; (1992)
- copyright 1992 Kluwer Academic Publishers, reprinted by permission of
Kluwer Academic Publishers; (1993b) - reproduced by permission of the
editor of the American Philosophical Quarterly; and (1995a) - copyright 1995
Kluwer Academic Publishers, reprinted by permission of Kluwer Academic
Publishers.
I received a great deal of help from a number of other people. It is a dis-
tinct pleasure to acknowledge my indebtedness to them here. They are:
Neera Badhwar, Randolph Clark, James Coley, Earl Conee, David Copp,
Ishtiyaque Haji, David Heyd, Thomas Hill, Jr., Thomas Hill, Sr., Joshua
Hoffman, John King, Jarrett Leplin, Alfred Mele, Gregory Mellema,
John Pollock, Gerald Postema, Philip Quinn, Andrews Reath, Gary
Rosenkrantz, Geoffrey Sayre-McCord, Ted Sider, Ernest Sosa, Lynne
Tirrell, and Gregory Velazco y Trianosky.
Finally, there are four individuals whom I wish to set aside for special
mention, since their contribution to the present work has been so
extensive.
Peter Vallentyne read a version of the entire manuscript and provided
me with numerous detailed comments. He stopped me from making a
number of errors and pointed the way to a number of improvements, for
all of which I am most thankful.
Terrance McConnell read various versions of the entire manuscript,
some portions several times, and always provided helpful and judicious
comments. He has set himself a dangerous precedent. As an expert on grat-
itude, he knows how grateful to him I should be. I think that in this
instance, if not in others, I have met the demands of morality.
Roderick Chisholm has been a source of tremendous inspiration to me
throughout, and indeed prior to, my professional career. He gave me sound
advice and staunch encouragement when I needed them most. His support
has been invaluable, and I extend to him my deepest thanks.
Fred Feldman has shaped this work more than any other person. His
work on obligation is the source of a great many of the ideas expressed in
this book; anyone acquainted with his work will immediately see how
much I have lifted practically wholesale from it. In addition, he is a won-
derful teacher and an extremely incisive critic ("cruel but fair," as Monty
Python might say). Even though he will doubtless find much still to
criticize in this book, it is to him that I dedicate it with great gratitude and
respect.

xiv
 1

Groundwork: Some distinctions

I believe that one ought to do the best one can. You may be inclined to
disagree. If you are, I think it's because you don't yet know what I mean.
Let me try to explain.

1.1 M O R A L AND N O N M O R A L O B L I G A T I O N

I am using the term "ought" to express moral obligation. It is of course

sometimes used to express what is required not by morality but by pru-
dence, or law, or aesthetics, or the rules of chess, or whatever. "You ought
not to have used so much red," for example, is likely to constitute aesthe-
tic rather than moral criticism. "You ought to watch your pennies" is like-
ly to constitute prudential rather than moral advice.
This well-recognized fact gives rise to a profound puzzle that I shall
mention only to put aside. Suppose that you have investigated your options
and reached the disconcerting conclusion that, although morality requires
you to do one thing, prudence requires another, and the law yet a third (and
aesthetics a fourth, and so on). You might find it hard to decide what to do.
You might turn to a friend for advice and say: "This is my situation. What
ought I to do?" But what do you mean? "What ought I to do from the moral
point of view?" Surely not, for you've already settled on an answer to that
question. Similarly, you've already settled on answers to the corresponding
questions having to do with the various nonmoral points of view.
Presumably, then, you mean, "What ought I to do, period?" This is very
puzzling. It appears to presuppose that the various points of view, moral
and nonmoral, are in some manner commensurate with one another, so
that they may be weighed one against the other and your question receive
a principled answer. The answer is given in terms of this "ought, period,"
which, it seems, is supposed in some way to transcend each of the various
points of view, moral and nonmoral.1 For my part, however, I find that I
1 Cf. Feldman (1986), pp. 212-15.

1
have no grasp of any such transcendent "ought"; I do not understand
how it could be that the various points of view, moral and nonmoral, are
commensurate with one another.
But perhaps this is just a deficiency in me. At any rate, as I said, I shall
put the matter aside, for even if it makes sense to say on occasion that one
ought, period, not to do what one ought morally to do, this would of course
not alter the fact that one still ought morally to do what one ought moral-
ly to do. My concern here is exclusively with what one ought morally to do,
and my contention is that, morally, one ought to do the best one can.
In saying this, I don't mean to deny that nonmorally one ought to do
the best one can (in some sense), but only to assert that morally one ought
to do the best one can (in some sense to be elaborated on later). It may well
be that nonmoral obligation is, in many instances, to be analyzed in terms
similar to those in which moral obligation is to be analyzed.2 Indeed, it
would be surprising if this were not so. But whether or not it is so is not my
concern here.

1.2 BINDING AND NONBINDING

SENSES OF "OUGHT"
"Ought" doesn't always express requirement or obligation, even within
morality. Sometimes it expresses an ideal, a desideratum rather than a moral
necessity. Here we might use "should" instead of "ought"; we would
probably not use "must." Examples of this use of "ought" are: "Everyone
ought to be happy," and "Little children ought not to have to suffer."
Sometimes the ideal, nonbinding sense of "ought" is called the "ought-
to-be," while the binding sense of "ought" is called the "ought-to-do."3
This is suggestive but a bit misleading. The idea is that, when discussing
what is ideal or desirable, we can express our thoughts in the following way
(to use the illustrations just presented): "It ought to be that everyone be
happy," and "It ought to be that little children not have to surfer"; where-
as, when discussing what is obligatory, we can express our thoughts in the
following way (to employ fresh illustrations): "Tim ought to do what he's
been told," and "Pru ought to do what she promised." The trouble is that
the "It ought to be that" locution can be applied to what is done, too (as
in: "It ought to be that Helen do what she can to help out"), and anyway
it is surely arguable that certain actions are sometimes morally desirable but
2 Cf. Feldman (1986), Chs. 5 and 7.
3 See, e.g., Broad (1985), pp. 225-35; Chisholm (1974), pp. 11-13; Castaneda (1975),
p. 207. Cf. Marcus (1966) on "evaluative" and "prescriptive" senses of "ought."
not obligatory.4 In addition, it seems that what is obligatory need not be
something that is done (as evidenced in: "Mary ought to be with her ailing
mother").5 It is better, therefore, to talk of what is binding or obligatory
and to contrast it with what is nonbinding or ideal.
Many philosophers have contended that there is no nonbinding sense of
"ought (morally)," or that, at best, such a nonbinding sense is derivative
from the binding sense (as in, perhaps: "God ought to prevent the suffer-
ing of little children").61 am inclined to think that this is mistaken, to think,
that is, that there is a perfectly respectable nonbinding sense of "ought
(morally)" that is not to be analyzed in terms of some binding sense of
"ought (morally)."7 Consider, for example, what Roderick Chisholm has
to say about supererogation: "The status of supererogation might be sum-
marized in this way. If I seek advice, concerning an act which would be
supererogatory, and ask 'Shall I do this?', I may well be told, 'You ought to,
but you don't have to' — it is advisable, but not obligatory." Of course, there
are other readings to be given to "You ought to, but you don't have to."
Someone might mean by this, "You are nonmorally (e.g., prudentially)
required to do so, but you are not morally required to do so," or "You are
morally required to do so, but no one has any moral claim on you that you
do so." Fair enough. (I shall have more to say about the latter possibility in
a moment.) What I don't see is that someone cannot also quite properly
mean, "It would be morally desirable if you were to do so, but you are not
morally required to do so (or even to try to do so, or to do something
related to doing so, or...)."
Here, though, someone might object: "What can it mean to say that I
ought morally to do something if I am not morally obligated to do it?" The
idea underlying this objection is not that it is improper to use "ought" in
an ideal, nonbinding sense, but that such an "ought" cannot be said to be
a moral "ought." If "Little children ought not to have to suffer" implies no
moral obligation, how is its "ought" a moral one? The objection concerns
the proper extension of the term "moral." This is a huge issue, which, once
again, I mention only to put aside. I have no ax to grind here and, indeed,
have some sympathy with the objection, although on balance I am inclined

4 Cf. Mellema (1993).

5 Broad himself notes these facts in Broad (1985), pp. 233-4. Cf. also Humberstone (1991),
p. 146.
6 See Moore (1922), p. 319; Prichard (1949), p. 4; Ross (1939), pp. 45-6; Broad (1985),
p. 235.
7 Cf. Castafieda (1975), p. 207.
8 Chisholm (1968), p. 417.
to think it perfectly legitimate (whether or not it is true) to say that there
are certain goals which, from a moral point of view, are ideal or desirable
but which do not imply any obligation on anyone's part. Whether or not
this is legitimate, however, is not germane to the present inquiry. However
the ideal "ought" is to be analyzed9 and whether or not it is properly said
to be a moral "ought," I shall confine my attention to the "ought" of moral
obligation.
But ought the ideal "ought" to be so casually set aside? Might it not be
that, while it is not to be analyzed in terms of the "ought" of obligation,
the reverse is true? Of course, I concede that if, as some contend,1 moral
obligation were to be so analyzed, the notion of an ideal "ought" could not
be so easily dismissed in such an inquiry as this. But I deny the contention.
The account of moral obligation that I shall give will not make any
reference to the ideal "ought."
Might there nonetheless not be some interesting connection between
these two "oughts," so that a full account of the "ought" of obligation
would still require that greater attention be given to the ideal "ought" than
I propose to give? Consider this example provided by Bernard Williams.
Suppose that

(1.1) someone ought to help that old lady,

but also that
(1.2) Jones is the only person who can help her;
it follows (Williams claims) that
(1.3) Jones ought to help her.
Williams contends that (1.1) does not express an obligation but that (1.3)
does, and that (1.3)'s derivation from (1.1) and (1.2) indicates that there is
some very close connection between the binding and nonbinding senses of
"ought."11 If he is right, then my failure to pay attention to the ideal
"ought" in my attempt to account for the "ought" of obligation might well
seem rash. But I don't think that he is right. An indication that something
has gone awry in the derivation of (1.3) from (1.1) is that, as Williams
intends for them to be understood, (1.1) (which involves the nonbinding
9 Forasamplingofanalyses,see:Chisholm(1974),p. 11; Jackson(1985);Feldman(1986),
Section 8.2.
10 E.g., Chisholm (1974), p. 13; Jackson (1985), p. 193.
II Williams (1981), p. 116.
sense of "ought") is what might be called "passive-transformable," whereas
(1.3) (which involves the binding sense) is not. (Williams mentions this fact
but appears not to regard it as a problem for the derivation.) That is, (1.1)
may be transformed, without loss of meaning, into its passive equivalent
as follows:
(1.1') that old lady ought to be helped by someone;
but (1.3) cannot be similarly transformed without loss of meaning as
follows:
(1.3') that old lady ought to be helped by Jones,
for it is Jones who has the obligation, not the old lady. Of course, (1.1) and
(1.3) could be read differently, either as neither being passive-transformable
or as both being (more on the former possibility in Chapter 9), and the
argument's validity might thereby be salvaged, but this will not provide
Williams with the link that he wants between the binding and nonbinding
senses of "ought."

1.3 OVERALL AND PRIMA FACIE OBLIGATION

Ever since the publication in 1930 of The Right and the Good by W. D. Ross,
it has been common practice to distinguish between two species of moral
obligation. Ross himself used a variety of terms to refer to these species. On
the one hand there is what he called "absolute obligation," "actual obliga-
tion," "duty proper," or "duty sans phrase" \ on the other hand there is what
he called "prima facie obligation," "prima facie duty," "conditional duty,"
"duty ceteris paribus," or "what tends to be duty."12 None of these terms is
particularly felicitous, as Ross himself acknowledged. More appropriate,
perhaps, is to talk of "overall obligation" or "obligation all things consid-
ered" for what Ross meant by "absolute obligation," and this is what I shall
do. It would also be more appropriate to talk of "obligation some things
considered" (or "in some respects") or of "pending" or "pro tanto"13 obli-
gation for what Ross meant by "prima facie obligation." Nevertheless,
since the term "prima facie obligation" is by now so well entrenched, it will
be easiest to use it.
The distinction between the two species of obligation is very familiar,
even if not very well understood. One can find oneself in a situation where
a variety of moral considerations pertain. What one ought to do insofar as
12 Ross (1930), pp. 18-20, 28, 30.
13 This is the suggestion in Kagan (1989), p. 17.
some of these are concerned may not be what one ought to do insofar as
others of them are concerned. What one ought on balance to do will, pre-
sumably, be some function of these various "insofar as" "oughts"; to put it
differently, what one's overall obligation is will be some function of what
one's various prima facie obligations are (if any).
It is worth repeating that what is presently at issue is two species of moral
obligation. To say that overall obligation has to do with what one ought
"on balance" or "all things considered" to do is, in this context, to say that
it has to do with what one ought to do all morally relevant things consid-
ered.14 The question that was briefly raised above concerning the com-
mensurability of various points of view, both moral and nonmoral, is not
at issue here. We are now working entirely within the moral point of view.
(This is not to say that there is no question about the commensurability of
different moral considerations; there is. It is one that I shall address in the
next chapter.)
My primary concern in this work is to provide an account of overall
moral obligation, although I shall also provide an account of prima facie
obligation in Chapter 5, an account according to which it is closer to the
truth to say that prima facie obligation may be understood in terms of over-
all obligation rather than the reverse.15
Sometimes moral obligations or duties are said to correspond to moral
rights or claims. Ross distinguishes four different theses having to do with
such alleged correlativity. They are:
(1.4) a right of A against B implies a duty of B to A;
(1.5) a duty of B to A implies a right of A against B;
(1.6) a right of A against B implies a duty of A to B;
and
(1.7) a duty of A to B implies a right of A against B.16
Each thesis has two main readings, one where "duty" expresses a prima
facie obligation (and hence, presumably, "right" expresses a prima facie
right), the other where "duty" expresses an overall obligation. Ross him-

14 Indeed, it is to say that it has to do with what one ought to do all moral-obligation-
relevant things considered, inasmuch as some moral considerations, as noted in the
last section, are relevant not to what is obligatory but to what is ideal.
15 Two possible attempts to provide such a "reverse" account are to be found in Ross
(1930), pp. 41 and 46-7, and Broad (1985), pp. 171-2.
16 Ross (1930), p. 48. Cf. Feinberg (1973), p. 61.
self (who employs the "prima facie" reading) says that (1.4) is clearly true
but that, since we can have duties to (nonhuman) animals but no rights
against them, (1.7) is false.17 Our having duties to animals but their having
no duties to us also leads Ross to note that not both (1.5) and (1.6) can be
true. He is uncertain which to reject but is inclined to say that animals, not
being moral agents, do not have rights, and therefore that (1.5) is false. (1.6),
however, he accepts as true.18
This rejection of (1.5) and (1.7) may be questioned. We may agree that
we have duties towards or concerning animals, but do we have duties to them
in the requisite sense? Ross thinks so, saying: "I suppose that to say we have
a duty to so-and-so is the same thing as to say that we have a duty, ground-
ed on facts relating to them, to behave in a certain way towards them."19
But it may be that this account of what it is to have a duty to someone or
something is too broad. It would appear to imply that certain potential ben-
eficiaries (such as the homeless) have a right not just to nonmaleficence but
to beneficence, if it is agreed that we have a duty to act not just nonmalef-
icently but also beneficently towards them. Although some may not balk
at this (either because they reject the view that we have a duty to act bene-
ficently towards such potential beneficiaries, or because, like Ross, they
accept the view that such persons have a right to our beneficence), others
will balk at it and claim that it is possible to have a duty towards or con-
cerning somebody or something without owing that duty to that person or
thing.21 In this way one might try to support not just (1.4) and (1.6) but also
to defend (1.5) and (1.7) against Ross's argument. For one could argue that
all and only moral agents are such that duties are owed to them (in the req-
uisite sense), and all and only moral agents are such that they owe duties to
others.
My own, tentative view is neither that of Ross nor that just mentioned.
Contrary to Ross, and in keeping with the view just mentioned, I believe
that there is indeed a difference between having a duty towards or con-
cerning someone or something and having a duty to someone or some-
thing. The latter, I believe, is correlative to a right held by the someone or
something in question, whereas the former is not. Hence I accept both (1.4)

17 Ross (1930), pp. 48-9.

18 Ross (1930), pp. 49-50, 53-4.
19 Ross (1930), p. 49.
20 Ross (1930), pp. 52-3.
21 Cf. Lyons (1979c), pp. 60-1; Feinberg (1979), p. 79. Also cf. Hart (1979), pp. 17-19,
Feinberg (1980), p. 139, and Hill (1979), pp. 122-3, on the ambiguity of "to" in such
contexts.
and (1.5). But I reject both (1.6) and (1.7), because, contrary to Ross and
to the view just mentioned, I'm inclined to think that, although one must
be a moral agent to have duties to others, one need not be a moral agent to
have rights against others. In particular, certain animals appear to me to have
rights against us but duties to no one.
At this point it might be objected that, while it is possible that one ought
(morally) to do something without one's having a (moral) duty or obliga-
tion to someone (or something) to do that thing, it is not possible to be obli-
gated to do something without one's owing that obligation to someone to
do that thing. Consider this example, provided by Joel Feinberg:
[CJonsider how the word "ought" differs from the word "duty." Suppose a
stranger approaches me on a street corner and politely asks me for a match.
Ought I to give him one? I think most people would agree with me that I
should...
Now suppose that Jones is on the street corner and another stranger polite-
ly requests a light from him... He brusquely refuses to give the stranger a match.
I think we can agree that Jones's behavior ... does not constitute an ideal for
human conduct under such circumstances; ... that it was not what Jones ought
to have done.
If we reproach Jones, however, ... he may present us with a vigorous self-
defense. "Perhaps I was not civil," he might admit. "But surely I was under no
obligation to give a match to that man. Who is he to me? He had no claim on
me...; I don't owe him anything. It may be nice to do favors for people; but a
favor, by definition, is nothing that we are legally or morally required to do."22
In a comment on his own example, Feinberg says:
Jones's defense makes me think no better of him. Still, ... [e] verything Jones
said ... was true. The moral I draw from this tale is that there are some actions
which it would be desirable for a person to do, and which, indeed, he ought to
do, even though they are actions he is under no obligation and has no duty to do.
It follows logically that to say that someone has a duty or an obligation to do X
is not simply another way of saying that he ought to do X.23
But this is all very complicated. There are a number of possible positions
to adopt concerning Jones's behavior (let alone Jones himself) in this story.
Even if we disregard the distinction between prima facie and overall oblig-
ation, still eight different positions suggest themselves. Using the terminol-
ogy that I have introduced, one may put these positions as in Table 1.1
(where A is the action of giving the stranger a match). Positions 3 and 7

22 Feinberg (1968), pp. 392-3.

23 Feinberg (1968), p. 393.
 Table 1.1

Jones ought, in Jones ought, in Jones has an

the nonbinding the binding obligation to the
Position sense, to do A sense, to do A stranger to do A

1 yes yes yes

2 yes yes no
3 yes no yes
4 yes no no
5 no yes yes
6 no yes no
7 no no yes
8 no no no

may be immediately dismissed, since one cannot have an obligation-to to

do something that is not something that one ought, in the binding sense,
to do. But which position is it that Feinberg accepts? Clearly he does not
adopt any position according to which Jones has an obligation to the
stranger to give him a match (where this is understood to mean that Jones
owes it to the stranger to do so), and so he does not accept any of positions
1, 3, 5, or 7. It seems clear, too, that he does not accept any of positions
5-8, according to which it would not be ideal (under the circumstances)
for Jones to give the stranger a match.24 This leaves positions 2 and 4, and
I don't know which of these it is that Feinberg accepts. His remarks seem
to me ambiguous between the two of them, and perhaps for good reason;
for even if one agrees that the stranger has no claim on Jones that Jones give
him a match, but that something has gone awry, morally, when Jones fails
to give him a match, still it is hard to know which of positions 2 and 4 to
accept. (Position 2, with its binding "ought," implies that Jones did wrong in
not giving the stranger a match, whereas position 4 does not.) Of course, if
one rejects the possibility that there be a binding "ought" without there
being an obligation-to, then one will reject position 2.
As before, however, I raise all this only to put it aside. Feinberg wishes
to reserve the term "obligation" for what I have called "obligation-to"; if
he accepts that position 2 is possible, he would still not express it by saying

24 Some people might now believe that common courtesy would require not giving the
stranger a match, due to considerations of health. It is worth noting that Feinberg first
presented his example in 1960, long before health-related condemnation of smoking
became common.
that Jones has an obligation, but not an obligation-to, to give the stranger
a match. That is fine by me. I choose, however, to use the term "obliga-
tion" simply as a nominalization of the binding "ought" and to say that, even
on position 2, Jones has an obligation to give the stranger a match. One
must adopt a certain terminology, and this is the terminology that I, like
many others, adopt. What matters in the end is not what terminology one
adopts but that one get clear as to what the concepts at issue are. That is
what I am trying to do. My purpose in this work is to clarify the concept
expressed by the binding "ought." It is important to recognize that, for pre-
sent purposes, I am neither presupposing nor denying the view that posi-
tion 2 (or position 6) is possible, that is, that it is possible that there be an
obligation (whether prima facie or overall) that is not an obligation-to
(although, as noted earlier, I am in fact inclined to believe that this is pos-
sible). My concern here is to provide an account of obligation without ven-
turing into this territory of "to," that is, to provide an account of what may
be called "mere obligation." If it is possible to have an obligation that is not
an obligation-to, then my account is designed to say what this amounts to.
I shall write henceforth under the assumption that this is possible. But if it
is not, still every obligation-to is of course an obligation, and my account
can then be construed as being designed to say at least to that extent what
obligation-to amounts to.

1.4 OBJECTIVE AND SUBJECTIVE OBLIGATION

So far I have, in effect, been noting the ambiguity of "ought" and trying to
narrow down that sense of it that is my main concern here. This process of
specification may be represented as in Chart 1.1.25 We have made our way

moral nonmoral
binding nonbinding
overall prima facie
mere obligation obligation-to
Chart 1.1
25 This diagram is incomplete, in that the various subspecies that have been identified on
the left-hand side have analogues that are not shown on the right-hand side. Thus, non-
moral obligation may be subdivided into binding and nonbinding obligation, nonbind-
ing obligation may be subdivided into overall and prima facie obligation, and prima facie
obligation may be subdivided into mere obligation and obligation-to.

10
to mere obligation. Does this suffice? According to some, it does not; there
is, they allege, ambiguity even here.
T o assess this claim of ambiguity, consider a distinction that Bertrand
Russell draws between different categories of acts. W e may distinguish, he
says, between

(1.8) that act which will have the best consequences

and

(1.9) that act which will probably have the best consequences.
He calls the first act the one that is, under the circumstances, the "most for-
tunate," while the second is the "wisest." 26 Ross and H. A. Prichard draw
a similar distinction between

(1.10) that act which best suits the objective features of the agent's situa-
tion (that is, which best suits the agent's actual situation)

and
(1.11) that act which best suits the subjective features of the agent's situa-
tion (that is, which best suits the agent's situation as the agent con-
ceives his situation to be). 27

If we ignore the question as to whether what makes an act "best" has to do

with its consequences or with its suitability to the agent's situation (or,
indeed, with something else), we get this threefold distinction:

(A) that act which is best,

(B) that act which is probably best,

and

(C) that act which the agent believes to be best.

(1.8) and (1.9) correspond to (A) and (B), respectively, whereas (1.10) and
(1.11) correspond to (A) and (C), respectively. (This is all decidedly rough,
but further refinement is not pertinent to our present purpose.)
It is on the basis of such distinctions that some philosophers have claimed
that "ought" is ambiguous, even when it has been narrowed down to mere
obligation. Russell, for example, distinguishes between what he calls
"objective obligation" and "subjective obligation." Perhaps surprisingly,

26 Russell (1970), pp. 12-13.

27 Ross (1939), pp. 146-7; Prichard (1949), p. 18.

11
what Russell says corresponds to objective obligation is not (A) but (B), and
what he says corresponds to subjective obligation is not (C) but this:

(D) that act which the agent believes to be objectively obligatory.

Others, too, have adopted this terminology, but sometimes it is applied dif-
ferently. For example, Fred Feldman says that it is (A) and not (B) that cor-
responds to objective obligation and that what corresponds to subjective
obligation is neither (C) nor (D) but this:
(E) that act which it is most reasonable for the agent to believe is objec-
tively obligatory.29
Others have used somewhat different terminology. C. D. Broad, for
instance, distinguishes between what he calls "material obligation," "for-
mal obligation," and "subjective obligation," saying that (A) corresponds
to what is materially obligatory, whereas (C), read in one way, corresponds
to what is formally obligatory and, read in another way, corresponds to
what is subjectively obligatory.30 And it might seem that Ross and Prichard
would say that (A) corresponds to what is objectively obligatory and (C)
corresponds to what is subjectively obligatory.
Interestingly, though, Prichard does not say this, and Ross appears to
deny it too. At one point Ross says the following:
It is clear that when we call an act right we sometimes mean that it suits the
objective features of the situation, and sometimes that it suits the subjective fea-
tures. And when people express different opinions about the rightness or
wrongness of an act, the difference is often due to the fact that one of them is
thinking of objective and the other of subjective rightness. The recognition of
the difference between the two is therefore in itself important as tending to rec-
oncile what might otherwise seem irreconcilable differences of opinion. But
the question remains, which of the characteristics — objective or subjective
rightness — is ethically the more important, which of the two acts is that which
we ought to do.31

Here Ross seems to grant the distinction between objective and subjective
obligation (or, as he puts it in the quoted passage, "rightness"; he draws no
distinction between rightness and obligatoriness) only to snatch it away
again by asking what the agent (really) ought to do. Prichard is more force-

28 Russell (1970), pp. 13-14.

29 Feldman (1986), p. 46.
30 Broad (1985), pp. 125-7, 131-2, 144, 172, and Broad (1946), p. 557.
31 Ross (1939), pp. 146-7.

12
ful still. He distinguishes between what he calls the "objective" and
"subjective" views of obligation and insists that the former view is mis-
taken and the latter correct.32 He thus denies that (A) captures a (proper)
sense of "obligation," insisting that only (C) does. More recently, Judith
Thomson has also insisted on unambiguity here, although her view is that
there is no subjective sense of "ought" (which in her view would, ifit exist-
ed, correspond to a variation on (D)).33 But others have insisted just as
strongly on ambiguity.34 What are we to think?
I'm inclined to think that we should side with Prichard and Thomson
(and perhaps Ross) here and deny that there are several senses of "ought"
at issue. Certainly there is no need to think that each of (A)-(E) (and per-
haps other act-categories) corresponds to some distinct sense of "ought."
Accepting that (A)-(E) are distinct (as one ought to do) does not require
accepting that each expresses some sense of "ought." Still, I grant that more
than one sense of "ought" may be at issue here. Consider the claim, which
is not uncommon, that one ought to do what one thinks one ought to do.
Sometimes this is offered as advice to someone seeking to discover what he
ought to do. As such, it seems to me likely to be singularly unhelpful,
although it may be that it is on occasion appropriate; for it may be that the
advisor takes the advisee, despite the latter's uncertainty, to be a fine judge
as to what he ought to do and is simply telling him that, in this case as in
others, his (the advisee's) judgment can be trusted. At other times, how-
ever, the claim seems to be offered as an account of what it is to be such that
one ought (in some sense) to do something. Here the claim seems to be
this: one ought (in some sense) to do what one thinks one ought (in some
other sense) to do. If (D) expresses a sense of "ought" - a "subjective" sense,
let us say - then it constitutes one version of this claim. But why think that
there is any sense of "ought" such that one ought (in that sense) to do
what one thinks one ought (in some other sense) to do? Is it appropriate
ever to speak in this way? I'm inclined to think not. It seems to me that
anyone who puts matters this way would do far better to put them as
follows: one is to blame if one fails to do what one thinks one ought (in the
only respectable sense) to do. Rather than distinguish two senses of "ought,"
as (D) does, and define one (the "subjective" sense) in terms of the other
(the "objective" sense), it seems to me preferable to declare that there is
only one sense of "ought" (at this level of inquiry — the level of "mere
obligation") and to say that blameworthiness (and other concepts of
32 Prichard (1949), Ch. 2.
33 Thomson (1986), p. 179.
34 See Broad (1985), pp. 126-7.

13
responsibility) are to be understood in terms of it.35 Still, if it is insisted that
there are two senses of "ought" here, we need only say in reply that we are
presently concerned just with one of them, namely, that in terms of
which the other is allegedly to be understood. If we were using the terms
"subjective obligation" and "objective obligation" to draw this distinction,
then the point could be put this way: our present concern is to give an
account only of objective and not also of subjective obligation.36
But I don't like this way of putting it, and not just because I'm inclined
to deny the ambiguity of "ought" at this level. I don't like it because it pre-
supposes that "ought" (in the sense that we are concerned with) is in some
important way "objective" (whatever that may mean, exactly), and,
although I do believe this, it is not something that I wish to presuppose for
present purposes. For Prichard's question is still an open question: even if
"ought" (in the sense that we are concerned with) is not to be understood
in terms of (D) (or (E), or any other account that itself makes reference to
"ought"), how is it to be understood? For all that I've said, it may be that
Prichard is right and that it is to be understood in terms of (C); and it would
be highly misleading- as Prichard would surely agree - to say that "ought,"
so understood, is "objective."
Here, though, it may seem to you that I've made a mistake. I began by
saying that one ought to do the best one can. Surely, you may think, this
corresponds to (A) and not (C) (and not (B) either), and so the "ought" that
I'm after is properly called objective. But if you think this, you probably
still don't know what I mean. Let me try to set matters straight.
I am using the term "best" very liberally. In saying that one ought to do
the best one can, I am contending merely that there is some way in which
what one ought to do is superior to any other of one's alternatives. I shall
call such superiority deontic. My claim, then, is that, to have the deontic sta-
tus of being obligatory, an act's deontic value must be greater than that of any
other alternative action. (Note that I am here restricting the use of "deon-
tic" to the moral realm, whereas others sometimes use the term more wide-
ly. An act's deontic value, then, although not necessarily itself a species of
moral value, is that which, in conjunction with the deontic values of its

35 Zimmerman (1988a) contains a detailed elaboration of this view.

36 It should be acknowledged at this point that some philosophers accept that "ought" is to
be defined in terms of a belief about "ought" and yet deny that there are two senses of
"ought" at issue. Strawson (1986), for example, claims that, as a matter of definition, one
cannot do wrong unless one believes that one is doing wrong — where the term "wrong"
is used univocally as the contrary of "ought" (p. 220). This claim strikes me as incoher-
ent, but there is no space to pursue the matter here.

14
alternatives, determines the act's deontic status — as morally obligatory,
right, or wrong — which is a moral feature of the act.) For reasons that will
emerge shortly, I shall steadfastly decline to specify what it is that deontic
value consists in; I wish merely to commit myself to the formal view that
one ought to perform an action if and only if that action is, in some way,
ranked first relative to all one's other options. This very abstract (and still
very rough) account of "ought" is compatible, not just with the view that
"ought" corresponds to (A), but also with the views that it corresponds to
(B) and (C), each of which involves some more substantive, and so less lib-
eral, construal of "best." To see this, suppose that we adopt, merely for the
purpose of illustration, the position that "best" in (A), (B), and (C) refers to
the act's instrumental value. Then, according to the view that "ought" cor-
responds to (A), what is deontically superior is what is instrumentally best,
while according to the view that "ought" corresponds to (B) what is deon-
tically superior is what is probably instrumentally best, and according to the
view that "ought" corresponds to (C) what is deontically superior is what
is believed instrumentally best by the agent. On this understanding of
"best," then, the view involving (A) incorporates the substantive moral
view that deontic value just is actual instrumental value, while the view
involving (B) incorporates the substantive moral view that deontic value
just is probable instrumental value, and so on. Insofar as what is actually
instrumentally best need not be what is probably instrumentally best, which
itself need not be what is believed instrumentally best by the agent, these
three views will on occasion issue in three different deontic rankings of the
agent's options. Nonetheless, each of these substantive views is a species of
the purely formal view that an act is obligatory if and only if its deontic value
is greater than that of (that is, the act is deontically superior to) any of its
alternatives.
Note how modest this formal view is. It merely states that what is oblig-
atory is unique relative to its alternatives and that this uniqueness consti-
tutes some sort of morally pertinent superiority; there is no commitment to
any substantive view as to what this superiority consists in.37 In this book I
shall deal in detail only with this formal claim; I shall not inquire whether,
on a more substantive level, it is to (A) or (B) or (C) (or, indeed, something
else) that "ought" corresponds.

37 Cf. Moore (1903), p. 147: "It is plain that when we assert that a certain action is our
absolute duty, we are asserting that the performance of that action at that time is unique
in respect of value." Of course, Moore is here talking about the instrumental value of the
act in particular, whereas, as I have just tried to explain, I am not committing myself to
the view that deontic superiority consists in instrumental superiority.

15
 Still, perhaps I should say in passing that I find the view that "ought"
corresponds to (C) wholly implausible. Prichard argued for this view,38 as
I have noted, and Ross, persuaded by Prichard, abandoned his earlier view
(which was that (A) captured what "ought" consists in)39 in favor of
Prichard's view.40 The reasons given by both Prichard and Ross seem quite
inadequate, but there is no need to go into this here, since the present
account does not presuppose that "ought" does not correspond to (C).4 I
suspect that, frequently at least, someone who is attracted to (C) in this
regard confuses it with (D) and also fails to distinguish between blame wor-
thiness (and other concepts having to do with moral responsibility) on the
one hand and wrongdoing (and other concepts having to do with moral
obligation) on the other.
I find it much harder to decide whether it is more plausible to say that
"ought" corresponds to (A) or that it corresponds to (B). What is it that a
conscientious person, who is trying to determine what he ought to do,
seeks to discover: that act which is best (in some substantive sense) or that
act which is probably best? There is reason to think the former. Suppose that
he believes that he has a duty to help those in distress and that he comes
upon the victim of a car accident. He must choose whether to leave the
victim where he is or to pull him free from the wreck. Suppose that he pulls
the victim free, believing that this is probably best. Does he believe that he
is doing what he ought to do, or does he believe only that he is probably
doing what he ought to do? Quite possibly the latter. After all, he may later
discover that it would have been better to leave the victim where he was
and thus come to believe, regretfully, that (however innocently, and how-
ever much the evidence pointed the other way) he in fact did the wrong
thing.
Yet there is also reason to think that it is more plausible to say that
"ought" corresponds to (B) rather than (A). Consider this case recently pro-
posed by Frank Jackson: Jill, a physician, can prescribe one of three drugs,
X, y, and Z, for a certain illness and knows that X will probably relieve but
not cure the ailment, while Y will either cure it or kill the patient and Z
will either cure it or kill the patient; but Jill doesn't know which of Y and
Z will cure and which kill.42 If "ought" corresponds to (A), then presum-

38 Prichard (1949), Ch. 2.

39 Ross (1930), p. 31: "There is therefore much truth in the description of the right act as
a fortunate act."
40 Ross (1939), Ch. 7.
41 For a lucid exposition of what goes wrong with Ross's case for (C), see McConnell
(1988b).
42 Jackson (1991), pp. 462-3.

16
ably Jill ought not to prescribe Xbut either ought to prescribe Yor ought
to prescribe Z. Jackson claims that this proposal is "intuitively wrong."43
I'm not sure that he is right, but neither am I sure that he is wrong.44
It is tempting to try to resolve such a puzzle by diagnosing ambiguity.
But, again, I resist this here. Surely Jill, as a conscientious physician, is ask-
ing just one question when she says: "What is it that I ought to do in this
case?" But if there is ambiguity, then I shall simply say this: the "ought" that
I'm after is the one that conscientious people like Jill are concerned with;
also, it is the one which is such that, if someone fails to do what he believes
he ought to do, then he is to blame.45
Here, though, I don't have to decide between these views about (A) and
(B). As I've noted, the account of "ought" that I shall give is compatible
with each of them. In fact, it is compatible with a wide variety of other
views not yet mentioned. I asked whether it is more plausible to say that
"ought" corresponds to (A) or to say that it corresponds to (B), but perhaps
some other view is more plausible still. Certainly other views have been
proposed. According to some philosophers, what one ought to do is a func-
tion not simply of what one can do but of what one knows that one can do
(so that, for example, one does not do wrong in failing to rescue someone
of whose plight one is nonnegligently ignorant).46 According to other
philosophers, what one ought to do is a function of what one intends, and
this may or may not have anything to do with what is, or is probably, or is
believed to be, best (in some less-than-fully-liberal sense). Kant is, of
course, the obvious figure to cite here,47 but one needn't be a Kantian to
agree that intended harm constitutes a more serious wrong than unintend-
ed harm - a matter of adding insult to injury48 - and it is difficult to sub-
sume this satisfactorily under either (A) or (B) or (C), at least as these would
ordinarily be understood.49 Again, many philosophers have contended that
what one ought to do is a function of respecting persons' rights;5 others
have contended that what one ought to do is a function of certain so-called

43 Jackson (1991), p. 466.

44 Cf. Prichard (1949), p. 29; Ross (1939), p. 152; Thomson (1986), Ch. 11.
45 See Zimmerman (1988a), Ch. 3.
46 Cf. Prichard (1949), pp. 22-5; Lemos (1980), p. 301.
47 Kant (1964), pp. 67-8.
48 Cf. Hume (1888), p. 349; Rousseau (1992), p. 144.
49 Intention also plays a central role in the oft-invoked principle of double effect, a princi-
ple which, as commonly interpreted, concerns the (im)permissibility of acting in certain
ways.
50 Almost any deontologist could be cited here, such as Kant (1964), p. 96, Donagan (1977),
p. 65, or Fried (1978), pp. 28-9 and 81-6.

17
agent-relative values;51 and so on; and again it seems strained, at best, to try
to subsume these under (A) or (B) or (C), at least as these would ordinari-
ly be understood. But my account, I claim, in abstracting from all that is
substantive in any ordinary sense of "best," is compatible with (almost)52
any such view; for (almost) any such view is a view according to which what
is obligatory is superior in some morally relevant way to its alternatives. For
present purposes, then, the principle by which alternatives are to be ranked
can be left wholly unspecified, as long as what is morally obligatory is
acknowledged as morally unique (in some positive way) relative to its
alternatives.
It is important to note that I am not seeking here to effect some sort of
reconciliation between consequentialist and deontological views, as might
be attempted, for example, by a consequentialist who ascribes intrinsic
value to justice, or promise keeping, or the respect of rights.53 That would
be a substantive moral enterprise. Rather, I am simply claiming that, on a
more abstract level, there is something common to (almost) all substantive
theories of obligation, whether consequentialist or deontological (or some-
thing else, if "deontological" is narrower than "nonconsequentialist"), and
that is that they share this view about the superiority of what is obligatory
to what is nonobligatory. That which is superior maximizes some value and,
I contend, (almost) all substantive theories of obligation can be appropri-
ately cast in a maximizing mold. Thus, besides consequentialism ("It is
obligatory to do that which maximizes intrinsic value"), each of rule utili-
tarianism ("It is obligatory to do that which maximizes rule utility"), Ross's
early theory ("It is obligatory to do that which maximizes prima facie strin-
gency"), Kant's theory ("It is obligatory to do that which maximizes com-
pliance with the Categorical Imperative"), the divine command theory ("It
is obligatory to do that which maximizes compliance with God's com-
mands"), and rights theory ("It is obligatory to do that which maximizes
nonviolation of persons' rights"), and so on, can be put in terms of maxi-
mization - in terms, that is, of doing what is "best" in some sense.54
This is not a new observation.55 Still, it might strike some as peculiarly
unhelpful to represent (almost) all substantive theories of obligation in
terms of maximization. Why do it? Because, as I hope to show in ensuing
chapters, it is extraordinarily fruitful; we can come to a finer understand-

51 E.g., Nagel (1986), Ch. 9.

52 The reason for the qualifier "almost" will be given in Chapter 2.
53 Cf. Feldman (1986), Section 3.3; contrast Slote (1989), pp. 91-2.
54 Cf. Lewis (1973), p. 76.
55 See Nozick (1974), p. 29n. Cf. Sen (1983), p. 128ff., and Dreier (1993), pp. 24-5.

18
ing of the conceptual structure of obligation if we conceive of it in maxi-
mizing terms. Michael Stocker claims that, if putting a theory of obligation
in terms of maximization is to be informative, then the value that it is said
should be maximized must be specifiable independently of the determina-
tion of what one ought to do. This condition appears to be satisfied by cer-
tain theories (such as consequentialism, where in principle what is
intrinsically valuable is determined first, and then it is said that what ought
to be done is that which maximizes intrinsic value) but not by others (such
as Kant's theory, which presumably accords no value to a certain maxim's
being universalizable independently of its being such that one ought to act
on it; or such as the divine command theory, which presumably accords no
value to certain actions independently of their being commanded by God).
And Stocker says:
It might seem that the most that could be charged against maximizers making
this move [of viewing all substantive theories of obligation in maximizing
terms] is that they have trivialized maximization by making it conceptually true.
Many maximizers would ... take this as an encomium, not a charge.
But I think they would be mistaken in seeing matters this way. For maxi-
mizers are now open to the charge of shirking the needed honest toil... After
all, one important task in ethics.. .is finding which objects of care are such that
more of them is better.56

But while I certainly grant that "making maximization conceptually true"

does indeed leave much work in moral theory left to be done, I deny that
it amounts to "trivialization." For much work can be accomplished by this
move.
Some have charged that such a move is not just trivial but "gimmicky." 57
The term appears to be a pejorative one. But while I concede that the move
does not do much to illuminate certain substantive theories with respect to
how they are to be contrasted with others (does it help us to understand
what is distinctive about Kant's theory to put it in terms of maximization?
Surely not), I do believe that it helps illuminate the concept of obligation
itself. And I concede too that, if a certain theory (such as Kant's, perhaps,
or the divine command theory) implies, when put in terms of maximiza-
tion, that that which is obligatory is infinitely deontically superior to that
which is wrong (so that, we might say, that which is obligatory has a value
of 1 and all other alternatives a value of 0), then nothing is really gained in
this instance by using the terms of maximization. But neither is anything lost.

56 Stocker (1990), pp. 299-300.

57 Nozick (1974), p. 29n.; Vallentyne (1988).

19
Moreover, what such a lack of gain shows is simply that such theories are
in fact extremely impoverished, conceptually. We shall see that, where
finer deontic rankings of alternatives are permitted by a theory, we can pro-
vide an account not just of mere obligation but also of conditional obliga-
tion, prima facie obligation, and what I shall call "subsidiary obligation,"
and also that what I shall call the "dynamics" of obligation are illuminated.
Theories with minimal rankings (theories that rank what is obligatory first,
all else last) are themselves limited, in that they cannot express these other
aspects of the concept of obligation.

1.5 ACT-EVALUATION AND O T H E R TYPES

OF EVALUATION

Much of ethics has to do with the evaluation of agents, their characters,

their motives, and so on. Some of ethics has to do with the evaluation of
acts. An inquiry, such as the present one, into the concept of obligation
might be classified as an inquiry into the evaluation of acts. It is certainly
quite distinct from an inquiry into some types of evaluation of agents, their
characters, their motives, and so on. Still, the lines are hard to draw with
precision. We might describe the present work as an inquiry into what it is
for an act to be obligatory, and then the classification of it as an inquiry into
the evaluation of acts might seemjustified. But, although I shall often write,
and have already written, as if such classification is appropriate, we might
equally well describe the present work as an inquiry into what it is for an
agent to be obligated to do some act, and then such classification might seem
less appropriate.58 The fact is that I am not really concerned with what the
proper classification in these terms is; I hope simply that I have done enough
by now to give you a fair idea of what the subject of inquiry is. I propose
now to turn to that inquiry.

58 Cf. Prichard (1949), pp. 36-7; Ross (1939), p. 56.

20
 Moral obligation: An analysis

2.1 VALUE MAXIMIZATION

2.1.1 Two problems
At his doctor's request, Sam sticks out his tongue and says: "Aah." In order
to vote, Vincent drives his car to the polling station. Puzzles lurk within
these mundane examples.
The claim that one ought to do the best one can (where, as explained in
the last chapter, "ought" expresses what I have called "mere obligation,"
and "best" refers to deontic value) is perhaps most naturally, indeed stan-
dardly, understood to mean the following: of all those actions which one
can perform (one's "alternatives," as it is often put), one ought to perform
that action (if any) whose deontic value would be greater, if it were per-
formed, than the deontic value of any other, if it were performed. But there
are at least two problems with so understanding it.1
The first problem was exposed by Hector-Neri Castaneda.2 Suppose
that it is deontically best that Sam comply with his doctor's request.
Suppose, that is, that
(2.1) Sam ought to stick out his tongue and say: "Aah."
Then presumably
(2.2) Sam ought to stick out his tongue, and Sam ought to say: "Aah."
But, on the current understanding of what it is to be obligated to do the
best one can, the first part of (2.2) implies that
(2.3) the deontic value of Sam's sticking out his tongue would be greater,
if it were performed, than the deontic value of his saying "Aah," if it
were performed;

1 For a detailed discussion of this issue, see Feldman (1986), Ch. 1.

2 Castaneda (1968).

21
while the second part of (2.2) implies that
(2.4) the deontic value of Sam's saying "Aah" would be greater, if it were
performed, than the deontic value of his sticking out his tongue, if
it were performed.
But (2.3) and (2.4) are manifestly inconsistent with one another.
There are five main responses to this puzzle. One might, first, deny that
it is legitimate to talk of compound events (such as that of Sam's sticking
out his tongue and saying "Aah") as actions or, at least, as the sort of thing
one ought to do. But this is surely a revisionary proposal. Second, one
might deny that (2.1) implies (2.2). But, although some do deny this, it is
surely counterintuitive to do so.3 Third, one might deny that Sam's stick-
ing out his tongue is a genuine alternative to his saying "Aah." Although I
have some sympathy with this position, it is of no help to the thesis that one
ought to do the best one can, on its present interpretation. For, on this
interpretation, the thesis concerns simply all those actions that one can per-
form, and certainly each of these actions is such that Sam can perform it.
Perhaps it is a mistake to call these actions "alternatives," but that is strict-
ly irrelevant. Fourth, one might deny that (2.3) and (2.4) are inconsistent.
But this seems patently false. Fifth, one might hold that the thesis that one
ought to do the best one can requires a somewhat different interpretation
from the standard one. This, I believe, is the correct response to the
puzzle.
The second problem reinforces this response. Suppose that the deontic
value of Vincent's voting would be greater, if he were to perform it, than
the deontic value of any other action that he can perform, if he were to per-
form it. In short,
(2.5) Vincent's voting would be deontically superior to his not voting.
Then the theory, on its current interpretation, implies that
(2.6) Vincent ought to vote.
But suppose also that, in order to get to the polling station, Vincent must
drive his car. That is,
(2.7) Vincent cannot both vote and not drive his car.
Suppose, finally, that
(2.8) Vincent's driving his car would be deontically inferior to his not
driving it,
3 More on this later, in the present chapter and in Chapter 6.

22
and thus that, on the theory's current interpretation,
(2.9) it is not the case that Vincent ought to drive his car.
If we also assume that
(2.10) if one ought to perform a certain action A and cannot do so with-
out performing a certain action B, then one ought to perform B,
we reach a contradiction.
Again, a number of responses are possible. One might deny the gener-
al principle embodied in (2.10), but this is surely counterintuitive.4 It is
worth noting here that (2.10) implies the claim that (2.1) implies (2.2);
denial of this claim would therefore require denial of (2.10).5
Second, one might question the consistency of (2.5), (2.7), and (2.8).
How can it be that Vincent's voting is better than his not voting, and his
driving worse than his not driving, when he cannot vote without driving?
The answer must, of course, rest on just what deontic value consists in, but
certainly this is perfectly possible on some substantive views about such
value. Consider, just by way of example, the view that an action's deontic
value is a function of the sum intrinsic value of its consequences. It could
well be that the consequences of Vincent's driving (including those conse-
quences which are the action of voting and its consequences) would be
intrinsically worse than those of his not driving, whereas the consequences
of his voting (now that the driving, for good or ill, has occurred) would be
intrinsically better than the consequences of his not voting. Or consider,
by way of another example, the view that an action is deontically superior
to its omission if and only if the action's maxim is universalizable whereas
the omission's is not. As far as I can tell, there is no guarantee that, if the
maxim for Vincent's voting were universalizable and its omission's not,
then the maxim for his driving would be universalizable and its omission's
not. Or, finally, consider the view that an action is deontically superior to
its omission if and only if the action does not involve the infringement of
anyone's rights whereas the omission does. Again, one might wonder
whether there is any guarantee that, if Vincent's voting involved no such
infringement, his driving would also involve no such infringement.

4 Again, more on this later and in Chapter 6.

5 Let A be the compound action of Sam's sticking out his tongue and saying "Aah" and B
be the action of his sticking out his tongue. Since Sam cannot perform A without per-
forming B, (2.10) implies that, if he ought to perform A, then he ought to perform B.
6 Some rights-theorists in fact insist on such a guarantee. See Thomson on the "Means
Principle for Claims" in Thomson (1990), pp. 156—7. This is in contradistinction to her
earlier view, in Thomson (1986), p. 109, that there is no such guarantee.

23
 A third response to the second problem is this. One might take note of
the passage of time involved and contend that it is a mistake to claim that
both (2.6) and (2.9) are true. At the time at which it is true to say that it is
not the case that Vincent ought to drive his car, it is not yet, and may never
be, true that Vincent ought to vote. Whether or not he ought to vote will
depend on whether he drives. Only if he drives will it be the case that he
ought to vote. But while this response may seem plausible, there are rea-
sons to think it mistaken; I shall go into these in Chapter 3. More to the
point here, the response is irrelevant, since the problem at issue does not
rest on the passage of time. Consider a parallel case in which

(2.5') Vincent's voting for Smith would be deontically superior to his vot-
ing for no one, which itself would be deontically superior to his vot-
ing for Jones,

(2.7') Vincent cannot vote without pulling a certain lever,

and
(2.8') Vincent's pulling the lever would be deontically inferior to his not
doing so
are all true, and where the reason that (2.8') is true is that, if Vincent were
to pull the lever, he'd vote for Jones.
A fourth response is this. We should not talk of individual acts being
obligatory; we should talk only of entire courses of action being obliga-
tory. Hence it is not the case that Vincent ought to vote. Rather, either he
ought to drive-and-vote-and-X or he ought to not-drive-and-not-vote-
and-Y (presumably we can rule out its being the case that he ought to drive-
and-not-vote-and-Z), where X and Y (and Z) represent whatever the
remainders are in the respective courses of action. But this is surely much
too revisionary a proposal. It would imply that almost all, and probably
all, of our everyday assertions concerning what we ought to do are simply
false (since they do not concern entire courses of action but only segments
thereof).
A final response is this: we should revise the standard interpretation of
the thesis that one ought to do the best one can. Once again, it is this
response that I endorse. The revision that I think these (and other such)
problems call for is the following.7

7 What follows borrows heavily from Feldman (1986), Ch. 2.

24
 2.1.2 The analysis
We should say that what one ought to do is a function of the deontically
best course of action (and concomitant events) open to one. We should
note that, whenever we are in a position to make a choice as to how to act,
it's not just that we can do action A or action B or action C (or some com-
bination thereof), but that we can affect how the world will be. This may
sound grand, even melodramatic, but it's the truth. For example, if I can
break a certain store window or not (by throwing a brick through it, or not)
— if I have a choice as to whether or not the window breaks; if it is up to
me whether or not the window breaks — then it's also true that I can make
the world be such that it contains a certain broken window (or not) — I have
a choice as to whether or not it contains this; whether or not it contains this
is up to me. Of course, much - most - of what happens in the world is not
up to me, but that does not alter the fact that how the world as a whole will
be is (often) up to me. Another, and by now fairly well-established way of
putting this point is as follows: at any moment of choice a number of dif-
ferent possible worlds are accessible to me; which of these will be the actual
world is up to me.
The situation may be very roughly illustrated as in Figure 2.1. Here the
agent, 5, has a choice at time T1 between doing A and not doing A. (Already
this is oversimple, in that there would presumably be a number of ways of
not doing A — by doing E, or F, or G, and so on, instead. Note that how-

Figure 2.1
25
ever not doing A is effected, it of course involves the failure to do A. Doing
something else, E, may not be sufficient for not doing A, since it may be
that S can do both A and E; thus, if not doing A is effected by doing E, E
must be done instead of A.) If S does A at Tu he will have a choice at T2 as
to whether or not to do B. If S doesn't do A at Tt, his only "choice" at T2
will be not to do B. If S does B at T2, he will have a choice at T3 whether
or not to do C, but his doing D will be a foregone conclusion. If S doesn't
do B at T2, his not doing C will be a foregone conclusion, but he will have
a choice at T3 whether or not to do D. The nodes in the figure thus repre-
sent points of choice and the paths W-W6 represent different possible
worlds (each sharing the same history up to T?) accessible to S as of Tx by
means of his choosing one way or the other. Of course, huge amounts of
information are left out of the picture: what happens before T?, what hap-
pens after T3, what happens between T, and T3 apart from the small num-
ber of actions to which reference is made. Nonetheless the picture is
helpful, I think, in bringing home two important points. First, as noted, at
any moment of choice our choice is, in effect, a choice between worlds.
Second, as time progresses, our options never increase but in fact grow
narrower, in that with each choice we cut ourselves off from worlds which
had hitherto been accessible to us; what is presently accessible to us always
was accessible to us (during our lifetime), but it may not remain so for
long.
Clearly, what has just been said is far from precise and needs further
work. I shall attend to that shortly. But at this point I want to lay out my
interpretation of the thesis that one ought to do the best one can; for then
the refinements to what has just been said will, I think, be more readily
appreciated. The main gist of the interpretation is simply this: an agent
5 ought to do an act A if and only if (a) 5 can do A, (b) 5 can refrain
from doing A9 and (c) any accessible world in which S does not do A
is deontically inferior to some accessible world in which S does A. More
precisely:

(I) agent 5 ought, at time T i n world W, to do action A at time T' iff8

(a) there is a world W' such that W' is accessible to S from Wat Tand S
does A at T' in W'\
(b) there is a world W" such that W" is accessible to S from Wat Tand
5 does not do A at T ' in W"\ and
(c) for all worlds W" such that W" is accessible to S from Wat Tand S
does not do A at T ' in W'\ there is a world W' such that

8 "Iff' is short for "if and only if."

26
 (1) W' is accessible to 5 from Wat T,
(2) 5 does A at T' in W\
(3) the deontic value for S at Tof Wf is greater than the deontic value
for 5 at T of W , and
(4) there is no world W"' such that
(i) W"' is accessible to S from Wat T,
(ii) 5 does not do A at T' in W"\ and
(iii) the deontic value for 5 at Tof W"r is greater than the deon-
tic value for S at Tof W'.9

Clauses (a) and (b) combined imply that A is optional for S, in the sense that
S can do A and am also refrain from doing it. We may call this the personal
sense of "optional." Of course, these clauses do not imply that A is morally
optional for S, in the sense that 5 may, morally, do A and may also refrain
from doing it. (On the contrary, if S ought to do A, then it is not the case
that S may refrain from doing it, as will be made clear below.) Clause (c)
doubtless appears more complex than necessary. In particular, is subclause
(c4) needed? Yes, it is (or, at least, caution on a certain matter dictates it). I
shall explain why this is so in Subsection 2.2.6 below.
Perhaps it is as well to repeat here that "ought" is intended to express
overall moral obligation and that no substantive theory is being presup-
posed as to what deontic value consists in. The latter fact is important, since,
as explained in Section 1.4, (I) is intended to accommodate almost all sub-
stantive theories of obligation. Thus, once again, not only is (I) intended to
be compatible with consequentialist theories, but it is intended also to be
compatible with a wide range of deontological theories. It is for this reason
that the values of the worlds are said to be the values that they have^br 5 at
T. The values are thus both agent- and time-relative; whether or not they
are significantly so depends on the particular substantive theory in question.
If the theory is consequentialism, then the relevant value is intrinsic value.
Such value is standardly understood to be both agent- and time-neutral.

9 Note that clauses (b) and (c) jointly imply clause (a), so that, strictly speaking, the latter
is redundant. Nonetheless I include clause (a) in order to emphasize the fact that one
ought to do the best one can.
Also, it may be that S should be said to be not just an agent but a moral agent (what-
ever exactly that amounts to), for I can imagine that some people might want to say the
following: some (nonhuman) animals in fact have a choice between various alternative
actions, and these actions are in fact such that some are better (even deontically) than oth-
ers, but it is false that such animals morally ought to choose the best alternatives, because
such animals are not moral agents. (See, e.g., Thomson, 1990, p. 215.) There may be
something to this; I shall not commit myself one way or the other here.

27
Here, then, the relativization of the values o£W', W", and W'" is insignif-
icant and can be ignored, insofar as the values for S at T of these worlds are
the values that they have for any possible agents at any possible times. But
on other theories the relativization may be significant.
Consider Kant's theory, for example. It might at first be thought that
such a theory resists accommodation by (I), but this is in fact not so. If the
substantive mark of obligation is compliance with the Categorical
Imperative, then the deontically best world open to an agent is simply one
in which such compliance is maximized. This statement is perfectly legiti-
mate and should not be thought to be at odds with what a Kantian would
want to say. To see this clearly, consider the (oversimplified) situation
depicted in Figure 2.2. Suppose that each of A, B, and C would constitute
compliance with the Categorical Imperative by the agent, whereas none of
~^4, ~B, and ~C would. A "minimal" Kantianism might declare any fail-
ure by the agent to comply with the Categorical Imperative categorically
wrong and thus deem W t infinitely deontically superior to any of the other
worlds, so that (as we might put it) it has a deontic value of 1 whereas all
the others have a deontic value of 0.10 A somewhat richer Kantianism might
declare every noncompliance as deontically bad as any other but allow for
them nonetheless to "add up," in which case the deontic ranking would be
this: first, Wt (no noncompliances); second, W2, W'3, and W5 (one non-
compliance each); third, W4, W6, and W7 (two noncompliances each); and
last, W8 (three noncompliances). A still richer Kantianism would declare

Figure 2.2
10 Cf. Vallentyne (1992), p. 120, including n. 10.

28
some noncompliances deontically worse than others, and this would allow
for a still finer deontic ranking of the worlds. This way of picturing matters
is also compatible with saying (as I think almost all Kantians would) that it
can sometimes happen that a person can avoid someone else's noncompli-
ance only by way of his own noncompliance. Think, for instance, of
Bernard Williams's celebrated case ofJim and the Indians: Jim cannot get
the captain not to kill the Indians except by killing one himself.11 A Kantian
would say: either Jim will fail to comply with the Categorical Imperative
or the captain will. In such a case, a Kantian would presumably say that it
is right for one to comply and wrong for one not to comply (even at the
cost of the other's noncompliance). This means that the relativization of
the values of worlds to agents is significant, in that S's noncompliance has a
special disvalue relative to S that it does not have relative to someone else.
This point perhaps requires elaboration. I am envisaging the Kantian as
saying the following. Jim's killing the Indian would constitute a violating-
by-Jim of the Categorical Imperative, and hence it would be morally
wrong for Jim to do it. Admittedly, Jim's not killing the Indian would guar-
antee (in the sense that Jim could not avoid them, even if the captain could)
several violatings-by-the-captain of the Categorical Imperative, and thus
would guarantee that (in some sense) more serious wrong is done, but it
would not itself be a wrongdoing by Jim. It is for this reason that the agent-
relative valuation of worlds is significant. In some nonrelative sense we
might well want to say that some Jim-accessible world in which Jim kills
the Indian is better than any such world in which he doesn't, since the
former involves only one violation of the Categorical Imperative whereas
the latter involves several. But for the Kantian such a valuation of worlds,
while perhaps accurate and informative for certain purposes, is not the rel-
evant valuation for determining what it is obligatory for Jim to do. When
it comes to determining Jim's obligations, some Jim-accessible world in
which Jim does not kill the Indian is better than any such world in which
he does.12
Thus all, or almost all, Kantians will say that deontic value is significantly
agent-relative. It may be that they would also say that it is significantly time-
relative; this will depend on whether they accept that it is possible that a
person can avoid his own noncompliance with the Categorical Imperative
only by way of some other noncompliance by him. Suppose, for example,
that Joe can at T1 do A at T2 and can at T1 do B at T3 but cannot at Tt do

11 Williams (1973), p. 98.

12 See Scheffler (1984) for an interesting internal critique of this view.

29
neither. Suppose also (if this is possible) that A would constitute a minor
injustice and B a major injustice, and that, for this reason, each would con-
stitute a violation of the Categorical Imperative. I take it that the Kantian
who accepts that this is possible would say that Joe may not do A, even if
only thereby can he avoid the greater injustice, B. This means that any
accessible world containing his not doing B at T3 is deontically worse, rela-
tive to Joe at Tj, than some accessible world containing his not doing A at
T2; and this is so even if, by some other measure, every ~^4-world accessi-
ble to Joe at Tj is worse than some ~B-world accessible to him then. In such
a case, then, the time-relativization of deontic value is significant; agent-
relativization alone would not suffice to yield the deontic inferiority of
every ~B-world to some ~^4-world, since the actions in question are both
Joe's.
This last case also serves to reinforce the need to talk in terms of the val-
uation of worlds rather than that of acts. An art-based theory might well
imply that Joe ought at Tt to avoid A at T2, since avoiding A would be bet-
ter than doing A, and that he ought at Tt to avoid B at T3, since avoiding
B would be better than doing B. But, as with the case of Vincent present-
ed in the last subsection, this would be problematic, since Joe cannot at Tt
both avoid A at T2 and avoid B at T3. The world-based theory developed in
this chapter implies that it cannot be both that Joe ought at Tt to avoid A at
T2 and that he ought at Tt to avoid B at T3, since {ex hypothesi) he cannot
at Tj avoid both.
Thus Kantianism appears perfectly compatible with (I). So too do all
sorts of other deontological theories. Still, at this point, certain limitations
to the analysis must be acknowledged. First, it is at best only a partial analy-
sis. I have said that the value at issue in (I) is deontic value; I have not tried
to give an account of this notion in turn. Nor shall I. Clearly, deontic value
is a type of morally relevant value, but there are other types of morally rele-
vant value.13 A fuller analysis would give an account both of how morally
relevant value is to be distinguished from morally irrelevant value and of
how deontic value in particular is to be distinguished from other types of
morally relevant value. I have no such account to offer. Second, it must be
recognized that (I) does not accommodate all substantive theories of oblig-
ation. After all, consequentialism has traditionally been put in terms simi-
lar to those given at the outset of this chapter, and it was argued that such
13 What I have in mind here is the possibility that two worlds be such that thefirstis supe-
rior to the second with respect to deontic value, whereas the second is superior to the
first with respect to some other morally relevant value - for example, with respect to the
incidence of virtue and vice.

30
terms are unacceptable, involving a misunderstanding of the very concept
of moral obligation. Perhaps, too, a similar problem besets Kant's empha-
sis on the universalizability of maxims, as this is traditionally conceived. At
any rate, any substantive theory based on such a misunderstanding will of
course not be compatible with (I). Nonetheless, (I) is surely compatible
with the basic view underlying traditional consequentialism (namely, that
obligation is some rather straightforward function of intrinsic value), and I
see no reason to think that the basic view underlying Kantianism and other
deontological theories should not be thought to be similarly compatible.
Still, there is this limitation. Consider the sort of theory that says that one
ought to do what one believes to be best — in some less than fully liberal sense
of "best." This can be accommodated by (I) only if the theory is under-
stood to concern what, among those actions the agent can in fact perform,
is believed by the agent to be best; (I) does not accommodate a theory that
is concerned with what, among those actions the agent believes he can
perform, is believed by the agent to be best, unless the agent's beliefs
concerning his options are accurate.

2.1.3 Extension of the analysis

The thesis that moral obligation is to be understood in terms of the com-
parative values of possible worlds is by no means new. Jaakko Hintikka,
among others, has made such a proposal; and Holly Goldman, Michael
McKinsey, and, especially, Fred Feldman have provided accounts of oblig-
ation that are very closely related to the account given in (I). Each is some-
what different, though, and, I believe, in some way inferior.14
For a further illustration of the application of (I), consider Figure 2.3.
Let the numbers 1—3 on the right denote the value-rankings of the respec-
tive worlds.15 (Worlds la and lb are thus of equal deontic value, relative to
the agent at Tv) Then, given (I), we may say the following: the agent, 5,
ought (at Tt) to do A (at Tt) and also ought (at Tt) to do C (at T3). Ought
(at Tj) he to do B (at T2)? No, for clause (c) of (I) is not satisfied (in that,
although there is a top-ranked world containing B, there is also a top-
ranked world containing ~B).

14 See Hintikka (1970), pp. 70-1; Goldman (1978), p. 202; McKinsey (1975), p. 385; and
Feldman (1986), p. 38. There is no space here to make a detailed comparison between
each of these accounts and the account that I have to offer.
15 (I) does not presuppose that there is in every case an uppermost limit to the values of
accessible worlds, but it is easier, for purposes of illustration, to suppose this.

31
 -C&D

Figure 2.3

Figure 2.3 clearly contains additional information, though. To help us

express it, let us develop the account further.
First of all, we can say that S may do A if and only if (a) S can do A,
(b) S can refrain from doing A, and (c) any accessible world in which S does
not do A is not deontically superior to some accessible world in which 5
does A. (The term "may" thus allows for ties in what is deontically best,
whereas the term "ought" does not.) More precisely:

(II) S may, at T in W, do A at T ' iff

(a) there is a world W' such that W' is accessible to 5 from Wat Tand
S does ,4 at T ' i n W'\
(b) there is a world W " such that W " is accessible to 5 from Wat Tand
S does not do A at T" in W "; and
(c) for all worlds W " such that W " is accessible to 5 from Wax. Tand 5
does not do A at T' in H/ ", there is a world PF' such that
(1) W' is accessible to S from Wat T,
(2) 5 does A at T' in W ' , and
(3) the deontic value for S at To£W" is no greater than the deontic
value for S at T of W\

(The difference between (I) and (II) consists, first, in the difference in their
respective subclauses (c3) and, second, in the fact that (II) lacks a corre-
sponding subclause (c4). As to the former point: this concerns the issue, just
noted, of possible ties in maximal deontic value. As to the latter: I shall have

32
something to say about this in Subsection 2.2.6 below.) On Figure 2.3,
then, not only ought S to do each of A and C, he may do each of these as
well. Moreover, he may do B, too, although it's not the case that he ought
to do it; for he may also do ~J3, that is, fail to do, or refrain from doing, B.
(B is, in other words, not just personally but morally optional.)
Further:

(III) S ought, at Tin W, not to do A at T'

[i.e., S may not, at Tin W, do A at T'] iff
[as in (I), except that "does A' and "does not do y4" are reversed]

Thus, on Figure 2.3, S ought not to do D; also, he ought not to do ~A and

he ought not to do ~C.
Further:

(IV) S may, at Tin W, not do A at T' iff

[as in (II), except that "does A" and "does not do A" are reversed]
Thus, on Figure 2.3, S may (at Tj) not do B (at T2). Of course he also may,
because he ought, not do each of the following: D, ~A, and ~C.
It will already have become apparent that English terminology poses
certain problems in this context. "May not" is ambiguous, as is illustrated
in the statements of (III) and (IV); sometimes it is used to express the oblig-
ation not to do something, sometimes merely the permission not to do
something. Further problems of this sort lie ahead. To help us deal with
them, some symbolism will prove useful. Using O for "is obligated" (or
"ought") and P for "is permitted" (or "may"), we may say that (I) gives an
account of what is expressed by O(A); (II) gives an account of what is
expressed by P(A); (III) gives an account of what is expressed by O(~y4);16
and (IV) gives an account of what is expressed by P(~A).
Conspicuous for their absence from the foregoing are the terms "right"
and "wrong." How do they fit in? Fairly easily, I think.17 What it is wrong
for one to do is that which one ought not to do; in symbols: W{A) iff

16 It is perhaps worth reemphasizing that ~A expresses the failure to do A, not merely the
doing of something other than A (which may be compatible with the doing of A). Thus
O(~A) expresses the obligation to refrain from doing A, not merely the obligation to do
something other than A.
17 Here I am concerned exclusively with the adjectival form of these terms. The nominal
forms of "right" and "wrong" (as in "Rachel has a right to be left alone" and "You have
done Rachel a wrong in not leaving her alone") and the verbal form of "wrong" (as in
"You have wronged Rachel") are pertinent to moral theory but are not at issue here,
although they will be discussed in Section 5.5.

33
O(~A). What it is right for one to do is not quite so clear. Sometimes peo-
ple talk of "the right" thing to do under the circumstances; here what is
meant, I think, is that which one ought to do. Sometimes people talk ofwhat
it is "all right" to do; here what is meant, I think, is that which one may do.
Sometimes, perhaps, what people mean by "right" occupies some middle
ground between what is expressed by O(A) and what is expressed by P(A);
if so, I'm not sure what precisely is meant, and I shall therefore not seek to
accommodate it explicitly in the present account.
The account may be extended further. First:

(V) O(A&B):
S ought, at T i n W, to do both A at T" and B at T " iff
(a) there is a world Wf such that W' is accessible to S from Wat Tand
S does A at T" in W' and S does B at T " in W'\
(b) there is a world W" such that W" is accessible to S from Wat Tand
5 does not both do A at T ' in W" and do B at T " in W"\ and
(c) for all worlds W" such that W" is accessible to S from Wat Tand S
does not both do A at V in W" and do B at T " in W", there is a
world W' such that
(1) W' is accessible to S from Wat T,
(2) S does A at T ' in W' and 5 does B at T " in W\
(3) the deontic value for S at Tof W' is greater than the deontic value
f o r S a t T o f W", and
(4) there is no world W'" such that
(i) W'" is accessible to S from Wat T,
(ii) S does not both do A at T ' in W"" and do B at T " in W " ,
and
(iii) the deontic value for S at T of W'" is greater than the deon-
tic value for 5 at Tof W'.

That is, roughly: S ought to do both A and B if and only if (a) S can do them
both, (b) S can refrain from doing them both (whether by doing neither of
them or by doing one but not the other), and (c) any accessible world in
which S does not do them both is deontically inferior to some accessible
world in which S does them both. (Clearly this can be extended even fur-
ther to accommodate not just two actions but any number of actions.) O n
Figure 2.3, then, not only ought S to do each of A and B and C (that is, not
only is each of these such that he ought to do it); he ought to do all of them
(that is, he ought to do (A&B&C)). It might be asked what the difference
really is between saying, for example, that O(A) and O(B) and saying that

34
O(A&B). Well, there is a difference, even though, on my account, the two
are close to being strictly equivalent; that is, it is almost true that (O(A) &
O(B)) iffO(A&B).18 But this near-equivalence may mask the difference, so
consider first of all this:
(VI) P(A&B):
S may, at T i n W, do both A at T ' and B at T " iff
[as in (II), with changes as from (I) to (V)]

O n Figure 2.3, S may of course do (A&B&C); and it is also the case that he
may do A, he may do B, and he may do C. But notice, as mentioned ear-
lier, that he may also do ~B. Clearly, though, it's not the case that he may
do (B&~B) (because he cannot). Thus it is not in general the case that P(A)
& P(B) iff P(A&B). This inequivalence obviously marks a difference, and
I hope now that it will be clear how it is that there is a difference between
saying that O(A) & O(B) and saying that O(A&B), even though they are
nearly equivalent.
W e may of course also say the following:

(VII) O(~(A&B)) [W(A&B)]i

S ought, at T i n W, not to do both A at T ' and B at T "
[i.e., S may not, at T i n W, do both A at T ' and B at T"]
[i.e., it is wrong, at T i n W, for S to do both A at T ' and B at T"]
iff
[as in (III), with changes as from (I) to (V)]
(VIII) P(~(A&B)):
S may, at T i n W, not do both A at T ' and B at T " iff
[as in (IV), with changes as from (I) to (V)]
Thus, on Figure 2.3, S ought not to do both ~B and D; this is because he
ought not to do D (although it remains true that he may do ~B). Also, S
may not do both A and B (in the sense of (VIII); he is permitted not to do
them both); this is because he may do (A&~B) (although it remains true
that he ought to do A).
A final set of four statements concerns the tricky connective "or." As far
as I can tell, the following best captures prevailing English usage:

(IX) P(AorB):
S may, at T i n W, do either A at T ' or B at T " iff
(a) S may, at Tin W, do A at T'; and
(b) S may, at Tin W, do B at T " ;

18 See the discussion of (2.40') in Subsection 2.3.1 below.

35
 (X) O(AorB):
S ought, at Tin W, to do either A at T' or B at T " iff
(a) S may, at Tin W, do either A at T ' or B at T " ; and
(b) 5 ought, at Tin W, not to do both -A at T ' and ~B at T "
[i.e., it is wrong, at T i n W, for S to do neither A at T ' nor B
at T " ] ;
(XI) O(AnorB):
S ought, at Tin FF, to do neither A at T ' nor B at T "
[i.e., S may not, at Tin W, do either ^4 at T ' or B at T"] 1 9 iff
(a) S ought, at Tin W, not to do A at T '
[i.e., it is wrong, at Tin W, for S to do A at T']; and
(b) 5 ought, at Tin W, not to do B at T "
[i.e. it is wrong, at T i n W, for S to do B at T " ] ;
(XII) P(AnorB):
S may, at Tin W, do neither ^4 at T ' nor B at T " iff S may, at Tin
W, both not do A at T ' and not do B at T". 20

Thus, on Figure 2.3, we may say each of the following (among other
things): S may do either A or B; S ought to do either (A&B) or (^&~B>); 5
ought to do neither ~ C nor D; and S may (in the senses of both (XI) and
(XII)) do neither ~ C nor D.
Some other points concerning English usage should be made here. First,
it seems especially important to point out that the statements "S may do
either A or B" and "S ought to do either A or B" are to be standardly under-
stood as interpreted in (IX) and (X), respectively, and not, as some logicians
would have it, in terms of mere disjunction.21 For this reason I have used
"or" and "nor" as part of the symbolism rather than resort to the usual log-
ical symbol for disjunction, v. The point here is this. For any propositions
p and q, \ip is true, then so is p or q (p V q). Thus, where p and q concern
actions (as where p is the proposition that S does A and q is the proposition
that 5 does B), then, if S does A, it is also true that either 5 does A or S does
B. Some have gone on to infer from this that, if S does A, then S does either
A or B. Now, it may be that there is a sense of "or" that permits this infer-
ence, but, even if this is so, it seems to me quite contrary to normal usage

19 Another possible English equivalent: 5 may, at T in W, do neither A at T' nor B at T".

But what this phrase means here is to be sharply distinguished from what it means in (XII)
below. In such cases, the symbolism that has been provided is particularly helpful.
20 That is, P(AnorB) is equivalent to P(~A&~B) and not simply to P(~Aor~B).
21 See, e.g., Castafieda (1981), p. 64.

36
to go on to say that, if S may (or ought to) do A, then S may (or ought to) do
either A or B.
Second, frequently, where P(AorB) or O(AorB) is true, so too is ~O(A)
& ~O(B). But this is not necessarily so. Similarly, it is frequently, but not
necessarily, also the case that 5 cannot do both A and B.
Third, (X) can be used to capture what is at issue in such common claims
as "Robin ought to return the book by next Wednesday," "Robin ought
to return the book between Tuesday and Wednesday," and so on. Let A be
the act of returning the book and Tx... Tn be a (perhaps discontinuous) series
of times, with T7 being the time of "ought" and Tn being Wednesday.
Then what's at issue in such examples is this: Robin ought at T} to do either
A at T2 or A at T3 or.. .or A at Tn. Similarly, (V) can be used to capture what
is at issue in such common claims as "Marsha ought to maintain silence
throughout (or during) the exam." Let A be the act of maintaining silence and
T2... Tn be a series of times, with T2 being the beginning of the exam and
Tn being the end. Then what's at issue in this example is this: Marsha ought
at T1 both to do A at T2 and.. .and to do A at Tn.
Fourth, the English term "ought" is peculiar with respect to tense. First,
it has no future tense. But this is of no philosophical significance; one can
simply use "will be obligated" to express future obligations.22 Second,
"ought" appears to have no past tense.23 But, again, this is of no philo-
sophical significance; one can use "was obligated" or "ought to have" to
express past obligations.24 At least one philosopher has been misled by these
grammatical peculiarities of "ought" and been moved to formulate the
"Principle of the present-tenseness of ought," according to which, the verb
"ought" always being in the present tense, it follows that, once S ought to
do A, always thereafter 5 ought to have done A. Although this may seem
correct (indeed is correct, given that "ought to have" expresses a past oblig-
ation), what in fact is meant here is that, if 5 ought at Tt to do A at T2, then
5 ought at all times subsequent to Tt to do A at T2.25 This, I submit, is patent
nonsense, and it is ruled out in (I) by clause (b); after A is done at T2, there
is no longer an accessible world in which A is not done at T2.
22 See Broad (1985), p. 225.
23 I say "appears," because I'm not exactly sure how to classify grammatically the phrase
"ought to have."
24 An alternative to "ought to have," perhaps more revealing but nonetheless nonstandard,
is "didn't ought," which some British speakers sometimes use to express past obligations.
25 See Castafieda (1981), p. 61, where this example is given: "Yesterday is when I ought,
now, to have visited Mary." (Cf. Castafieda, 1975, Section 7.12.) Feldman is commit-
ted to something like this, too, but not on the basis of a misunderstanding of what "ought
to have" means. See Feldman (1986), p. 43.

37
 2.1.4 Perfect and imperfect duties
Sometimes a distinction is drawn, following Kant, between "perfect
duties" and "imperfect duties." Kant's famous four examples of duty con-
cern the duty not to commit suicide, the duty not to deceive others by
virtue of making a false promise, the duty to develop one's talents, and the
duty to act charitably.26 The first two are said to be perfect duties (the first
to oneself, the second to others) and the last two are said to be imperfect
duties (the first to oneself, the second to others). Just what the distinction
is supposed to consist in is controversial. In a note Kant says that a perfect
duty is one that admits of no exception in the interests of inclination,
whereas an imperfect duty admits of such exception, but just what this
means is unclear.
The matter seems to me very complex. First, it is worth remarking that
it is natural to express the perfect duties negatively (as the duties not to do
such-and-such), as I have done, while the imperfect duties are naturally
expressed positively. Is this significant? It's hard to tell. Certainly, some
have thought that there is a significance to the distinction between positive
and negative duties, but it's not easy to see why one should think so.
Indeed, it might seem that perfect duties could be expressed positively (in
this case, as the duties to preserve one's life and to behave honestly towards
others) and imperfect duties negatively (in this case, as the duties not to let
one's talents go to waste and not to behave meanly towards others),
although it must be acknowledged that these alternative renderings might
not quite hit the mark: isn't there, after all, a difference between (positive)
action and (negative) omission or refraining?28 Whatever the truth of the
matter, the present account is designed to accommodate both positive
duties (see (I)) and negative duties (see (HI)).29
Something else that Kant might have had in mind is this. A perfect duty
is fully specific, whereas an imperfect duty leaves it to the agent to deter-
mine just how the duty is to be fulfilled. For example, a duty not to kill
oneself is just that, whereas a duty to be charitable can be satisfied either by
being charitable towards A or by being charitable towards B, and so on.

26 Kant (1964), pp. 89-91.

27 E.g., Foot says that negative duties are more stringent than positive ones, in Foot (1980),
p. 162ff.
28 The literature on this topic is huge. For just a small sampling, see Brand (1971), Gorr
(1979), Moore (1979), Walton (1980), and Zimmerman (1984), Ch. 8.
29 If a distinction is to be drawn between simply not doing A and refraining from doing A
and/or omitting to do A, etc., so be it. Then the formula "S ought to refrain from doing
A (or omit to do A, etc.)" can be understood in terms of (I), with "refrain from doing
A" being substituted for "do A."

38
This distinction, I think, is problematic. Hardly any duty is fully specific in
the sense indicated; after all, one can refrain from suicide by doing one thing
(e.g., by going to the movies instead), or by doing another, and so on. 30 At
any rate, the present account is adequate to the expression of what is at issue
here. If S ought not to commit suicide, then every accessible world in
which 5 commits suicide is one that is deontically inferior to some world
in which he does not. If 5 ought to act charitably, then every accessible
world in which S fails to act charitably is deontically inferior to some world
in which he acts charitably. If there is more than one top-ranked accessible
world in which S acts charitably, then S may enjoy a certain sort of latitude
with respect to how to satisfy the duty to act charitably. Perhaps in one such
world he is charitable towards A but not towards B, in another towards B
but not towards A, and so on. In this case S ought either to act charitably
towards A or to act charitably towards B (or...) (see (X)), but it's not the
case that he ought to act charitably towards A and it's not the case that he
ought to act charitably towards B. Thus it is not (morally) up to S whether
to act charitably, but how to do so is up to him.
Another thing that Kant might have had in mind is this: a perfect duty
is a duty-to (with a correlative right), but an imperfect duty is merely a
duty. 31 The present account is not designed to accommodate this distinc-
tion explicitly. As noted in Section 1.3,1 seek here only to account for what
I have called "mere obligation," that is, for what it is to have an overall duty
(or obligation) to do something; how to supplement this with an account
of what it is to have a duty-to (or obligation-to) is something that I shall
leave aside.
Sometimes it is said that what Kant had in mind is this: a perfect duty is
a duty to perform some act (or to refrain from doing so), whereas an imper-
fect duty is a duty to adopt a certain maxim for action. 32 It may seem that
the present account cannot accommodate this, since it has been put in terms
of " 5 ought to do ^4," etc. I shall address this point in the next subsection.
Finally, it might be thought that what Kant had in mind is this: a perfect
duty ought always to be satisfied, but an imperfect duty only ought some-
times to be satisfied.33 What is meant by this? Presumably the following. To
say that S has a perfect duty at Ti to do A is to say, on this understanding,
that
(2.11) for all times T (at which 5 exists) not earlier than Tu S ought at Tt
to do A at T.
30 Cf. Chisholm (1968), p. 416, n. 6.
31 This is how Mill understands the distinction in Mill (1957), Ch. 5.
32 Cf. Hill (1971), p. 57.
33 Cf. Hill (1971), pp. 56-7.

39
To say that S has an imperfect duty at T1 to do A would then be to say that
(2.12) for some but not all times T (at which S exists) not earlier than Tu
S ought at Tt to do A at T.
(This way of putting things presupposes that S can do A on more than one
occasion - that is, that A is "repeatable." This view of action is controver-
sial; I shall address it in Subsection 2.2.5 below.) Of course, Kant presum-
ably meant that the duties of which he spoke are duties had by everyone (or
at least by everyone capable of having duties of the sort in question). In this
sense, a perfect duty to do A would be understood to consist in the truth
of the following:
(2.13) for all agents S and times T' and T (such that Tis not earlier than
T' and S exists at these times), 5 ought at T' to do A at T.
Indeed, a so-called absolutist such as Kant34 might mean that a perfect duty
is one which ought always to be satisfied by everyone "no matter what."
This phrase is unclear but presumably expresses some sort of necessity.
Thus, on this understanding, a perfect duty to do A consists in the truth of
the following:
(2.14) necessarily, for all agents 5 and times T' and T (such that Tis not
earlier than T" and S exists at these times), 5 ought at T' to do A at
T.
The present account shows how these statements are to be understood.

2.2 UNDERLYING METAPHYSICAL AND

CONCEPTUAL ISSUES
2.2.1 "Can" and control
According to the analysis of obligation just presented, S ought to do A only
if S can do A and 5 can refrain from doing A. To be obligated to do A, then,
S must be in control of whether or not he does A; it must be personally
optional for S. That "ought" implies "can" is of course controversial; I shall
address the matter in Section 3.1. Here I want to attend to a preliminary
question. The present account may seem to restrict itself to obligations
concerning actions ("5 ought to do A"; "5 can do A"). Is such restriction
warranted?
34 Or Donagan (1977) or Fried (1978).

40
 Strictly, it is not. The sort of control, or freedom, expressed by "can" in
this context may be either direct or indirect. It is controversial just what
sort or sorts of item are such that an agent may have direct control over
them, but it is clear that there must be some such sort or sorts for there to
be any control at all. Consider the example given earlier: that of breaking
a certain store window by throwing a certain brick through it. Let us say
that this is both something that I can do on a particular occasion and some-
thing that I can refrain from doing. I thus have control over whether or not
the window breaks. This must surely be only indirect control, though; I
can achieve the window's being broken only by way of achieving some-
thing else. What is this something else? Moving my arm, for one thing; and
releasing the brick; and so on. On my view (although the truth of this claim
is not essential to the present account of obligation), the first item in this list,
that is, that item over which I have direct control, is a choice or decision to
throw the brick (or some suitably related decision — one that will result in
the window's being broken). I have direct control over the decision and
thereby indirect control over its consequences. These consequences include
my arm's moving, the brick's being released, the brick'sflyingthrough the
air, the brick's hitting the window, the window's breaking, the alarm's
sounding, and so on.
It is worth noting that none of the items on the list given in the last two
sentences is an action - unless a decision is itself an action, which I suppose
we may grant in a minimal sense of "action."35 Yet I have control over all
of them. Where do nonminimal actions themselves fit in? As follows, I
believe.36 The action that is my moving my arm just is (roughly) my decid-
ing to do something relevant and this decision's causing my arm's moving.
Thus I have a sort of "hybrid" control over this action: I have direct con-
trol over the decision involved and (thereby) indirect control over its con-
sequence, and my action just is this decision plus this consequence.
Similarly, my releasing the brick is this same decision coupled with the con-
sequence of the brick's being released; my breaking the window is this same
decision coupled with the consequence of the window's breaking; and so
on; and I have a sort of hybrid control over all these actions.
This applies to omissions, too. Perhaps I have made a promise to stay
indoors and water the plants; when I go outside and throw bricks, I omit
to water the plants. I have control over this omission. How? Once again, I
believe, my control can be traced to a decision that I make, a decision whose
35 Some contend that only decisions are actions (see, e.g., Prichard, 1949, p. 190, and Davis,
1979, p. 41). This is surely too restrictive.
36 See Zimmerman (1984), Ch. 4, and (1988a), Ch. 2.

41
consequences are incompatible with my remaining indoors and watering
the plants. My not remaining indoors and watering the plants may thus be
said to be a consequence37 of this decision.
It may be that certain items over which we have control don't fit easily
into this decision-plus-consequence picture. I am thinking primarily of
certain mental phenomena, such as desires, motives, intentions, and other
such attitudes, as I shall call them. (Obviously, there are important distinc-
tions to be drawn among these phenomena, but doing so is not necessary
for my purposes here.) Sometimes, it is clear, the adoption of an attitude is
the result of acting in a certain way (I may have had to school myself, at
length, in order to gain my newly acquired tolerance towards others who
are in certain ways different from myself; I may have had to work hard to
rid myself of my desire for cigarettes; and so on), and this fits in well enough
with the picture in question. But it may seem that at other times I can adopt
a certain attitude directly, that is, without doing so by way of deciding or
acting in a certain way. Perhaps, when Kant urges the adoption of a certain
maxim, he means (in part) that we ought to adopt a certain attitude - an
attitude of preparedness to act in certain ways. (I shall say something simi-
lar concerning cooperation in Chapter 9.) And perhaps it is possible to do
this directly. I suspect not, but I'm not sure. At any rate, to repeat, it is not
essential for present purposes that the foregoing decision-plus-conse-
quence picture be accepted. What is essential is only that the distinction
between direct and indirect control be accepted, and this distinction seems
to me unassailable. Surely one cannot have control over anything without
having direct control over something; and surely not everything one has
control over is such that one has direct control over it. Moreover, it is sure-
ly plausible to think of some things (on my account, nonminimal actions)
being such that one's control over them is a hybrid of direct and indirect
control.
We thus have three categories of control: direct, indirect, and hybrid.
And so we have three candidates for what is obligatory: those items over
which we have direct control, those items over which we have indirect
control, and those items over which we have hybrid control. I see no rea-
son in principle to reject any of these candidates.38 That is, it seems to me
perfectly proper to say, for example (and now I'll resort to the account of
action just given), that it is obligatory for me that I decide to stay indoors
37 Of a certain sort. See Zimmerman (1984), pp. 58-9, and (1988a), pp. 23-4, 30, and 93.
38 As Feldman (1993) demonstrates, on that type of account of obligation that both he and
I advocate, no inconsistency arises if the class of items that can be obligatory is claimed
to be wider than the class of actions.

42
and water the plants; it is obligatory for me that I stay indoors and that I
water the plants; and it is obligatory for me that I be indoors and that the
plants receive water. Similarly, it is obligatory for me that I not decide to
break a certain window; it is obligatory for me that I not break the win-
dow; and it is obligatory for me that the window not break. Part of the rea-
son why all of these things are obligatory for me is that I have control over
them, although only some of these things are actions — in the minimal sense,
those over which I have direct control, and, in the nonminimal sense, those
over which I have hybrid control; those over which I have only indirect
control are not actions in any normal sense of the term.
Nonetheless, I shall continue to talk in the abstract in such terms as "S
ought to do A," "S can do A" and so on, where A ranges over things we
do. The reasons for this are first that doing so simplifies exposition, and sec-
ond that, even though we cannot properly be said to "do" the conse-
quences of our decisions, they are nonetheless something that we bring
about — and it cannot ever be the case that a certain consequence be oblig-
atory for S without its being the case that it is obligatory for S to bring it
about, that is, so to act (whether nonminimally or only minimally) that the
consequence occurs.
At this point, though, some would demur. In particular, H. A. Prichard
and W. D. Ross would claim that only those items (in their view, what they
call "volitions" or "self-exertions") over which an agent has direct control
are such that they may be said to be obligatory. Prichard and Ross at times
talk of actions as being obligatory, but they claim that, in the strict sense,
only volitions are actions; items that include the consequences of volitions
(as, on my view, watering the plants would include the plants' receiving
water) are not, strictly, actions39 and not, strictly, obligatory.40 The reason
that they say that only volitions can be obligatory is apparently that they
believe that only volitions are, strictly, in an agent's control.41 But this can-
not be right; after all, the window's being broken (or not) is in my control,
and this isn't a volition. Of course, this isn't in my direct control, and I think
that what Prichard and Ross have in mind is that anything that is only in an
agent's indirect control is notfully in the agent's control. For example, after
I make the decision to throw the brick, what happens is not up to me but
rather, as it is sometimes put, "up to nature."42 And nature may not coop-
erate. I may decide to throw the brick, but I may be stricken by a sudden
39 Prichard (1949), pp. 20, 190.
40 Prichard (1949), pp. 34-5.
41 Ross (1939), p. 160.
42 See Davidson (1980), p. 59.

43
paralysis, or the brick may be heavier than I thought, or the brick may slip,
or my aim may be off, or I may have miscalculated the distance 'twixt hand
and window, or the window may be made ofplexiglass, and so on, in which
case the window will not break despite my decision to throw the brick; and
whether or not any of these things is so is not (at the present time) in my
control.
Now, it is surely correct to say that whether or not the window breaks
is, in some sense, not just up to me but "up to nature." There is a sense,
then, in which the window's breaking is not fully in my control; whether
or not it breaks is, to some extent, a matter of luck. But so what? Under the
circumstances (which include the fact that nature is cooperative), whether or
not the window breaks is up to me. Why not, then, say that I ought not to
break it? Prichard and Ross appear to think that what's obligatory must be
fully in one's control, and they seem to think that that and only that which
is in one's direct control is fully in one's control. But this is a mistake.43
Whether or not I exert myself to do something is itself to some extent a
matter of luck, just as whether or not my exertions bear fruit is to some
extent a matter of luck; for such self-exertion requires the cooperation of
all sorts of things over which I lack control (e.g., that I not be distracted,
that I not suffer a heart attack, that I be in a position to exert myself, and so
on).44 If'full control were required for obligation, nothing would be obliga-
tory; but some things are obligatory, and so full control is not required.
Some measure of control is still required, however; but both direct and
indirect control constitute some measure of control, and so I see no need
to say that, strictly, only those items over which we have direct control are
obligatory.45
The obvious fact that our control is limited (not simply in the sense that
we are never fully in control of anything, but in the sense that some things
are such that we are not in control of them at all) means that we may divide
the future into two portions. Castaneda calls these the "Future Framework"

43 Broad makes this mistake too. See Broad (1985), pp. 133-4.
44 See Zimmerman (1988a), Section 4.11.
45 The same holds true of moral responsibility. One may be morally responsible (e.g., to
blame) for those things over which one has (had) direct control and for those things over
which one has (had) only indirect control. Still, there is an important difference here.
Only those things over which one has (had) direct control are relevant to the extent or
degree to which one is to blame — or so I argue in Zimmerman (1988a), p. 54ff. and
Sections 4.3 and 4.11. There is no analogous claim to be made about obligation, for it
does not come in degrees in this way.
For more on inadequacies in the position of Prichard and Ross, see McConnell
(1989).

44
and the "Future Zone of Indeterminacy."46 That which lies in the Future
Framework is that over which we have no control; it is fixed; it is featured
in every world accessible to us. That which lies in the Future Zone of
Indeterminacy is that over which we have (some measure of) control; it is
not fixed; it is featured in some but not every world accessible to us. Of
course, what's fixed for one person may or may not be fixed for another.
It is fixed for each of us that the sun will rise tomorrow; but although it is
(probably) fixed for me what Bill Clinton will do tomorrow, it is not fixed
for him what he will do tomorrow.47 This is an important point: what oth-
ers do may not be fixed for them, but it may be fixed for me, from which
it would follow that what worlds are accessible to me is (in part) up to them.
(This is crucial to the question of cooperation, as we shall see in Chapter
9.) Of course, sometimes what others do may not be fixed for me; I may
be able to influence their actions. In general, then, the actions of others may
be a part of my Future Framework or be a part of my Future Zone of
Indeterminacy, just as many other things may be (such as a certain win-
dow's being broken).48

2.2.2 "Can" and accessibility

What exactly is the relation between "can" and accessibility? So far, I have
hardly mentioned the latter when discussing the former, and yet it is in
terms of the latter that the formal account of obligation has been given.
In one respect, the relation is straightforward. We can say:
(2.15) S can at Tdo A at T' in W iff there is a world accessible to S from
Wat Tin which S does A at T'.
Or more generally (to cover those items over which one may have control
but which aren't strictly actions):
(2.16) it is personally possible for S at Tin Wthatp be true iff there is a
world accessible to S from Wat Tin whichp is true.
(Here p may itself be an entire world, that is, a possible proposition that
entails every proposition or its negation.) Although true, however, these
statements fail to reveal certain underlying complexities.
The concept at issue — that which I have called "personal possibility" —
is a very familiar one. It is crucial to a number of philosophical debates,
46 Castaneda (1975), p. 135.
47 Barring, of course, the truth of fatalism or hard determinism.
48 See Feldman (1986), p. 22; compare Lyons (1965), pp. 75, 82-3.

45
including the debate concerning freedom and determinism (which I shall
not discuss) and the debate concerning whether "ought" implies "can"
(which I shall discuss in Section 3.1). "Can" can, of course, be used to
express a number of different concepts of possibility, and so it would be
desirable if some further account of the particular concept of personal pos-
sibility were given here. Unfortunately, I haven't much to offer. I've
already indicated that, when it is personally possible for S to do A and it is
also personally possible for 5 to refrain from doing A, then S is in control of
his doing A. Still, doubtless a variety of concepts of control might be dis-
tinguished. Perhaps an illustration will help. There is a sense in which Carl
Lewis can run 100 meters in under 11 seconds but in which I cannot. But
it would be surprising if, in the sense of personal possibility, Lewis can do
this right now; for that to be true, there must be no obstacle in the way of
his doing so right now. Still, it is true, even in the sense of personal possi-
bility, that he can do so later (or, more accurately, that he can now do so
later); for, I assume, he can, right now, so act that he will later, right then,
run 100 meters in under 11 seconds. In other words, although there is no
world now accessible to me in which at some time (now or later) I run 100
meters in under 11 seconds, there is a world now accessible to Carl Lewis
in which he does so.
But these are just other words. I don't mean, by virtue of (2.15) and
(2.16), to define personal possibility in terms of accessibility. Although
some come close to advocating this,491 do not. If anything, the notion of
accessibility seems to me a philosopher's artifice that is itself to be under-
stood in terms of the notion of personal possibility. I suggest the following,
which is explicitly based on the view (already acknowledged to be contro-
versial, and certainly not to be defended by me here) that direct control
concerns decisions.
To say that someone can (in the sense of personal possibility) do some-
thing is to say this: either he can directly do it, or he can indirectly do it.
And to say that he can indirectly do it is, roughly, to say this: there is some-
thing else that he can directly do such that, if he were to do that, then he'd
be one step closer to doing this. That is, taking the first step would enable
him directly to take a second step, and doing this would enable him direct-
ly to take a third step, and so on, for however many steps are needed. This
is in keeping with the branching pattern of Figures 2.1-2.3: a move by the
agent from one point of choice takes him to the next, where he is able to
make another choice; a move from this takes him to another; and so on.

49 See Feldman (1986), pp. 17, 23. Certainly Feldman shrinks from defining accessibility in
terms of personal possibility.

46
 A first stab at rendering this more precise is the following:50
(2.17) S can at T do A at T" in W iff there are times Tv..Tn and
decisions Di...Dn such that:
(a) Tis identical with T?;
(b) T ' is identical with Tn;
(c) Tj is not later than T2 and.. .and Tn_i is not later than Tn;
(d) S can directly at Tt make Dj at T? in W7;
(e) if Dj is not identical with Dn, then for any D, and 7) (where i < n),
if 5 were to make D, at 7], then S could directly at Ti+1 make Di+1
at T)+J; and
(f) if S were to make Dn at Tw, then S would thereby do A at Tn.
This is pretty complicated; unfortunately, it requires further complication.
There are two problems with the proposal as it stands.
First, it may be that someone can now do something later without being
now able to do anything now. After all, there are worlds accessible to me
even as I sleep; for I will wake up, and then, at some point, I will be direct-
ly able to do something that will enable me to do something else, and so
on. This point is, I think, easily handled by revising clause (a) so that T is
said not to be identical with Tt but simply not later than it (as long as 5 exists
atT).
The second problem is trickier. Suppose that Chris can cross the road;
he can do this because he can make a decision to step off the curb, and so
on. But suppose that Chris is an inveterate daredevil and in fact would make
the decision to cross the road only if there were a lot of traffic about, in and
out of which he could (he thinks) dart; and suppose that, if Chris were to
make this decision under those circumstances, he would in fact get run over
and never reach the other side of the road. In this case, clause (e) of (2.17)
is not satisfied. What we need to say, I think, is that there are certain con-
ditions (that in fact obtain) such that, if Chris were to make the decision to
cross under these conditions, then that would enable him to do something
else, and so on.
For these reasons, I think that (2.17) should be replaced by the follow-
ing:

(2.18) 5 can at T d o A at T" in W iff there are times Tv..Tn9 decisions

Dv..Dtt, and conditions Ct...Cn such that:
(a) S exists at Tand Tis not later than T7;
(b) T' is identical with Tn;

50 Compare Chisholm (1976), pp. 62—4, and Zimmerman (1984), pp. 235—6, for similar
approaches.
47
 (c) Tj is not later than T2 and.. .and Tn_1 is not later than Tn;
(d) iS can directly at T1 make Dt under Ct at T1 in W;
(e) if Dj is not identical with Dn, then for any D(, C,, and Tj (where
i < n), if S were to make D( under C, at TJ, then S could directly at
Ti+1 make Di+1 under Ci+1 at Ti+1; and
(f) if 5 were to make Dn under Cn at Tn, then 5 would thereby do A
atTn.
This statement concerns only actions, of course, whereas it has been noted
that other sorts of things may be under an agent's control. To accommo-
date this, we may generalize (2.18) as follows:
(2.19) S can at Tso act at T' in W that p is true iff
[as in (2.17) through clause (e), and]
(f) if S were to make Dn under Cn at Tn, then p would be true.
We may also say:
(2.20) it is personally possible for 5 at Tin W that j9 be true iff for some
time T', S can at Tso act at T' in Wthatp is true.
And, whether or not p is an entire world, we can also put this as follows: p
is accessible to 5 from Wat T. Thus personal possibility just is accessibili-
ty, and (2.20) declares it to be analyzable in terms of "can." Furthermore,
this analysis establishes the truth of (2.15) and (2.16), once it is recognized
(as noted in Subsection 2.1.2) that, where S can do A, S can so act that the
world contains his doing A, and vice versa.
I have said that nothing ever becomes accessible that wasn't already
accessible. In a sense, then, whatever one can do one could always do. But
this might seem to fly in the face of an obvious fact: one can learn to do
things one couldn't do before; one can gain an ability. But here we need
only note that personal possibility is to be relativized to times; in particu-
lar, "can" has a double time index. What is directly in one's control is, tem-
porally, immediately so;51 but what is indirectly in one's control is,
temporally, only remotely so.52 Given this, the acquisition of a "new abili-

51 At least, this is typically so with respect to its initiation. Whether it is necessarily so is

unclear to me. Perhaps even the exercise of direct control is or can be a noninstantaneous
process, such that one can have direct control at some time over an event that begins to
occur only at a later time.
52 At least, this is typically so. Again, whether it is necessarily so is unclear to me. Perhaps
it is possible that one have immediate direct control over some event E, that E have F as
a consequence, and that E and F occur simultaneously, so that one's indirect control over
F is not remote but immediate.

48
ty" may be understood in terms of something's becoming immediately pos-
sible that was hitherto only remotely so (or, perhaps, in terms of some-
thing's becoming close to immediately possible that was hitherto a long way
from being so). If I can now speak French now, having learned to do so, I
could always speak French now. Of course, this doesn't mean that, at all
earlier times of my existence, I could then speak French then.

2.2.3 "Can" and knowledge

Sometimes "can" is linked with "know-how." I can't open the safe because
I don't know the combination to the lock; I can't get my VCR to work
because I don't know which buttons to push in what sequence; and so on.
I think this comes very close to saying that one cannot do something unless
one can do it intentionally.
There is certainly such a sense of "can." The illustrations just given con-
firm that. But it is not this sense of "can" that I'm working with here. On
the contrary, I can open the safe, and I can operate the VCR, in the pre-
sent sense of "can," and this is simply because I can (easily) perform those
physical movements which would result in these things being done.53 Of
course, there is a sense in which I'd be very lucky if I managed to do these
things; but in this sense of "luck" it is not that their accomplishment is
strictly out of my control, but rather that, among the very many things that
I can apparently do, I have no reason to choose to make this particular set
of movements (that would result in the safe's being open or in the VCR's
working) over the many other sets of movements (that would be fruitless).
So I can open the safe. Can I likewise win the lottery? Perhaps I can win
it; that depends on how it works. If I would win it by inscribing a certain
number on a card, and I can inscribe the number, then I can win it. But if
to win the lottery I have to purchase a certain ticket with the winning num-
ber already imprinted, then it is highly unlikely that I can win it. It is high-
ly unlikely, that is, that there is a world accessible to me in which I win the
lottery; for there are only a relatively few tickets that I can purchase (given
that my purchasing power is limited and that there are only a relatively few
stores that sell the tickets and that I can reach in time), and the likelihood
that one of these tickets has the winning number is very small. There is a
distinction to be drawn, then, between its being unlikely that I will do one
of the many things that I can do (as in the safe, VCR, and card-inscribing
cases) and its being unlikely that I can do a certain thing (as in the pre-
imprinted ticket case).
53 Cf. Feldman (1986), pp. 24-5.

49
 Many, indeed most, of the things that I can (in the present sense) do are
things I know nothing about. Those worlds that are presently accessible to
me are such that, at best, I only know a tiny fraction of their features.
Perhaps I have a fair idea about several (though certainly not all) of my
immediate and close-to-immediate personal possibilities; but the more
remote the possibilities, the less I will tend to know about them. Unless
I die within a year, I can now do many things a year hence; but, except
for the most mundane among them, I have little idea what these things
may be.54
In saying that the "can" that "ought" implies has no interesting tie to
knowledge, I am attempting to accommodate those substantive theories of
obligation (act utilitarianism, as traditionally understood, is one; certain
rights-theories are others) that allow for the possibility that an act be oblig-
atory even though the agent doesn't know that he can perform it or how
to perform it. Thus it is not consistent with the present approach to insist,
as some do,55 that an act can be obligatory only if the agent knows that he
can perform it or how to perform it. Still, it is consistent to make a claim
that comes close to this, namely, that unknown personal possibilities con-
fer no deontic value on the accessible worlds that include them and thus do
not affect the relative rankings of these worlds. If this were true, then,
although some such possibilities could nonetheless be obligatory, their
obligatoriness would be wholly parasitic on the obligatoriness of known
possibilities.

2.2.4 "Can" and control again

What precisely is the relation between "can," control, and freedom?
A person must do (cannot not do) something, as of a certain time, if he
does it in every world accessible to him as of that time. A person does do
something if he does it in the actual world. And a person can do something,
as of a certain time, if he does it in some world accessible to him as of that
time. Hence, since the actual world is accessible to everyone, whatever
a person must do he does do, and whatever he does do he can do. No
surprises so far: whatever is necessary is actual, and whatever is actual is
possible.
But there is a puzzle lurking here. Sometimes, when talking of what is
personally possible, people use the term "free to" instead of "can." "S can
54 See Feldman (1986), p. 24.
55 Cf. Prichard (1949), pp. 22-5; Ross (1939), pp. 149-51; McCloskey (1973), p. 62;
Lemos (1980), p. 301.

50
do A" and "5 is free to do A" are sometimes used interchangeably. Is this
appropriate? The answer must be no, I think, at least if the following two
points are conceded: first, that "can" is related to "muse" and "does" in the
way just indicated; and second, that, if a person does what he is free to do
(while he is free to do it), then he does it freely. For if one puts both points
together, and if one uses "can" and "free to" interchangeably, then one
obtains the result that whatever anyone must or does do he does freely. And
this is surely unacceptable in any standard use of "does freely" — and here I
mean to be invoking that common, if not very well understood, sense of
the term that is at issue in the debate concerning freedom and determin-
ism.56 And yet I declared earlier, in Subsection 2.2.1, that the "can" implied
by "ought" expresses control or freedom. How can this be so, given what
has just been said?
The fact is that what I said earlier was said too loosely, and I must now
put matters right. I have already noted (and this turns out to be crucial) that
there are two "can"-statements implied by the fact that 5 ought to do A,
namely, that S can do A and that S can refrain from doing A (i.e., there are
a world accessible to S in which S does A and a world accessible to S in
which S doesn't do A).57 For what I think is true is the following:
(2.21) S is in control at T of his doing A at T' iff 5 can at Tdo A at T'
and also can at T refrain from doing A at T';
(2.22) 5 is in control at T of his doing A at T" iff S is free at T to do A at
T' and is also free at Tto refrain from doing A at T'.
Thus, perhaps surprisingly, even though "can" in the present sense does not
imply "free to," the following is true:
(2.23) S can at T do A at T" and can also at T refrain from doing A at T"
iff S is free at T to do A at T' and is also free at T to refrain from
doing A at T'.
And so, I believe, "ought" implies "can," not just in that sense of "can"
with which I have so far been concerned and which I have put in terms of

56 Actually, I believe that there is more than one sense of the term at issue in the debate,
but there is no need to pursue this, since the present point holds, I think, on all relevant
senses.
57 Feldman's account of obligation (1986, pp. 38, 43) does not have this same implication.
According to him, "ought" implies "can do" but doesn't imply "can refrain." This leads
him to say that whatever is inevitable is obligatory - including all that lies in the past. This
is very odd. Still, he does also note (p. 43) that a second, related sense of "obligation" can
be accounted for simply by including the point about "can refrain."

51
accessibility, but in a new, stronger sense of "can" that might be put in
terms of "free to." And, given the relation between "free to" and "does
freely" mentioned earlier, we may also say:
(2.24) if S ought at Tto do A at Tand S does A at T, then 5 does A freely
atT.
(This concerns immediate obligation. The corresponding claim concern-
ing remote obligation, where Tis distinct from T', is this:
(2.25) if S ought at Tto do A at T' and 5 does A at T', then S does A
freely at T'.
This statement is not true; for, although S may be free at Tto do A at T',
he may lose this freedom in the interim and hence, by the time he comes
to do A, not do it freely.)
It might be thought that the reason why (2.23) is true, even though
"can" (as used here) doesn't imply "free to," is simply this:
(2.26) S is free at Tto do A at T' (or to refrain from doing A at T') iff S
can at T do A at T' and also can at T refrain from doing A at T'.
In other words, "free to" is two-sided, as it were, implying optionality,
whereas "can" is just one-sided. But I am not sure that this is so. For it seems
to me that a person can freely do something even though he cannot refrain
from doing it (in other words, even if it is not personally optional; even if
he is not in control of doing it).58 If this is so, and if, as I suspect, when some-
one does something freely he is (or was) free to do it, then (2.26) is not true.
That someone can do something freely without being able to refrain
from doing it seems to me to be one of the main lessons to be learned from
Harry Frankfurt's ground-breaking discussion of the relation between
moral responsibility and alternate possibilities.59 Frankfurt argues — success-
fully, I think — that a person can be morally responsible for what he has
done, even though he couldn't have done otherwise. (One example might
be this: Larry cannot leave the room, because the door is locked; he is
nonetheless morally responsible for staying in the room, in part because he
is unaware of the door's being locked.) What I think this shows is not that
a person can be morally responsible for what he has done even though he
didn't do it freely, but rather that a person can have done something freely
even though he couldn't have done otherwise. (Thus Larry freely remains
in the room, even though he cannot leave it.) It is for this reason that moral
58 At least in "standard control" of doing it. Perhaps he is in some sort of "curtailed con-
trol" of doing it. See Zimmerman (1988a), pp. 32-3.
59 Frankfurt (1969).

52
responsibility does not require alternate possibilities. The link between
responsibility and freedom remains intact.60
Let us suppose that freely doing something entails being, or having been,
free to do it. Then, as I have said, (2.26) is false, given the fact that freely
doing something does not require being (or having been) able to refrain
from doing it. Why, then, is (2.23) true, given that "can" does not imply
"free to"? My answer is simply this: doing something, while being able to
do otherwise, is sufficient for doing it freely;6 doing something freely is
sufficient for being (or having been) free to do it; and being free to do (or
to refrain from doing) something is sufficient for being able (where "able"
expresses what the "can" of personal possibility expresses) to do (or to
refrain from doing) it. I am not sure just how each of these claims might
itself be justified.
At this point, though, it is natural to ask: if moral responsibility doesn't
require alternate possibilities, why think that moral obligation does? I shall
consider this question in Subsection 3.1.5.

2.2.5 The ontological status of actions

A distinction is commonly drawn between two different ways of conceiv-
ing of actions: as types, or as tokens. Types are properties or relations;
tokens are instances of the exemplification of properties or relations.
Consider killing. Killing is an act-type, whereas a specific killing is a token
of this type. It seems that obligatoriness, rightness, and wrongness may be
predicated of both tokens and types. Allie may say that a certain killing was
wrong; Barbara may say that killing is wrong (period). When faced with a
formula of the form
(2.27) S ought at Tto do A at T (in W),
then, we must ask this question: how is "do A" to be understood, as a
variable ranging over act-types, or as a variable ranging over act-tokens?62
The short answer is: it doesn't really matter. The analyses (I)—(XII)

60 See Zimmerman (1988a), Section 4.10, for a fuller discussion of this issue.
61 In some very important, but perhaps in some ways minimal, sense of "freely." Richer
senses can be obtained by adding certain conditions, such as lack of coercion (see
Zimmerman (1988a), pp. 24—6, including n. 22, on the distinction between "strict
freedom" and "broad freedom"), or competence (crucial to certain conceptions of
autonomy), and so on.
62 It was said earlier that in fact not just actions but other events too may be obligatory. The
more general question, then, is whether we're concerned with event-types or event-
tokens. But I shall continue to talk just of actions, for simplicity's sake. What I say can be
generalized to apply to all events.
53
provided earlier in this chapter apply for the most part equally well to
actions conceived as types or as tokens. Consider (2.27). On one reading,
this can be reexpressed in terms of a four-term relation (actually five-term,
if worlds are included) as follows:
(2.28) doing A at T" is obligatory for S at T.
Here "doing A" may be understood to be a variable ranging over act-types.
On another reading, however, (2.27) can be reexpressed in terms of a two-
term relation (actually three-term, if worlds are included) as follows:
(2.29) 5 doing A at T' is obligatory at T.
Here "S doing A at T" " may be understood to be a variable ranging over
act-tokens. As substitution instances for each of (2.28) and (2.29), respec-
tively, consider these illustrations:
(2.30) helping that old lady on Monday is obligatory for Jones on Sunday;
(2.31) Jones's helping that old lady on Monday is obligatory on Sunday.
As far as I can tell, the corresponding substitution-instance for (2.27),
namely,
(2.32) Jones ought on Sunday to help that old lady on Monday,
is neutral with respect to the ontological presuppositions of (2.30) and
(2.31).63
It may seem, however, that one should not remain neutral on this issue.
When we're concerned about what we ought to do (when Jones asks:
"Should I help that old lady or shouldn't I?"), aren't we concerned about
whether a particular action is obligatory? After all, Jones isn't inquiring into
the morality of helping old ladies in general, but simply into the morality
of this particular helping. And doesn't this indicate that what he's really ask-
ing is whether or not an act-token is obligatory, rather than an act-type?
The question is confused. Of course, we are frequently concerned about
what to do here and now (or then and there) rather than in general, but this
doesn't favor one reading over the other. The particularity that is built into
the act-token o£jones's helping that old lady on Monday is attached external-
ly to the act-type of helping that old lady by means of the relata of Jones and
Monday. (2.31) enjoys no advantage over (2.30) on this score.

63 In fact, (2.32) is compatible with yet another reading, namely: "Jones helping that old lady
on Monday is obligatory on Sunday' where "Jones helping that old lady" refers to a cer-
tain state ofaffairs. There are a number of reasons for thinking it preferable to conceive of
actions as states of affairs, rather than as types or tokens. I provide such a treatment of
action in Zimmerman (1984). There is no need to enter this territory here, though.

54
 In fact, there is good reason to think that our inquiries into what we
ought to do are frequently concerned with act-types. Whether or not this
is so depends on how act-tokens are to be individuated. There are two
questions to be asked here. First, how are act-tokens to be individuated
across worlds? Second, how are they to be individuated within worlds? I
shall consider each in turn.
Suppose that Jones ought (on Sunday) to help that old lady on Monday
and that he does what he ought. Thus, we may assume, a certain act-token,
X, to which one might refer by the description "Jones's helping that old
lady on Monday," occurs in the actual world. Suppose that in fact what
Jones does is to help the old lady at noon on Monday. It would have been
equally permissible for him to help her at six o'clock in the evening. If he
had done this instead, would the act-token involved have been X, or would
it have been some distinct token, Y ? I'm not sure what we should say here,
but if Xis distinct from Y (which is what I suspect most people would say),
and if Y would have been equally permissible, then X wasn't obligatory.
Thus "Jones ought to help that old lady on Monday" would not be prop-
erly understood in terms of a certain act-token's being obligatory. Rather
it would be best understood in terms of there being an obligation with
respect to the disjunction X or Y or Z... (for an indefinite number of
tokens) — in terms, that is, of analysis (X) above rather than analysis (I).64
This wouldn't mean that (I) is strictly inapplicable to tokens, but it would
mean that, when we say "5 ought at T to do A at T'," we frequently, and
perhaps always, mean something else, namely, that S ought at T to do A at
T" or B at T"", or... (for an indefinite number of tokens and times, where
all of the tokens are of some common type and all of the times are contained
in T'). It seems to me unlikely that we should systematically misexpress
ourselves in this way, and thus likely that, when we say: "S ought at Tto
do A at T"," we do not take "do ^4" to refer to tokens - at least, not if we
take Xto be distinct from Yin the case just given.
Of course, one way to try to avoid the foregoing problem is to insist that
Xis identical with Y after all. But such coarse individuation of tokens across
worlds gives rise to a further problem, at least when coupled with coarse
individuation within worlds (of the sort advocated by Elizabeth Anscombe,
Donald Davidson, and others).65 Suppose, as before, that X occurred; that
is, Jones helped the old lady (at noon on Monday). Suppose further that, in
helping her, Jones startled her, and that this is not something he should have
64 Cf. Mellema (1991), p. 31, and Vallentyne (1989), p. 308.
65 See Anscombe (1969) and Davidson (1980). Contrast, among others, Alvin I. Goldman
(1976) and Kim (1976).

55
done. Call Jones's startling her Z. Those who advocate coarse individua-
tion within worlds claim that Xis identical with Z. But then how could X
be obligatory when Z was not? One response would be to say that X wasn't
obligatory; if Jones had helped the old lady without startling her, as he
should have done, then it wouldn't have been Xbut some other token that
would have occurred. But this response is not consistent with coarse indi-
viduation across worlds. Another response is to say that X (i.e., Z) was
obligatory under the description "helping the old lady" but not under the
description "startling the old lady." Such a response would echo what
coarse individuators have said before in other contexts (such as the context
of intentional action: some actions are claimed to be intentional under one
description but not under another),66 but the price of accepting this
response would be that of declaring such formulae as
(2.27) S ought at Tto do A at T' (in W)
crucially incomplete. Their completion would require mention of descrip-
tions. I doubt whether this accords with our common understanding of
obligation.
An additional problem with saying that it is act-tokens that are obliga-
tory is this. If a token exists only when it occurs, then what is said to be
obligatory need not, indeed need never, exist. It is not clear that it makes
sense to talk of the obligatoriness of something that does not exist.67
One reason for thinking that we nonetheless do commonly conceive of
obligation as concerning act-tokens is this. Many people want to say that
an act is obligatory only if its consequences satisfy a certain condition; but
types don't have consequences, only tokens do. I think this view is mis-
taken. Of course, we might say:
(2.33) Jones's helping that old lady at noon on Monday caused that old
lady's being pleased at six o'clock on Monday,
and this makes apparent reference to a causal relation between two event-
tokens. But I think we could equally well say:
(2.34) helping that old lady caused (relative to Jones, noon on Monday,
that old lady, and six o'clock on Monday) being pleased,
and here there is no apparent reference to tokens.
66 See Anscombe (1969), p. 11 and elsewhere; Davidson (1980), pp. 5, 46, and elsewhere.
67 Prichard appears to doubt that it makes sense to talk this way (see Prichard (1949), p. 37),
whereas Broad does not (see what he says about what he calls an "agibile" in Broad, 1985,
p. 226).

56
 This tactic might prompt the following response. The move from (2.33)
to (2.34) involves a rethinking of the relation of causation.68 It involves
"complicating" causation (by thinking of it as a six-term, rather than two-
term, relation) in order to keep obligation "simple." This is no less objec-
tionable than "complicating" obligation (by introducing descriptions) and
thereby keeping causation "simple." I suspect that this response is unsatis-
factory, but I shall not pursue the issue here. For, to repeat, the account of
obligation that has been given is intended for the most part to be adequate
both to a conception of actions as types and to a conception of actions as
tokens.
Why only "for the most part"? One reason has already been given: those
who individuate tokens coarsely both across and within worlds and who
claim that it is tokens that are obligatory must supplement the formulae
given with mention of descriptions (or employ some similar device).
Another reason is this: statements (2.11)—(2.14) cannot be understood (as
they stand) to concern act-tokens; for, as mentioned in Subsection 2.1.4,
they presuppose the repeatability of actions, and tokens are not repeatable.
Consider (2.11) again, which was supposed to capture one sense of what it
is to have a perfect duty:
(2.11) for all times T (at which S exists) not earlier than Tu S ought at Tt
to do A at T.
This presupposes that S can do A on more than one occasion, which is per-
fectly intelligible where A refers to an act-type but not where it refers to an
act-token. Still, even here the proponent of understanding obligation in
terms of tokens need not concede defeat; he need only insist on revising
such formulae as (2.11). Perhaps this would do:
(2.35) for all times T (at which S exists) not earlier than Tu S ought at Tt
to do either A at T or B at T or... (for an indefinite number of
tokens, all of which are of some common type).
Similarly, an "absolute" injunction against performance of a certain act-
type might be recast in terms of all tokens of that type being wrong.

2.2.6 Indefiniteness
Indefiniteness can infect obligation in a number of different ways. Here I
shall mention four.
68 I advocated a similar rethinking in Zimmerman (1984), Section 3.1, where, as remarked
in note 63 to this chapter, actions are treated not as tokens or types but as states of affairs.

57
 First, it might be thought that certain deontically relevant values are
incommensurate with one another.69 For example, it might be thought that
liberty and equality are both morally valuable, and that for that reason any
world that contains one or the other of them is, to that extent, deontically
good, but that they are nonetheless incommensurate, in that it is impossi-
ble to say that one is more or less valuable than the other or that they are
equally valuable.70 Of course, this rough statement might be taken to mean
a number of different things. It might be understood only to say that we are
never in a position to tell which is more valuable, liberty or equality. Such
epistemological difficulty need not point to any incommensurability in
principle. Or the statement might be understood only to say that there is
no general rule as to which is more valuable, liberty or equality, since the
former is sometimes more valuable, sometimes less valuable, than the other.
This too would not point to any incommensurability in principle; on the
contrary, it presupposes commensurability. Or the statement might be
understood to mean that on some but not all occasions liberty is neither
more nor less valuable than, nor equal in value to, equality. This would be
an instance of what Walter Sinnott-Armstrong calls "limited ^compara-
bility."71 Or the statement might be understood to mean that on all occa-
sions liberty is neither more nor less valuable than, nor equal in value to,
equality. This would be an instance of what Sinnott-Armstrong calls
"extreme incomparability."72
I suspect that certain morally relevant values may indeed be incom-
mensurate and that, as a result, it can happen that one is faced with acces-
sible worlds that are deontically incomparable, that is, worlds of which it
must be said that one is, deontically, neither better nor worse than, nor
equal in value to, the other.73 What does the present account imply in such
cases? That depends on just where the deontic incomparability applies.
Remember that, according to (I), S ought to do A if and only if S can do
A, S can refrain from doing A, and any accessible world in which S does
not do A is deontically inferior to some accessible world in which 5 does
69 This is a distinct matter from that mentioned in Section 1.1, where the (incommensu-
rability of moral with nonmoral requirements was at issue.
70 Cf. Berlin (1979), p. 151.
71 Sinnott-Armstrong (1988), p. 62ff.
72 Sinnott-Armstrong (1988), p. 59ff.
73 This is not to say that the incommensurateness of deontically relevant values must result
in the deontic incomparability of worlds that contain these values. After all, even if
liberty and equality are incommensurate, it may be that one world contains more of each
than another, so that the former world can be compared (favorably) with the latter.
Cf. Seung and Bonevac (1992), pp. 800-1.

58
A. (I) does not require that there be some particular A-world that is deon-
tically superior to every ~^4-world. Rather, it requires that, for every ~A-
world, there be some y4-world that is deontically superior to it. This allows
for deontic incomparability between worlds. For instance, it may be that S
is faced with two deontically incomparable sets of worlds, but as long as,
within each set, every ~^4-world is deontically inferior to some ^4-world,
5 ought to do A. (This even allows for deontic incomparability within sets
between A-worlds and yl-worlds and between ~y4-worlds and ~v4-worlds.)
But if the deontic incomparability at issue is between all A-worlds and some
~^4-world and all ~v4-worlds and some ^4-world, then clearly (I) and (III)
imply that it is neither the case that S ought to do A nor the case that 5
ought not to do A. What's left? Obviously this: 5 may do A and S may not
do A (in the sense of (IV)); that is, A is morally optional.74 In other, rougher
words: if you have a choice to make between two deontically incompara-
ble actions, you may take your pick; neither is obligatory, each is permis-
sible. This is what my account implies, and it seems to me correct.
The second type of indefmiteness that I want to acknowledge is that of
vagueness. It may be that certain moral norms are vague, in that there are
borderline cases where it is not clear whether or not they apply or whether
or not they have been satisfied. Suppose, for instance, that Gertrude ought
to be grateful to Lucy but that just what constitutes an adequate display of
gratitude is unclear. Perhaps it is clear that simply sending her a thank-you
note would be to do too little, while giving her a brand-new BMW would
be to go far beyond what gratitude alone demands. But just where the min-
imum and maximum fall may not be clear. Would sending her a bunch of
daisies suffice, or would it have to be lilies? It may be tempting to say that
the difficulty here is merely epistemological, that gratitude's boundaries are
well defined in every case, even if we sometimes cannot tell what they are.
But this won't do, I think. For there certainly appear to be cases where
what's required of someone is in fact indeterminate, not just hard for us to
determine. One such case is this: I promise Gail that I will help her with
her gardening around noon tomorrow. Suppose that this gives rise to an
overall moral obligation. Since the content of the promise is irremediably
vague, it seems that the content of the obligation is also. Clearly, if I turn
up by noon and help Gail as much as I can until all the gardening is done,
I will have done what I ought (and perhaps even more than I was required
to do). But what if I don't turn up until 12:15, or 12:30, or 1:00, or 2:00.
74 Here the phrase "no greater than" in (II) and (IV) is important; "at least as great as" would
not do, since it implies comparability. Cf. Vallentyne (1989), p. 304; McConnell
(1993b), p. 250; and Pietroski (1993), p. 508.

59
and what if I don't help Gail as much as I can but merely lend a hand, or a
finger, every once in a while? Will I have failed to do what I ought? Where
is the line to be drawn?
I think that we must accept that there is no precise point at which to
draw the line, and hence that obligations can be vague in the way indicat-
ed. My account is not intended to imply otherwise. It is intended, rather,
to be compatible with the claim that there are borderline cases with respect
to whether or not a certain action, A, is featured in the best accessible
worlds. Obviously this is a somewhat disturbing thought, but it is not a
problem for my account in particular. Vagueness is disturbing generally;
how to account for it satisfactorily is a general problem. Whatever the prop-
er account of vagueness, I see no reason to think that it would not be con-
sistent with what I have to say here about obligation.
The third type of indefiniteness concerns the indeterminacy of options,
in the following sense. I have said, in (2.18), that S can do A only if a cer-
tain counterfactual is true, one that concerns what would be the case if S
were to make a certain decision. But can't it happen that such a counter-
factual lacks a determinate truth-value? Perhaps it can.76 If it can, then in
such cases the statement that S can do A also lacks a determinate truth-value
and so, too, each of the statements that S ought to do A, S may do A, S
ought not to do A, and S may not do A lacks a determinate truth-value.
I think that we may well have to accept that obligation can be indeter-
minate in this way. Again, my account is not intended to imply otherwise.
But again, if the thought is disturbing, it doesn't point to a special problem
with my account. Rather, it is an issue concerning counterfactuals in gen-
eral. Of course, one could try to avoid the problem in the present context
by denying that obligation is tied in this way to counterfactuals, but I think
that this would be incorrect.77
The final type of indefiniteness concerns the possibility (which I shall
accept for argument's sake)78 that an agent be faced with an infinite num-
ber of accessible worlds and that each world is such that there is a deonti-
cally better one. The account that I have given is adequate to this possibility.
I have said that S ought to do A as long as (roughly) every accessible ~A-
world is deontically inferior to some ^4-world. This is quite compatible
with there being no deontically best world; it just means that, if S ought to

75 For a limited discussion of vagueness and obligation, see Sorensen (1991).

76 Cf. Lewis (1973), pp. 66-7, 91-4; Mendola (1987), p. 127ff.
77 Feldman tries to do just this, in Feldman (1986), p. 17, by denying that accessibility is to
be understood in terms of counterfactuals.
78 I am in fact quite skeptical of this possibility, however. See Zimmerman (1990a).

60
do A, then, at a certain point, all worlds that are higher in the ranking must
be A -worlds. Here, by the way, is where the final subclauses of (I) and (III)
are required. It is conceivable, I suppose, that 5 be faced with an infinite
number of accessible worlds so ordered that, although every ~y4-world is
worse than some ^4-world, it is also true that every ^4-world is worse than
some ~A-world. Without their respective subclauses (c4), (I) and (III)
would imply, in such a case, both that 5 ought to do A and that S ought
not to do A. This is unacceptable. With their respective subclauses (c4), (I)
and (III) imply that it is not the case that 5 ought to do A and that it is not
the case that S ought not to do A.79 Notice that (II) and (IV) imply, even
in the absence of any such subclause, both that S may and that S may not
(in the sense that S is permitted not to) do A. This seems to me precisely
what should be said in such a case.
Feldman calls the sort of situation where 5 is faced with an infinite num-
ber of increasingly good accessible worlds "delightful."80 But he overlooks
an implication that is somewhat disturbing, the implication that, no matter
what S does, he cannot avoid doing something wrong.81 For if an accessi-
ble world, Wu is distinct from another, W2i it must be because S acts dif-
ferently in W1 than in W2;S2 and if W1 is deontically inferior to W2, then S
must act wrongly in Wt.
One way to try to avoid this result would be to amend the account of
wrongdoing provided above. As (III) now stands, it implies that it is possi-
ble that S ought not to do A (that is, it would be wrong for 5 to do A), even
though there is no other act B that S ought to do instead; for it may be that
5 ought not to do B, either. Indeed, this is precisely the case in the sort of
situation presently envisaged. Such an implication may well seem odd, and
it could be obviated by adding to (III) the condition that there is an acces-

79 Feldman's account of obligation contains no clause corresponding to subclause (c4) of

(I) and is nonetheless able to accommodate the possibility that S ought to do A even
though S is faced with an infinite number of increasingly good accessible worlds. (See
Feldman, 1986, pp. 37—8.) This is because his account says (roughly) that 5 ought to do
A just in case some ^4-world is better than every -~v4-world. In order to accommodate
deontic incomparability, however, my account (as noted above) says only that every ~A-
world is worse than some A -world. Such an account requires supplementation with sub-
clause (c4), given the possibility in question.
80 Feldman (1986), p. 37.
81 Cf. Slote (1989), p. 110.
82 This is because, on my view (presented in Subsection 2.2.1), worlds are accessed through
choices. Given that choosing is a (minimal) way of acting, if one accessible world differs
from another, it will be (at least) because the agent makes a choice in one (namely, the
choice by which he accesses that world) that he does not in the other.

61
sible world, or set of accessible worlds, deontically superior to all other
accessible worlds. Thus, we could say, one can do wrong only if there is a
best that one can do. To avoid the appearance of being ad hoc, this sug-
gestion could be coupled with the observation that the account already
implies a parallel thesis, namely, that one can be obligated only if there is a
less than best that one can do. (For, in the limiting case where all one's
accessible worlds are such that none is deontically superior to any other,
everything is permissible, nothing is obligatory; there is not even the dis-
junctive obligation to take one's pick, since — as (X) declares — such an
obligation implies that some choices would be wrong.)
But I am disinclined to adopt this amendment, for three reasons. First,
it destroys the symmetry of the treatment of O(A) in (I) with that of O(~A)
in (III), a symmetry that seems natural to me. Second, there is a pair of par-
allel theses to which appeal can be made other than the pair just mentioned
in the last paragraph. These are the theses that one can do wrong only if one
can do better, and that one can be obligated only if one can do worse. These
two theses are implied by my account as it already stands, and it is arguable
that whatever intuitive plausibility the first pair of theses has is itself to be
derived from the plausibility of this second pair. And third, the suggestion
of course has the implication that, in the sort of case under discussion, one
can do no wrong. This seems unacceptable to me, for it gives one license
to actualize some deontically abysmal accessible world, even though there
is a far (indeed, infinitely) superior accessible world that one could choose
to actualize.
I think, then, that it is best to accept that, in the very special sort of sit-
uation envisaged, it is indeed the case that S cannot avoid doing something
wrong. Thus the present account (under the assumption that an agent may
be faced with an infinite number of accessible worlds, each of which is
deontically inferior to some other) does not rule out moral dilemmas of a
certain sort. Notice, however, that the sort of dilemma at issue may be called
(following Peter Vallentyne) a. prohibition dilemma: all of S's options involve
wrongdoing. This is not the same as saying that obligation dilemmas are pos-
sible, that is, that it can happen that 5 ought to do some act A and ought to
do some act B but cannot do both.83 The present account implies that such

83 See Vallentyne (1987a), p. 113, and (1989), pp. 302 and 305. I am here extending
Vallentyne's own use of "prohibition dilemma" somewhat. According to him, a prohi-
bition dilemma has the form ~P(A) & ~P(~A). The account given in (II) and (IV) rules
out the possibility that ~P{A) & ~P(~A) be true. But, as just noted, it does not rule out
the possibility that all of an agent's options involve wrongdoing, and the term "prohibi-
tion dilemma" seems to me aptly applied to such cases.

62
dilemmas are impossible (barring certain complications). This is an issue
that I shall address more fully in Chapter 7.

2.3 IMPLICATIONS OF THE ANALYSIS

2.3.1 Certain deontic implications
Anyone who has some familiarity with the work of deontic logicians will
be aware of the vast number of formulae which have been touted as either
axioms or theorems of one or another system of deontic logic. These for-
mulae are usually alleged to express certain truths about obligation.
Consider the following (which have been selected both because they are
representative of the sorts of formulae in question and because it will be
especially useful to comment on them):

(2.36) Op D Pp;
(2.37) Op = ~P ~p;
(2.38) OpZ>~O~p;
(2.39) O(pv~p);
(2.40) O(p&q) = (Op & Oq);
(2.41) P(p v q) = (Ppv Pq);
.
(2.42) O(p Dq)D(Op DOq);
(2.43) (pD q) D (Op D Oq);
(2.44) Op D OOp;
(2.45) Pp D OPp.

Each of these formulae seems to suggest a proposition that looks reason-

able, but, if you're at all like me, you'll be wondering just what proposition
is supposed to be expressed and just how reasonable it is.
I think it's fair to say that in many cases the efforts of deontic logicians
have been expended prematurely. In many cases they have attempted to
construct formal systems on the basis of considerations too weak to bear the
weight imposed upon them. More work needs to be done on the basic con-
cept of obligation before such systems can provide genuine philosophical
insight. Nevertheless, such systems are certainly not without their philo-
sophic uses. If formulae such as those just presented are at all plausible, then
they can serve as a gauge of the adequacy of an account of obligation such
as the one that I have proposed. If the account departs from the formulae,
it ought to be able to explain why.
First, though, the formulae must be interpreted, and this is no easy task.
Standardly, the O and P are intended to express obligatoriness and permis-
63
sibility, respectively, but what are the variables p and q supposed to range
over? Presumably propositions. This is so for two reasons. First, the con-
nectives ~, D, =, &, and v have a familiar interpretation when applied to
propositions and propositional variables, but not otherwise. Second, and
less importantly, certain parallels with alethic modal logic (the logic of
necessity and possibility, taken by many to be analogous in various ways to
obligatoriness and permissibility, respectively) may then be drawn. The
trouble is that, although such treatment of the variables lends itself nicely
to the expression of such claims as that it ought to be that little children not
have to suffer, it is hard to see how it can adequately handle such claims as
that Tim ought to do what he's been told, given that "It ought to be that
Tim do what he's been told" does not express the same claim. (See Section
1.2 on binding versus nonbinding senses of "ought.") One way around this
would be to read Op as "5 ought to see to the occurrence ofp," but this
has difficulties of its own. Suppose that Roger ought to raise his hand. Just
what event or proposition is it that Roger ought to see to the occurrence
of, his raising his hand or his hand's rising? In neither case is it clear that we
get an equivalent claim; in one case too much seems to be said, in the other
too little. To say that Roger ought to see to it that he raises his hand seems
to say too much, in that it suggests that he ought to do something over and
above just raising his hand. To say that Roger ought to see to it that his
hand rises seems to say too little, in that it appears compatible with his ful-
filling his obligation by way of ensuring that his hand rises in such a way
that it is not he but someone or something else that raises it.
In light of these considerations, I think it best to rewrite the formulae
under consideration so that they conform with the account of obligation
that I have proposed. Thus first of all we should consider this claim:

(2.36') if O(A), then P(A); that is,

if S ought [at T] to do A [at T in W], then S may [at T] do A [at
T i n W].
It is clear that my account - specifically, (I) and (II) above - implies that this
is true. If all ~vl-worlds [accessible to 5 from Wat T] are worse than (that
is, deontically inferior, for 5 at T, to) some ^4-world [accessible to S from
Wat T], then certainly the former are no better than the latter.
The next claim to be considered is this:

(2.37') O(A) iff ~P(~A).

84 Feldman employs this reading in Feldman (1986), p. 37.

64
This breaks down into two further claims, the first of which is this:
(2.37'a) ifO(,4),then~P(~,4).
It is clear that my account also implies this to be true. If all ~A-worlds are
worse than some /1-world, then it is not the case that all v4-worlds are no
better than some ~v4-world. The other claim at issue here is this:
(2.37'b) ]f~P(~A), then O(A).
This turns out not to be true on my account, for ~P(~^4) may be true due
to a lack of personal optionality, that is, true in virtue of there being either
no /1-world accessible to S or no ~A-world accessible to S (that is, in virtue
of S's not both being able to do A and being able to refrain from doing A),
in which case O(A) will be false. For example, no one can jump over the
moon; hence no one may refrain from doing so; but it doesn't follow that
everyone ought to do so. Still, the following is true on my account (where
C expresses the "can" of personal possibility):
(2.37'c) i£C(A)&C(~A) &~P(~A), then O(A).
If there are both an ^4-world and a ~v4-world, and if it is not the case that
all ^4-worlds are no better than some ~/l-world, then all ~y4-worlds are
worse than some /1-world. Furthermore, in light of the facts that obliga-
tion requires personal optionality and that, where there is personal option-
ality, acting in a manner that is not permissible constitutes acting wrongly,
we may indeed accept the following:
(2.37") O(A) iff W(~A).
Next is this claim:

(2.38') if O(A), then ~O(~,4).

My account implies that this is true. (Indeed, it follows from (2.36') and
(2.37'a).) If 5 ought [at T] to do A [at T' in W], then it is not the case that
S ought [at T] not to do A [at T' in W]. This is because, if all ~y4-worlds
are worse than some ^4-world (than which there is no better ~^4-world),
then it is not the case that all yl-worlds are worse than some ~^4-world. Of
course, (2.38') is controversial (as are, perhaps, some of the other claims
already considered). I shall discuss it and related claims in detail in Chapter 7.
It is not clear just how the next formula, (2.39), is to be rewritten so as
to conform with the present account. It suggests the following:
(2.39') S ought to do either A or -A,
65
but just what this is supposed to mean is not clear. In accordance with (X),
we might try interpreting it as follows:
(2.39") 0(Aor~A).
This is false, on my account, since it implies that, for any act A, S may do
A and S may not do (that is, may refrain from doing) A. But I think it is
plain that proponents of (2.39) have had something else in mind. Just as
p v q is standardly taken to be equivalent to ~(~p&~q), so too, I think, we
can understand (2.39) to be rewritable as follows:
(2.39'") O(,4v~,4);thatis,
O~(~A&A).
Even if this does capture what is intended, though, it turns out to be false
on my account. This is because its truth would imply the truth of
C(~A&A), and this is true of no agent.
The next claim is this:
(2.40') O(A&B) iff (O(A) & O(B)),
which breaks down into these two:
(2.40'a) if O(A&B), then O(A) &O(B);
(2.40'b) if O(A) & O(B), then O(A&B).
Thefirstof these is not true on my account, for it may happen that, although
(A&B) is personally optional, A (or B) is not; if so, then although O(A&B)
may be true, O(A) (or O(B)) won't be.
To this observation there are a number of possible responses. First, one
might claim that every genuine action is indeed personally optional; it can-
not be, for instance, that S acts in moving his arm unless he could have done
otherwise. (Certainly, it can happen that one's arm moves irresistibly, but
then it is not oneself who is moving it.) Thus, where O(A&B) is true, each
of O(A) and O(B) will be too. But this is problematic, for two reasons. First,
it seems false that action, to be genuine action, must be optional, for three
reasons: hard determinism may rule out the possibility of free action, but it
doesn't rule out the possibility of all action whatsoever;85 certain cases of
actual action are plausibly understood to involve irresistible impulses, cases
such as, perhaps, certain compulsive or obsessive actions, actions due to
addiction or hypnosis, and sleepwalking; and Frankfurt-type cases, where

85 Unger (1977) notwithstanding, where the argument founders on an equivocation on

"cause."

66
one does something "on one's own" but where one would have been
forced to do it anyway, seem possible.86 Secondly, and more importantly,
it must be remembered that, in the formula O(A) and its ilk, the variables
are not restricted to ranging over actions. (See the discussion in Subsection
2.2.1 above.) And certainly some sorts of events that can be optional for an
agent don't have to be. For instance, O(A) might on occasion express "it is
obligatory for me that I be indoors," and yet there can clearly be times when
my being indoors is not personally optional for me (either because I can-
not get indoors or because I cannot get outdoors).
A second response might be to acknowledge the second point just made
and yet to insist that, if O(A&B) is true, then each of O(A) and O(B) must
be too, whether or not what is obligatory is said to be an action. Since my
account doesn't imply this, it ought to be revised. Specifically, (V) should
be revised so that it implies, not just that (A&B) is optional, but that each
of A and B is too. But this response is problematic, too. First, it seems that
a revision would be called for not just for (V) (the analysis of O(A&B)) but
also for (I) (the analysis of O(A)). Consider, for instance, the claim that
Chuck ought to open a checking account. (I) implies that opening a check-
ing account is optional for Chuck. But, on the present suggestion, it ought
to say more. For opening a checking account is surely a complex action.
Suppose that it can be analyzed in terms of actions X, Y, and Z, so that we
may now say that Chuck ought to do (X&Y&Z). Then (V), revised accord-
ing to the present suggestion, would imply that each of X and Y and Z is
optional for Chuck, whereas (I), left unrevised, does not have this implica-
tion; (I) implies only that (X&Y&Z) is optional for Chuck. Thus (I) would
need to be revised to say that each component part of A is optional for S;
but a precise expression of this would surely be very complicated and dif-
ficult to accomplish. A second problem with this second response is simply
this: it's just not clear that, whenever O(A&B) is true, so too is each of O(A)
and (OB). Consider, for example, the claim "You ought not to stand idly
by while there is poverty." Why wouldn't it be correct to analyze this in
the terms set out in (V), namely: (a) there is a (not-standing-idly-by-and-
poverty)-world accessible to you; (b) there is a not-(not-standing-idly-by-
and-poverty)-world accessible to you; and (c) all worlds of the latter sort
are worse than some world of the former sort? Surely this can be true, even
though every world accessible to you is a poverty-world; that is, nothing
you can do will eradicate poverty. Similarly, someone might claim,
"Whatever you do, you ought to do it without complaining." Some of the

86 See Frankfurt (1969).

67
things one does may not be avoidable, although whether or not one com-
plains while doing them is. This is enough to give the claim some plausi-
bility. Whether or not the claim turns out to be true, we should not dismiss
it simply on the grounds that it implies statements of the form O(A&B)
where A is conceded not to be optional.
Thus (2.40'a) is indeed to be rejected. But a third response is to note that
three very closely related claims are nonetheless true on my account. The
first two are these:
(2.40'c) i£C(~A) & O(A&B), then O(A);
(2.40'd) if C(~B) & O(A&B), then O(B).
These are true because, if all ~ (A &B) -worlds are worse than some (A&B)-
world, then (since any ~^4-world is a ~(A&B)-world and any (v4&B)-world
is an ^4-world) all ~^4-worlds are worse than some yl-world; similarly, all
~B-worlds are worse than some B-world.87 The third implication is this:
(2.40'e) i£O(A&B), then either O(A) or O(B).
This is true, because it cannot happen that (A&B) is optional without at
least one of A and B being optional.
Let us now turn to the second "half" of (2.40'), namely:
(2.40'b) if O(A) & O(B), then O(A&B).
This claim, sometimes called the "principle of agglomeration" (and to be
discussed further in Chapters 6 and 7), is true on my account. Suppose that
all ~^4-worlds are worse than some /1-world. Such an A-world must be
either an (A&B)-wov\d or an (A&~B) -world. But if we suppose also that
all ~B-worlds are worse than some B-world, then all ~y4-worlds are worse
than some (A&B)-world. Similarly, all ~B-worlds are worse than some
(A&B)-world. Now, a ~(A&B)-wov\d is either a ~^4-world or a ~B-world.
Hence all ~ (A &B)-worlds are worse than some (A &B) -world.
The next claim is presumably to be rewritten as follows (interpreting the
v as before):
(2.41') P(A v B) iff (P(A) v P(B)); that is,
P~ (-A&-B) iff (P(A) v P(B)).
This breaks down into two claims. The first is this:
(2.41'a) if P~(~A&~B), then P(A) v P(B).
87 It was thus on the implicit understanding of the personal optionality for Sam of each of
the actions of sticking out his tongue and saying "Aah" that I claimed at the outset that
(2.1) implies (2.2).

68
My account implies that this is true. If P~ (~A&~B) is true, then all
(~A&~B)-worlds are no better than some ~(~A&~B)-world; but if nei-
ther P(A) nor P(B) is true, then the best worlds will all be (-A&-B)-
worlds. The second claim is this:
(2.41'b) if P(A) v P(B), then P~ (-A&-B).
My account implies that this is false. Suppose that P(A) is true, and hence
that P(A) v P(B) is true, but suppose that P(B) is false because C(~B) is false
(that is, because S cannot refrain from doing B). If C(~B) is false, then so
too is C(~A&~B); but from this it follows that P~ (~A&~B) is also false.
Inasmuch as p D q is equivalent to ~(p&~q), I think we may rewrite
(2.42) as follows:
(2.42') if O(A D B), then, if O(A), then O(B); that is,
if O~(A&~B), then, if O(A), then O(B).
My account implies that this is true. Suppose that all ~vl-worlds are worse
than some ^4-world. Either B is performed in the y4-world or it is not. If it
is not, then, given the truth of O~ (A&~B), there is a still better world in
which both A and B are performed, and all ~B-worlds will be worse than
this world.
It is not clear just how (2.43) should be rewritten. Perhaps we can put
it as follows:
(2.43') if~(,4&~B), then, if 0(^4), then O(B),
where we understand this to express the following: if S doesn't do A with-
out doing B, then, if he ought to do A, then he ought to do B. My account
implies that this is false. So what if S doesn't do A without doing B ? Perhaps
he ought to. That is, it may be that the best ^4-worlds are ~B-worlds; and,
if the best worlds are ^4-worlds, then O(B) will not be true. As an illustra-
tion, let A be the action of helping Hilda and B be the action of humiliat-
ing Hilda. Perhaps Laura ought to lend Hilda a hand and refrain from her
usual caustic comments; if she helps Hilda but also humiliates her, it hard-
ly follows that, in humiliating her, she has done what she ought.
A strengthened version of (2.43') may appear more plausible:
(2.43") if ~C(A&~B), then, i£O(A), then O(B).
That is: if S cannot do A without doing B (whether this is out of logical
necessity, physical necessity, or something weaker), then, if he ought to do
A, then he ought to do B. The foregoing example about humiliation is
inapplicable here. If S cannot do A without doing B and the best accessi-
69
ble worlds are A- worlds, then afortiori they are B- worlds as well. But in fact
(2.43") isn't quite right, either; for it may be that 5 cannot do A without
doing B because he simply cannot refrain from doing B, in which case O(B)
will not be true. So B must be said to be optional, and what is needed is this:
(2.43'") if ~C(A&~B) & C(~B), then, if O(A), then O(B).SS
This is implied by my account and will be featured in discussion later in this
chapter and in Chapters 6 and 7.
The final two formulae are difficult to rewrite to conform with the pre-
sent account.89 Perhaps the most suitable renditions are these:
(2.44') if S ought to do A, then S ought so to act that S ought to do A;
(2.45') if 5 may do A, then S ought so to act that S may do A.
If these are to make sense, I think that it must be recognized that the time
of S's acting so that S ought to or may do A is prior to the time of S's actu-
ally doing A. Thus these propositions apply at best when there is an implic-
it clause in their consequents to this effect: if S has time so to act. On this
understanding, (2.44') is true, on my account, but (2.45') is not. As to the
former: if all ~/l-worlds are worse than some ^4-world and then S so acts
that this is no longer true, then S clearly departs from the high road and
goes wrong. Consider Figure 2.1 again, now under the assumption that W\
is deontically superior to all the other worlds. Given this assumption, S
ought at T1 to do C at T3, and he ought at T} so to act that he ought at T2
to do C at T3. If S does ~A at Tu or if he does ~B at T2, then it will no
longer be true that he ought (prior to T3) to do C at T3. But this will be
because he went wrong at either Tt or T2.1 shall address this issue in greater
detail in the next chapter. As to (2.45'): this is not true, even given the
implicit extra clause, and that is simply because it is permissible to make cer-
tain preparations to do what is itself permissible, even if these preparations
rule out doing something else that is permissible. For instance, perhaps I
may either go to the theater or go to the opera this evening, although I can-
not do both. If, as I may, I spend my money on tickets to the theater, then
I have rendered myself unable to go to the opera; hence it is not the case
any longer that I may now go to the opera; hence it is not the case that I
ought so to have acted that I may now go to the opera.
It is interesting to note that the truth of (2.44') implies that there is a
sense in which we have an obligation to be morally perfect. This sense may
88 It was thus on the implicit understanding of the optionality of B that I claimed at the out-
set of this chapter that we may assume the truth of (2.10).
89 Problems with the interpretation of such formulae are explored in Marcus (1966).

70
be called that of "perfection in action," in that (in this sense) an agent is
morally perfect if and only if he never acts wrongly.90 Note that to state that
we have an obligation to be perfect in this sense is not to set an impossibly
(or even unreasonably) high standard. It does not imply that we have an
obligation to be perfect in some other sense (for instance, in the sense that
our characters, or motives, or whatever, be beyond all moral criticism)
where it might be impossible for us to achieve such perfection (or unrea-
sonable to claim that we should do so). Nor does it imply an impossibly
high ideal for action. It can of course happen that an agent can perform each
of a number of very good deeds and yet cannot, for one reason or another,
perform all of them. Stating that this agent, like all agents, has an obligation
to be perfect (in the present sense) does not commit me to stating that he is
obligated to perform all of these very good deeds.91 On the contrary, since
he cannot perform them all, my account commits me to saying that it is not
the case that he ought to perform them all. What he ought to do, of course,
is to avoid wrongdoing. This might well involve his taking his pick among
the various good deeds in question, performing some and not performing
others. As long as he does avoid wrongdoing, he will have satisfied the
obligation to be perfect in the present sense.
There are of course many other formulae that we could examine here,
but I don't think that there is any need to do so. The ten that have been
discussed are representative of how my account handles such matters. I have
explained why some of the formulae are not implied by the account, and I
hope that the explanation serves to confirm the acceptability of the account.
Some of the formulae that are implied by the account are admittedly con-
troversial; as noted, these will be discussed in later chapters. In addition,
there are of course implications of my account that are not purely deontic.
Foremost among these is surely the following:
(2.46) if O(A), then C(A); that is,
if S ought [at T in W] to do A [at T'], then S can [at T in W] do
A [at r ' ] .
This controversial claim will be examined in Chapter 3. When conjoined
with other implications, (2.46) of course yields yet further implications. For
example, the principle of agglomeration — (2.40'b), above — conjoined with
(2.46) gives us this:
(2.47) if O(A) & O(B), then C(A&B),

90 See McGinn (1992), p. 33.

91 Contrast Clark (1993), pp. 54-5.

71
and this is true, on my account, even though (as just noted) the following
is of course to be rejected:
(2.48) if C(A) & C(B), then C(A&B).
But this, too, will be discussed in later chapters.

2.3.2 Certain deontic paradoxes

Formal systems of deontic logic have been plagued by certain so-called
paradoxes. These paradoxes are not formal inconsistencies; rather, they are
cases where the system's representations of certain "ought"-statements
seem separately plausible but jointly implausible, since when conjoined
they have, or appear to have, counterintuitive implications. Some such
paradoxes will be treated in Chapter 3 (when self-imposed impossibility is
discussed) and in Section 4.3 (when conditional obligation is discussed).
But two (or three, depending on how you count) may be discussed now.
The first paradox is known as Ross's Paradox.92 In standard form, it may
be put as follows. Given
(2.49)
and
(2.43)
we may infer
(2.50) OpDO(pvq).
It has been thought that the premises are true but the conclusion false, since
the latter seems to warrant saying (for instance) that, if Matt ought to mail
a letter, then he ought to either mail or burn it, and this seems to grant Matt
license to burn the letter. It might be tempting to try to undermine the para-
dox simply by pointing out that (2.43) is false (as argued above), but in fact
this won't do; for a strengthened version of (2.43) has been acknowledged
as true, and the argument can be reformulated around it. So let me refor-
mulate it in this way, now rewriting it so as to conform with the present
account. Accordingly, we may say that, given
(2.49') ~C(A&~(AvB))
and
(2.43"') if~C(,4&~B) & C(~B), then, if O(A), then O(B),

92 See Ross (1941), where the subject is imperative rather than declarative sentences.

72
we may infer
(2.50') if O(A) & C(~(A v £)), then O(A v B).
Now, (2.49') is necessarily true: one cannot do some act A and yet do nei-
ther A nor B (for any act B). And I have said that my account implies that
(2.43'") is true. Hence this account implies that (2.50') is true: if 5 ought
to do A and can refrain from doing either A or B, then 5 ought to do either
A or B. But isn't this to be rejected? As noted, it would appear to give
employees of the postal service license to burn rather than mail the letters
entrusted to them.
The resolution of this minor puzzle is of course to be found in the dis-
tinction between O(A v B) and 0(AorB). Briefly, it is only the latter that
gives S license to do B (as spelled out in (X) above), but it is only the for-
mer that is at issue in (2.50'). O(A v B) means simply that 5 ought not
to do neither A nor B, and clearly this is true if 5 ought to do A. End of
problem.93
Formula (2.43) gives rise, though, to another, much more interesting
paradox, commonly known as the Good Samaritan Paradox.94 Let p be the
proposition that Harry helps Vera, who has been robbed by Rita. This
entails q, the proposition that Vera has been robbed by Rita. Suppose that
Harry ought to help Vera. (2.43) would seem to imply that it follows that
Rita ought to have robbed Vera - a disturbing conclusion. Now, some sys-
tems of deontic logic do imply this conclusion and so seem properly criti-
cizable on this score. The present account, however, bars this conclusion,
for two reasons. First, (2.43"') is to be read as follows: if S cannot do A
without doing B and S can refrain from doing B, then, if S ought to do A,
then S ought to do B. Obviously, this thesis is restricted to inferring what
5 ought to do from what S ought to do; it will not in general permit infer-
ring an obligation on the part of Rita from an obligation on the part of
Harry. Second, if Harry ought now to help Vera, and Rita's robbery ofVera
lies in the past, then once again no inference with respect to an obligation
to rob is forthcoming; for "ought" does not apply to the past but only to
that which is not presently inevitable (as the insertion of C(~B) in (2.43'")
guarantees).
But although this response is accurate, it does not put the Good
Samaritan Paradox to rest. For the paradox can be recast so as to concern
only a single agent's obligations and only the future.95 Suppose, then, that
93 See Follesdal and Hilpinen (1970), pp. 21-3; Castafieda (1981), pp. 63-5.
94 See Aqvist (1967).
95 Cf. Castafieda (1981), pp. 52-3.

73
A is the action of Harry's helping the woman (Vera) whom he (Harry) will
rob at T3, and B is the action of Harry's robbing Vera. Suppose that Harry
ought at Tj to do A at T 2 . Given (2.43'"), it seems to follow that Harry
ought at Tt to do B at T3; for, it seems, Harry cannot at Tt do A at T 2 with-
out doing B at T 3 , and Harry can at Tt refrain from doing B at T3. Clearly
this is an unwelcome result.
The proper response is this. The claims that Harry ought to do A and
that Harry cannot do A without doing B are jointly unacceptable. Which
is to be rejected depends on just what A is taken to be. Perhaps this is best
explained by explicitly invoking worlds in the description of actions. If A
is the action of Harry's helping in world H^the woman whom he robs at
T3 in the actual world, then it is evident enough that Harry can do A without
doing B. For certainly there are worlds accessible to Harry at T1 in which
Harry helps Vera at T 2 without robbing Vera at T3, even though it is true
that Harry will in fact (that is, in the actual world) rob Vera at T3. If A is
the action of Harry's helping in world Wthe woman whom he robs at T3
in W, then it is evident that Harry cannot do A without doing B. But then
we should not say that Harry ought at Tt to do A at T2. For B does not
occur in the best worlds accessible to Harry, and if Harry cannot do A with-
out doing B, then A doesn't occur in the best accessible worlds either.
This last point may be a bit hard to grasp, and I think that this is in part
because the example lends itself so readily to an understanding of A and B
according to which doing A does not entail doing B. Let me turn, then, to
a version of the Good Samaritan Paradox where it is clear that doing a cer-
tain thing 4 does entail doing another thing B. This is the version commonly
known as the Paradox of the Knower. Suppose that Greg, a security guard,
ought to provide truthful reports of any wrongdoing done on his watch and
hence (to keep the discussion to a single-agent case) ought to provide a
truthful report of any wrongdoing committed by himself. (The paradox
was originally put in terms of knowing rather than truthful reporting, 96 but
the latter is more suitable to single-agent cases.) Suppose that Greg ought
not at Tj to do A at T2 but in fact will do A at T 2 . Is it the case that Greg
ought at Tj to provide at T3 a truthful report of his doing A at T2? Clearly
not, and this is because his doing so entails his doing A at T 2 . The best worlds
accessible to Greg at Tt are worlds in which he does not do A at T 2 and
hence does not provide at T3 a truthful report of his doing A at T2.97 This
is of course perfectly compatible with saying that, once he has done A

96 See Aqvist (1967), p. 366ff.

97 Cf. Feldman (1990), pp. 327-8.
74
at T2, Greg ought then to provide at T3 a truthful report of his doing so.
Such shifts in obligation will be the topic of Chapter 3.

2.3.3 Analysis and neutrality

I take it to be a point in my account's favor that it satisfactorily handles the
various deontic paradoxes. (Again, some of these paradoxes are to be dis-
cussed in later chapters.) On the other hand, certain implications of the
account have been acknowledged to be controversial. How, then, can I
propose the account as an analysis of the concept of obligation? Aren't
analyses supposed to be neutral between various conflicting conceptions of
a concept? Shouldn't, for example, an analysis of "ought" neither imply
that "ought" implies "can" nor imply that it doesn't?98
This is a tricky issue, which could be circumvented simply by declaring
that (I)-(XII) are intended to be understood as necessary truths and leaving
it at that, without declaring in addition that they are intended to be taken
as analyses. But this would be misleading, for I do intend them to be taken
as analyses.
It is a mistake to think that analyses must be neutral in the sense at issue.
Consider what is probably the most celebrated philosophical analysis, that
of the concept of knowledge as true belief that is (in some way) justified.
This is clearly not neutral in the sense at issue. It implies that knowledge
entails belief, that knowledge entails truth, and so on, and each of these
implications has been contested.
But consider what Feldman says. Although he conceives of his project
as one of "laying out the logical structure of the concept of moral obliga-
tion,"99 which matches a description that I have given of my project, he
denies that his account is intended as an analysis of this concept. He says:
I propose that we understand the principle that we ought to do the best we can
in the following way:
MO: MOs, t, p is true at w iff (3u/)[As, t, w' &p is true at w' & ~(3u>")
{As, t, w" &~p is true at w" & W(w") > IV(u/)}]

On this formulation,.. .s morally ought, as of a time, t, to see to the occurrence

of a state of affairs, p, iff p occurs in some world accessible to s at t, and it is not
the case that ~p occurs in any accessible world as [intrinsically] good as (or
better than) that one.100
98 For discussion of a closely related issue, see Sayre-McCord (1986).
99 Feldman (1986), pp. x-xi.
100 Feldman (1986), p. 38. Some typographical changes have been made.

75
This, of course, is structurally very similar to the account that I have pro-
posed. Later, though, Feldman considers this principle:
(2.51) (s)(t)(p)(s morally ought, as of t, to see to p if and only if MOs, t, p),
and he has this to say of it:
Some readers...may have expected that I would take (2.51) to be an empty
tautology. If my definition, MO, had been intended as an "analysis" of the con-
cept of moral obligation, then this expectation would perhaps be justified. But
the definition was not intended to be an analysis of some concept we already
have. It was intended to be a stipulative definition of a concept to which I
wanted to draw attention.
If we compare (2.51) with the following principle, I think the difference
between moral obligation and MO will emerge:
(2.52) (s)(t)(p)(MOs,t,p if and only if MOs,t,p)
If we grant (as I think we must) that MOs,t,p has a meaning, then we should
grant that (2.52) is an empty tautology. It tells us nothing of interest about
morality. Even those who think that my approach is hopelessly wronghead-
ed.. .will accept (2.52). On the other hand, (2.51) is intended to tell us how we
ought to behave. It tells us that we morally ought to bring about all and only
those states of affairs that fall under the concept MO... So even if (2.51) is a
necessary truth, it is not an empty tautology. It is a fundamental moral doc-
trine...
Since some will disbelieve the proposition expressed by (2.51) while not dis-
believing the one expressed by (2.52), these propositions are distinct...
This "open question" argument serves to emphasize the fact that my defin-
ition, MO, is not an analysis of the concept of moral obligation.101
I think these remarks are erroneous. I think we must grant that (2.51) and
(2.52) are distinct propositions, but we should not infer from this that (2.51)
is not an analysis of the concept of moral obligation. If we did, then we
would have to say that the following proposition is not an analysis of the
concept of knowledge:

(2.53) 5 knows thatp iff S has a justified, true belief that p ,

since it is distinct from this proposition:
(2.54) S has a justified, true belief that p iff S has a justified, true belief
that p .
101 Feldman (1986), pp. 207-8. The numbering of propositions has been changed to con-
form with the numbering that has been used in this chapter.

76
 Of course no one will reject either (2.52) or (2.54), while some will
reject either or both of (2.51) and (2.53). But analyses need not be uncon-
troversial; they need not be neutral with respect to various conceptions of
the concept that is analyzed. Perhaps some people102 believe that analyses
must be neutral and uncontroversial because they believe that analyses are
analytic and that analytic statements are trivial, and hence neutral and
uncontroversial. But this would be a mistake. If (2.53) is a correct analysis,
then the following statement is analytic:
(2.55) if S knows that]?, then S believes that p.
This statement derives its analyticity from (2.53). All analyticity is derived in
this way from analyses;103 thus analyses are not themselves analytic. Is (2.55)
uncontroversial? Hardly. It is "trivial," in the sense that, given (2.53) (or
something like it), belief is in a way conceptually "contained" in knowl-
edge. But since (2.53) is not itself uncontroversial, there is no reason to
expect (2.55) to be, even if it is analytic.
Similarly, if my proposed analysis of obligation is correct, then so too is
the following claim:
(2.46) if 5 ought to do A, then S can do A.
Indeed, (2.46) will be analytic; but its being so is no reason to expect it to
be uncontroversial or in any way neutral with respect to all possible con-
ceptions of obligation. In saying this, I am of course committed to saying
that anyone who denies (2.46) is making some sort of conceptual error; but
this is not to say that the denial of (2.46) involves a logical error, at least not
in the sense that it involves an inconsistency in and of itself, as would, for
example, the denial of this claim:
(2.56) if S can do A, then S can do A.
But it is conceptual truths, and not logical truths, that I am after.
Two huge questions loom here. I can only point to them; I cannot do
them justice. The first is this. If analyses are not themselves analytic but syn-
thetic, how are they to be distinguished from other synthetic necessary
equivalences that are not analyses? I wish I knew the answer to this ques-
tion, for it would mean that I could clearly distinguish conceptual errors
from errors of some other sort (and thereby, perhaps, clearly distinguish
metaethics from normative ethics). Consider, for instance, this claim:
(2.57) S ought to do A iff A best pleases S.

102 I am not saying that Feldman is among these people.

103 See Chisholm (1989), pp. 33-4.

77
Someone might propose this as a necessary truth. If so, I am sure that he
would be mistaken. I wouldn't accuse him of a conceptual error, though, but
rather of a substantive error. Perhaps it is easy to see why this is so; for (2.57)
in no way attempts to "break down" the concept of obligation into com-
ponent concepts, while this is what an analysis attempts to do. But other
claims in this vein are not so easy to judge. Consider Feldman's own
account again, with MO and (2.51) merged thus:
(2.58) (s)(t)(p)(s morally ought, as of t, to see to p iff @w')[Asy t, w' &p
is true at w' &~(3w"){As, t, w" &~p is true at w" & W(w") >

Surely there is some sort of "breaking down" involved here. Nevertheless,

I think that Feldman is correct to deny that this constitutes (pure) analysis.
The explicit invocation of intrinsic value seems to me to render the claim
substantive, and not purely conceptual. Even if Feldman is correct about
the link between obligation and intrinsic value, those who deny that he
is are making a substantive error; if he is incorrect, then he has made a
substantive error — he hasn't made an error concerning the "structure of
the concept" of obligation. Or so it seems to me, but I confess I don't know
how to argue for this or against the claim that the errors mentioned are
indeed conceptual rather than substantive.
The second question is this. I say that someone who denies (2.46), for
example, is making a conceptual error. But might it not be that someone
whom I take to be denying (2.46) is not in fact denying it but rather some-
thing else? For might not this person be using "ought" (or "can") to express
a concept different from that which I take it to express, so that the proposi-
tion that he takes to be expressed by "if S ought to do A, then S can do A"
is indeed false? The answer, of course, is that this may indeed be so. Perhaps
there are ways of trying to rule this out — I can test him, for example, on
just when he takes it to be appropriate to use "ought" (or "can"), skillful-
ly varying the contexts so as to get a fix on his meaning — but the disturb-
ing possibility remains that those who affirm what they take (2.46) to
express and those who deny what they take (2.46) to express are not in
reality disagreeing with one another but rather talking at cross-purposes.

78
 The dynamics of obligation

3.1 "OUGHT" AND "CAN"

There are few propositions more widely discussed in the literature on
philosophical ethics than the proposition that "ought" implies "can." I have
noted that my account of obligation implies this proposition to be true,
indeed to be analytic, when understood as follows:

(2.46) if 5 ought [at Tin W] to do A [at T'], then 5 can [at T in W] do

A [at T'].

That my account implies the truth of this proposition will no doubt incline
many readers to reject the account.
In this section I shall consider a number of reasons for denying (2.46),
proceeding from objections that seem to me quite weak to objections that
seem to me stronger (but still unsuccessful). In Subsection 2.2.4 I distin-
guished two senses of "can" such that "ought" implies "can." The weaker
sense was just that of personal possibility or accessibility; it is this sense of
"can" that I have employed in this work. The stronger sense was that of
being free to perform an action (stronger in that a person's being free to
perform an action implies, but is not implied by, that action's being per-
sonally possible for that person). In what follows, it is "can" merely in the
sense of personal possibility that is explicitly at issue - except when alter-
nate possibilities are discussed in Subsection 3.1.5, where it is "can" in the
sense of "free to" that is at issue.

3.1.1 An internal criticism

First, it might seem that (2.46) should be deemed unacceptable by my own
lights, in that the account that I have proposed is avowedly intended to
accommodate such views as that what is obligatory is what is probably best

79
(in some substantive sense of "best"), or what is apparently best, and so on.
And might it not easily be that what is probably or apparently best is in fact
not open to the agent, even when what is best is open to the agent?
The answer is that there are two understandings of the claim
(3.1) ^4 is probably (or apparently) the best (course of) action open to the
agent.
On one reading what this means is the following:
(3.1/) of all actions (or courses of action), whether open to the agent or
not, A is that which is probably (or apparently) the best one open
to the agent.
Here, it is true, there is no guarantee that A is indeed open to the agent.2
The second reading is this:
(3.1") of all actions (or courses of action) that are open to the agent, A is
that which is probably (or apparently) the best.
Here, of course, it is implied that A is open to the agent. (By way of anal-
ogy, suppose that someone, looking down from a balcony at a room full of
people, proclaims
(3.2) Rex is probably (apparently) the richest Republican here.
This may mean one of two things. First, it may mean that
(3.2') of all the people in this room, Rex is probably (apparently) the rich-
est Republican.
Now this may be true and yet Rex not even be a Republican. But the claim
in question might instead be this:
(3.2") of all the Republicans present, Rex is probably (apparently) the
richest.
This cannot be true if Rex is not a Republican.) The fact is that, while my
account accommodates theories according to which (3.1") corresponds to
what is obligatory, it does not accommodate theories according to which
(3.1') corresponds to what is obligatory. On my view, anyone who thinks
that (3.1') corresponds to what is obligatory is making a conceptual error
(just as, for instance, anyone who thinks of obligation along the lines men-
tioned in the second paragraph of Chapter 2 is making a conceptual error).
1 See the discussion of (A)-(C) in Section 1.4.
2 Cf. Frankena (1963), p. 159.

80
The error has to do with the fact that, unlike (3.1"), (3.1') is consistent with
the denial of

(2.46) if 5 ought to do A, then S can do A.

Of course, my insistence that the denial of (2.46) constitutes a conceptual

error is not itself proof of (2.46). But so far I have only tried to show why
(2.46) is not to be deemed unacceptable by my own lights.

3.1.2 "Can" and "can avoid"

A second reason for denying (2.46) is this.3 We shouldn't say that it is moral-
ly permissible for Trudy not to travel faster than light; for what is physical-
ly necessary is not the sort of thing to which moral permissibility pertains.
But, as we all know, an act is obligatory if and only if it is not permissible
not to do it, and so it turns out that Trudy is obligated to travel faster than
light, even though she cannot.
This is a distinctly odd argument. Let A be Trudy's traveling faster than
light. The argument may then be represented as follows:4

(3.3) ~P(~,4);
(3.4) O(A) iff ~P(~A);

hence,
(3.5) O(A);
but

(3.6) ~C(A);

hence,

(3.7) it is not the case that, if O(A), then C(A).

In assessing this argument, we should ask why anyone would want to

assert its first premise, (3.3). The answer, of course, is that we should not
say that something (such as not traveling faster than light) is permissible
if it is impossible to avoid. In other words, (3.3) is itself based on the
following claim:

(3.8) if P(~A)9 then C(A).

3 What follows is inspired by Shaw (1965).

4 Here I use symbolism introduced in Subsections 2.1.3 and 2.3.1.

81
How odd that an argument against the claim that "ought" implies "can"
should rest on the claim that "may" implies "can avoid"! Clearly, some-
thing has gone wrong here. Some might diagnose (3.3) as the culprit, but
I do not. I think it's quite right to say that "may" implies "can avoid." My
account is committed to this. Where the argument goes wrong, I believe,
is in asserting (3.4). Far from being something that we all know to be true,
(3.4) is, on my account, false. I pointed this out in Subsection 2.3.1. (3.4)
is the same proposition as (2.37'), which itself implies both
(2.37'a) if O(A), then ~P(~A)
and
(2.37'b) if ~P(~A), then O(A).
The culprit is (2.37'b). This is false. What's true is this:
(2.37'c) if C(A) & C(~A) &~P(~A), then O(A).
And so I do affirm the following:
(3.4') O(A) iff C(A) & C(~A) & ~P(~A).
But of course this cannot be used in an argument against the claim that
"ought" implies "can," since it implies that "ought" does imply "can."

3.1.3 An alleged redundancy

A quick argument against (2.46) has been provided by Alan White. He says:
It is...because what we ought to do is not necessarily something that we can
do that we have occasion to use such an expression as 'We ought to deal with
this matter now, if we can'. The qualification 'if we can', like the qualification 'if
we have time', would be simply redundant if we ought to do only what we can.5
The allegation of redundancy is inaccurate. Sometimes we are unsure
whether we can do something and whether, if we can, we ought to do it.
The answer "We ought to do it, if we can" is partially helpful. While not
speaking to the first question, it does speak to the second.

3.1.4 Past and present obligations

Walter Sinnott-Armstrong has proposed rejecting the claim that "ought"
implies "can" on the following grounds.6 He asks us to consider the
statement
5 White (1975), p. 148.
6 Sinnott-Armstrong (1985).

82
(3.9) I ought to have mowed your grass.
He says that this admits of two interpretations:

(3.9') I now ought to have mowed your grass in the past;

and

(3.9") it was true in the past that I ought to mow your grass.
Sinnott-Armstrong favors the first interpretation, saying:
[S]uppose I promise on Monday to mow your grass by Saturday. (3.9") is still
true on Thursday and on all future days, because it was true on Tuesday that I
ought to mow your grass by Saturday. In contrast, (3.9') is not true on
Thursday, because it is not yet true that I ought to have mowed your grass in
the past (before Thursday), since I promised only to mow it by Saturday.7

Sinnott-Armstrong then proceeds to provide two analogous interpreta-

tions of

(3.10) I can have mowed your grass.

They are:

(3.10') I now can bring it about that I did mow your grass;

and

(3.10") it was true in the past that I could mow your grass.

He then argues as follows. If "ought" implies "can," then (3.9) implies

(3.10). The proper interpretation of (3.9) is (3.9'); analogously, that of
(3.10) is (3.10'). But (3.90 doesn't imply (3.10'), since (3.9) will be true
even when the promise in question is not fulfilled by Saturday and thus
even when (3.10) is false. Hence "ought" does not imply "can."
This is intriguing but, I fear, quite muddled. We should note, first of all,
that Sinnott-Armstrong's argument that (3.9') is preferable to (3.9") as an
interpretation of (3.9) is unpersuasive, in that it overlooks a third interpre-
tation of (3.9) which I believe is clearly preferable to both, and that is:

(3.9'") it was true in the past that, for some time T, I ought to mow your
grass at T, and Tis past.

In the example given, Tis Saturday, and it is for this reason that (3.9) is not
true when uttered prior to Saturday. Now, if "ought" implies "can," then
7 Sinnott-Armstrong (1985), p. 46. The numbering of propositions has been changed to
conform with the numbering that has been used in this chapter.

83
we should expect that (3.9"') implies this:
(3.10"') it was true in the past that, for some time T, I could mow your
grass at T, and Tis past.
And, I believe, this implication does indeed hold. The fact that, as Sinnott-
Armstrong correctly says, (3.9) can be true whereas (3.10') is false, is irrel-
evant; for (3.10') is not the proper interpretation of (3.10).
To bolster his claim that (3.9') is the proper interpretation of (3.9),
Sinnott-Armstrong says this:
[Someone might respond] that judgments like (3.9') are never true. However,
such a universal claim cannot be justified without begging the question.
Furthermore, it often seems natural to say that someone ought to have done
something, even if he or she did not do it. We need to be able to make present
judgments about what ought or ought not to have been done in the past in
order to justify present punishment for past acts or failures to act. The correct
interpretation of such judgments cannot have the form of (3.9") — [or, I am
sure Sinnott-Armstrong would add, (3.9'")] - because [such] judgments.. .are
true in many cases where punishment is not justified, such as when it was no
longer true at the time at which the agent did the act that he or she ought not
to do it. Thus, it must be judgments like (3.9') that seem natural and needed.8

But this, too, is muddled. I don't know whether it's begging the question
for me to insist, as I do,9 that, if S ought at T to do A at T', then T" is not
earlier than T, but I can certainly give an account of the propriety of "pre-
sent judgments" that differs from Sinnott-Armstrong's. Sinnott-Armstrong
seems to suggest that punishment is justifiable only ifjudgments like (3.9')
can be made, for judgments like (3.9") (and (3.9'")) can be true even when
punishment is not justifiable. It is of course true that punishment may not
be justifiable even when judgments like (3.9'") are true, but this does not
warrant the inference that it is justifiable only when judgments like (3.9')
are true. On the contrary, judgments like (3.9') are never true. It is judg-
ments ofresponsibility, not (simply) ofobligation, that justify punishment, and
it is certainly possible that someone should now be responsible for a past
action, even though it is not possible that someone should now be oblig-
ated to perform a past action.10

8 Sinnott-Armstrong (1985), p. 47. The numbering of propositions has been changed as

before.
9 See the discussion of "ought to have" in Subsection 2.1.3, including note 25.
10 See Zimmerman (1988a), Ch. 3, on what is there called "retrospective" responsibility.

84
 3.1.5 Alternate possibilities
The question was raised in Chapter 2: if moral responsibility doesn't require
alternate possibilities (as Frankfurt has argued, and as I have concurred),
why think that moral obligation does? I shall now address this question,
doing so by way of a response to an argument made by David Widerker,
according to which the claim that moral responsibility doesn't require
alternate possibilities in fact implies the claim that "ought" does not imply
"can."11
One cautionary note: up to now, "can" has been understood in terms
merely of personal possibility or accessibility, but in the present discussion
of alternate possibilities it is necessary that it be understood in terms of
freedom. (On this matter, see the second paragraph of this chapter.) To
make this explicit, let us use C' to symbolize this stronger sense of "can."
The principle that is presently at issue, then, is this revised version of (2.46):
(2.46/) if O(A), then C(A).
The principle concerning moral responsibility and alternate possibilities
that Frankfurt and I and possibly Widerker12 reject is this:
(3.11) a person is morally responsible for what he has done only if he
could have done otherwise; that is (ignoring matters of tense),
if R(A), then C'(~A)U
Frankfurt argues that the rejection of this principle does not require rejec-
tion of (2.46').14 But Widerker argues that Frankfurt is mistaken, since
(2.460 entails (3.11), or at least that part of (3.11) that deals with moral
blame:
(3.11a) S is morally blameworthy for doing A only if 5 could have done
other than A; that is,
if B{A), then C'{~A).
Widerker's argument is simple. He claims that (3.11a) follows from (2.46'),
given the following proposition, which he claims to be a necessary truth:
(3.12) 5 is morally blameworthy for doing A only if 5 ought to have done
other than A; that is,
if B(A), then O(~A).
11 Widerker (1991).
12 Widerker is not fully explicit on this issue.
13 Frankfurt (1969), p. 829ff.
14 Frankfurt (1988), pp. 95-6.

85
 Of course, (3.11a) does follow from (2.46') conjoined with (3.12).
(Widerker is relying on that version of (2.46') which says that, if S ought
to do other than A, then S can do other than A.) But (3.12), far from being
necessarily true, is false. This is because moral responsibility is a function of
what one believes about one's moral obligations and not of what one's moral
obligations actually are.15 One can be blameworthy for what one has done,
even if one was justified in doing it, as long as one believed at the time that
one was not justified in doing it. Thus blame worthiness does not entail
wrongdoing, and to justify an action is not to exculpate its agent. (Failure
to recognize this fact too often prematurely terminates our moral inquiries.)
It is also the case (although this is not strictly germane to the present dis-
cussion) that wrongdoing does not entail blame worthiness; this is widely
accepted. Wrongdoing and blameworthiness are thus logically indepen-
dent of one another (though not conceptually so; the latter is to be analyzed
in terms of the former), and Widerker's argument against Frankfurt fails.
It could, of course, still be that Widerker's conclusion is true. But I
believe otherwise. I think that (2.46') doesn't imply (3.11a), because the
latter is false but the former true. The reason is this. (3.11a) is a contraction
of two other principles:

(3.13) S is morally blameworthy for doing A only if S could have done

A; that is,
if E(.4), then C(.4);
and

(3.14) S could have done A only if 5 could have done other than A; that is,
ifC'(,4),thenC'(~,4).
What Frankfurt's argument against (3.11a) shows is that (3.14) is false;
(3.13) is left untouched.16 (Here, by the way, is where it is most obvious
that we are dealing with the stronger "can" of freedom — C' — and not
merely that of personal possibility — C. For it would hardly require the
insightful efforts of Frankfurt to show that the following proposition is false:
(3.14') ifC(,4),thenC(~,4).
Similarly, there would hardly be any need to belabor the truth of this
proposition:
(3.13') if B(A), then C(A).)

15 I have argued for this in Zimmerman (1988a), Sections 3.1 and 3.6.
16 See Zimmerman (1988a), Section 4.10, for discussion of this.

86
In other words, Frankfurt has not severed the link between moral respon-
sibility and freedom, a link that, I believe, holds of conceptual necessity. In
this regard, the concept of moral blameworthiness is just like many other
fundamental moral concepts. For example, in addition to (3.13), we may
also accept:

(3.15) S is morally praiseworthy for doing A only if S could have done

A; that is,
i£Pr(A), then C'(A).
More importantly, (2.46') is just one more instance of this link between morali-
ty andfreedom, as is the following principle:

(3.16) it is morally wrong for 5 to do A only if S can do A; that is,

if W(A), then C(A).

Now, none of these principles themselves directly concerns alternate

possibilities. But when we note that

(2.37") O(A) iff W(~A),

we can immediately derive from (3.16) and (2.46') these further principles
(with appropriate substitutions of ~A for A):

(3.17) S ought to do A only if 5 can do other than A; that is,

ifO(,4),thenC'(~,4);

and

(3.18) it is morally wrong for S to do A only if 5 can do other than A;

that is,
if J/F(A),thenC'(~,4).

Thus the link between obligation and alternate possibilities is preserved by

an appeal to (2.37"), not (3.14), and this leaves room for the cogency of
Frankfurt-type cases against the link between responsibility and alternate
possibilities by means of an attack on (3.14).
Some questions may remain, though. One is this. Even if it is granted
that (3.14) is false, might not the link between responsibility and alternate
possibilities be restored by means of an argument similar to the one that I
have given that aims to preserve the link between obligation and alternate
possibilities? In particular, might there not be a responsibility analogue to
(2.37") that functions as (2.37") does in my argument? The answer is that
I know of no such analogue. In particular, it is clearly not the case that

87
(3.19) 5 is morally praiseworthy for doing A iff S is morally blame-
worthy for doing other than A; that is,
Pr(A) iff B{~A).

For S cannot both do A and fail to do A. Indeed, it is not even the case that,
where 5 has not yet acted, S would be praiseworthy for doing A if and only
if 5 would be blameworthy for failing to do A. For more is required for
praiseworthiness than the avoidance of blameworthiness, and more is
required for blameworthiness than the failure to act in a praiseworthy
fashion.17
Another question is this. Doesn't it simply seem that Frankfurt-style
counterexamples work just as well when applied to obligation as when
applied to responsibility? Consider, for example, this case. Murphy freely
commits murder but, for Frankfurt-type reasons, couldn't have done other
than commit murder. Surely Murphy does something wrong. But, given
(2.37"), it follows that (2.460 *sfelseafter all; t n a t is> it follows that "ought"
does not imply "can." A similar point can be made concerning (3.17), the
claim that "ought" implies "can avoid." Suppose that Ralph freely refrains
from committing murder, although again, for Frankfurt-type reasons, he
couldn't have done other than so refrain. Surely he does what he ought;
but then it follows that (3.17) is false after all.
The situation here may be put as follows. I have argued:
(2.37") O(A) iff W(~A);
(2.460 ifO(,4),thenC'(,4);
hence,
(3.18) if W(A), then C'(~A)\
and also:
(2.37") O(A) iff W(~A);
(3.16) if W(A), then C'{A)\
hence,
(3.17) if O(A), then C'(~A).
The critic accepts thefirstpremise of each argument but, because he rejects
their conclusions, rejects their second premises. One alternative would of
course be to reject theirfirstpremise instead, thus leaving the link between
"ought" and "can" and between "wrong" and "can" intact. But of course
17 See Zimmerman (1988a), Ch. 3.

88
I reject this alternative, for I take the arguments to be sound. Yet hasn't the
critic given good, Frankfurt-type reasons for rejecting their conclusions? I
don't think so. I would readily grant that Murphy does something that is in
some way morally bad, and that Ralph avoids doing this; but, because nei-
ther agent can act otherwise, it is incorrect to say that Murphy does some-
thing wrong or that Ralph acts as he ought. (Similarly, it is incorrect to say
that it is or would be wrong for Ralph to commit murder, where this is
understood to include the rider: under the circumstances. I can of course agree
that it would be wrong for Ralph to commit murder if he had the option of
doing so.)
This is not a particularly satisfactory situation. It may seem that, when
the critic makes a point, I simply dig in my heels; when I make a point, he
digs in his heels. Is there a way to resolve this? That is unclear to me. I can
point to what is surely the initial intuitive plausibility of (2.46') and (3.16);
I can point to the general kinship between (2.46'), (3.16), (3.15), and
(3.13); I can (as I do in other parts of this chapter) defend (2.46) (and, by
implication, (2.46') and (3.16)) against other attacks. All of this seems to me
to favor my position rather than the critic's, but I don't suppose it will win
over everyone.

3.1.6 Psychological incapacitation

Having mentioned "can avoid" as well as "can," I shall turn now to an argu-
ment against the claim that "ought" implies "can avoid," reverting to the
weaker sense of "can" that merely involves personal possibility (for there
is no longer any need to stick with the stronger sense). The claim now at
issue, then, is this:

(3.17') ifO(^4),thenC(M).
And the argument is simply this. It can happen that a person's moral train-
ing during childhood is very rigorous, so rigorous that, in later years, he is
psychologically incapable of acting otherwise than in accordance with his
moral obligations. This implies that (3.17') is false.
While I agree that moral training can be very rigorous, I am doubtful
that it can lead to such incapacity. Still, perhaps it can. George Washington
declared himself (while still a child) unable to tell a lie (although he was
apparently quite capable of destroying someone else's personal property).
Suppose that this was literally true, and true as a result of his moral training.

18 Adapted from van Rijen (1993), pp. 265-6.

89
In avoiding mendacity, did he act as he ought to have done? Well, maybe
he did; for perhaps he was obligated, at a time prior to his becoming unable
to tell a lie, not to tell a lie at that later date when he had become unable to
do so. But this of course is quite consistent with (3.17') and is, presumably,
not what the critic has in mind. No, the critic means that, even after he had
become unable to tell a lie, Washington was obligated then not to do so.
But I deny this. I would grant, as I did with Ralph, that Washington avoid-
ed doing something that was in some way morally bad, but he did not avoid
doing something wrong; he did not act as he ought to have done.
It may again seem that I am just digging in my heels. But I am remind-
ed here of a lovely remark attributed to Mark Twain: "I am morally supe-
rior to George Washington. He couldn't tell a lie. I can and I don't."
There is a moral insight here. One who can tell a lie but refrains from doing
so is to be distinguished, morally, from one who "refrains" from doing so
because he is unable to do so.20 Similarly, one who feels fear but nonethe-
less acts courageously is to be distinguished, morally, from one who acts
fearlessly. And so on. Of course, it might be argued that the moral distinc-
tions in question have to do not with obligation but with something else
(character, perhaps). I would agree that they have to do with something
else, but I don't see it as digging in my heels to say that they have also to do
with obligation.

3.1.7 The obligatory and the ideal

One of the stronger arguments against the claim that "ought" implies "can"
comes from considerations of such common claims as the following:

(3.20) I ought not to have felt so covetous;

(3.21) you ought to feel gratitude;
(3.22) he ought to be ashamed;23
(3.23) she ought to stop laughing at her friend's new haircut.24

19 As quoted in van Inwagen (1983), pp. 63-4.

20 It is arguable that it is possible on occasion to act freely on a certain impulse even when
the impulse is irresistible. (I explore this issue in Zimmerman, 1995b.) Even if this is pos-
sible, I would say that, given the irresistibility of the impulse, such action cannot be either
obligatory or wrong, although it can of course be morally significant in some other way.
21 See Moore (1922), p. 319.
22 See Frankena (1963), p. 150.
23 See White (1975), p. 198.
24 See Sinnott-Armstrong (1988), p. 116.

90
In these cases, and in indefinitely many others like them, it seems that what
the agent ought to do he cannot do. A standard response to cases of this sort
is to say that there are two senses of "ought," one of which implies "can"
and the other of which doesn't. The claims just given involve a sense of
"ought" that does not imply "can," while the sense of "ought" at issue in
the account of obligation provided in Chapter 2 does imply "can."25
This response is frequently ridiculed. White, for example, says that "it
offers no reason, other than its alleged ability to explain failing to feel what
one ought, for supposing that there is a second sense of 'ought'."26 Sinnott-
Armstrong similarly complains that the response is question-begging.27 But
while I grant that, if no independent reason for distinguishing two senses
of "ought" were given other than preservation of (2.46) in face of the
examples given in (3.20)—(3.23), then this would constitute a begging of
the question against opponents of (2.46), the fact is that, despite what White
and Sinnott-Armstrong say, such independent reason is forthcoming.
Moore himself distinguishes that sense of "ought" having to do with what
is ideal (or desirable) from that which has to do with duty (or obligation).
There is surely reason to acknowledge this distinction; indeed, it is the very
distinction that I introduced in Section 1.2, when contrasting what I called
the "binding" and "nonbinding" senses of "ought." I made no mention
there of the issue of whether "ought" implies "can"; nevertheless, I submit
that the distinction had intuitive appeal. If any further evidence of the dis-
tinction is needed, simply recall the fact, mentioned in Section 1.2, that cer-
tain "ought"-statements (those which express the nonbinding sense of
"ought") are what I called "passive-transformable," whereas others (those
which express the binding sense of "ought," that is, obligation) are not. In
the nonbinding sense of "ought,"
(1.1) someone ought to help that old lady
expresses exactly the same proposition as
(1.1') that old lady ought to be helped by someone,
while, in the binding sense of "ought,"
(1.3) Jones ought to help that old lady
expresses a quite different proposition from that expressed by
(1.3') that old lady ought to be helped by Jones.
25 See Moore (1922), p. 318fF. Cf. Frankena (1963), pp. 150-1.
26 White (1975), p. 150.
27 Sinnott-Armstrong (1988), p. 119.

91
The charge that this response begs the question cannot, therefore, be
sustained.
It may of course be true on occasion that someone is in (indirect) con-
trol of his emotions; perhaps he can suppress his covetousness, or summon
up gratitude, or induce shame, and so on.28 If so, then he may well be oblig-
ated to exhibit such control. The point being made here is simply this:
(3.20)-(3.23) can all be true, even if the agent cannot control whether or
not he has the emotions in question; but if they are true in this way, then
the "ought" that they contain does not express obligation.
It is worth noting also that the distinction between the two senses of
"ought" is not confined to matters having to do with the emotions (such
as covetousness, gratitude, shame, and amusement). On the contrary, the
case just given of Jones and the old lady has to do with an action, that of
helping. Other cases are easily imagined. For instance, we might say that
(3.24) Smith ought to give up smoking,
or that
(3.25) White ought to leave that wound alone,
even though neither Smith nor White can do what he ought.29 Here, too,
it is the nonbinding, ideal sense of "ought," rather than the binding, oblig-
ation sense, that is operative.
A final point is this. What has been said about uncontrollable emotions
applies also to uncontrollable motives. Ross is well known for making this
point:
That action from a good motive is never morally obligatory follows.. .from the
Kantian principle...that 'I ought' implies 'I can'... I can act from a certain
motive only if I have the motive; if not, the most I can do is to cultivate it by
suitably directing my attention or by acting in certain appropriate ways so that
on some future occasion it will be present in me, and I shall be able to act from
it. My present duty, therefore, cannot be to act here and now from it.30
Others have made a similar point.31 We must take care, however, not to
make unwarranted inferences from this observation. For example,
although it may be that, imbued with malice, I cannot summon up any
sympathy and hence that it is not true that I ought (am obligated) to act

28 Cf. Moore (1922), p. 316.

29 Both cases are borrowed from White (1975), p. 148.
30 Ross (1930), pp. 4-5.
31 E.g., Prichard in Prichard (1949), pp. 129-30.

92
from sympathy, still it might be true that I ought not to act maliciously. For
while having a certain motive may not be in our present control, it is, I think,
often possible to exercise control over whether or not we act/rom a motive
that we do have. In saying this I am, like Ross, clearly not using "motive"
to refer simply to that which does infact "move" one to action; rather, I use
it to refer to that which has the capacity to do so. Motives, in my sense, may
thus compete with one another for the role of "actual motivator," and, I
am claiming, which one plays that role is frequently up to the agent.

3.1.8 Excuses
It may be argued that, even if the distinction between what is obligatory
and what is ideal is accepted, still there are actions which are obligatory, and
not ideal, and yet which cannot be performed. Alan Montefiore gives this
example: "I ought to be at the meeting tonight, but unfortunately I cannot
manage it."32 Of such a case White says: "We don't blame him because we
accept his inability as an excuse. But it wouldn't be an excuse for not doing
what he ought to do if it could make that cease to be what he ought to
do."33 Similarly, Sinnott-Armstrong says:

[I]f a bride cannot get to her wedding, then she ought at least to let someone
know and to excuse her absence... [I]f it were not first true that she ought to
go to her wedding, she would have nothing to excuse when she failed to go to
her wedding. Excuses are distinguished from justification in that to justify an
act is to show that it is right, whereas to excuse an act is to admit that it is wrong
but to deny responsibility.34

Here, I think, the opponents of the claim that "ought" implies "can"
are in the grip of a dogma, one attributable to J. L. Austin, who declared
that excuses are to be distinguished from justifications in just the way that
Sinnott-Armstrong mentions.35 But what Austin said seems wrong.
Excuses aren't offered only where wrongdoing (that is, the failure to satis-
fy an obligation) is admitted. Consider this case, provided by Terrance
McConnell:
Suppose I am at a lake when a man drowns. Another accuses me of wrongdo-
ing for failing to save the man. But I reply that I am not able to swim (or that

32 Montefiore (1958), p. 24.

33 White (1975), p. 152.
34 Sinnott-Armstrong (1988), p. 113, emphasis added.
35 See Austin (1968), pp. 19-20.

93
the man was on the other side of the lake and I was not able to get there in
time).36
It is surely clear in such a case that the appeal to inability does not presup-
pose any admission of wrongdoing. And yet it would be quite natural to
say that the agent in the example offered his inability as an excuse.
N o w , there are two ways to go at this point. One might simply insist
that excuses do presuppose wrongdoing, just as Austin claims. Indeed,
McConnell himself says this. But then, as McConnell states, it would be
inaccurate to say of his example that in it an excuse is being offered. This
would mean that what many find natural to say about his example is in fact
incorrect. I have no real quarrel with this. The moral would be that it is also
incorrect, even if natural, to say of Montefiore's meeting case and Sinnott-
Armstrong's wedding case that an excuse is being offered. Hence there is
no grist after all for the mill of the opponent of the claim that "ought"
implies "can."
Another way to go here — and one that I prefer — is simply to reject
Austin's claim that excuses presuppose wrongdoing and to assert that what
it seems natural to say in Montefiore's and Sinnott-Armstrong's and
McConnell's cases is in fact correct. For what is an excuse? An excuse is
offered in response to an accusation.37 It is a defense to a charge. It is a denial
of responsibility when responsibility has been imputed. This can occur in
a variety of settings, including those in which wrongdoing is admitted, but
also (as I mentioned when responding to Widerker's argument in
Subsection 3.1.5) including those in which it is not. Given this under-
standing of what an excuse is, we may admit that excuses have been offered
in the three cases just mentioned without thereby being committed to
denying that "ought" implies "can."
A slight complication here is the following. In the meeting case, don't
we have an admission of wrongdoing? After all, the protagonist has said: "I
ought to be at the meeting tonight." But this statement is open to various
interpretations. First, he may mean simply that he ought in the ideal, non-
binding sense to be at the meeting. Or he may mean that he has been
charged with being at the meeting. O r he may mean (and I think this is very
common) that, although he can (literally) manage to go (and, perhaps, has
a prima facie obligation to go), he cannot do so without serious inconve-
nience, to himself or others (and therefore, perhaps, without violating some
more stringent prima facie obligation). If the protagonist means any of these

36 McConnell (1989), p. 438.

37 Austin says just this on p. 19 of Austin (1968).

94
things, he says nothing inconsistent with (2.46), the claim that "ought
(overall)" implies "can." And if he means none of these, if he means that
he ought (overall) to be at the meeting although he literally cannot be there,
then, I say (barring a complication to be discussed next), he is simply mis-
taken; for it cannot be that he ought (overall) to do what he literally
cannot do.

3.1.9 Self-imposed impossibility

Even if the foregoing discussion of excuses is plausible, though, it may seem
insufficient to salvage the claim that "ought" implies "can." For it seems
that there just are cases where a person ought (is obligated) to do something
but cannot (literally) do it, and these are cases where the person appears to
have no excuse for not performing the action. Here I have in mind cases of
what may be called "self-imposed impossibility." Many philosophers have
appealed to such cases in an effort to prove (2.46) false.38 A typical case is
this. Bruce borrows some money on Friday, promising to repay it on
Sunday. Since nothing else of moral significance conflicts with such repay-
ment, he is therefore overall obligated to repay the money on Sunday. But
he spends all the money on Saturday night, thereby rendering himself
unable to repay it on Sunday. Surely, even though Bruce cannot repay the
money then, he is obligated to do so. To declare otherwise smacks of para-
dox,39 for what an easy escape Bruce would then have from the bonds of
obligation — what an easy escape we would all have! We need only disable
ourselves in the relevant way and, voila, we're off the hook.
This seems to me to constitute the strongest challenge to the claim that
"ought" implies "can." A similar argument can also be made against the
claim that "ought" implies "can avoid." It has to do with what may be
called "self-imposed necessity." Suppose that Betty, like Bruce, borrows
some money on Friday, promising to repay it on Sunday. Unlike Bruce,
Betty is conscientious. To ensure that she keeps her promise, she takes the
money on Saturday to the room in which she is due to repay it on Sunday,
enters the room, locks the door, and throws away the key, thereby thwart-
ing, in the manner of Ulysses resisting the Sirens, the temptations of
Saturday-night spending sprees. Let us suppose that this action renders
Betty incapable of not repaying the money on Sunday. Must we then
conclude that it is not true after all that she ought to repay it then? Surely
38 See, e.g., Henderson (1966), p. 106; Robinson (1971), p. 252; Stocker (1971), p. 314;
White (1975), p. 149; Richman (1983), p. 85; Sinnott-Armstrong (1988), pp. 116-17.
39 Cf. the discussion of "Suzy Mae" in Powers (1967), pp. 385-6.

95
not. Hence "ought" does not imply "can avoid" any more than it implies
"can."
The proper response to this latest challenge requires careful develop-
ment.

3.2 IMMEDIATE AND REMOTE OBLIGATION

A crucial component in the account of obligation that I have proposed is
the use of two time indices. The canonical phrase form at issue is this:
(3.26) S ought, at Tin W, to do A at T".
This reflects the fact, discussed in Subsection 2.2.1, that control may be
either direct or indirect. The example given there had to do with my
throwing a brick through a store window. I had control over whether or
not the window broke. Such control was indirect, however, being para-
sitic on the direct control that I had over my decision to throw the brick.
The control that I had over my breaking the window was hybrid, con-
joining the direct control over the decision with the indirect control over
the window's breaking, which was a consequence of that decision.
Let's embellish the story a little. Suppose that breaking the window was
a necessary preliminary to my snatching the jewels, and I snatched them.
Then the decision to snatch was something over which I had direct con-
trol, the disappearance of the jewels was something over which I had indi-
rect control, and the snatching was something over which I had hybrid
control. But that's not all. At the time I made the decision to throw the
brick I had indirect control over each of the items just mentioned - the
decision to snatch, the disappearance of the jewels, and the snatching. Thus,
although (on the account that I favor) one can have direct control only over
one sort of thing (decisions) and one can have hybrid control only over one
sort of thing (nonminimal actions), one can have indirect control over all
sorts of things, including decisions and nonminimal actions, since these can
themselves be among the consequences of decisions.
Now, as I stated in Subsection 2.2.1, this decision-plus-consequence
picture is not essential to my account of obligation; what is essential is the
distinction between direct and indirect control. And this distinction gives
rise to a closely related distinction between immediate and remote control.
This has to do with the times involved. The canonical phrase form here is
this:

(3.27) S can, at Tin W, do AatV.

96
If T is distinct from (that is, earlier than) T', then S's control over A is
remote; otherwise it is immediate.40 Thus we should understand the claim
that "ought" implies "can" as the claim that (3.26) implies (3.27). That is,
to put it as before, but now with the times emphasized:
(2.46) if S ought, at Tin W, to do A at Y, then 5 can, at Tin W, do A
atV.
This is what my account of obligation implies. And just as control may be
immediate or remote, so too may obligation. If S ought at T to do A at T',
and Tis distinct from (that is, earlier than) T', then S's obligation is remote;
otherwise it is immediate.
Some philosophers have rejected the distinction between immediate
and remote obligation,41 others have accepted it,42 and still many others
have of course been silent on the matter.4 The distinction seems to me of
first importance. Not only does it solve the problems of self-imposed
impossibility and necessity, but it provides the wherewithal for a detailed
account of the dynamics of obligation - the shifting of obligation and its
supersession. It is to this that I now turn.

3.2.1 Self-imposed impossibility

Consider again the puzzle concerning Bruce, who borrows money on
Friday, spends it all on Saturday night, and thus fails to repay it on Sunday.
Not only does the distinction between immediate and remote obligation
solve the problem here, it helps explain what we want to say about the case,
and that is the following. Prior to the Saturday-night spending, Bruce
ought to repay the money on Sunday (in part because he can, prior to
Saturday night, repay the money on Sunday). After he has spent all the
money, though, and so no longer can repay the money on Sunday, it is no
longer true that Bruce ought to repay the money on Sunday. On the con-
trary, he is faced with a new (more restricted) set of options, and his oblig-
ations have shifted accordingly (he ought to apologize, perhaps, and so on).

40 On the relation between immediate and remote control, on the one hand, and direct
and indirect control, on the other, see Subsection 2.2.2, including notes 51 and 52 to
Ch.2.
41 See, e.g., Prichard (1949), pp. 182-3.
42 See, e.g., McKinsey (1975), Castafieda (1977), Goldman (1978), Feldman (1986),
Ch.2.
43 Even when they recognize the distinction between immediate and remote control. Cf.
Alvin I. Goldman (1976), pp. 208-9.

97
 7",

Figure 3.1

The situation may be pictured as in Figure 3.1 (where Tt is Friday, T2 is

Saturday night, T3 is Sunday, and where B, S, R, and A stand for "borrow,"
"spend," "repay," and "apologize," respectively).44 This picture captures
the following (among other things): Bruce ought on Friday, all the way up
to the spending on Saturday night, to repay the money on Sunday; Bruce
ought (by default, as it were) at those same times not to apologize on Sunday
for not repaying the money;45 but, after the spending, Bruce ought then
to apologize on Sunday for not repaying the money; it is not true that
Bruce ought, at these post-spending times, to repay the money on
Sunday (because he cannot, at these times, do this); nor is it true that he
ought, at these times, not to repay it (because he cannot, at these times,
avoid this).
Some philosophers resist this treatment of cases of self-imposed impos-
sibility — even though they accept that the temporal indexing of "ought"
and "can" is legitimate — because they wish to insist that (in my example)
Bruce still ought on Sunday to repay the money.46 It is not clear to me why
they wish to insist on this. I suspect that it is because they wish to insist that
Bruce does wrong in not repaying the money and that they believe that my
sort of treatment of "ought" denies this.47 But, if so, they are mistaken, not

44 As in Figure 2.3, the numbers 1-3 on the right denote the value-rankings of the respec-
tive worlds. Worlds 2a and 2b are thus of equal value, as are worlds 3a and 3b. That these
worlds be of equal value is stipulated merely for purposes of illustration.
45 By default, because he ought at those times not to act in such a manner that there's some-
thing to apologize for.
46 See in particular Sinnott-Armstrong (1988), p. 117.
47 Cf. Stocker (1971), pp. 314-15.

98
in insisting that Bruce does wrong in not repaying the money, but in think-
ing that my account implies otherwise. On the contrary, when an obliga-
tion is not satisfied, a wrong is done, whether or not the obligation is
remote. If Bruce ought on Friday to repay the money on Sunday, and
Bruce doesn't repay the money on Sunday, then Bruce does wrong. And
the wrong isn't just that he spends the money on Saturday night. True, he
ought on Friday, and up to Saturday night, not to spend the money on
Saturday night, and so his spending is wrong — an immediate wrong, since
the final time at which he ought not to spend coincides with (or is imme-
diately prior to) the time of spending. But it is also true that Bruce ought
on Friday, and up to Saturday night, to repay the money on Sunday, and
so his not repaying is wrong, too - a remote wrong, since the final time at
which he ought to repay does not coincide with the (pertinent) time of
nonrepayment. Indeed, in this case it seems that the remote wrong is in a
way more significant than the immediate wrong. Even though there is a
sense in which the remote wrong of nonrepayment is committed by virtue
of the immediate wrong of spending (the remote wrong would not have
been committed, in the manner in which it was committed, had not the
immediate wrong been committed), nonetheless the moral primacy rests
with the remote wrong, in that (let us assume)48 the spending is wrong
because and only because the nonrepayment that it necessitates is wrong (the
spending would not have been wrong had not the nonrepayment been
wrong).
Although this moral primacy of a remote wrong over the immediate
wrong by virtue of which it is committed helps to dramatize the fact that
remote wrongs are genuine wrongs, it is in fact not crucial that the moral
primacy rest with the remote wrong. Suppose, for instance, that Bruce's
spending were to take the specific form of gambling. It might then be that
the gambling is wrong not just because it necessitates the nonrepayment
but for some other reason also; it might even be that the gambling is more
seriously wrong than the nonrepayment, quite independently of the fact
that it necessitates the nonrepayment. Still the nonrepayment would itself
be wrong. Indeed, a remote wrong might be wrong only because some-
thing else is wrong, and yet it would of course still be a wrong. Perhaps,
after spending the money, Bruce hitches a ride to Boston, thereby render-
ing himself unable to be in New York, the site of the repayment scheduled
for Sunday. Bruce ought (on Friday) to be in New York (on Sunday); his

48 It is on this assumption that the value-rankings in Figure 3.1 are based. See note 44 to
this chapter.

99
not being there constitutes a remote wrong, even though the wrongness of
this is wholly parasitic on the wrongness of not repaying the money.49
In sum, then, it will be perfectly true to say on Monday that Bruce ought
to have repaid the money on Sunday and so did wrong in not repaying it,
for what this amounts to saying is that there was a time T such that Bruce
was obligated at Tto repay the money on Sunday. It will also be true to say
on Monday that Bruce could not have repaid the money on Sunday, if this
is taken to mean that there was a time T (not later than Sunday) such that
Bruce could not at T repay the money on Sunday, but not if this is taken to
mean that there was no time Tsuch that Bruce could at Trepay the money
on Sunday. (2.46) is therefore preserved.

3.2.2 The time at which a wrong is done

Some might feel uneasy with the foregoing because it isn't clear just when
remote wrongs are supposed to be done. If Bruce ought on Friday to repay
the money on Sunday, then it is wrong on Friday for Bruce not to repay
the money on Sunday; but if Bruce doesn't repay the money on Sunday,
it isn't on Friday that he does wrong in this regard. The (pertinent) nonre-
payment occurs on Sunday, after all.50 But the trouble is, it also seems prob-
lematic to say that Bruce does wrong on Sunday in this regard, since by that
time it is (according to the proposed account) no longer wrong for him not
to repay the money then. So, if he does a wrong — a remote wrong —just
when is it that this wrong is committed?
It seems to me useful to draw a distinction here between the time at
which an agent goes wrong and the time at which the wrong occurs. "Do
wrong" may be used in either case. The wrong that is done is the (perti-
nent) nonrepayment. This occurs on Sunday.51 Thus the wrong is, in this
sense, done on Sunday, even though by then it is no longer wrong for Bruce
to fail to repay the money. In general, then, we may say this:

(3.28) the wrong constituted by S's not doing A at T' occurs at Tin W
iff

49 I have no precise account to offer ofwhat I have called the "moral primacy" of one oblig-
ation or wrongdoing over another, nor of the corresponding concept of one obligation
or wrongdoing being "parasitic" on another. I hope that what I say here is intuitively
plausible.
50 Clearly, nonrepayment occurs at other times, too; the point is that repayment is sched-
uled for Sunday.
51 See the last note.

100
 (a) Tis identical with T';
(b) for some time T", S ought, at T " in W, to do A at T'; and
(c) 5 does not, in W, do ^4 at T'.
This is intended to capture the sense in which the wrong done by Bruce
itself occurs on Sunday. (Here, talk of "doing wrong" can sometimes seem
a bit forced, for one can conceive of cases where a wrong is "done" in this
sense even though the agent is no longer alive. For example, perhaps S
ought at Tj to do A at T3 but commits suicide at T2.)
But even if this is, in one sense, when the wrong is done, it is still true
that, in some other sense, Bruce doesn't do wrong on Sunday with respect
to the nonrepayment. For the point at which he goes wrong is Saturday
night, when he spends all the money and thereby ensures the nonrepay-
ment on Sunday. To capture this, and to accommodate straightforward
immediate wrongdoings (where the times at which an agent goes wrong
and at which the wrong occurs coincide), it might therefore seem that what
we should in general say is this: an agent goes wrong with respect to failing
to perform some obligatory action either at the time at which the action is
"scheduled" to be performed or at the time at which the action is rendered
impossible. That is:

(3.29) S goes wrong, at T i n W, in not doing A at T iff

(a) 5 ought, at Tin W, to do A at T\
(b) S does not, in W, do A at T; and
(c) it is not the case that S can, after Tin W, do A at T'.

This would yield the desired result in cases of immediate wrongdoing

(where T a n d T' are identical), since no one can change what has already
taken place. It would also yield the desired result in the spending case. If we
look at Figure 3.1, it seems clear that Bruce goes wrong on Saturday night
(this is where he takes a turn for the worse), not just with respect to spend-
ing the money on Saturday but with respect to not repaying the money on
Sunday, since such repayment is rendered impossible at the moment of
spending.
But (3.29) is in fact not adequate to other cases of remote wrongdoing.
Consider this case. O n Friday, Mandy tells Phil that she will be able to help
him move furniture on Sunday, thereby making a sort of informal com-
mitment to do so; on Saturday, however, Mandy promises Bill that she will
babysit his child on Sunday, thereby making a formal commitment to do
so. Note that, although Mandy can still on Saturday afternoon help Phil on
Sunday, she cannot both do that and babysit the child. Let us assume that,

101
 Figure 3.2

because of the informal commitment, Mandy ought on Friday to help Phil

on Sunday, but also that, because of the formal commitment, Mandy ought
on Saturday to babysit for Bill on Sunday. Suppose that what Mandy in fact
does is to babysit on Sunday. Where does she go wrong with respect to not
helping Phil move furniture?
The situation may be pictured as in Figure 3.2 (where Tt is Friday, T2 is
Saturday, T3 is Sunday, and where T, P, H, and B stand for "tell,"
"promise," "help," and "babysit," respectively).52 On this picture, Mandy
ought on Friday to help Phil on Sunday. By taking Track 2b, Mandy does
wrong in not helping Phil. This wrong occurs on Sunday. But Mandy does
not go wrong on Sunday in not helping Phil, for Track 2b is preferable to
Track 3. Rather, it seems clear, it is at noon on Saturday that Mandy goes
wrong in this respect. But this is not what (3.29) implies. Rather, (3.29)
implies that Mandy goes wrong on Sunday in this respect, since at no time
prior to Sunday does it become impossible for Mandy to help Phil then.
(This case, like the spending case, is supposed to be a case where the imme-
diate wrong is wrong only because the remote wrong is wrong. That is, the
formal commitment to Bill would not be wrong were it not for the infor-
mal commitment to Phil. This is intended to dramatize the fact that remote
wrongs are genuine wrongs. But, as before, there is in fact no necessity that
remote wrongs have such primacy in order to be genuine wrongs. If, rather
than promise Bill to babysit his child, Mandy were instead to kidnap the
child, thereby rendering it obligatory for her to return the child to Bill on

52 Again, the value-rankings of the worlds are provided for purposes of illustration only.

102
Sunday and so not help Phil move furniture then, her not helping Phil
would still constitute a remote wrongdoing.)
It is at noon on Saturday that Mandy goes wrong, for that is where she
takes a turn for the worse. Helping Phil is not then rendered impossible,
but it is rendered relatively undesirable, so that it is no longer the case that
Mandy ought to help Phil. It therefore seems plausible to replace (3.29) by
the following:

(3.29') 5 goes wrong, at T i n W, in not doing A&tT' iff

(a) S ought, at Tin W, to do A at T';
(b) S does not, in W, do A at T'; and
(c) it is not the case that S ought, after Tin W, to do A at T'.
The suggestion here is this. To find out where an agent goes wrong with
respect to not performing an action, retrace his steps back down the track
which he in fact took until you reach a juncture, a point of choice, where
it is true that he ought then to perform the action - where, that is, there is
a top-level track leading out from that point and on which the act is fea-
tured. In Figure 3.2, we need only go back along Track 2b to the point of
choice at T2 to find a point where it is true that Mandy ought then to help
Phil move furniture. Hence that is where Mandy goes wrong with respect
to not helping Phil. Similarly, if we suppose that Bruce takes Track 2b in
Figure 3.1, we need only retrace his steps to T2 to find a point where it is
true that Bruce ought then to repay the money. Hence that is where Bruce
goes wrong with respect to the nonrepayment.
While I think that what has just been said is very close to the truth, it is
still not quite adequate. And that is because, even if, when a wrong is done,
it occurs only once, nonetheless the agent may have^one wrong with respect
to it on several occasions. Consider this case. Victor is in a position at To
to visit each of, but not both of, Alf and Bert at T4. He ought to visit Alf.
But at T1 he calls Bert on the phone, making Bert so eager to see him that
he now ought to visit Bert. However, at T2 he calls up Alf, making Alf so
eager to see him that he now ought to visit Alf once again. And at T3
he calls up Bert again, so that he now ought, once again, to visit Bert.
Nonetheless, he visits Alf. The situation may be pictured as in Figure 3.3
(where it is assumed, for simplicity's sake, that the third call would not have
been possible had not the second occurred and the second would not
have been possible had not the first occurred, and where Cu C2, C3t A, and
B, stand for "make the first call," "make the second call," "make the
third call," "visit Alf," and "visit Bert," respectively).53 Victor takes

53 As before, the value-rankings of worlds are provided for purposes of illustration only.

103
 Figure 3.3

Track 5. The question at issue is this: where does he go wrong in not

visiting Bert at T4? (3.29') implies that he goes wrong at T4 in this respect,
and I don't wish to deny that this is so. After all, his immediate obligation
is to visit Bert. But what (3.29') doesn't reveal is that Victor also goes wrong
earlier in this respect. He goes wrong at T2. For from Tt to T2 he ought to
visit Bert, and yet at T2, by calling up Alf, he deviates from this obligation.
This fact is not accounted for by (3.29').
For another example, suppose that, instead of Track 5, Victor takes
Track 4a. Where does he go wrong in not visiting Alf at T4? At T4, but
also at Tt\ for at both these points there is a top-level track on which visiting
Alf at T4 is featured and which Victor chooses not to take.
To capture this possibility of multiple lapses, of multiple deviations from
the strait and narrow with respect to one and the same action, I propose
that we accept this revision of (3.29'):
(3.29") S goes wrong, at T i n W, in not doing A at T' iff
(a) S ought, at Tin W, to do A at T';
(b) S does not, in W, do A at T'; and
(c) it is not the case that S ought, just after Tin W, to do A at T'. 54

54 By clause (c) I mean that, for any time T* after T such that S ought, at T* in W, to do
A at T\ there is an earlier time T** after T such that it is not the case that 5 ought, at
T** in IV, to do A at T.

104
 In general, then, one goes wrong whenever one takes a turn for the
worse. Whenever one does this one of course commits an immediate
wrong, but one may also be committing a remote wrong. Thus no remote
wrong can be done without an immediate wrong being done (Bruce's
wrongly not repaying the money can be traced to his wrongly spending;
Mandy's wrongly not helping Phil can be traced to her wrongly promising
to babysit for Bill), although it sometimes happens that an immediate wrong
is wrong only because some remote wrong is wrong (as in the cases
depicted in Figures 3.1, 3.2, and 3.3).

3.2.3 The time at which an obligation is satisfied

Just as commissions of wrongs are sometimes remote, so too are satisfac-
tions of obligations. Recall the case of self-imposed necessity where Betty
locks herself in a room on Saturday, thereby ensuring repayment on
Sunday. After she throws away the key, she is no longer obligated to repay
the money, for she cannot avoid doing so. Nonetheless, in repaying the
money, she does what she ought. The situation may be pictured as in Figure
3.4 (where Ti is Friday, T2 is Saturday, T3 is Sunday, and where B, T, and
R stand for "borrow," "throw away," and "repay," respectively).
I propose that we treat this matter in a manner parallel to that in which
remote wrongdoing has been treated. Just as "do wrong" does double duty,
so too does "satisfy an obligation." Just as the time of going wrong may be
distinguished from the time at which the wrong occurs, so too the time of
maintaining an obligation may be distinguished from the time of its fulfill-
ment. In Figure 3.4, the time at which Betty fulfills her obligation to repay
the money is Sunday (T3), but she maintains this obligation only up until
Saturday (T2). In general, then, I propose the following:

7",

Figure 3.4
105
(3.30) S's obligation to do A at T is fulfilled at T i n W iff
(a) Tis identical with T';
(b) for some time T", S ought, at T " in W, to do A at T'; and
(c) 5 does, in W, do ^4 at T';
(3.31) 5 maintains, from Tuntil T' in H^, his obligation to do A at T " iff
(a) S ought, from T until T" in W, to do ^4 at T";
(b) S does, in W, do A at T"; and
(c) it is not the case that S ought, just after T' in W, to do A at T".

Let us apply these formulae to Figure 3.2. Suppose now that Mandy takes
Track 3; that is, that she doesn't babysit the child but does help Phil move
furniture. According to the formulae, Mandy thereby fulfills (on Sunday)
and maintains (up until Saturday) an obligation to help Phil move furni-
ture. Or again, with respect to Figure 3.3: suppose now that Victor takes
Track 4b. In this case he fulfills at T4 his obligation to visit Bert, an oblig-
ation which he maintains from T1 until T2 and from T3 until T4.
Return to Figure 3.2. It would be accurate to observe that Mandy could
have done worse - she could have taken Track 5 - and that she would then
have satisfied no obligation at all. But we should not be misled by this obser-
vation. It would be a mistake to infer that one satisfies an obligation as long
as one avoids the worst track open to one or that one satisfies no obligation
if one takes the worst track open to one. Neither of these is necessarily so.
Consider the sort of case represented in Chart 3.1, where the numbering
reflects the ranking of the options, or tracks, open to an agent, S. In such a
case, if S takes Track 2, he satisfies no obligation, even though he could
have done worse. (This claim will be qualified in Chapter 4.) On the other
hand, if S takes Track 4, he does satisfy an obligation, namely, the obliga-
tion to do A. (This claim will be discussed further in Chapter 6.) The moral
is that we should not necessarily be impressed if someone does something
that he ought, as long as he also does not do something else that he ought;
for some ways of failing to do what one ought are superior to some ways of
doing what one ought. Thus, even though it is accurate to say of Mandy that,
in taking Track 3 in Figure 3.2, she satisfies her obligation to help Phil, this
need not constitute praise for her action; for it does not tell the whole story.

1: A& B&...
2: ~A& -B&...
3: ~A& B&...
4: A& ~B&...
Chart 3.1
106
 3.2.4 The cancellation of obligation

I have said that, when an obligation is not satisfied, a wrong is done. This
is clearly true of immediate obligations, but some may balk at accepting this
claim when it is remote obligations that are at issue. Perhaps it will be agreed
that Bruce does wrong in not repaying the money and that Mandy does
wrong in not helping Phil (or in not babysitting the child, depending on
which version of that story we're working with), but aren't there other
cases where it is accurate to say that a remote obligation is not satisfied and
yet no wrong is done? What if the reason for Bruce's not repaying the
money had not been that he spent it all but that he had been mugged (or
struck by lightning) on his way to repaying it? What if the reason for
Mandy's not helping Phil had not been that she was fulfilling a wrongly
made promise to babysit for Bill but that Phil had released her on Saturday
from her informal commitment to help him (or that Phil had died from a
heart attack on Saturday)? Isn't the pertinent obligation canceled in such
cases, and doesn't this show that not all failures to satisfy an obligation
constitute wrongdoing?
I grant the intuitive plausibility of this objection. It is interesting to note
that talk of such cancellation of obligation, if taken literally, would appear
to presuppose the possibility of remote obligation; for how in general could
an obligation to do A at T2 be canceled at T1 unless it were already in exis-
tence at Tj? Nonetheless, I reject the possibility of such cancellation.
Although it may be, for instance, that Bruce would not have been obligat-
ed to repay the money if he had been mugged, or that Mandy would not
have been obligated to help Phil if Phil had released her from her commit-
ment, this doesn't imply that their obligations would thereby have been
canceled. For consider Bruce. Either he could on Friday have avoided the
mugging (or lightning) on Saturday or he couldn't. (It is extremely likely
that he could; he could have walked down Elm Street rather than Oak
Street, and then he wouldn't have been mugged or struck by lightning.)
Similarly, either Mandy could on Friday have avoided Phil's releasing her
from her commitment (for example, by pleading with him not to be so gen-
erous, or condescending, or whatever) or she couldn't; or again, either
Mandy could have avoided (prevented) Phil's having a heart attack or she
couldn't (and perhaps here it is likely that she couldn't). In every case what
we should say is this. If the agent could not have avoided the event that
allegedly canceled the original obligation, then he (or she) did not have the
original obligation after all, although it may have been quite reasonable (for
him or someone else) to believe that he did. (If Bruce could not on Friday

107
have avoided the mugging or the lightning, then there was no world acces-
sible to him on Friday in which he repaid the money on Sunday, and so he
was not obligated on Friday to repay the money on Sunday, although it
may have been quite reasonable to believe that he was. So too for the other
cases.) On the other hand, if the agent could have avoided the "canceling"
event, then, if he was already obligated to perform the pertinent action, he
may well have been obligated to avoid the "canceling" event itself. (If
Bruce could on Friday have avoided the mugging, and if Bruce was obli-
gated on Friday to repay the money on Sunday, then Bruce was obligated
on Friday to avoid the mugging on Saturday. If Mandy could on Friday
have avoided Phil's releasing her from her commitment,55 and if Mandy
was obligated on Friday to help Phil on Sunday, then she may well have
been obligated on Friday to avoid Phil's releasing her from her commit-
ment on Saturday - although not necessarily: if the deontically best that
Mandy could on Friday do on Sunday was to help Phil, whether or not Phil
released her from her commitment, then Mandy was not obligated on
Friday to avoid this release.) Thus every failure to satisfy an obligation does
indeed constitute wrongdoing.56
But, it may be objected, even if Bruce could literally have avoided the
mugging, he surely could not have known he would be mugged, or even
had reason to believe that he would be. Or, even if he did have reason to
believe this, surely it is not he but his muggers who did wrong. To say
otherwise is to blame the victim rather than the perpetrators of a wrong-
doing — a reprehensible practice.
The proper response is this. I am not blaming Bruce in this version of the
case; I am merely saying that he did wrong. Doing wrong is not sufficient

55 Recall from Subsection 2.2.1 that others' actions enjoy no special status among events
when it comes to their accessibility.
56 Cf. Thomason (1981), pp. 178-9. Contrast, perhaps, McConnell, who says that debts
can sometimes be annulled, and gives this illustration (McConnell 1993a, p. 80): "Ifyou
have a financial debt to a person and she steals from you the amount of money owed
(or more), it seems that you no longer owe her any money." I would say here that, while
there may be a sense in which, prior to the theft, you owed your creditor the money,
it is not the case that you had an overall moral obligation to pay it to her, unless you had
an overall moral obligation to avoid her stealing it from you. Did you have an obliga-
tion to reimburse her before she stole it? If so, you had, ipso facto, an obligation to avoid
her reimbursing herself in this way; if not, then, unless for some other (and presumably
rather odd) reason you had an obligation to avoid the theft anyway, it was permissible
for you so to act that she reimbursed herself in this way (even if not permissible for her
so to act in this way), and so permissible for you not to see to her reimbursement in more
customary fashion by paying her yourself.

108
(nor is it necessary) for being to blame. The two must not be confused.57
And anyway I am not insisting that Bruce did wrong in not avoiding the
mugging. I am saying that, if he was obligated to repay the money, then he
did wrong in not avoiding the mugging. Recall the distinction, discussed
in Section 1.4, between theories according to which what is obligatory is
what is best (in some substantive sense) and theories according to which
what is obligatory is what is probably best. It was best (let us assume) for
Bruce to walk down Elm Street rather than Oak Street; hence, according
to the first sort of theory, that's what he ought to have done. But it might
well be that it was (at the time) probably better for Bruce to walk down
Oak Street, in which case, according to the second sort of theory, he did
what he ought, even though that involved being mugged and not
repaying the money. But, if so, then Bruce was not obligated to repay the
money after all (although it might be true that he was obligated to repay the
money if he didn't walk down Oak Street - a matter of conditional obliga-
tion, to be discussed in Chapter 4).58
Another, similar case might help to drive home this point. W. D. Ross
discusses returning a borrowed book through the mail. At one point59 he
declares that the obligation in question is satisfied only if the book in fact
reaches the hands of the person from whom it was borrowed, although the
agent may well not be to blame if the obligation is not satisfied (perhaps
blame will rest with some employee of the postal service). Later Ross
declares that the obligation in question is satisfied if the agent "exerts him-
self" to return the book, even if the book gets lost in the mail. I prefer the
first view, but that is not the present point. The present point is simply this.
If the agent is obligated so to act that the book arrives at its final destina-
tion, then he is obligated not so to act that the book gets lost in the mail; if
the agent is not obligated not to use the mail in this way, then he is not
obligated so to act that the book arrives at its final destination. I am not
insisting (although I believe) that the agent is obligated so to act (given that
he can so act) that the book arrives at its final destination, but only that, if
he is so obligated, then he ought to avoid getting the book lost in the mail.
57 See Zimmerman (1988a), pp. 40ff. and 65.
58 It might be thought that it was both probably best for Bruce to walk down Oak Street
and probably best for him to repay the money, so that each is something that he ought
to have done. But this is either to take an act-based, rather than world-based, approach
to obligation, or to take a world-based approach where the worlds in question are only
probably rather than actually accessible. Both approaches have already been rejected; see
Subsections 2.1.2 and 3.1.1 above.
59 Ross (1930), p. 42ff.
60 Ross (1939), p. 153ff.

109
Similarly, if Bruce is obligated to repay the money, then he is obligated to
avoid getting mugged. This is not to blame Bruce if he gets mugged, just
as it is not to blame the book-borrower if the book gets lost in the mail.
Nor is it to let the mugger, or the postal worker, off the hook. These latter
agents may well have done wrong themselves and may well be to blame for
what they have done. But, if either of these judgments about the mugger
and the postal worker is correct, their being so is wholly independent of the
judgments to be made about Bruce and the book-borrower.
In denying that obligation can be canceled, I have been arguing that
agents can sometimes be obligated in cases where many may think that they
are not. But uneasiness with my treatment of remote obligation and wrong-
doing may come, as it were, from the opposite direction. There are cases
where I am committed to saying that the agent is not obligated to perform
a certain action but where it might seem that we should say that he is obli-
gated to do so. For instance, I have said that Bruce's spending on Saturday
night renders it impossible for Bruce to repay the money on Sunday, so that
it is not true that he ought, from Saturday night on, to repay the money
on Sunday. But what if Bruce were luckily to stumble across some
extra money on Sunday morning? What if he were to win the lottery?
Surely then he ought, as of that moment, to repay the borrowed money
after all.
Here the correct response is simply this. Either Bruce can so act on
Saturday night that, despite spending the money then, he will stumble
across some extra money on Sunday morning, or he cannot. (See Sub-
section 2.2.3 above.) If he can, then his spending on Saturday night does
not after all render it impossible for him to repay the money on Sunday,
and so he still ought, after spending, to repay the money. If he cannot, then
this is a case of self-imposed impossibility, just as originally described and
intended.61 (Notice that it might be true in some sense that Bruce might win
the lottery even though he cannot, where the lottery is of the pre-imprint-
ed ticket variety. This just means that, for all that we or anyone else knows,
Bruce has the winning ticket.)
A final case is this. Suppose that Pru makes a promise that she cannot,
and never could, keep. (A variation on this: suppose that Pru misrepresents
herself - as a competent lifeguard, say.) My account implies that Pru
was never obligated to keep her promise, and hence that Pru committed
no wrong with respect to not keeping her promise. Surely this is counter-
intuitive.

61 Cf. a variation on this sort of case in Sinnott-Armstrong (1988), p. 119.

110
 Here I would reply that where Pru goes wrong is with respect to the ini-
tial act of misrepresenting herself (her intentions, her abilities). This is
avoidable and ought to be avoided. Sinnott-Armstrong agrees that Pru
ought not to make the promise but argues that it is also true that, having
made it, she ought to keep it, on the grounds that, if it were not true that
Pru ought to keep the promise, then there would be no reason why she
ought not to make it in the first place.62 Not so. There is very good reason
why she ought not to make the promise, even though it is not the case that
she ought to keep it, and that is that she cannot keep it. She is misrepresent-
ing herself; she is misleading others; she is creating false beliefs and certain-
to-be-frustrated expectations. Under the circumstances, this is (we may
assume) wrong.

3.2.5 The supersession of obligation

That obligation cannot be canceled does not imply that it cannot be super-
seded. Obviously it can be. Every obligation is eventually superseded, if
only because, if 5 ought at T to do A at T', T' will eventually arrive. The
interesting cases of supersession, however, are those involving remote
obligation, where the obligation at T to do A at T' is superseded prior to
r.
One thing that the previous discussion reveals is that obligation, even
when overall and not merely prima facie, can be overridden. Such overrid-
ing is dynamic and to be distinguished from the sort of static overriding
involved in one conditional obligation's overriding another and in one
prima facie obligation's overriding another (issues to be discussed in the
next two chapters). For example, Mandy's obligation to help Phil move
furniture is overridden by her obligation to babysit. In general we may say:
(3.32) S's (overall) obligation to do A at T" is overridden, at T i n W, by
his (overall) obligation to do B at T' iff
(a) 5 ought, just prior to Tin W, to do A at T';63
(b) it is not the case that S ought, just prior to Tin W, to do B at T';
(c) S ought, at Tin W, to do B at T;
(d) it is not the case that 5 ought, at Tin W, to do A at T'; and
(e) S can, at Tin W, do A at T.

62 Sinnott-Armstrong (1988), pp. 118-19.

63 By this I mean that, for any time T* prior to T such that it is not true that S ought, at
T* in W, to do A at T', there is a later time T** prior to T such that S ought, at T**
in W, to do A at T.

Ill
 T4

Figure 3.5

Notice four things. First, if S's obligation to do A at T' is overridden, it

nonetheless survives, in the sense that, if he does A at T", he at least satis-
fies that obligation. (As pointed out in the last section, however, this may
be no cause for celebration.)
Notice also that it is possible that an obligation that has been overridden
be restored, in the sense that the agent may once again become such that he
ought overall to satisfy it. For example, ifMandy were to murder Bill's child
on Saturday night, it might be that she ought then once again to help Phil
move furniture on Sunday. The situation might be pictured as in Figure 3.5
(where Tt is Friday, T2 is now Saturday noon, T3 is now Saturday night,
and T4 is Sunday, and where T, P, H, and B are to be understood as before
in Figure 3.2, with Mstanding for "murder"). It is in principle possible for
an obligation to be overridden and restored an indefinite number of times.
If this happens, though, each moment of overriding and restoration marks
a fresh turn for the worse by the agent.
Notice, thirdly, that, although in many cases of the overriding of an
obligation to do A by an obligation to do B, the agent 5 cannot do both A
and B, this is not necessarily the case and is not implied by (3.32). Although
it is true of Mandy (in that she cannot both help Phil move furniture and
babysit for Bill), it is not true of Gertrude in the following case. At Tt
Gertrude ought, from gratitude, to send Lucy lilies at T3; it is not the case
that she ought to send her daisies - they would be a poor second; and it is
112
emphatically not the case that she ought to send both lilies and daisies, since
this would greatly disturb Lucy (bringing back certain horrible childhood
memories). Suppose that, at T2, Gertrude spills weed killer on the lilies that
she had intended to deliver to Lucy at T3. This changes things. N o w she
ought to send Lucy daisies, for the lilies have shriveled. The obligation to
send lilies has been overridden by the obligation to send daisies. Gertrude
can, of course, still send both lilies and daisies, but this remains something
that she clearly ought not to do. 64
Notice, finally, that not all supersession of a remote obligation consists
in its being overridden. When Bruce's obligation to repay the money on
Sunday is superseded by his spending it on Saturday, the obligation is not
overridden; for it does not survive but is rather extinguished, in that he
cannot any longer satisfy it. So too with Betty's obligation to repay the
money; it is superseded by her throwing away the key on Saturday, but it
is extinguished rather than overridden. W e may in general say:

(3.33) S's obligation to do A at T' is extinguished at T i n W iff

(a) S ought, just prior to Tin W, to do A at T'; and
(b) either 5 cannot, at T in W, do A at T', or 5 cannot, at T in W,
refrain from doing A at T'.

Once extinguished, an obligation cannot be restored. Notice that the ful-

fillment or nonfulfillment of an immediate obligation trivially counts as its
extinction. But extinction can occur remotely, too, as in Bruce's case
(where his spending renders him unable to repay the money) and in Betty's
case (where her throwing away the key renders her unable to refrain from
repaying the money).

64 It might be objected that Gertrude cannot any longer send lilies that are in a suitable con-
dition, and that it is this obligation that has been overridden; thus it is true after all that,
when an obligation has been overridden by another, not both can be satisfied. This is
confused. I grant that Gertrude had an obligation at Tls one that she no longer has at T2,
to send unshriveled lilies; but an obligation to send unshriveled lilies entails an obligation
to send lilies (see (2.43'"))' a n d so my remarks stand.

113
 Conditional obligation

4.1 VARIETIES OF C O N D I T I O N S ON OBLIGATION

So far the focus of inquiry has been the conceptual structure of statements
of the form "5 ought [at Tin W] to do A [at T ' ] " and various related state-
ments. These are intended to be understood as statements of uncondition-
al overall moral obligation. But statements of conditional overall moral
obligation are just as common. These have the form "S ought to do A, if..."
It is the purpose of this chapter to provide an account of the conceptual
structure of such statements.
The most complete account of conditional obligation that has yet been
given is that provided by Fred Feldman.1 I shall not seek to reproduce
the full details of this account here but wish rather to build on its insights.
However, a brief overview of some of Feldman's findings is in order.
Feldman notes that statements of the form "S ought to do A, if..." are
ambiguous. He distinguishes four principal interpretations. The first inter-
pretation treats the statement as one involving material implication. Using
the symbolism introduced earlier, we may represent such statements as
having this form:

(4.1) p D O(A).

Feldman remarks on a number of features of such statements. Here I will

draw attention to just one, that which he calls "factual detachment."2 Such
detachment consists in the obligation's obtaining unconditionally when the
condition in question is satisfied. Statements of the form of (4.1) allow for
such detachment, in that, given both (4.1) and

(4.2) p,

1 Feldman (1986), Ch. 4.

2 He borrows this term from Greenspan (1975), Section 2.

114
we may of course infer

(4.3) O(A).
Feldman also remarks that he is doubtful that any sentence of ordinary
English in fact represents the assertion of a statement of the form of (4.1).3
This may be, but it seems to me that such statements as that mentioned by
Alan White and discussed in Subsection 3.1.3, namely,

(4.4) we ought to deal with this matter now, if we can,

are perhaps open to such interpretation. (Remember that (4.1) is equiva-

lent to
(4.1') ~pvO(A).

It seems to me that (4.4) might plausibly be recast as

(4.4') either we ought to deal with this matter now or [it's not the case that
we ought, because] we cannot.)

Nonetheless, it may be that such statements are really just subjunctive con-
ditionals in disguise.
The second interpretation treats statements of the form "S ought to do
A, if..." as involving strict implication. In symbols:

(4.5) p=>O(A)
or, equivalently,
(4.5') D ( ? D O(A)).
Here, too, factual detachment applies. And here there are, I think, many
clear uses of statements of this sort. For instance, each of the following
involves strict implication:

(2.37'c) if C(A) & C(~A) &~P(~A), then O(A);

(2.43'") if ~C{A & ~£) & C(~B), then, \£O{A), then O(£);

(2.44') if S ought to do A, then S ought so to act that S ought to do A.

In addition, as Feldman notes,4 a utilitarian will typically intend the fol-
lowing statement to be so understood:

(4.6) if it would maximize utility, then he ought to do it.

3 Feldman (1986), pp. 78-9.
4 Feldman (1986), p. 79.

115
Similarly, a Kantian will typically intend for the following statement to be
so understood:
(4.7) if it would violate the Categorical Imperative, then he ought not to
doit.
The third interpretation treats statements of the form "S ought to do A,
if..." as involving a subjunctive conditional. This may be symbolized as
(4.8) p > O(A)
and understood to say: "if it were the case that/?, then 5 would be obligat-
ed to do A " Clearly, factual detachment once again applies. Equally clear-
ly, many English sentences typically represent assertions of such statements.
Feldman gives this example:5
(4.9) if Smith were to win a million in the lottery, he would have an oblig-
ation to give generously to charity.
Another example would be this:
(4.10) if Tom is (or were) a crook, then you ought to vote for Dick.
Let us embellish this last example, for it will prove instructive. Suppose
that there are three candidates for election, Tom, Dick, and Harry, and that
they are to be ranked in that order.6 Let us therefore suppose that your
options are to be ranked as in Chart 4.1 (where T, D, and H stand for "vote
for Tom," "vote for Dick," and "vote for Harry," respectively). In that
case, what (4.10) amounts to is this. If Tom were a crook, then your options
would not be ranked as in Chart 4.1 but in some other way, such as in Chart
4.2. In other words, your options would be reordered, deontically.

1: T & ~D & ~H...

2: ~T & D & ~H...
3: ~T & ~D & H...
4: ~T & ~D & ~H...
Chart 4.1

1: ~T & D & ~H...

2:-7(5 ~D & H...
3:~T& ~D & ~H...
4: T & ~D & ~H...
Chart 4.2
5 Feldman (1986), p. 85.
6 The example is borrowedfromJackson and Pargetter (1987), pp. 77—8.

116
 Contrast this case with another. Suppose that
(4.11) if you don't vote for Tom, then you ought to vote for Dick.
It is clear that this is not to be analyzed in terms of a subjunctive conditional.
For (4.11) does not amount to this: if you don't (or weren't to) vote for
Tom, then your options will (or would be) reordered as in Chart 4.2.
What's at issue in (4.11) is quite different from what's at issue in (4.10);
whereas (4.10) warrants factual detachment, (4.11) does not. That is, sup-
pose that it turns out that Tom is a crook; then (4.10) implies that you ought
to vote for Dick. But suppose that you won't in fact vote for Tom; it would
not follow from (4.11) that you ought to vote for Dick. You ought to vote
for Tom, not Dick. To say that you ought to vote for Dick just because you
won't vote for Tom is to make things either impossibly difficult for you or
ridiculously easy. It makes things impossibly difficult if the claim that you
ought to vote for Tom is retained; for, on the assumption that you cannot
both vote for Tom and vote for Dick, you would then be faced with a moral
dilemma (and, whatever one thinks about the theoretical possibility of such
dilemmas, surely no one thinks that they are so easily generated - more
on this in Chapter 7). It makes things ridiculously easy if the claim that
you ought to vote for Tom is abandoned; for you cannot change the
fact that you ought to vote for Tom simply by not voting for him (com-
pare the objection concerning self-imposed impossibility discussed in
Chapter 3).
Insofar as (4.11) does not warrant factual detachment, it is to be distin-
guished not just from statements of the form of (4.8) but also, of course,
from statements of the form of (4.1) and (4.5). (4.11) concerns what I shall
call conditional obligation proper, and it is to the analysis of this that I now
turn.

4.2 THE ANALYSIS

There is no doubt that statements of conditional obligation proper play a
large role in pronouncements about morality. Examples abound:
(4.12) if Pru is going to fail to keep her promise, then she ought not to
make it in the first place (although she ought both to make it and
to keep it);
(4.13) if Charlie changes lanes, then he ought to accelerate (although he
ought neither to change lanes nor to accelerate);7

7 This example is borrowed from Goldman (1978), p. 186.

117
(4.14) if Audrey administers medicine B on Monday, then she ought to
administer it on Tuesday too (although she ought in fact to admin-
ister medicine A instead on both Monday and Tuesday);8
(4.15) if Sally sins, then she ought to repent (although she ought not to sin
in the first place);
(4.16) if Alan fails to attend the 10:00 A.M. meeting on the first floor, then
he ought to attend the 10:00 A.M. meeting on the second floor
(although what he ought to do is attend the meeting on the first
floor and not the one on the second);

and so on. In every case, we have a statement that (if used typically) is to be
understood as a statement of conditional obligation (proper).9
I propose (following the lead of others)10 that conditional obligation be
analyzed in a manner that parallels the analysis already given for uncondi-
tional obligation. The main gist of the latter was this: S ought to do A if and
only if (a) 5 can do A, (b) 5 can refrain from doing A, and (c) any accessi-
ble world in which S does not do A is deontically inferior to some accessi-
ble world in which S does A. More precisely:

(I) S ought, at T in W, to do A at V iff

(a) there is a world W' such that W' is accessible to 5 from Wr at T and 5
does A at T ' in W'\
(b) there is a world W" such that W" is accessible to S from P'Fat Tand
S does not do A at T ' in W"; and

8 This example is borrowed from Feldman (1986), p. 52.

9 Henceforth, I shall usually omit the designation "proper."
10 In particular that in Feldman (1986), Ch. 4. It is perhaps worth noting explicitly here
something to which Feldman pays scant attention: that statements of conditional oblig-
ation proper cannot be satisfactorily analyzed by means of a material conditional within
the scope of the obligation operator. For example, (4.11) cannot be satisfactorily
analyzed as follows (where T and D stand for "vote for Tom" and "vote for Dick,"
respectively):
O{~T D D).
For this statement is to be understood (see Subsection 2.3.1) as equivalent to the fol-
lowing:
O~(~T&~D);
and this follows directly from the truth of
O(T).
Yet it is clear that (4.11) doesn't follow simply from the fact that you ought to vote
for Tom. Compare the discussion of Ross's paradox in Subsection 2.3.2.

118
 (c) for all worlds W" such that W" is accessible to S from Wat Tand S
does not do A at T ' in H7", there is a world Pi7' such that
(1) W' is accessible to S from Wat T,
(2) S does ^4 at T ' in W7',
(3) the deontic value for S at T of H 7 ' is greater than the deontic value
for S at T o W , and
(4) there is no world W'" such that
(i) W'" is accessible to S from Wat T,
(ii) S does not do A at T ' in W7'", and
(iii) the deontic value for S at T of W"' is greater than the deon-
tic value for S at T of Wf.
Conditional obligation is to be analyzed in analogous fashion. I propose,
roughly, that S ought to do A, on the condition that p is true, if and only if
(a) S can do A while p is true, (b) S can refrain from doing A while p is true,
and (c) any accessible j?-world in which S does not do A is deontically
inferior to some accessible p-world in which S does A. More precisely:
(XIII) S ought, at T in W, to do A at T", on the condition that p is true
iff
(a) there is a world W' such that W' is accessible to 5 from Wat T
and S does A at T ' in W' and^ is true in W'\
(b) there is a world W" such that H/~" is accessible to S from WatT
and S does not do A at T ' in W " andj? is true in W"; and
(c) for all worlds W" such that H7"" is accessible to 5 from WatT and
5 does not do A at T' in W" avAp is true in W", there is a world
Wr such that
(1) W7"' is accessible to S from W at T,
(2) S does ,4 at T ' i n W7',
(3) p is true in W\
(4) the deontic value for S at T of W' is greater than the deontic
value for 5 at TofW", and
(5) there is no world W'" such that
(i) W'" is accessible to S from W at T,
(ii) 5 does not do A at T ' in H7"",
(iii) p is true in H7"'", and
(iv) the deontic value for S at T of W'" is greater than the
deontic value for S at TofW'.
Applied to (4.11) and Chart 4.1, what this analysis of conditional obliga-
tion tells us is that, given the restriction to ~ T-worlds, Option 2 is deontically
best. This is of course consistent with saying that Option 2 is not the best

119
option, that is, that the best world is not a ~T-world.11 Morever, it is of
course consistent with saying that, even if the actual world is a ~T-world,
the best world is not a D-world; hence, factual detachment is not validat-
ed. Thus we cannot validly infer an obligation to vote for Dick simply from
the fact that (4.11) is true and that you will in fact not vote for Tom.
Similar remarks pertain to the other examples. Concerning (4.12): the
best world accessible to Pru in which she fails to (keep a promise to) do A
is one in which she doesn't promise to do A; nonetheless, the best world
accessible to Pru is one in which she both promises to do A and does A.
Concerning (4.13): the best world accessible to Charlie in which he
changes lanes is one in which he accelerates; nonetheless, the best world
accessible to Charlie is one in which he neither changes lanes nor acceler-
ates. Concerning (4.14): the best world accessible to Audrey in which she
administers medicine B on Monday is one in which she administers it on
Tuesday also, although the best world accessible to her is one in which she
administers medicine A on both Monday and Tuesday. Concerning (4.15):
the best world accessible to Sally in which she sins is one in which she
repents, although the best world accessible to her is one in which she
doesn't sin. And concerning (4.16): the best world accessible to Alan in
which he fails to attend the meeting on the first floor is one in which he
attends the meeting on the second floor, although the best world accessi-
ble to him is one in which he attends the meeting on the first floor and not
the meeting on the second floor.
Siblings of (XIII) may be generated in a manner analogous to that in
which siblings of (I) were generated. For example, just as (XIII) is analo-
gous to (I), so the following is analogous to (II):
(XIV) 5 may, at T in W, do A at T', on the condition that p is true iff
[as in (XIII), except that there is no corresponding subclause (c5)
and subclause (c4) goes as follows:
the deontic value for 5 at TofW" is no greater than the deontic
value for S at T of W'.]
Mercifully, I shall refrain from giving explicit statements of the analogues
to (III)—(XII). Nonetheless, one further sibling of (XIII) deserves explicit
recognition, for it has no direct analogue among (I)-(XII). Consider this
statement:
(4.17) you ought to take that item with you only if you pay the shop-
keeper.
11 Such talk of a "best world" is rough. (See note 15 to Ch. 2.)

120
As typically uttered, this seems to me equivalent to the following:

(4.17') you ought not to take that item with you if you don't pay the
shopkeeper.

And so in general I propose the following:

(XV) 5 ought, at T in W, to do A at T", only on the condition that p is

true iff S ought, at T in W, not to do A at T", on the condition
thatp is not true (that is, on the condition that ~p is true).

Statements of biconditional obligation are of course possible. Consider this

example:

(4.18) you ought to go to the party if and only if you accept the
invitation.

This is equivalent to the conjunction of the following two statements:

(4.18a) you ought to go to the party if you accept the invitation;
(4.18b) you ought not to go to the party if you don't accept the
invitation;
and it leaves open whether you ought to go, ought not to go, or neither
ought to go nor ought not to go (and so, too, with respect to accepting).
Three other features of conditional obligation proper should be noted.
First, in addition to its failure to validate factual detachment, conditional
obligation also fails to validate what Feldman calls "augmentation."12 That
is, from the truth of

(4.19) 5 ought to do A, on the condition that p is true

we may not in general infer

(4.20) S ought to do A, on the condition that p & q is true.

You ought to vote for Dick if you don't vote for Tom; but you ought to
vote for Harry if you don't vote for Tom and you don't vote for Dick.
Secondly, however, conditional obligation does validate a restricted
form of detachment that may be called "necessity detachment."13 That is,
from the truth of (4.19) and that of

(4.21) S cannot so act that p is not true,

12 Feldman (1986), p. 91.

13 See Feldman (1986), pp. 90-1; Greenspan (1975), p. 265.
121
we may indeed infer the truth of
(4.3') 5 ought to do A.
For if S cannot so act thatp is not true, then all worlds accessible to S arep-
worlds, and so the deontically best worlds accessible to S are p-worlds; thus,
if the best accessible p- worlds are yl-worlds, then the best accessible worlds
are y4-worlds. Suppose, for example, that Charlie cannot avoid changing
lanes (perhaps because some other car is physically forcing him to do so).
In that case, it is inaccurate to state that he ought neither to change lanes
nor to accelerate. On the contrary, given that
(4.13') if Charlie changes lanes, then he ought to accelerate,
we may infer
(4.22) Charlie ought to accelerate.
Finally, conditional obligation also validates what Feldman calls
"deontic detachment."14 That is, from the truth of (4.19) and
(4.23) 5 ought so to act that p is true,
we may indeed infer (4.3'). For if S ought so to act that p is true, then the
deontically best worlds accessible to 5 arep-worlds; and so, again, if the best
accessible p-worlds are y4-worlds, then the best accessible worlds are A-
worlds. Suppose, for example, that you ought to accept the invitation; then
from (4.18a) we may infer that you ought to go to the party. On the other
hand, if you ought not to accept the invitation, then from (4.18b) we may
infer that you ought not to go to the party.

4.3 CERTAIN DEONTIC PARADOXES

Various so-called deontic paradoxes have already been discussed. In
Subsection 2.3.2 there was discussion of those paradoxes known as Ross's
Paradox, the Good Samaritan Paradox, and the Paradox of the Knower; in
Chapter 3 a paradox concerning self-imposed impossibility (namely, that
such impossibility can rid one of certain obligations) was addressed.15 All
the other standard paradoxes have to do with conditional obligation, and
so I propose now to address some prominent examples.16

14 Feldman (1986), p. 91. Again, he is here following Greenspan (1975), Section 2.

15 Here the paradox known as the Suzy Mae Paradox is relevant. See note 39 to Ch. 3.
16 Much of what follows is handled in a similar fashion by Feldman in Feldman (1986),
(1989a), (1989b), and (1990).

122
 Let me begin, however, with a type of case that, as far as I know, has not
received discussion before. Suppose it is true that
(4.24) you ought to pay the shopkeeper if you take that item with you.
According to certain accounts of the logic of "ought," 7 this is equivalent
to its being the case that
(4.17') you ought not to take that item with you ifyou don't pay the shop-
keeper.
Such contraposition may seem unobjectionable at first, and yet it seems on
inspection that (4.24) can be true while (4.17') is false. Consider a case
where the item in question is food which your child desperately needs and
for which you can afford to pay. We may assume that (4.24) is true. Might
it not be, though, that, because of your child's need, you ought to take the
food even if you don't pay for it? If so, (4.17r) is false (as an expression of
overall obligation; it may be true as an expression of primafacie obligation).
The account of conditional obligation that I have given allows for the
possibility that (4.24) be true but (4.17') false. Suppose that your options
are to be ranked as in Chart 4.3 (where T stands for "take" and P for
"pay"). In this case, the following are all true:
(4.25) you ought to take that item with you;
(4.26) you ought to pay the shopkeeper;
(4.24) you ought to pay the shopkeeper if you take that item with you;
(4.27) you ought to take that item with you if you pay the shopkeeper.
But true, too, is the following:
(4.28) you ought to take that item with you (even) if you don't pay the
shopkeeper,
and, correspondingly, (4.17') is false.
Since (4.17') is false, so too is its equivalent
(4.17) you ought to take that item with you only if you pay the shop-
keeper.
1: T & P ...
2: T&-P...
3: ~T & P ...
4: ~T & ~P ...
Chart 4.3
17 For example, those that would symbolize (4.24) in some such way as O(T D P), where
D signifies a material conditional. Compare note 10 to this chapter.

123
And since (4.27) and (4.28) are true, so too are their equivalents:
(4.27') you ought not to take that item with you only if you don't pay the
shopkeeper;
(4.28') you ought not to take that item with you only if you do pay the
shopkeeper.
But I confess that (4.27') and (4.28') would be a pretty odd way of making
one's point. Still, (4.27) and (4.28) are somewhat odd too, and they're odd
because (4.25), a statement of straightforwardly unconditional obligation,
is true. When such statements are true, then statements of conditional
obligation, such as (4.27) and (4.28) and their equivalents, are generally
otiose, even if true.
At any rate, what it is important to note here is that, although (4.24)
is true, (4.17) is not. Hence it would have been a mistake to offer the
following analysis:
(4.29) S ought, at T in W, to do A at T", only on the condition thatp is
true iff S ought, at T in W, so to act that p is true, on the condi-
tion that S does A at T'.
We might symbolize this proposal as follows: O (A only ifp) iff O(p if A).
This is to be rejected. What is to be accepted is (XV), which states: O(A
only if;?) iff O(~A i£~p). (4.29) and (XV) are not equivalent.
The remainder of the paradoxes to be addressed all have to do with the
matter of factual detachment. Hector-Neri Castaneda has introduced what
he calls the Paradox of the Second Best Plan.18 In this context, consider
Audrey again, of whom it was said:
(4.14) if Audrey administers medicine B on Monday, then she ought to
administer it on Tuesday too (although she ought in fact to admin-
ister medicine A instead on both Monday and Tuesday).
Suppose it happens to be true that
(4.30) Audrey will administer medicine JB on Monday.
In that case, don't we wish to say that

(4.31) Audrey ought to administer medicine B on Tuesday?

But how can we say both this and something else that is implied19 by (4.14),
namely,
(4.32) Audrey ought to administer medicine A on Tuesday?

18 Castaneda (1981), pp. 58-60.

19 Bywayof(2.40 / c).
124
The answer is that we may not infer (4.31) from (4.14) and (4.30); such fac-
tual detachment is not warranted. Of course, once it has happened that
Audrey has administered medicine B on Monday, it may well then be true
that she ought to administer it on Tuesday also. This is just one more
instance of the sort of shift in obligations that was the topic of discussion in
the last chapter. But prior to Monday it will not be true to say that Audrey
ought then to administer medicine B on Tuesday, even if it is true then to
say that she will administer it on Monday.
In the case just discussed, there is a sense in which (4.31), though not
true on Sunday, becomes true on Monday (once medicine B has been
administered). ° But cases of this sort are possible where there is no time for
such a shift in obligations, and hence no time at which a statement such as
(4.31) becomes true.21 Consider again

(4.13) if Charlie changes lanes, then he ought to accelerate (although he

ought neither to change lanes nor to accelerate).
Suppose that the changing of lanes and the acceleration, were either or both
to occur, would occur at the same moment in time. In that case, even if it
is true that
(4.33) Charlie will change lanes,
it is not true, at any time, that
(4.22) Charlie ought to accelerate.
For at no time does the best world accessible to Charlie contain his accel-
erating. Charlie's options may be ranked as in Chart 4.4 (where C stands
for "change" and A for "accelerate"). At no time is Charlie faced with a set
of options where A is featured in the top-ranked course of action.

1: ~C & ~A
2: C & A
3: C & ~A
4: ~C & A
Chart 4.4

20 Actually, this seems to me strictly false. What I would prefer to say is that (4.31) is ambigu-
ous (depending on what the time of obligation is that is at issue) and that on one reading
(where the time in question is Sunday) it is and will always remain false, whereas on
another reading (where the time in question is Monday, after medicine B has been
administered for the first time) it is and will always be true.
21 Cf. Feldman (1990), p. 338, n. 23.

125
 Let me return now to the Paradox of the Knower, first discussed in
Subsection 2.3.2. This paradox may now be presented more perspicuous-
ly, in light of the analysis of conditional obligation provided in this chap-
ter. Recall Greg, the security guard. We may now understand him to have
the following conditional obligation: to provide a truthful report of his per-
forming certain (wrongful) actions, if he does perform them. Consider one
such action, A. Thus:

(4.34) Greg ought to provide a truthful report of his doing A, if he does

A.

The problem here is this. Suppose that

(4.35) Greg will in fact do A.

Then, according to certain accounts of the logic of "ought,"

(4.36) Greg ought to provide a truthful report of his doing A.

But his providing a truthful report of his doing A entails his doing A and,
given that he ought to provide such a report, certain accounts of the logic
of "ought" (including mine: see (2.43"')) imply that

(4.37) Greg ought to do A.

But how can it be that he ought to do A when A is wrong?

The resolution of this problem may now be stated simply. Since condi-
tional obligation does not warrant factual detachment, we cannot infer
(4.36) from (4.34) conjoined with (4.35). For although it may be that the
best v4-worlds accessible to Greg are worlds in which he provides a truth-
ful report of his doing A, still the best worlds accessible to him are worlds
in which he does not do A (and so also does not provide a truthful report
of his doing so).
The so-called Paradox of Reparation may be handled in exactly analo-
gous fashion. Recall that

(4.15) if Sally sins, then she ought to repent...

Suppose also that

(4.38) Sally will sin.

Some may infer that therefore

(4.39) Sally ought to repent.

126
But let us suppose that repentance for sinning entails having sinned (at least
when the repentance is sincere). Surely we don't wish to infer that
(4.40) Sally ought to sin.
But, of course, we are in no danger of being committed to (4.40), and that
is because the move from (4.15) and (4.38) to (4.39) is invalid. Such factu-
al detachment is unwarranted.
The so-called Paradox of Gentle Murder may be treated in the same
manner.23 Suppose that
(4.41) if Murphy murders Vic, then he ought to do it gently,
and that
(4.42) Murphy will murder Vic.
Some may infer that therefore
(4.43) Murphy ought to murder Vic gently.
But Murphy's murdering Vic gently entails his murdering Vic. Surely we
don't wish to infer that

(4.44) Murphy ought to murder Vic.

And of course we shouldn't. The move from (4.41) and (4.42) to (4.43) is
invalid.
Finally, a much-discussed paradox, concerning so-called contrary-to-
duty imperatives and first introduced by Roderick Chisholm, is subject to
similar resolution.24 Suppose that:
(4.45) Aida ought to go to the aid of her neighbors;
(4.46) if Aida goes to the aid of her neighbors, then she ought to tell them
that she's coming;
(4.47) if Aida doesn't go to the aid of her neighbors, then she ought not
to tell them that she's coming;
and
(4.48) Aida won't go to the aid of her neighbors.

22 Actually, this entailment appears false. Sincere repentance would seem at most to entail
that one believe that one has sinned.
23 This paradox was introduced in Forrester (1984). The treatment that I give is adequate
also to variations on the paradox presented in Goble (1991).
24 Chisholm (1963). The issue is there couched in terms of "ought-to-be."

127
Some may infer from (4.45) and (4.46) that
(4.49) Aida ought to tell her neighbors that she's coming.
But some may also infer from (4.47) and (4.48) that

(4.50) Aida ought not to tell her neighbors that she's coming.
Yet how could (4.49) and (4.50) both be true? As before, though, there is
no danger of being committed to the joint truth of (4.49) and (4.50).
Whereas the inference of the former is indeed valid (being an instance of
deontic detachment), the inference of the latter is not (being an instance of
factual detachment).
Castaneda has expressed dissatisfaction with this treatment of
Chisholm's Paradox, in that he thinks that the joint statement of (4.48) and
(4.49) is itself paradoxical.25 Feldman has replied that, although the joint
assertion of (4.48) and (4.49) might seem odd in the absence of the explic-
it insertion of (4.45), it is nonetheless true.26 Castaneda has expressed dis-
satisfaction with this reply, in that he contends that what makes the joint
assertion of (4.48) and (4.49) paradoxical is this. Suppose that Aida is intent
on not going to the aid of her neighbors but is otherwise keen to minimize
her wrongdoing. Thinking along Feldman-type lines (lines that I have fol-
lowed in this chapter), she therefore decides to satisfy the obligation alleged
in (4.49) and tells her neighbors that she's coming. But this, far from min-
imizing her wrongdoing, compounds it.27 Feldman has responded to this crit-
icism, too, in a manner that seems to me appropriate,28 but he overlooks
what I take to be the most effective response to it. This response makes
reference to subsidiary obligation, a matter to which I now turn.

4.4 SUBSIDIARY O B L I G A T I O N

4.4.1 Factual detachment

You may be uneasy with the solutions just proposed to the paradoxes con-
sidered. Should we dismiss factual detachment so summarily in the case of
conditional obligation (proper)? Should we really deny that

(4.22) Charlie ought to accelerate,

(4.36) Greg ought to provide a truthful report of his doing A,

25 Castaneda (1989a), p. 9ff.

26 Feldman (1989b), p. 35.
27 Castaneda (1989b), p. 167.
28 Feldman (1990), p. 331ff.

128
(4.39) Sally ought to repent,

(4.50) Aida ought not to tell her neighbors that she's coming?
If we do deny these statements, what then really is the point of making the
corresponding statements of conditional obligation, namely,
(4.13) if Charlie changes lanes, then he ought to accelerate.
(4.34) Greg ought to provide a truthful report of his doing A, if he does
A,
(4.15) if Sally sins, then she ought to repent...,
and
(4.47) if Aida doesn't go to the aid of her neighbors, then she ought not
to tell them that she's coming?
It is of course true that detachment of an unconditional obligation, when
the condition of the corresponding conditional obligation is satisfied, is
sometimes warranted. We have already seen that both necessity detachment
and deontic detachment are valid. Other such restricted detachment rules
may seem plausible. As a variation on necessity detachment, it may seem
that, in cases where the action that is conditionally obligatory is "sched-
uled" to occur after the condition in question is "scheduled" for satisfac-
tion, if and when the condition is satisfied, the action then becomes
obligatory; that is, it may seem that, if 5 ought at T1 to do A at T3, on the
condition that some event E occurs at T2, and if E occurs at T2, then S
ought at T2 to do A at T3. For example, if Audrey ought on Sunday to
administer medicine B on Tuesday, on the condition that she administers
it on Monday, and if she does administer it on Monday, then she ought on
Monday (once medicine B has been administered) to administer it on
Tuesday. But in fact this does not hold without exception. I said in Section
4.3 that it "may well" be that Audrey ought on Monday to administer med-
icine B on Tuesday, but I didn't say that this was certain. For suppose that,
while administering the medicine on Monday, Audrey also gives her
patient a lethal injection; presumably it would not then be true that she
ought on Monday to administer the medicine again on Tuesday. (This is
another illustration of the failure of augmentation for conditional obliga-
tion.) The situation may be represented as in Chart 4.5 (where AM, ATi BM,
BT, and LM stand for "administer medicine A on Monday," "administer
medicine A on Tuesday," "administer medicine B on Monday," "admin-

129
 1: AM& ~BM & ~LM & AT& ~BT ...
2: ~AM & BM & ~LM & ~AT & BT ...
3a: AM & BM & ~LM & AT & BT ...
3b: [all other combinations with ~/_M
4a: ~AM & BM & LM & ~AT & BT ...
4b: [all other combinations with LM]
Chart 4.5

ister medicine B on Tuesday," and "give a lethal injection on Monday,"

respectively). Suppose that these are Audrey's options as of Sunday. If she
takes Track 2, it will indeed be true that she ought on Monday (once she
has administered medicine B) to administer medicine B on Tuesday. But
if she takes Track 4a, it will not be true that she ought on Monday to admin-
ister medicine B on Tuesday.
Mark Vorobej has offered the following restricted detachment rule (a
variant on the rule of deontic detachment): when the condition in ques-
tion is satisfied and its nonsatisfaction is not obligatory (rather than, as
before, its satisfaction being obligatory), then the obligation in question
may be detached.29 This, too, is unacceptable. Suppose again that

(4.18a) you ought to go to the party if you accept the invitation.

But suppose also that
(4.51) it is not the case that you ought to accept the invitation,
(4.52) it is not the case that you ought not to accept the invitation,
and
(4.53) you don't accept the invitation.
Given (4.51), we surely do not want to say that
(4.54) you ought to go to the party,
and yet Vorobej's proposal implies that (4.54) follows from (4.18a), (4.52),
and (4.53).
Are there other restricted detachment rules that are acceptable? Quite
possibly. But the fact remains that the sort of unrestricted detachment that
has here been called factual detachment is not valid for conditional obliga-
tion (proper), given my analysis of such obligation. And, I have argued, this
is just as it should be. After all, it simply is not true that you ought to vote

29 Vorobej (1986), p. 20.

130
for Dick; for Tom is the best candidate. Moreover, rejection of factual
detachment allows for a straightforward solution to many of the deontic
paradoxes.
Nonetheless, there is still the undeniable inclination to say that, in some
sense, you ought to vote for Dick because you won't vote for Tom; that
Charlie ought to accelerate because he will change lanes; that Sally ought to
repent because she will sin; that Aida ought not to tell her neighbors that
she's coming because she won't go to their aid; that Alan ought to attend the
meeting on the second floor because he won't attend the meeting on the first
floor; and so on. I think this inclination is correct. Although I have just reject-
ed factual detachment for conditional obligation, I want now to suggest that
there is a sense in which such detachment is acceptable after all. We can
have our cake and eat it too, as long as we recognize that the sort of oblig-
ation that is detached is, while of course unconditional and overall,
nonetheless subsidiary.

4.4.2 Levels of obligation

The phenomenon that I have in mind has been noted by others. In partic-
ular, Michael McKinsey has attempted to account for it.30 He has talked of
"levels of obligation" and said that obligations may be "primary," "sec-
ondary," "tertiary," or indeed "«-ary" (where n > 1). McKinsey is con-
cerned with just the sort of cases that we are concerned with. The meeting
case brings this out well. In this case, McKinsey wants to say, Alan oughtt
to attend the 10:00 A.M. meeting on the first floor; given that he fails to do
this, he ought2 to attend the 10:00 A.M. meeting on the second floor; and
so on. Here "ought/' expresses a primary obligation and "ought2" a sec-
ondary obligation. Concerning such levels of obligation, McKinsey says:
By saying that an obligation is secondary (or tertiary, or...), I do not mean that
it is any less of an obligation than a primary one. In my view it is just as incum-
bent upon a person to fulfill his secondary obligations, as it is incumbent upon
him to fulfill his primary ones. For notice that if a person does not do his best,
we will blame him also for not doing his second best, just as we blame him for
not doing his best.31
This is both puzzling and promising. It is puzzling, in that it is not clear
what it means to say that a non—primary obligation is "just as incumbent"
as a primary one. Such talk seems to make things either impossibly difficult

30 McKinsey (1975).
31 McKinsey (1975), p. 391.

131
or ridiculously easy, in the manner indicated earlier in Section 4.1. Yet
McKinsey's proposal is nevertheless promising, for he has latched onto an
important fact, and that is that, if one fails to do what one ought to do but
does what is "second best," one does less wrong32 than if one also fails to do
what is "second best"; and one who fails to do what is "third best" does still
more wrong; and so on. Alan does wrong if he fails to attend the 10:00 A.M.
meeting on the first floor; but he does more wrong still if he also fails
to attend the 10:00 A.M. meeting on the second floor (and this despite the
fact that it is never the case that he ought - in the sense analyzed in (I) -
to attend the meeting on the second floor).33 It should further be noted,
however, that, if Alan attends the meeting on the first floor, he does no
wrong — he violates no obligation — in not attending the meeting on the
second floor.
We have noted that, if Alan fails to attend both meetings, he does a dou-
ble wrong, even though he violates only a single obligation (in the sense of
(I)). And this is clearly connected with the fact that he has a conditional
obligation to attend the meeting on the second floor if he fails to attend the
meeting on the first floor. I propose that we say that Alan violates a prima-
ry obligation to attend the meeting on the first floor and a secondary oblig-
ation to attend the meeting on the second floor. (I) captures only the
concept of primary obligation; it is only with this concept that we have
hitherto been working. We need an account of non-primary obligation -
that is, subsidiary obligation.
For this purpose, the simplest assumption that one can make is the fol-
lowing: all worlds accessible to the agent are comparable in deontic value
and at least one of these worlds is such that no other of them is more valu-
able than it. Given this assumption, which I shall call the Simple
Assumption, we may say that some world (or worlds) accessible to the agent
at a certain time has (or have) deontic value-rank 1 (relative to the agent at
that time) and all other accessible worlds have a corresponding subordinate
deontic value-rank n (where n is some positive integer greater than 1). It is
on the basis of this assumption that the various diagrams and charts so far
provided have been drawn up.34
Working with this assumption, we may modify the accounts of uncon-
ditional and conditional obligation that have been given. First, we should

32 I would prefer to put it in this way rather than, as McKinsey does, in terms of blame.
See Zimmerman (1988a), pp. 41-3.
33 This is assured by the fact that both meetings occur at 10:00 A.M.; hence there is no time
for a shift in obligations of the sort discussed in Ch. 3
34 Again, see note 15 to Ch. 2.

132
explicitly acknowledge that the account of unconditional obligation that
has been provided is an account of unconditional primary obligation by
modifying (I) as follows:

(I') S ought!, at T in W, to do A at T iff

[as in (I), but with the following addition to clause (c):
and (5) W* has deontic value-rank I.] 35

W e may then assign levels to conditional obligation by modifying (XIII) as

follows:

(XIII') S oughtM, at T in W, to do A at T', on the condition that p is true

iff
[as in (XIII), but with the following addition to clause (c):
and (6) W' has deontic value-rank n.]

And then I propose that we allow for unrestricted (that is, factual) detach-
ment of subsidiary obligation by providing this account of such obligation
(where n is greater than 1):

(XVI) S ought n , at T in W, to do A at T' iff there is a proposition^ such

that
(a) S oughtn, at T in W, to do A at T', on the condition that p is true;
and
(b) p is true in W.

Given (XVI), we may indeed say that you ought to vote for Dick because
you won't vote for Tom; that Alan ought to attend the meeting on the sec-
ond floor because he won't attend the meeting on the first floor; and so on.
These statements are true, given that they constitute statements of secondary
obligation; but they remain false if they are intended as statements of pri-
mary obligation. This is how we can have our cake and eat it too. You
ought! to vote for Tom, and so it is not the case that you ought! to vote for
Dick; otherwise things would be impossibly difficult or ridiculously easy.
Nonetheless, inasmuch as you won't vote for Tom, you ought 2 to vote for
Dick. Ifyou don't satisfy this subsidiary obligation, you commit a subsidiary
wrong. In this way, if you vote for neither Tom nor Dick, you do a double
wrong; you compound the primary wrongdoing involved in not voting for
Tom.

35 Given the Simple Assumption, this addition is strictly redundant, for every accessible
world in which S does not do A at T" must be less than maximally valuable. But I sup-
ply the addition for the sake of perspicuity.

133
 It is clear that wrongdoing may be compounded still further. If you vote
for neither Tom nor Dick nor Harry, you commit a primary and a sec-
ondary and a tertiary wrong. In principle, one may pile subsidiary wrong
upon subsidiary wrong ad indejinitum. Perhaps Alan oughtt to attend the
meeting on the first floor, ought2 to attend the meeting on the second floor
if he fails to attend the meeting on the first floor, ought3 to attend the meet-
ing on the third floor if he fails to attend either meeting on the first and sec-
ond floors, and so on for as many floors and meetings as you like. It is clear
that Alan has the opportunity to be a very naughty boy indeed. But note
that (XVI) also implies, quite correctly, that there is no naughtiness, no
wrongdoing, if Alan fails to attend any of the higher-floor meetings but does
attend the first-floor meeting.
At this point we may make what I take to be the most effective response
to Castaneda's concern about the Chisholm case. He said that it seems to
be a mistake to claim that

(4.49) Aida ought to tell her neighbors that she's coming

when it is also true that

(4.48) Aida won't go to the aid of her neighbors;

for, ifAida were to try to minimize her wrongdoing (short of making (4.48)
false) by satisfying the obligation at issue in (4.49), she would, paradoxical-
ly, compound it instead. It is clear that I can now agree that Aida would be
compounding her wrongdoing, for it is true that

(4.47) if Aida doesn't go to the aid of her neighbors, then she ought not
to tell them that she's coming.

(4.48) and (4.47) jointly yield the following subsidiary obligation:

(4.50') Aida ought not to tell her neighbors that she's coming.

And Aida violates this obligation if she satisfies the (implicitly primary)
obligation at issue in (4.49). It is worth repeating here a point made in
Subsection 3.2.3: we should not necessarily be impressed if someone does
something that he ought (that is, oughtj) to do; for some ways of failing to
do what one ought to do are superior to some ways of doing what one
ought to do.
All of this has been predicated on the Simple Assumption, though, and
it may be that, on some substantive accounts of obligation, this assumption
is false. Such falsity will be due to one or other of two factors: either not all
accessible worlds are comparable in deontic value, or no accessible world
134
has a maximal deontic value.36 If either possibility were to arise, then fully
specific judgments of subsidiary obligation would appear to be impossible.
Although the accounts of unconditional and conditional obligation given
in (I) and (XIII) would still stand, either possibility would render a fully spe-
cific deontic ranking of all accessible worlds unachievable, so that (XIII')
and (XVI) would be inapplicable as originally intended. Nonetheless, even
if we therefore couldn't identify a certain wrong as secondary or tertiary,
and so forth, there would still be sense to the claim that, in doing such-and-
such, a certain agent has compounded his wrongdoing. One way to make
such a judgment would be arbitrarily to assign the rank of 1 to some acces-
sible world, then rank other deontically inferior worlds accordingly, and
then apply (XIII') and (XVI), thereby identifying certain wrongdoings as
relatively secondary, or tertiary, and so forth. This would still give some
measure of the extent of wrongdoing committed by an agent, although the
judgment would be merely a relative one, either because some deontical-
ly incomparable worlds will have been ignored or because, given that there
is no accessible world with a maximal deontic value, the agent will, in an
absolute sense, have done infinite wrong.

4.4.3 Overriding
In Subsection 3.2.5 a type of dynamic overriding of one (unconditional)
overall obligation by another was explicated by means of (3.32). N o w that
levels of overall obligation have been distinguished, a different, though
related, type of overriding - a static type - may be explicated. We may say
that one conditional overall obligation overrides another if and only if the
former is of a higher rank than the latter. Or more precisely:

(4.55) S's (overall) obligation^, at T in W, to do A at T", on the condition

that p is true, is overridden by S's (overall) obligationm, at T in W,
to do B at T', on the condition that q is true, iff
(a) 5 oughttt, at T in W, to do A at T', on the condition that p is true;
(b) S oughtm, at T in W, to do B at T', on the condition that q is true;
and
(c) m is less than n.
A related account of one subsidiary obligation's overriding another may, of
course, also be given:
(4.56) S's (overall) obligation^, at T in W, to do A at T" is overridden by
S's (overall) obligationw, at T in W, to do B at V iff
36 See the discussion of both possibilities in Subsection 2.2.6.

135
 (a) 5 oughtM, at T in W, to do A at T';
(b) S oughtw, at T in W, to do B at T'; and
(c) m is less than n.

Note that, if a subsidiary obligation to do A is overridden by a subsidiary

obligation to do B, then the latter, overriding obligation is not satisfied. For
the former, overridden, lower-level obligation "kicks in," as explained
above, only if the higher-level obligation is violated.
It may be that the concept of overriding just accounted for is a rather
technical concept, in that talk of "overriding" is more commonly applied
to the sort of dynamic overriding accounted for in (3.32) and to the sort of
static overriding of prima facie obligation to be accounted for in Subsection
5.3.4. Nonetheless, I think that it may on occasion happen that the termi-
nology is applied to what has been accounted for here, and so it is desirable
to have such an account. Some might say, for instance, that, although there
is admittedly an obligation (of a sort) to vote for Harry, it is overridden by
an obligation to vote for Dick, which is itself overridden by an obligation
to vote for Tom. This may be understood in the manner just indicated.
It is worth noting that, as in the case of dynamic overriding accounted
for in (3.32), the sort of static overriding of an obligation to do A by an
obligation to do B accounted for in (4.55) and (4.56) doesn't require that
the agent be unable to do both A and B. The voting case is most naturally
understood as a case in which you cannot vote for more than one candi-
date, but it needn't be understood in this way. It may be that you can vote
for more than one, but any such vote would be deontically inferior to a
single vote or no vote at all. In that case, although the obligation to vote
for Harry is overridden by the obligation to vote for Dick, it is nonetheless
possible for you to both vote for Harry and vote for Dick.

4.4.4 A rival account

The only other account of subsidiary obligation of which I am aware is the
one provided by McKinsey. It too is explicitly predicated on the Simple
Assumption. 37 His account is at odds with mine both with respect to what
counts as a subsidiary obligation and with respect to what does not.

37 See McKinsey (1975), pp. 391-2. Greenspan (in Greenspan (1975), p. 266f., and (1978),
p. 81) apparently thinks that subsidiary obligation is merely conditional obligation.
Clearly, I believe this to be mistaken. Goldman (in Goldman, 1978, p. 211, n. 3)
mentions this issue, but she reserves judgment and provides no account of levels of
obligation.
136
 What McKinsey says is this:

(4.57) x oughtM at t to do Ax if and only if

(1) A{ is contained in every <f>xt of rank n; and
(2) for every (j>xt which has a rank m higher than n (i.e., where m < n),
there is an Aj such that cf>xt contains A- and x will not do Aj.3S

Here x is an agent, t a time, Ai and Aj are actions performed at times t{ and

tj, respectively, and <j>xt is a "life-sequence" of actions open to x at t.
The case of Alan and the meetings can be used to show that McKinsey's
account is more liberal than mine. The case may be partially represented as
in Chart 4.6 (where M1 stands for "attend the meeting on the first floor"
and M2 for "attend the meeting on the second floor"). Here it is assumed
that just one world has rank 1 and just one world has rank 2. We may of
course assume that there are lower-ranked accessible worlds in which M,
occurs or in which M2 occurs, but there are none in which both occur. My
account of subsidiary obligation implies that Alan oughtj to attend the
meeting on the first floor and, given that he does not do this, ought2 to
attend the meeting on the second floor. My account also implies that Alan
ought j not to attend the meeting on the second floor. But it does not imply
that Alan ought2 not to attend the meeting on the first floor, given that
he goes to the meeting on the second floor. (This would require that M^
and M2 be jointly personally possible for Alan, and they're not.) But
McKinsey's account does imply this. Such a claim strikes me as incorrect,
indeed bizarre.
McKinsey's account is also more conservative than mine. Consider
a (rather strange) variation on the paying-the-shopkeeper case that may
be represented as in Chart 4.7. Here there is only one top-ranked world

1: M1 & ~M2 ...

2: ~M1 & M2 ...
Chart 4.6

1: T& P...
2a: T & ~P...
2b: ~T & P...
3: ~T & ~P...
Chart 4.7

38 McKinsey (1975), p. 392.

137
but there are two second-ranked worlds. My account warrants saying
all of the following:
(4.25') you ought! to take that item with you;
(4.26') you ough^ to pay the shopkeeper;
(4.28") you ought2 to take that item with you if you don't pay the shop-
keeper;
and
(4.58) you ought2 to pay the shopkeeper if you don't take that item with
you.
In short, it is best for you to take and pay, but doing either one is prefer-
able to doing neither, although (rather strangely) which one you do, if you
don't do both, is a matter of indifference. Further, my account implies that,
given that the relevant condition is satisfied in either (4.28') or (4.58), one
or other of the following would be true:
(4.59) you ought2 to take that item with you;
(4.60) you ought2 to pay the shopkeeper.
In this way, if you take Track 3, you will quite properly be said to have
done wrong on two levels. Yet McKinsey's account allows us to say neither
of (4.59) and (4.60), since in each case clause (1) of his account, (4.57), is
violated.
It is noteworthy that McKinsey seems not to be especially concerned
with conditional obligation and detachment when giving his account of
levels of obligation.39 Perhaps it is this lack of concern that has, at least as I
see it, led him astray. For the two are intimately related, as I have sought to
show.

4.4.5 Level of wrongdoing and seriousness of wrongdoing

It might be thought that my proposal is unnecessary. Why trouble ourselves
with levels of obligation when all we need do is note that, the further one
departs from Track 1, the more serious the wrong that one does? Consider
again the voting case, represented in Chart 4.1. Is there really any need to
insist that you commit wrong on two levels if you take Track 3? Two
wrongs, yes: ~T, and H. But why talk of two levels? Doesn't this multiply
wrongdoing beyond necessity, in that, in addition to the primary wrongs
committed in the performance of ~T and H, it is being claimed that you
39 He does mention this issue in a footnote (McKinsey, 1975, p. 395, n. 10), but it is not a
matter that occupies his attention in the rest of his paper.

138
commit a secondary wrong by performing ~D? Can't we express all that
we need to express by noting the (primary) wrongs involved and also not-
ing that Track 3 is a greater departure from Track 1 than Track 2 would
have been?
I think not. While I have no wish to deny that taking Track 3 is more
seriously wrong than taking Track 2, we must still contend with the incli-
nation to infer that you really ought to vote for Dick, given that you won't
vote for Tom. My proposal accounts for this; the simpler counterproposal
does not.
As to seriousness of wrongdoing, it should be observed that, on those
substantive accounts according to which the deontic ranking of options
reflects some value that is specifiable independently of its relation to mat-
ters of obligation,40 simply noting that taking Track 3 is more seriously
wrong than taking Track 2 does not tell the whole story. Suppose that
Track 1 has a value of 100, Track 2 a value of 90, and Track 3 a value of 10
(these numbers are of course purely ad hoc). Then taking Track 3 is very
much more wrong than taking Track 2, a fact not revealed by their relative
rankings alone. Nonetheless, it remains true that, in taking Track 3, you do
wrong on two levels.
Level of wrongdoing is to be distinguished from seriousness of wrong-
doing in that these constitute two different measures of the extent of
wrongdoing. Or better, there are two different extents. I said earlier that it
is clear that Alan has the opportunity to be a very naughty boy indeed, but
this use of "very naughty" was merely an acknowledgment of all the dif-
ferent levels of obligation applicable to Alan. Perhaps there was nothing that
he could do that would have been deontically very bad, and so nothing that
would have been very seriously wrong, in which case he did not have the
opportunity to be "very naughty" in another sense. Conversely, an agent
may have the opportunity to do very serious wrong and yet not have many
levels of obligation apply to him. Both senses must be allowed for, and the
simplifying counterproposal overlooks this.

4.4.6 Level of wrongdoing and level of obligation

It may be thought that, contrary to what I have intimated, the account that
I have proposed does not guarantee that level of wrongdoing corresponds
straightforwardly with level of obligation. What I have in mind here can
perhaps be best put by means of an illustration. Consider Sally, the sinner.

40 On this issue, see Section 1.4.

139
 1: ~S & ~R ...
2: ~S & R ...
3: S & R ...
4: S & ~R ...
Chart 4.8
Her situation may be represented as in Chart 4.8 (where S stands for "sin"
and R for "repent"). (By including Track 2, I am now allowing for the
possibility of repenting without having sinned.)41 Suppose that Sally takes
Track 3. In this case we may say each of the following:
(4.61) Sally oughtj not to sin;
(4.62) Sally oughtj not to repent;
(4.63) Sally ought2 not to sin (even) if she repents and, given that she does
indeed repent nonetheless, (still) she ought2 not to sin;
and
(4.64) Sally ought3 to repent if she sins and, given that she does sin, she
ought3 to repent.
What may seem puzzling is this. Sally's obligation not to sin, mentioned in
(4.61), is said to be primary, yet the obligation to repent, mentioned in
(4.64), given that Sally sins, is said to be tertiary. But hasn't there been just
one wrongdoing — namely, sinning — and so just one level of wrongdoing? If
so, then there is no straightforward correspondence between level of
wrongdoing and level of obligation.
The answer is this. When Sally sins, she plummets to Level 3. Why?
Because there is an intermediary level. As (4.63) shows, Sally not only
oughtj not to sin, she also ought2 not to sin. Thus she does indeed do a
double wrong — wrong on two levels — when she sins, and that is why her
obligation to repent is tertiary.
In general, the point can be put this way. Ifwe keep our eye on the "big"
picture, we can say that an agent has a primary obligation to take some first-
level track, a secondary obligation to take some second-level track if he fails
to take a first-level track, a tertiary obligation to take some third-level track
if he fails both to take a first-level track and to take a second-level track, and
so on. It is through wrong being done on one level that an obligation on
the next level kicks in. A straightforward correspondence between level of
wrongdoing and level of obligation is thus assured.

41 Cf. note 22 to this chapter.

42 Cf. note 45 to Ch. 3, concerning Bruce's obligation not to apologize when he does noth-
ing for which he owes an apology.

140
 Prima facie obligation

5.1 A TALE OF TWO OBLIGATIONS

It was the best of acts, it was the worst of acts. Steve stopped to render first
aid to Carl, who had had a car accident. But he thereby failed to keep his
dinner date with Dave, who had badly wanted Steve to sample his latest
culinary concoction. What a sorry business!
The story is a familiar one. W. D. Ross would say1 that Steve had a prima
facie obligation (of beneficence) to tend to Carl and a conflicting prima
facie obligation (of fidelity) to keep his date with Dave. Under the cir-
cumstances, Steve was quite justified in satisfying the former obligation at
the expense of the latter; indeed, he was overall obligated to do so.2 The
term "overall obligation" is not Ross's; he uses instead "absolute obliga-
tion," "actual obligation," "duty proper," and "duty sans phrase," but none
of these seems to me as felicitous as "overall obligation." The term "prima
facie obligation" is Ross's; I don't think that it is particularly felicitous either
(nor did Ross),3 but, as noted in Chapter 1, since Ross's term is by now so
well entrenched, it seems on balance advisable to continue to use it.
The main question to be addressed in this chapter is simply this: how is
the concept of prima facie obligation to be understood? This of course pre-
supposes that there is indeed such a concept in the first place. This has some-
times been denied, and denied by philosophers from both ends of the
normative spectrum. Alan Donagan, for example, a dyed-in-the-wool
Kantian, takes Ross to task for claiming that there is a prima facie obliga-
tion of fidelity that is overridden by a prima facie obligation of beneficence
in the sort of case just given. Donagan says:
If a man accepts an invitation to dinner, it would be absurd for his host to under-
stand him as having promised not to prevent a serious accident, or not to bring
1 See Ross (1930), p. 18ff.
2 Broad would say that Steve had conflicting "component" obligations and that his "resul-
tant" obligation was to act as he did. See Broad (1985), p. 150.
3 See Ross (1930), p. 20.

141
relief to the victims of one, if to do these things would prevent him from din-
ing. It is a promiser's duty to express any condition to his promise which the
promisee might misunderstand; but there would be no misunderstanding in
such a case... To any relatively trivial promise there are a host of tacit condi-
tions, all of which will normally be satisfied, which both promiser and promisee
must and do understand; and when, as occasionally happens, such a condition
is not satisfied, the promiser treats his obligation to the promisee as annulled.
He has no need to consider himself as having a responsibility to fulfil it that is
outweighed by a heavier responsibility.4
I think Donagan is wrong. When Steve helped Carl, he did not keep his
promise to Dave. Of course, Steve didn't promise Dave to have dinner with
him even if he (Steve) could otherwise bring relief to the victim of an acci-
dent; to that extent Donagan is right. Nonetheless, nor did Steve promise
to have dinner with Dave only if no such emergency arose; for then he
would not have failed to keep his promise to Dave when he helped Carl.
The content of Steve's promise was not fully specific, even implicitly. He
simply promised to have dinner with Dave, and this promise was silent,
even implicitly, on the issue of conflicting emergencies. And it is at least in
part because he broke his promise, while having a (prima facie) obligation
not to break his promise, that Steve has the residual obligation to offer Dave
an explanation as to why he didn't show up - at least, that is what many,
including Ross, would say, and I cannot see that Donagan has undermined
this view.5
It might be thought that, even if Donagan were correct about there
being no prima facie obligation of fidelity in cases of the sort just discussed,
still that would not provide sufficient reason to reject the concept of prima
facie obligation altogether. I agree. But Donagan seems to think otherwise,
for immediately after the passage just quoted he says the following:
It would be unjust to Ross not to record that, in his Foundations of Ethics, he
treated a number of casuistical problems about promising on these traditional
lines [i.e., the lines just advocated by Donagan]. However, he did not see that
such a treatment deprives his theory in The Right and the Good of its raison
d'etre.6
This seems to suggest that, if no prima facie obligation of fidelity is to be
found in cases of the sort just discussed, then no prima facie obligations are
to be found anywhere (for it must surely be agreed that the analysis and
identification of prima facie obligations is indeed the "raison d'etre" of

4 Donagan (1977), p. 93.

5 Cf.McConnell(1995).
6 Donagan (1977), p. 93.

142
Ross's theory of right in The Right and the Good). No argument for this bold
claim is given, but certainly Donagan does not concern himself with the
notion of prima facie obligation in the presentation of his own theory. At
any rate, both the claim and its antecedent strike me as false.
At the other end of the normative spectrum are utilitarians, also well
known for their failure to employ the notion of prima facie obligation.
Most are silent on the issue of what prima facie obligations there are or what
prima facie obligation itself might be, and their silence is eloquent: they dis-
miss the notion. One who is not silent on this matter is Fred Feldman (who
proposes a modified version of traditional utilitarianism). Feldman explic-
itly dismisses the concept of prima facie obligation, proclaiming it a
"mirage,"7 by which he means that there really is no such concept, so that
terms that may appear to express it either are meaningless or express some
other concept. Again, no argument is given for this (other than the con-
cept's recalcitrance with respect to Feldman's efforts to analyze it).
I believe that the concept ofprima facie obligation is genuine and, more-
over, that it is amenable to analysis. Even if I am right, this of course does
not mean that either Donagan's or Feldman's normative theory is false; for
it is possible that the concept of prima facie obligation be a respectable con-
cept and yet that there be no actual (or at least no very interesting) prima
facie obligations. Nonetheless, in presenting my account of prima facie
obligation, I shall frequently rely, for purposes of illustration, on Ross's
view that there are indeed certain significant prima facie obligations, such as
those of fidelity, beneficence, and the like. This is a natural view to hold, I
think, and such reliance will render the analysis that I shall give easier to
follow.

5.2 DEFECTIVE ANALYSES

To many, prima facie obligation has seemed to be overall obligation that is
essentially subject to some sort of condition. Ross himself suggests the term
"conditional duty" as an alternative to the term "prima facie duty."8 And
this may seem quite natural. Consider, for instance, the claim that one
ought to keep one's promises, where this is taken to express a prima facie
obligation. We may express this as follows:

(5.1) if 5 makes a promise to do A, then 5 ought to do A.

How is this is to be analyzed?

7 Feldman (1986), p. 144.

8 Ross (1930), p. 19.

143
 It is clear that none of the following will do as an analysis:
(5.2) that S makes a promise to do A materially implies that 5 ought over-
all to do A;
(5.3) that S makes a promise to do A strictly implies that S ought overall
to do A;
(5.4) that S makes a promise to do A subjunctively implies that S ought
overall to do A.
None of these will do, for the simple reason that all such implication war-
rants factual detachment (see the discussion of (4.1), (4.5), and (4.8), respec-
tively, in the last chapter), whereas it is clear that the fact that one has made
or will make a promise to do something cannot by itself transform a prima
facie obligation to keep it into an overall obligation to do so. An addition-
al problem with (5.2) and (5.3), though not (5.4), is that neither material
nor strict conditionals are defeasible, where a conditional statement of the
form "if p, then r" is understood to be defeasible just in case it is possible
that it be true while "ifp and q, then r" is false. (That is, to use terminolo-
gy introduced in Chapter 4, both material and strict implication validate
"augmentation.") Such indefeasibility would appear difficult to reconcile
with the overridability of prima facie obligation. For while one's prima
facie obligation to keep a promise may on occasion also be an overall oblig-
ation, it needn't be; and it won't be, even though one has made a promise,
if in addition something more pressing requires one's attention.
We might try analyzing the prima facie obligation to keep a promise to
do A as follows:
(5.5) 5 ought overall either not to make a promise to do A or to do A.
(The idea here is to have the overall obligation apply to the entire condi-
tional, itself reformulated as a disjunctive statement, and not just to its con-
sequent.)9 But again this fails to accommodate the overridability of prima
facie obligation. Suppose that one makes a promise at Tt to do A at T3 and
that one therefore has a prima facie obligation at T2 to do A at T3. According
to the proposal, then, one ought overall at T2 either not to make the
promise at T1 or to do A at T3. But one cannot at T2 not make the promise
at TP Given the principle that, if one ought overall either to do Xor to do
Y and one cannot do X, then one ought overall to do Y, it follows that one
ought overall at T2 to do A at T3.10 But this may not be the case; for, even
9 Compare the analysis proposed in Hintikka (1970), p. 9Iff.
10 This principle, whether read in terms of 0(XorY) or in terms of O(X v Y), is implied
by my account of overall obligation.

144
though one has a prima facie obligation at T2 to keep one's promise and do
A at T3, it may be that one has no overall obligation at T2 to do so, because
of some more pressing, contrary prima facie obligation.11
The fact that conditional overall obligation, as analyzed in (XIII) in
Section 4.2, does not warrant factual detachment and is defeasible (that is,
does not validate augmentation) may suggest that it is well suited to an
analysis of the concept of prima facie obligation. But, as Feldman has noted,
since necessity detachment is valid for statements of conditional overall
obligation, we cannot analyze (5.1) as follows:
(5.6) 5 ought overall to do A, on the condition that he promises to do A.
For, where it is also true that S cannot so act that he does not promise to
do A (the most obvious type of case being one where the promise has
already been made), we may detach from such a conditional overall oblig-
ation the unconditional overall obligation to do A. But, as already noted,
we don't wish to say that S has an overall obligation to do A just because he
has a prima facie obligation to keep his promise and has already promised
to do A.12
In light of the fact that the concept of prima facie obligation resists any
of the foregoing analyses, Feldman concludes, as noted earlier, that it is
merely a "mirage." But this conclusion appears premature, as I shall now
seek to demonstrate.

5.3 THE ANALYSIS

5.3.1 "All else being equal"
The basic idea underlying the analysis that I shall propose is this. When one
has a prima facie obligation to do A, doing A is of greater deontic value,
taken in and of itself, than not doing A. Another, and common, way of
putting this is to say that doing A is deontically superior, all else being equal,
to not doing A.13 (Strictly, this characterization applies only to direct, and
not to indirect, prima facie obligations, as will be explained shortly.) This
way of putting matters is quite revealing, for it indicates that prima facie
obligation is indeed essentially overall obligation conditioned in some way

11 Further problems with Hintikka's sort of analysis are discussed in Purtill (1973) and
Feldman (1986), p. 182.
12 See Feldman (1986), pp. 130-1.
13 Cf. Pietroski (1993) for an extended investigation into this characterization ofprima facie
obligation.

145
 (and thus that Ross's alternative term "conditional duty" is an apt one).
Thus it is sometimes said: one ought (overall), all else being equal, to act
beneficently; one ought (overall), all else being equal, to keep one's promi-
ses. Nonetheless, the fact that prima facie obligation is overall obligation
conditioned in some way should not blind us to another fact that is fre-
quently overlooked, and that is that prima facie obligation may itself bo
either unconditional or conditional. The prima facie obligation of fidelity
should be construed as a conditional prima facie obligation: if one makes a
promise (or some other sort of commitment) to do A, then one has a prima
facie (and not necessarily overall) obligation to do A. For there is, it would
seem, no call for fidelity tout court; it is fidelity to a commitment that is
required, and, if there is no unconditional requirement to make a commit-
ment (as in general there would seem not to be), there is no uncondition-
al requirement of fidelity. That is, although commitment-plus-fidelity is
deontically superior, all else being equal, to commitment-plus-nonfideli-
ty, we should not say that fidelity is deontically superior, all else being equal,
to nonfidelity. For there are two ways to display nonfidelity - either by way
of commitment-plus-nonfidelity, or by way of noncommitment-plus-
nonfidelity - and only the former is tantamount to infidelity; there seems
to be no reason to think that commitment-plus-fidelity is deontically supe-
rior, all else being equal, to noncommitment-plus-nonfidelity.
Having taken note of the possibility that prima facie obligation be either
conditional or unconditional, we should note also that the detachment of
an unconditional prima facie obligation from a conditional one is analo-
gous to the detachment of an unconditional overall obligation from a con-
ditional one. That is, although factual detachment is not valid, both
necessity detachment and deontic detachment are valid. For example, even
if one is going to make a promise to do A, one does not yet have for this
reason alone an unconditional prima facie obligation to do A; rather, one
simply has a conditional prima facie obligation to do A if one promises to
do so. But if the promise is inevitable (for example, one has already made
it), then one has an unconditional prima facie obligation to keep it; or if
one has an unconditional prima facie obligation to make the promise, then
one has an unconditional prima facie obligation to keep it.
Again, however, whether a prima facie obligation is itself uncondition-
al or conditional, if it is properly characterized as an overall obligation "all
else being equal," then it is essentially an overall obligation that is condi-
tioned in some way; for, whether or not it is itself conditional, it carries
with it this extra condition of "all else being equal." How are we to under-
stand this extra condition?
146
 It would make matters very simple if this condition were some sort of
constant (call it E for "equal"). For then the obligation at issue would be
straightforward conditional overall obligation of the sort analyzed in (XIII),
and both necessity detachment and deontic detachment would be clearly
valid. Consider the claim that Faith has a fidelity-based prima facie obliga-
tion to pay Pam $10 (P), given her commitment (C) to do so. This would,
on the present suggestion, be understood to mean that she has an overall
obligation to do P on the condition that both C and E hold. Now consider
necessity detachment, where this involves C's being inevitable. If Faith
cannot so act that C does not hold, then all worlds accessible to her are C-
worlds; and so, if the best (C&E)-worlds accessible to her are P-worlds,
then the best E'-worlds accessible to her are P-worlds. As for deontic
detachment (where Faith has an unconditional prima facie obligation to do
C): if the best (C&E)-worlds accessible to her are P-worlds and the best
E-worlds accessible to her are C-worlds, then, again, the best E-worlds
accessible to her are P-worlds.
What makes matters awkward is that it is clear that "all else being equal"
is not to be interpreted as a constant. On the contrary, what it means varies
from context to context; for it is a sort of indexical, in that "else" means
"other than this." My suggestion is this: to say that 5 ought overall to do A,
all else being equal, is to say, roughly, that any world, W, in which S does
A is deontically superior to any other world, W\ which is "just like" W
except that in W' S does not do A. (As before, this in fact applies only to
direct, and not to indirect, prima facie obligation. Again, more on this in a
moment.) Of course, this cannot be precisely correct, for no worlds can dif-
fer from one another with respect to just one such fact. If S does A in W
but not in W\ then any proposition which is distinct from but implies the
proposition that S does A will also fail to be true in Wf. What we need to
say, then, is that W* is as "close" to W as possible, given that in it S does
not do A; that is, Wf is "minimally different" from W with respect to S
doing A.
This is not an unfamiliar notion. It, or something close to it, has been
put to good use by several philosophers in the analysis of counterfactuals.
To adapt what John Pollock has said:14 the basic idea is that Wf is a world
that is obtained from W by making minimal changes which suffice to make
it the case that S does not do A; truths in W that are in the appropriate sense
"irrelevant" to or "independent" of 5's doing A must also be true in W'\
gratuitous changes are disallowed. I would like to be able to provide a

14 Pollock (1976), pp. 70-1.

147
precise analysis of this crucial concept, but I do not know how to do so. It
might seem that all that needs to be said is this: W' is minimally different
from W with respect to S doing A if and only if S does A'mW but not in
W' and every proposition that is true in W but which does not imply that
S does A is also true in Wf. But this won't do. Consider any two proposi-
tions^ and q, neither of which alone implies that S does A but which joint-
ly imply this. Given the current proposal, each of p and q will be true in
W'\ but then it will be true after all that S does A in W'. Pollock has him-
self proposed an analysis of minimal difference that does not succumb to
this problem.15 It may be that this analysis is acceptable, but I am reluctant
to commit myself to it; for it is very complex and, besides, rests on the con-
troversial notion of a "simple" proposition. Moreover, the analysis (as is
typical in the treatment of counterfactuals) involves a past—future asymme-
try, so that similarity of the past is weighted more heavily than similarity of
the future; and I do not wish to presuppose any such asymmetry here.
Formally, then, I shall invoke the concept of the closeness ofpossible worlds
unanalyzed. I hope that my use of this concept will be sufficiently intuitive
to render what I have to say about prima facie obligation both plausible and
helpful, just as its use (or the use of some closely related concept) has proven
helpful in the analysis of counterfactuals.
For reasons (having to do with detachment) that will become apparent
shortly, I shall need to talk of those worlds that are minimally different for
SatT in Wwkh respect to S doing A; that is, the focus of attention will be
on those worlds that are accessible to S. The canonical phrase form in this
context is therefore this: W" is minimally different from W\ for S at Tin
W, with respect to 5 doing A at T". This may be understood in terms of
the closeness of worlds as follows:

(5.7) W" is minimally different from W\ for S at Tin W, with respect to

5 d o i n g s at T ' iff
(a) W' is accessible to 5 from W at T and S does A at T ' in W'\
(b) W" is accessible to S from W at T and 5 does not do A at T ' in
W"\ and
(c) there is no world W'" such that
(1) W'" is accessible to S from Wat T,
(2) S does not do A at T ' in W"\ and
(3) W" is closer to W than W" is.

15 Pollock (1976), p. 70ff.

148
I shall also make use of the related concept ofconditional minimal difference,
which may be understood as follows:
(5.8) W" is minimally different from W\ for S at T i n W, with respect
to S doing A at T ' , on the condition that p is true, iff
(a) W' is accessible to S from W at T and S does A at T ' in W' and/?
is true in W'\
(b) W7" is accessible to S from W 2X Tand S does not do A at T' in
W" and/? is true in W"\ and
(c) there is no world W'" such that
(1) W'" is accessible to 5 from W2X. T,
(2) S does not do A at T' in W'",
(3) /? is true in W"\ and
(4) W" is closer to W' than W " is.
Given this understanding of minimal difference, both unconditional and
conditional, we may proceed to an analysis of prima facie obligation. I have
suggested: to say that S ought overall to do A, all else being equal, is to say,
roughly, that all accessible worlds in which 5 does A are deontically supe-
rior to all accessible worlds that are minimally different in this respect. This
double use of "all" implies that there is something in general - "across the
board" - to be said in favor of doing A. It is this that indicates that doing A
is of greater deontic value, taken in and of itself, than not doing A. I pro-
pose initially that we say this (revision will be called for later):

(5.9) S has, at T in W, a direct unconditional prima facie obligation to

do A at T ' iff
(a) there are a world W' and a world W" such that W" is minimally
different from W\ for S at T in W, with respect to S doing A at
T'; and
(b) for all worlds W' and W",
if (1) W" is minimally different from W\ for S at T in W, with
respect to S doing A at T',
then (2) the deontic value for 5 at T of W' is greater than the
deontic value for S at T of W".
Similarly, we may say:

(5.10) 5 has, at T in W, a. direct conditional prima facie obligation to do

A at T ' , on the condition that/? is true, iff
(a) there are a world Wf and a world W" such that W" is minimally
different from W\ for S at T in W, with respect to S doing A at
T', on the condition that/? is true; and
149
 (b) for all worlds W' and W'\
if (1) W" is minimally different from W\ for 5 at T in W, with
respect to S doing A at T', on the condition thatp is true,
then (2) the deontic value for S at T of Wf is greater than the
deontic value for S at T of W".
Four points should be noted here.
First, instead of talking of prima facie obligations, we could talk of deon-
tic reasons; for there is a sense of "reason" according to which one has a rea-
son to do something just in case one has a prima facie obligation to do it.16
Note that reasons may be actual or merely potential; I take this to mean that
they may be unconditional or merely conditional.
Second (5.9) and (5.10) imply that, if 5 has a direct prima facie obliga-
tion to do A, then 5 can do A and S can refrain from doing A. A is, in other
words, personally optional for S. This is in keeping with the analysis of
overall obligation provided in Chapters 2 and 4.
Third, (5.9) and (5.10) are restricted to what I have called direct prima
facie obligations. (Ross uses the term "direct" similarly.)17 If Ben has a
direct prima facie obligation to act beneficently, this consists in every acces-
sible beneficence-world being deontically superior to any accessible world
that is minimally different with respect to beneficence. Similarly, if Faith
has a direct prima facie obligation to pay Pam $10, if she so promises, this
consists in every accessible promise-plus-payment world being deontical-
ly superior to any accessible promise-world that is minimally different with
respect to payment.
But not all prima facie obligations are direct. Some are indirect, and
there are two ways in which they may be so. First, some may be called (fol-
lowing Ross again)18 incidental. For example, if Ben cannot act beneficent-
ly without driving to Boston, then he has an incidental unconditional prima
facie obligation to drive to Boston. Note that, since it would be highly sur-
prising if all accessible worlds in which Ben drives to Boston were deonti-
cally superior to all accessible worlds that are minimally different in this
respect, we have no reason to think that Ben has a direct unconditional
obligation to drive to Boston. But, as long as he must drive to Boston to
display beneficence, Ben does have an incidental, beneficence-based obliga-
tion to drive to Boston. Similarly, if Faith cannot keep her promise with-

16 There is also another, nondeontic sense of "reason" that I shall not seek to characterize
here. Cf. Section 1.2 and also Ch. 2, n. 13.
17 Ross (1930), p. 46.
18 Ross (1930), p. 46.

150
out going to the bank and withdrawing some cash, then she has an inci-
dental conditional prima facie obligation to go to the bank and withdraw
some cash. The principle here is straightforward. It may be roughly stated
as follows: if S has a direct prima facie obligation to do A and cannot do A
without doing B, then S has an incidental prima facie obligation to do B.19
More precisely, let us say, first of all:

(5.11) p requires q, for S at T in W, iff for all worlds W' accessible to S

from W at T, if p is true in W\ then q is true in W'.

W e may then account for incidental prima facie obligations as follows:

(5.12) S has, at T in W, an incidental unconditional prima facie obliga-
tion to do B at T " iff for some act A and time T',
(a) S has, at T in W, a direct unconditional prima facie obligation to
do A at T';
(b) S doing A at T ' requires S doing B at T", for S at T in W\
(c) 5 doing A at T ' is distinct from S doing B at T"; and
(d) there is a world M/~' such that W' is accessible to S from Wat T
and S does not do B at T " in H/~';
(5.13) 5 has, at T in W, an incidental conditional prima facie obligation
to do B at T " , on the condition that p is true, iff for some act A
and time T ' ,
(a) 5 has, at T in H7, a direct conditional prima facie obligation to do
A at T', on the condition that p is true;
(b) S doing AatT' requires S doing B at T", for S at T in H^;
(c) S doing A at T ' is distinct from S doing B at T"; and
(d) there is a world H 7 ' such that W' is accessible to S from W at T
and 5 does not do B at T " in W'.
The third clause in each is required to ensure that not all direct prima facie
obligations are also said to be incidental. The fourth clause is required to
preserve the thesis that whatever one has a prima facie obligation to do is
personally optional.
The second way in which prima facie obligations can be indirect may
be illustrated as follows. Suppose that Ben has an incidental prima facie
obligation (grounded in beneficence) to drive to Boston and also has an
incidental prima facie obligation (grounded in nonmaleficence) not to go
to Boston without paying a visit to his mother, who lives there (in that,
19 This principle is no more vulnerable to the Good Samaritan Paradox than is the corre-
sponding principle concerning overall obligation, namely, (2.43'")- See the discussion in
Subsection 2.3.2.

151
although it would not be mean of him not to visit his mother, it would be
mean of him to go to Boston and not visit her). Then Ben has what I shall
call a compound prima facie obligation (grounded in beneficence and non-
maleficence) to visit his mother. The principle at issue here may be rough-
ly stated as follows: if S has a prima facie obligation (whether direct or
incidental or, indeed, compound) to do A and also has a prima facie oblig-
ation not to do A without doing B, then S has a compound prima facie
obligation to do B. More precisely:20

(5.14) S has, at T in W, a compound unconditional prima facie obliga-

tion to do B at T" iff for some act A and time T",
(a) S has, at T in IV, a prima facie obligation to do A at T';
(b) 5 has, at T in W, a prima facie obligation not to both do A at T'
and not do B at T"; and
(c) there is a world W' such that W' is accessible to S from W at T
and S does not do B at T ' in W'\
(5.15) S has, at T in W, a compound conditional prima facie obligation
to do B at T", on the condition that p is true, iff for some act A
and time T",
(a) S has, at T in W, a prima facie obligation to do A at T , on the
condition that p is true;
(b) S has, at T in W, a prima facie obligation not to both do A at T'
and not do B at T"; and
(c) there is a world Wr such that Wf is accessible to 5 from W at T
and 5 does not do B at T " in W'. 21

Note finally that, just as the account of overall obligation presented in

Section 2.1 included analyses of overall permissibility and wrongdoing, so
too the present account of prima facie obligation may be extended to
include analyses of prima facie permissibility and wrongdoing. Consider
wrongdoing first. This may be accounted for simply as follows:

(5.16) it is prima facie wrong, for S at T in W, to do B at T " iff 5 has, at

T in W, a prima facie obligation not to do B at T".
20 Compare the corresponding principle for overall obligation, (2.42'), discussed in
Subsection 2.3.1.
21 A still more precise account of compound prima facie obligation, both unconditional
and conditional, would (in order to avoid any appearance of circularity) distinguish
between level-1 compound obligations, which would themselves be generated by oblig-
ations that are not themselves compound, and other level-n compound obligations,
which would be generated by level-(« - 1) compound obligations.

152
This presupposes an account of what it is to have a prima facie obligation
not to do something. Such an account can be developed along lines exact-
ly analogous to those along which the account of what it is to have a prima
facie obligation to do something has been developed. That is, instead of
talking of a direct unconditional prima facie obligation to do A (see (5.9)),
we can talk of a direct unconditional prima facie obligation not to do A;
this would involve all accessible ~yl-worlds being deontically superior to
all accessible worlds that are minimally different in this respect. (For exam-
ple, it may be that Malcolm has a direct unconditional prima facie obliga-
tion not to act maleficently; this would involve all accessible non-
maleficence-worlds being better than all accessible worlds that are mini-
mally different in this respect.) Instead of talking of an incidental uncondi-
tional prima facie obligation to do B (see (5.12)), we can talk of an incidental
unconditional prima facie obligation not to do B; this would involve not
doing A requiring not doing B or, equivalently, doing B requiring doing
A. (For example, it may be that Malcolm cannot play a certain practicaljoke
without acting maleficently; given his nonmaleficence-based reason not to
act maleficently, he then has a nonmaleficence-based reason not to play the
practicaljoke.) And so on. An account of what it is to have a conditional
prima facie obligation not to do something may be likewise developed.
Developing an account of what it is for an act to be prima facie permis-
sible is slightly more tricky. Evidently this must be grounded in the notion
of one's not having a direct unconditional prima facie obligation not to do
A, but how precisely is it to be accounted for? Should we say simply: some
accessible ^4-world is at least as good as all worlds that are minimally differ-
ent in this respect? Or should we say: all accessible ^4-worlds are at least as
good as all worlds that are minimally different in this respect? I think it is
clear, on reflection, that we should say the latter; for the former is far too
weak. It may well be, for instance, that some walking-world accessible to
Willis is at least as good as all worlds that are minimally different in this
respect, but we shouldn't infer from this that it is prima facie permissible
for Willis to walk. For it may be that every direct prima facie obligation that
Willis has requires that he not walk. On the other hand, if all beneficence-
worlds accessible to Ben are at least as good as all worlds that are minimal-
ly different in this respect, we may indeed infer that it is prima facie
permissible, as far as beneficence is concerned, for Ben to act beneficently.
All of this could of course be laid out more formally, and it could be sup-
plemented with an account of both indirect and conditional prima facie
permission, but I shall forbear doing this here.
153
 5.3.2 Three objections
The foregoing account of prima facie obligation may appear objectionable
for at least three reasons.
First, it might be thought that the account implies that whatever is
directly prima facie obligatory is also overall obligatory (or at least overall
permissible), for the following reason. According to the account, 5 has a
direct prima facie obligation to do A just in case, for every two accessible
worlds that are just alike except in one S does A and in the other 5 doesn't
do A, the former is deontically superior to the latter. But this appears to
imply that the deontically best accessible world (or worlds), W, is one in
which S does A; for otherwise that accessible world that is just like W
except in it 5 does A would be better than W.
Clearly such an implication would be unacceptable, but fortunately the
present account doesn't have it. The term "just alike" is here not just rough
but inaccurate. (Note that, when giving my rough, intuitive account of "all
else being equal" above, I used the term "just like" and not the term "just
alike.") In particular, one cannot assume that, if W" is minimally different
from W' (for 5 at T in W) with respect to S doing A at T", then W' is mini-
mally different from W" with respect to S not doing A at T". Suppose, for
example, that Roy is faced with conflicting direct prima facie obligations
of reparation and beneficence. Let us say that the deontically best accessi-
ble world is one in which he satisfies the former but not the latter (he can-
not satisfy them both). Next best would be his satisfying the latter but not
the former. Worse still would be his satisfying neither. This may be repre-
sented as in Chart 5.1 (where R stands for "reparation" and B for "benefi-
cence") . Given a proper filling-in of "...," we can say that W2 is minimally
different from Wx with respect to Roy's not acting beneficently; and yet
W^ (and not Wr) is minimally different from W2 with respect to Roy's
acting beneficently. Thus what is prima facie obligatory need not also be
overall obligatory or permissible.
A second objection stems from consideration of cases such as the fol-
lowing. Suppose, as in Subsection 3.2.5, that Gertrude owes Lucy a debt
of gratitude. The most appropriate way for her to display gratitude would
be to send Lucy lilies; daisies would be a poor second. But even though

W,: R & ~S & ...

W2:~R & B & ...
W3:~R & ~B & ...
Chart 5.1
154
sending daisies would be inferior, it might seem that, insofar as it would
constitute some display of gratitude, Gertrude has some reason, and hence a
prima facie obligation, to send daisies; it's just that this obligation is over-
ridden by the obligation to send lilies. But my account does not warrant
our saying this. Presumably Gertrude has no direct prima facie obligation to
send daisies, and doing so is not necessary for a display of gratitude and hence
is not even incidentally prima facie obligatory (barring its being necessary
for the satisfaction of some other direct prima facie obligation). In fact, it
might appear that I cannot even say that Gertrude has an incidental prima
facie obligation to send lilies, since even this appears not to be necessary for
a display of gratitude (in that sending daisies would be an alternative, albeit
inferior).
What I think we should say here is this. Granting that gratitude comes
in degrees, let us say that "full" gratitude is gratitude to degree 10.
Gertrude's sending lilies would thus constitute a display of gratitude to
degree 10; her sending daisies would, let us say, only constitute a display of
gratitude to degree 5. Now, while it is plausible to say that a display of grat-
itude is better, all else being equal, than a lack of a display of gratitude, it
doesn't seem plausible to say that a display of gratitude to degree x is bet-
ter, all else being equal, than a lack of a display of gratitude to degree x; for
what if the lack of a display of gratitude to degree x is due to a display of
gratitude to a greater degree y? Still, it is plausible to say that a display of
gratitude to at least degree x is better, all else being equal, than a lack of a
display of gratitude to at least degree x. This observation allows us to say
that Gertrude does indeed have an incidental obligation to send Lucy lilies
(given that there is no degree of gratitude greater than 10 and that sending
lilies is the only way to display gratitude to that degree); and, although it
won't allow us to say that she also has an incidental obligation to send Lucy
daisies, it does allow us to say that she has an incidental obligation either to
send lilies or to send daisies (given that nothing else would constitute a dis-
play of gratitude to at least degree 5). And this indeed seems to be what
should be said.
In fact, the matter is even more complex than this. Suppose that
Gertrude ought, from gratitude, to send Lucy lilies, and also ought, from
gratitude, to send Chrissie chrysanthemums, but cannot do both. Here, it
seems, gratitude requires the impossible, and yet my account implies that
whatever is prima facie obligatory is possible. The solution, of course, is
that gratitude, more fully characterized, is gratitude-to-a-benefactor (or,
perhaps, would-be benefactor, or presumed benefactor), just as fidelity is
fidelity-to-a-promisee (or someone to whom another sort of commitment
155
has been made). (If we talk of "what gratitude requires" without qualifica-
tion, then, we must mean what gratitude to every benefactor requires; so
too for fidelity, reparation, and any other obligation which is owed to a par-
ticular person or persons.)22 My account implies, properly, that Gertrude
can satisfy each ofher prima facie obligations of gratitude-to-Lucy and grat-
itude-to-Chrissie; it does not imply, nor should it, that she can satisfy them
both.
A third objection is more troublesome. It is that the account implies that
certain direct prima facie obligations that are intuitively thought of as being
merely conditional turn out also to be unconditional. Consider fidelity.
This has been assumed (for purposes of illustration) to constitute a direct
conditional prima facie obligation, in that all commitment-plus-fidelity-
worlds have been assumed to be deontically superior to any worlds that are
minimally different with respect to fidelity. But notice that one cannot (it
seems) display fidelity in the absence of a commitment; that is, all accessi-
ble fidelity-worlds are also commitment-worlds. But then, if all accessible
commitment-plus-fidelity-worlds are deontically superior to their closest
commitment-plus-nonfidelity-worlds, it must also be true that all accessi-
ble fidelity-worlds are deontically superior to their closest nonfidelity-
worlds. In other words, the direct prima facie obligation of fidelity will be
not just conditional but unconditional. The same applies to other (alleged)
conditional direct prima facie obligations such as those of gratitude and
reparation. For one cannot (it seems) display gratitude in the absence of that
condition (namely, being a beneficiary) on which it is conditionally prima
facie obligatory; likewise, one cannot make reparation for a nonexistent
harm.
It is tempting at first simply to bite the bullet here but then go on to point
out that there are still certain intimately related direct prima facie obliga-
tions that are merely conditional. For example, even if Faith cannot display
fidelity to her promise to pay Pam $10 without making that promise, still
she can pay Pam $10 without promising to do so. Thus, while all accessi-
ble promise-plus-payment-worlds may be better than their closest accessi-
ble promise-plus-nonpayment-worlds, there is no need to think that all
accessible payment-worlds are better than their closest accessible nonpay-
ment-worlds. And isn't that what really matters here?
Well, no, it isn't. There are two problems with this. First, on some occa-
sions, it may not be possible to identify a conditionally obligatory act in such
22 Perhaps, for example, nonmaleficence is such an obligation, whereas beneficence is not.
This has to do with whether persons have a right to the performance of an obligatory act.
Rights are discussed further in Section 5.5.

156
a way that the condition is not included in the identification. Although
Faith can do what fidelity requires (namely, pay Pam $10) without promis-
ing to do it, it may be that Roy cannot do what reparation requires (mend-
ing a broken vase, say) without committing the relevant harm. Secondly,
if it is admitted that Faith has a direct unconditional prima facie obligation
to display fidelity (even if it is also pointed out that her direct prima facie
obligation to pay Pam $10 is merely conditional), it must then also be
admitted, given the foregoing account, that she has an incidental prima
facie obligation to make the relevant promise that is tmconditional.
Similarly, if it is admitted that Roy has a direct unconditional prima facie
obligation to make some reparation, then it must also be admitted that he
has an incidental prima facie obligation to commit the relevant harm that
is ^conditional. This is surely to admit too much. The original account
must therefore be revised.
We must deny not only that Faith has an unconditional prima facie
obligation to pay Pam $10 but also that she has an unconditional prima facie
obligation to display fidelity to the commitment to pay Pam $10 (given that
she has not already made this commitment and that there is no indepen-
dent unconditional reason to make it). (Similar remarks pertain to the prima
facie obligations of gratitude and reparation.) What I think warrants such
denial is an observation already made above: although commitment-plus-
fidelity is deontically superior, all else being equal, to commitment-
plus-nonfidelity, it is not deontically superior, all else being equal, to
noncommitment-plus-nonfidelity; for nonfidelity constitutes infidelity
only when there is an actual commitment. It is in virtue of this fact that the
prima facie obligation of fidelity is merely conditional. But how can this
be? After all, consider a situation where it is possible to display a great deal
of maleficence to one person while also displaying minor beneficence to
another. We may suppose in this case that maleficence-plus-beneficence is
deontically superior, all else being equal, to maleficence-plus-nonbenefi-
cence, and also that maleficence-plus-beneficence is not deontically superior,
all else being equal, to nonmaleficence-plus-nonbeneficence. Yet this
doesn't make the prima facie obligation of beneficence a merely condi-
tional one. So why should the corresponding fact about commitment-plus-
fidelity make the prima facie obligation of fidelity a merely conditional
one?
The answer is to be found in the fact that commitment is a prerequisite of
fidelity, whereas maleficence is not a prerequisite of beneficence. This is
the very feature about fidelity that has given rise to the present problem,
and it points to the proper solution. Commitment is a prerequisite of
157
fidelity, not just in the sense that there cannot be fidelity without a
commitment, but in the sense that fidelity requires that a commitment pre-
cede it. Commitment and fidelity thus come in a well-ordered "package,"
as it were, and the package must commend itself a6 initio if fidelity is to be
unconditionally obligatory. (Similar remarks pertain to gratitude and repa-
ration and their respective initial conditions.) To put this point more
precisely, let us first say this:

(5.17) W" is minimally different from W', for 5 at T in W, with respect

to S doing A at T", on the condition that p varies, iff
(a) W' is accessible to S from W at T and S does A at T' in W' andp
is true \x\W'\
(b) W" is accessible to S from W at T and S does not do A at T ' in
W" and |? is not true in W"; and
(c) there is no world W'" such that
(1) W'" is accessible to S from W at T,
(2) 5 does not do A at T ' in W"\
(3) p is not true in W'", and
(4) W'" is closer to W' than W" is.

With this in hand, we may now revise (5.9) by appending an extra clause
as follows:

(5.9') S has, at T in IV, a direct unconditional prima facie obligation to do

A at T iff
(a) there are a world W' and a world W" such that W" is minimally
different from W\ for S at T in W, with respect to S doing ^4 at T';
(b) for all worlds H/' and H/",
if (1) W" is minimally different from PF', for 5 at T in M^, with
respect to S doing A at T',
then (2) the deontic value for S at T of W7' is greater than the
deontic value for S at T of W"; and
(c) for all propositions p,
if ( 1 ) 5 has, at T in W, a direct conditional prima facie obliga-
tion to do A at T', on the condition that/? is true,
(2) for some event E and time T " earlier than T', /? is the
proposition that E occurs at T",
(3) S doing A2XT' requires p, for S at T in W, and
(4) there is a world accessible to 5 from W at T in which p is
not true,
then (5) for all worlds W' and W",

158
 if (i) W" is minimally different from W\ for S at T in W, with
respect to 5 doing A at T', on the condition that p varies,
then (ii) the deontic value for S at T of W' is greater than the
deontic value for 5 at TofW".
The "official" account of prima facie obligation is thus as follows: replace
(5.9) by (5.9') and leave (5.10) (which, indeed, is presupposed by (5.9')) and
(5.11)-(5.15)astheyare.
Finally, it should be noted that this account accommodates the possibil-
ity that deontic value be significantly agent- or time-relative, inasmuch as
the relevant value of the worlds is the value they have for S at T. Let me
now elaborate on this with respect to Ross's theory. I said earlier that at
least some of the values that ground prima facie obligations are thought by
Ross to be significantly agent-relative. Fidelity, for example, is clearly a
value of this sort. Ross says that each agent always has a prima facie oblig-
ation to see to it that he keeps his promises.231 take this to imply that each
of us always has a (conditional) prima facie obligation to see to it that we
ourselves display fidelity. Now it might be that Ross also believes that each
of us always has a prima facie obligation to see to it that others display fideli-
ty; for he might believe that such displays are intrinsically valuable, and he
does say that we do always have a prima facie obligation to do that which
will promote such value.24 But even if this is so, it does not detract from the
fact that it seems pretty clear that, according to Ross, for each of us a spe-
cial value attaches to our displaying fidelity. So too for gratitude and repa-
ration, it seems.25 Whether or not this is so for other prima facie obligations
that Ross lists is unclear to me. Does he believe, for instance, that there is
a special value for each of us to be attached to our not harming innocent

23 Ross (1930), p. 17ff.

24 On the obligation to promote intrinsic value, see Ross (1930), pp. 24 and 39. In fact,
however, the discussion in Ross (1930), Ch. 5, especially p. 140, suggests that Ross does
not believe that each of us always has, in virtue of the prima facie obligation to promote
intrinsic value, a prima facie obligation to see to it that others display fidelity. For there
he lists four intrinsic goods - virtue, pleasure, the allocation of pleasure to the virtuous,
and knowledge - and he explicitly says that he doubts the existence of any others (save
combinations of those mentioned). It might be thought that Ross takes fidelity to be a
virtue, but this would be a mistake. For he takes virtue to concern motive, and he explic-
itly divorces the duty of what he calls "fidelity" from motive. (See Ross, 1930, p. 22 and
also p. 5.) Still, Ross does say (1930, p. 36) that fidelity can be bonific - that is, instru-
mentally good — and so he clearly thinks that, on those occasions on which it is bonific, there
is a prima facie obligation to promote it.
25 See Ross (1930), pp. 27, 30, and 41-2.

159
people, even if there is also a value to be attached to preventing others from
harming innocent people? I am not sure, although certain of his remarks
suggest that he does believe this.26

5.3.3 Detachment
The move from (5.9) to (5.9') blocks factual detachment. Even if Faith will
promise to pay Pam $10, the prima facie obligation to display fidelity to this
commitment is merely conditional; for clause (c) of (5.9') is not satisfied.
To be acceptable, though, the account must not only block factual detach-
ment, it must also validate necessity detachment and deontic detachment.
And so it does. Consider necessity detachment first. If Faith cannot avoid
the commitment to pay Pam $10, then clause (c) of (5.9') is satisfied after
all, and trivially so; for there will then be no accessible worlds that are min-
imally different with respect to Faith's displaying fidelity to this commit-
ment, on the condition that her making this commitment varies. Thus her
conditional prima facie obligation to display fidelity is transformed into an
unconditional prima facie obligation to do so, and her obligation to pay
Pam $10 is likewise transformed.
Deontic detachment also holds, although the story here is more com-
plicated. Let us look first at deontic detachment having to do with direct
prima facie obligations. Suppose that Faith has already made a promise
(promise 1) to make another promise (promise 2) to pay Pam $10. Then,
by necessity detachment, she has an unconditional prima facie obligation
to make promise 2. Given the conditional prima facie obligation to pay Pam
$10 if she makes promise 2, it should turn out that Faith has an ^condi-
tional prima facie obligation to pay Pam $10. And it does. Consider the
world-schemata represented in Chart 5.2 (where P2 stands for "promise 2"
and P for "payment"). We already know (given a proper filling in of "...")
that, because of the direct unconditional prima facie obligation to make
promise 2, Wt is better than W3. We also know that Wt is better than W2,
for this is what the direct conditional prima facie obligation to pay Pam $10
if promise 2 is made amounts to. Thus (P2&P) is better, all else being equal,
than ~(P2&P), and so Faith has a direct unconditional prima facie obliga-

W,: P2 & P & ...

W2: P2 & ~P & ...
W3:~P2 & P & ...
Chart 5.2
26 See Ross (1930), p. 22.

160
 IV,: B & P &
W2: B & ~P &
W3:~B& P &
Chart 5.3

IV,: B & P2 & P &

W2: B & P2 & ~P &
Chart 5.4

tion to do (P2&P).27 Thus she also has an incidental prima facie obligation
to do P.
Consider, next, incidental prima facie obligations. Suppose that Faith's
prima facie obligation to make promise 2 is not grounded in any already-
made promise but rather in beneficence. Beneficence requires that she
make promise 2; her obligation to do so is thus incidental and not direct.
Consider the world-schemata represented in Chart 5.3 (where B stands for
"beneficence" and P for "payment"). Since beneficence is directly uncon-
ditionally obligatory, Wt is better than W3. In addition, since B requires P2
(where P2 stands for "promise 2"), Wt and W2 must, more precisely, have
the forms represented in Chart 5.4. Since Pis directly prima facie obliga-
tory on the condition that P2 occurs, Wi is better than W2. Hence (B&P)
is unconditionally obligatory, and so P is as well.
Consider, finally, compound prima facie obligations. Suppose that
beneficence requires that Faith visit her mother and that nonmaleficence
requires that she not visit her mother without making promise 2. Then she
has an unconditional compound obligation to make promise 2. Consider
the world-schemata represented in Chart 5.5 (where B stands for "benefi-

IV,: B & N & P &

W2: B & N & ~P &
IV3: B & ~N a P &
W4: ~B & N & P &
Chart 5.5

27 Clause (c) of (5.9') is satisfied, since the only relevant prerequisite of (P2 & P) is promise
1. Note that there is no need in this context to compare Wx with
W4:~P2&~P&...
For, although this world is a ~(P2 & P)-world, it is not as close to WA as either W2 or
W3 is. (Still, I think that Wi is better than W4, and that this could be satisfactorily
proven.)

161
 W,: B & N & 13 & P
W2 : B & N & 1O & ~P
Chart 5.6

cence," AT for "nonmaleficence," and Pfor "payment"). Given the direct

unconditional prima facie obligation of beneficence, we may say that Wi
is better than W4. Given the direct unconditional prima facie obligation of
nonmaleficence, we may say that Wt is better than W3. In addition, since
(B&N) requires P2 (where P2 stands for "promise 2"), Wt and W2 must,
more precisely, have the forms represented in Chart 5.6. Since P is prima
facie obligatory on the condition that P2 occurs, Wt is better than W2.
Hence (B&N&P) is unconditionally obligatory, and so Pis as well.

5.3.4 Overriding

A straightforward account of the overriding of one prima facie obligation

by another emerges from the foregoing. Roughly, what we should say is
simply this: S's prima facie obligation to do A is overridden by his prima
facie obligation to do B if and only if every accessible world in which he
does A is deontically inferior to some accessible world in which he does B.
Or more precisely:

(5.18) S's prima facie obligation, at T in W, to do A at T' is overridden

by S's prima facie obligation, at T in W, to do B at T' iff
(a) S has a prima facie obligation, at T in W, to do A at T';
(b) S has a prima facie obligation, at T in W, to do B at T'\ and
(c) for all worlds W" such that W" is accessible to S from W at T and
S does A at T ' in W"\ there is a world W' such that
(1) W' is accessible to 5 from W at T,
(2) S does B at T ' in W\ and
(3) the deontic value for S at T of Wf is greater than the deontic
value for S at T of W".
This applies to all prima facie obligations, whether unconditional or
conditional, direct or indirect.
Notice that it is possible that direct obligations override in a manner that
fails to match the manner in which their corresponding incidental obliga-
tions override. Suppose that Ben has a direct prima facie obligation of
beneficence and that on a certain occasion this overrides a direct prima facie
obligation of fidelity. Suppose also that he has the corresponding inciden-
tal obligations to drive to Boston and to drive to Baltimore, respectively. It

162
may be that his obligation to drive to Boston does not override his obliga-
tion to drive to Baltimore; indeed, it may even be that the latter overrides
the former. This would happen if there is yet another direct prima facie
obligation, say an obligation of reparation, that overrides the obligation of
beneficence and the fulfillment of which itself requires that Ben drive to
Baltimore.

5.4 R O S S O N P R I M A FACIE O B L I G A T I O N

Since it was Ross who introduced the concept of prima facie obligation to
us, I have relied fairly heavily on what he says in the presentation of my
own analysis of the concept. I have invoked his distinction concerning
direct and incidental obligations, and I have made free use of some of his
examples of prima facie obligation (fidelity, beneficence, gratitude, repara-
tion, and so on). It is appropriate at this point to try to determine to what
extent my account squares with Ross's, at least with regard to certain key
issues.

5.4.1 Conditional and unconditional prima facie obligation

First, it might be thought that Ross does not distinguish between condi-
tional and unconditional prima facie obligation, inasmuch as he sometimes
refers to all prima facie obligation by the term "conditional duty." But in
fact I think that Ross does make this distinction but simply uses different
terminology to do so. I have in mind his comments on what he calls "spe-
cial" and "general" prima facie obligations. Special obligations, he says,
rest on the performance of some prior act (such as the making of a promise,
the infliction of an injury, or the acceptance of a benefit); the obligations
(of fidelity, reparation, and gratitude, respectively) are therefore condi-
tional on the performance of these acts. General obligations (among which
Ross counts the obligations of justice, beneficence, and nonmaleficence)
do not rest on the performance of such prior acts and may therefore be said
to be unconditional.
At least, this is how I understand Ross's distinction between special and
general obligations and what he means by "rest on." Note that, in some
other, more liberal sense of "rest on," all obligations, both conditional and
unconditional, will rest on the satisfaction of some condition, for the per-
formance of any act requires that certain conditions be satisfied. For exam-
ple, just as no one can keep a promise unless he has made one and no one

28 Ross (1930), p. 27.

163
can make reparation for an injury unless one has been inflicted, so too no
one can administer justice unless there is occasion for doing so, and no one
can act beneficently unless someone is available as a potential beneficiary.
So what Ross means by "rest on" cannot be this. Rather it is, I suggest,
what I have indicated, namely, that, with respect to fidelity, not only does
fidelity require commitment, but noncommitment-plus-nonfidelity is no
worse than (that is, is not deontically inferior to) commitment-plus-fideli-
ty, and so fidelity isn't unconditionally obligatory; similarly, noninfliction-
of-harm-plus-nonreparation is no worse (indeed, it is better, given the
prima facie duty of nonmaleficence) than infliction-of-harm-plus-repara-
tion, and so reparation isn't unconditionally obligatory; and so on. But with
respect to beneficence, for example, there is no "prior act" like that of com-
mitment (or the infliction of harm) which it is prima facie permissible to
forgo, and so nothing that is morally on a par with noncommitment-plus-
nonfidelity. All nonbeneficence is morally on a par with infidelity (that is,
with commitment-plus-nonfidelity). Or so I understand Ross. Of course,
someone might take the view that beneficence is in general only condi-
tionally required, in that its obligatoriness rests on the performance of some
prior act (such as that of becoming a neighbor of the potential beneficiary,
in some sense of "neighbor" according to which not everyone else is auto-
matically one's neighbor); but I am sure that this is not Ross's view.

5.4.2 The self-evidence of prima facie obligations

Ross says that the direct prima facie obligations he mentions are self-evi-
dent. His remarks concerning this matter clearly indicate that he believes
that it is a necessary truth that certain acts are prima facie obligatory
(whether conditionally or unconditionally).29 Such a claim of course goes
well beyond what a mere conceptual analysis could be thought to imply,
but is it consistent with what I have said?
This is a complicated matter. The account so far provides an analysis of
what it is to claim that a particular agent, S, has a prima facie obligation at
a particular time. Unless S exists necessarily, no such claim can be a neces-
sary truth. Still, we can allow for necessary truths concerning prima facie
obligation by expanding the account that has already been given. We can,
in particular, say the following:30
29 Ross (1930), p. 29.
30 Here, as in the presentation of (2.11)—(2.14) at the end of Subsection 2.1.4, acts are treat-
ed as being "repeatable." But see again the remarks made in Subsection 2.2.5 on this
matter.

164
(5.19) S has a permanent unconditional prima facie obligation in W to do
A iff for all times T and T',
if (a) there is a world accessible to S from W at T in which S does
A at T', and
(b) there is a world accessible to S from H7 at T in which 5 does
not do A at T',
then (c) S has an unconditional prima facie obligation, at T in W, to
do A at T'.
Similarly, we may also say:
(5.20) S has a permanent conditional prima facie obligation in W to do A
iff for all times T and T', there is a proposition p such that,
if (a) there is a world accessible to S from J^F at T in which 5 does
A at T' and p is true, and
(b) there is a world accessible to S from W at T in which S does
not do A at T' and^ is true,
then (c) S has a conditional prima facie obligation, at T in W7, to do A
at T', on the condition that j? is true.
A prima facie obligation is thus a permanent one if it applies on all occa-
sions where it is applicable (rather than simply on all occasions, period,
which would imply that it is always applicable, that is, can always be acted
on; this would be an extremely restrictive, and hence not very useful,
understanding of what it is for a prima facie obligation to be permanent).
We can then say:
(5.21) there is a universal unconditional prima facie obligation in W to do
A iff for all persons S,
if (a) S exists in W,
then (b) S has a permanent unconditional prima facie obligation in W
to do A;
(5.22) there is a universal conditional prima facie obligation in W to do A
iff for all persons 5,
if (a) 5 exists in W,
then (b) S has a permanent conditional prima facie obligation in W to
do A
And, finally, we can say:
(5.23) there is a necessary unconditional prima facie obligation to do A iff
for all worlds W, there is a universal unconditional prima facie
obligation in W to do A;
165
(5.24) there is a necessary conditional prima facie obligation to do A iff
for all worlds W, there is a universal conditional prima facie oblig-
ation in Wto do A.
One can then go on to claim, as Ross appears to do, that, if ever an agent
has a prima facie obligation to do some act A, this is grounded in the fact
that there is a necessary direct prima facie obligation to do either A itself or
some other act for which A is necessary.31 Although the analysis that I've
provided does not imply this claim, it is certainly consistent with it.
Nonetheless, there are complications, one minor, one major, in trying to
reconcile what Ross has to say about the self-evidence of certain particular
prima facie obligations with the account that I have provided. The minor
complication concerns the (alleged) prima facie obligation of fidelity.
Although I have allowed myself to invoke this obligation for the sake of
illustration here, it is in fact doubtful that there is a necessary direct prima
facie obligation of fidelity of the sort that Ross alleges. For what he appears
to say is that every commitment necessarily givesriseto a prima facie oblig-
ation to fulfill that commitment. This cannot be right. For one thing, peo-
ple sometimes promise to do something that they in fact cannot do. The
account that I have given implies that there is no prima facie obligation to
keep such promises.33 Of course, even if this is what Ross says, there is an
easy qualification available, and that is to say that every commitment nec-
essarily givesriseto a prima facie obligation to fulfill it, if the agent can ful-
fill it. But there is reason to think that this, too, is false. For it often happens
that people promise to do what is in fact overall wrong. What reason is there
then to think that all commitment-worlds accessible to the agent in which
he displays fidelity are better than all accessible commitment-worlds that
are minimally different in this respect? For any world that is minimally dif-
ferent in this respect will at least involve his not committing the overall
wrong in question. Once again, though, a qualification is perhaps avail-
able, although it is not so easy to see just how it should go.
31 Or, we might add, some other acts which imply that one has a compound prima facie
obligation to do A. Ross does not appear explicitly to acknowledge the possibility of
compound prima facie obligations.
32 This is most evident at Ross (1930), pp. 37 and 40, but see also pp. 21, 24, 27, and 28.
33 Some philosophers would of course respond by denying that "ought" implies "can." I
shall not discuss this issue any further here. Others would respond by claiming that appar-
ent promises to do what is impossible are in fact not genuine promises, and then claim
that every genuine promise does in fact ground a prima facie obligation to fulfill it. (See,
e.g., Martinich, 1987.) This seems mistaken, in that it seems possible that a person should
sincerely, and thus genuinely, promise to do something, in the mistaken belief that he
can do that thing.
34 Compare what Donagan says about promising, in Donagan (1977), pp. 92-3.

166
 The major complication is this, and it applies to all the particular prima
facie obligations that Ross alleges to be self-evident, except perhaps the
prima facie obligation to promote intrinsic value.35 The account that I have
provided relies on the notion of minimal difference, where this is explicit-
ly tied to worlds that are accessible to the agent in question. But surely it is pos-
sible that some agents be such that they cannot avoid some evil's occurring
if they satisfy some Ross-alleged direct prima facie obligation; in such a
case, given that the evil is sufficiently great (so that it greatly diminishes the
deontic value of any world in which it occurs and renders any world in
which it does not occur very valuable, deontically, relative to any world in
which it does occur), it will not be the case that all accessible worlds in
which the alleged obligation is satisfied are deontically superior to all acces-
sible worlds that are minimally different in this respect. Hence this alleged
obligation does not hold for such people; hence it does not hold of neces-
sity; hence it is not self-evident that it holds. As an illustration of this, con-
sider the possibility that Martha will commit mass murder if and only if
Malcolm acts nonmaleficently, and that Malcolm can do nothing about
this. We may suppose (barring certain Kantian considerations) that not all
worlds accessible to Malcolm in which he acts nonmaleficently are deon-
tically superior to all worlds accessible to him that are minimally different
in this respect. In fact, it may well be that the very reverse is true, so that
Malcolm has (on my account) a prima facie obligation to act maleficently.
Thus it cannot be, on the account that I have provided, that there is a
necessary unconditional prima facie obligation to act nonmaleficently. So,
too, for beneficence and other obligations that Ross calls "general."
Similarly, the conditional obligations of fidelity, gratitude, and so on that
Ross calls "special" cannot hold of necessity, on the account that I have
provided.
If we are to accommodate what Ross says about the self-evidence of
such prima facie obligations, therefore, a different account must be given.
Let me indicate briefly here how I think this account should go. First, we
would need to revise our understanding of minimal difference, so as not to
restrict our focus to worlds accessible to the agent. We would simply revise
(5.7) by saying:

(5.7') W' is minimally different from W with respect to S doing A at T'

iff

35 The example I give below concerns an act that is plausibly thought to involve a great
intrinsic evil. But if there are acts that do not involve great intrinsic evils but which
nonetheless greatly diminish the deontic values of worlds in which they occur, then even
the prima facie obligation to promote intrinsic value is subject to this complication.

167
 (a) S does ,4 at T ' i n W\
(b) S does not do A at T ' in W'\ and
(c) there is no world W" such that
(1) S does not do A at T ' in W", and
(2) W" is closer to W than W' is.

(5.8) would be similarly revised. Then (5.10) would be revised as follows:

(5.10') S has, at T in W, a direct conditional prima facie obligation to do

A at T ' , on the condition thatp is true, iff
(a) there are a world W' and a world W " such that
(1) W' is accessible to S from W at T,
(2) W " is accessible to 5 from W at T, and
(3) H 7 " is minimally different from W' with respect to S doing A
at T', on the condition thatp is true; and
(b) for all worlds W' and W"\
if (1) W7" is minimally different from Wf with respect to 5
doing A at T', on the condition that p is true,
then (2) the deontic value for 5 at T of W' is greater than the
deontic value for S at T of W .

Changes to (5.9') and (5.11)—(5.24) would then be made accordingly. On

this approach, the personal optionality of what is prima facie obligatory is
preserved (see subclauses (al) and (a2) of (5.10')), but it could still be that
Malcolm has a prima facie obligation to act nonmaleficentry despite his
inability to avoid Martha's committing mass murder if he does so act.
Why, then, did I not simply proceed in this fashion in the first place?
The answer has to do with detachment. It is not that I think such a proce-
dure mistaken; on the contrary, it gives us an account of what might be
called prima facie obligations "in principle." W e could then say, with Ross,
that everyone necessarily has a reason, in principle, to act nonmaleficently;
as noted, this would apply even to Malcolm. But the problem is that,
although versions of necessity detachment and deontic detachment do
indeed hold for conditional prima facie obligations in principle, they do not
warrant certain judgments that we wish intuitively to make. For example,
the version of necessity detachment that concerns prima facie obligations
in principle involves inferences of this form: if all ^-worlds in which S does
A are better than all p-worlds that are minimally different in this respect,
and all worlds are p-worlds, then all worlds in which S does A are better
than all worlds that are minimally different in this respect. Note: "all

168
worlds," not just "all accessible worlds"; this is because minimal difference
is now being understood in terms of possible, and not just accessible,
worlds. Now, with necessity detachment so understood, we could not
employ it to infer, as we surely wish to, that, for example, someone who
has made a promise has an unconditional prima facie obligation to keep it.
For, although a promise is inevitable once made, it is not necessary, in the
stronger sense of "necessary" now at issue. Similarly, for deontic detach-
ment with respect to prima facie obligations in principle, the uncondition-
al obligation that triggers the detachment must itself be an obligation in
principle, and this fails to validate certain detachments that we intuitively
wish to accept (such as that of detaching an unconditional obligation to do
A by appealing to an unconditional obligation to promise to do A, the lat-
ter having itself been detached - by necessity detachment - owing to con-
siderations having to do merely with inevitability and not with necessity in
the stronger sense).
The fact is that there are, I think, two types of prima facie obligations,
and thus there are also two categories of detachment. There are prima facie
obligations in principle, and there are prima facie obligations in practice. Thus
there are also (necessity and deontic) detachment in principle (where the
obligations in question are prima facie obligations in principle) and
detachment in practice (where the obligations in question are prima facie
obligations in practice). The detailed account that I have provided is of
prima facie obligations and detachment in practice; the account that I have
sketched in the last two paragraphs is of prima facie obligations and detach-
ment in principle. We need both. We need the latter to accommodate what
Ross says about the self-evidence and necessity of prima facie obligations;
we need the former to accommodate common intuitions concerning the
detachment of prima facie obligations.

5.4.3 Prima facie obligation and overall obligation

Let me turn now to what Ross has to say about the relation between prima
facie obligation and overall obligation. Here he makes two important
claims. The first is this:
I suggest 'primafacie duty' or 'conditional duty' as a brief way of referring to the
characteristic.. .of being an act which would be a duty proper if it were not at
the same time of another kind which is morally significant.36

36 Ross (1930), p. 19.

169
This might be interpreted as follows:
(5.25) S has a prima facie obligation to do A iff S would have an overall
obligation to do A if any prima facie obligation that 5 has not to do
A were overridden.

Does my account imply that this is true?

It might be thought that any account of obligation that purports to cap-
ture Ross's concept of prima facie obligation must imply that (5.25) is true.
But the fact is that this thesis is not implied by my account (although they
are consistent with one another). The implication fails because my account
of prima facie obligation incorporates, as noted earlier, a certain generality,
and such generality, which I take to be essential to prima facie obligations,37
is not incorporated in my account of overall obligation. Does this show up
a failing in either of these accounts?
I don't think so. Ross seems to think that there are several basic, inde-
pendent values (such as those of fidelity, gratitude, and so on); that noth-
ing else is valuable (at least with respect to the determination of obligation)
unless it is a means to the promotion of these values; and that there is always,
indeed necessarily, a deontic reason to promote these values. If we grant all
this, and we also grant (as seems natural) that the deontic value of a world
is a direct function of those deontically relevant values which are promot-
ed at that world, then, given the foregoing accounts of prima facie and
overall obligation, it seems that we could argue cogently for (5.25); for
there will be no deontically relevant value that does not correspond to some
prima facie obligation.38 But whether or not we should grant all this seems
to me to go well beyond questions of conceptual analysis; it is a normative
question. (See Subsection 2.3.3.) This being the case, it is not a failure in
(I) that it fails to imply (5.25). In fact, it would be a failure in (I) if it betrayed
such a normative bias.
Nonetheless, (5.25) might seem promising as the basis of an analysis of
the concept of prima facie obligation, an analysis that is quite different from
the one that I have proposed. Of course, (5.25) could not itself be accept-
ed as such an analysis, because it would be circular. But consider this
proposal by Frank Snare:

37 Indeed essential, in some form or other, to reasons of all sorts.

38 This is not to say that it is easy to see just how this argument would go. There are a num-
ber of complications, having to do with the distinctions between conditional and uncon-
ditional obligations, obligations in principle and obligations in practice, permanent versus
universal versus necessary obligations, and with other matters.

170
(5.26) 5 has a prima facie obligation to do A iff for some set T of act-
types
(a) A is a token of some member of T, and
(b) if the omission of A were not also a token of some member of T,
then S would have an overall obligation to do A.39

Intuitively, T is supposed to be that set of morally significant act-types in

virtue of which an act-token may be prima facie obligatory. Circularity is
avoided, however, since such identification of T is not necessary for the
statement of (5.26). In addition, the proposal would appear to avoid the
normatively biased implication that whatever is overall obligatory is prima
facie obligatory, insofar as it requires that, to be prima facie obligatory, A
be a token of some member of T, whereas this is not said to be a require-
ment for an act's being overall obligatory.
But Snare's proposal is nonetheless problematic. It would seem to fall
afoul of the fact that subjunctive conditionals (of the sort employed in the
proposal) warrant factual detachment. For might it not happen that,
although the antecedent of clause (b) in (5.26) is satisfied, S ought so to act
that it is not satisfied? Consider this case. Ben is in a position at T1 to per-
form some beneficent act A at T 2 . Let us assume that acting beneficently is
a type of act that is a member of set T, and that Ben (therefore) has a prima
facie obligation at Tt to do A at T2. Let us suppose also that, if he omits to
do A at T 2 , he will not thereby perform an act that is of a type that is also a
member of T. In this case, (5.26) implies that Ben has an overall obligation
at Tt to do A at T2. But suppose that Ben ought overall at T1 to make and
then keep a certain promise, and that the keeping of it would require him
to omit to do A at T2. He doesn't make this promise, and that is why his
omitting to do A is not otherwise morally significant. (The omission is not
an instance of fidelity, for example.) But he ought overall both to make it
and to keep it, and so (contrary to what (5.26) implies) he ought overall not
to do A.
One might respond to this case by maintaining that Ben's omitting to
do A must be morally significant: even though it is not an instance of fidelity
(for he doesn't make the promise), it must nonetheless be an instance of
some member of T, since after all he does not do what he overall ought to
do. But this would be to revert to the normative bias of (5.25), for it would
be to insist that nothing is overall obligatory unless it exemplifies some
member of T, that is, unless it is prima facie obligatory.

39 Snare (1974), p. 237, somewhat revised. A similar proposal is made in Atwell (1978).

171
 Another response would be to try to fix Snare's proposal by tinkering
with its clause (b) so as to circumvent this objection,40 but I don't think this
would do. For the proposal in fact does not escape the normative bias of
(5.25) after all. This is because counterfactuals of the sort used in its clause
(b) are true if both their antecedent and their consequent are true. Now,
suppose that S has an overall obligation to do A; then (b)'s consequent is
true. Next, take any type of action of which A is a token and let T be the
set which has that type as its only member; then (a) is true and (b)'s
antecedent is also (likely to be) true. Hence both (a) and (b) are true, and
from this Snare's proposal allows us to infer that 5 has a prima facie oblig-
ation to do A.
For these and other42 reasons, then, I think that Snare's proposal is to be
rejected, and in general the project of giving an analysis of the concept of
prima facie obligation on the basis of (5.25) seems to me unpromising.
I began this subsection by saying that Ross makes two important claims
about the relation between prima facie obligation and overall obligation.
The second claim is this:
Every act..., viewed in some aspects, will be prima facieright,and viewed in
others, prima facie wrong, and right acts can be distinguished from wrong acts
only as being those which, of all those possible for the agent in the circum-
stances, have the greatest balance of prima facietightness,in those respects in
which they are prima facie right, over their prima facie wrongness, in those
respects in which they are prima facie wrong.43
This might be interpreted as follows:
(5.27) S has an overall obligation to do A iff S has a prima facie obliga-
tion to do A that overrides any prima facie obligation that S has not
to do A.
Does my account imply that this is true?
Again, the answer is that it does not. The reason for this is the same
as that given before when discussing (5.25): there is no guarantee that

40 For example, one might try modifying clause (b) so that it reads as follows: if the omis-
sion of A were not also a token of some member of T, andifit were not the case that S ought
[or, better, may] overall so act that the omission of A is a token of some member of T, then S
would have an overall obligation to do A.
41 Cf. Lewis (1973), p. 26; Pollock (1976), p. 38.
42 The presupposition that there is a single act-token corresponding to the omission of A
is problematic, as is the presupposition that obligatoriness applies to act-tokens. On the
latter point, see Subsection 2.2.5.
43 Ross (1930), p. 41.

172
the generality inherent in prima facie obligation is also present in overall
obligation.
Even though (I) implies neither (5.25) nor (5.27), it is certainly consis-
tent with them. With respect to (5.27), though, this claim might be dis-
puted, on the grounds that Ross conceives of it as an analysis.44 If overall
obligation is to be understood in terms of prima facie obligation, as this
interpretation of (5.27) contends, then it is not to be analyzed indepen-
dently of the concept of prima facie obligation, as (I) contends.
Now, I think it must be granted that, if (5.27) is accepted as an analysis
of overall obligation, my proposed analysis of overall obligation must be
rejected.451 therefore reject (5.27) as an analysis (though I grant that it may
express a necessary truth). How does Ross himself view (5.27)? The evi-
dence seems to me inconclusive: although some passages46 in The Right and
the Good suggest that he does accept (5.27) as an analysis, he never states this
explicitly; moreover, other passages47 suggest that he eschews any such
analysis. At any rate, if Ross does accept (5.27) as an analysis, I oppose him
on this point. I leave it to you to judge whether this indicates an inadequa-
cy in my account.
One reason to suspect that it does indicate an inadequacy might be this.
As has been noted, Ross claims that (certain of) one's prima facie obliga-
tions are self-evident, but he also claims that one's overall obligations are
not and that one may appeal to the former when trying to determine the
latter. This is surely plausible, but it may appear to indicate a priority of
prima facie obligation over overall obligation that is inconsistent with (I).
Not so. As Snare has pointed out, the logical or conceptual independence
of overall obligation from prima facie obligation is not inconsistent with
the former being epistemologically dependent on the latter.48 Similarly,
one might claim that sometimes it is appropriate to appeal to one's prima
facie obligations in order to explain what one's overall obligations are.49 But,
once again, the account given in (I) is not inconsistent with such priority
of prima facie obligation over overall obligation. For it may well be that
certain matters are, and that we find them to be, of fundamental moral
44 Others have certainly proposed analyses along the lines of (5.27). See, in particular,
Chisholm (1974), pp. 11-13. See also McCloskey (1963), pp. 336 and 343; Loewer and
Belzer (1991), pp. 365-6.
45 This claim presupposes that there cannot be two acceptable analyses of a single concept,
where one of these analyses itself employs concepts that the other does not.
46 Ross (1930), pp. 41 and 46-7.
47 Ross (1930), pp. 10-12 and 16.
48 Snare (1974), p. 244.
49 Cf. again Ross (1930), p. 46. Cf. also Pietroski (1993), p. 513.

173
significance —well-being, perhaps, or respect, or integrity, or fidelity, or jus-
tice, or freedom, or something else — such that our overall obligations turn
on, and we believe them to turn on, whether these values are promoted.
Far from wishing to deny this, I believe it to be true. No doubt, then, we
will agree that there are prima facie obligations, both in principle and in
practice, having to do with the promotion of these values, so that some-
times, when we deliberate about what we ought overall to do in some sit-
uation, we may bear in mind that these prima facie obligations apply in this
situation. But this is quite consistent with the analyses of overall and prima
facie obligation that have been proposed here.

5.4.4 Residual obligation

Steve missed his dinner date with Dave. Although he was perfectly justi-
fied in doing so - for Carl was in dire need of his ministrations - still there
was a cost involved, a moral cost. He broke his promise, and Dave was dis-
appointed. For this reason, Steve's helping Carl appears to have left a cer-
tain "moral residue," as it is sometimes put. Ross distinguishes two types of
moral residue, one involving emotion, the other action. He says:

When we think ourselves justified in breaking, and indeed morally obliged to

break, a promise in order to relieve some one's distress, we do not for a moment
cease to recognize a prima facie duty to keep our promise, and this leads us to
feel, not indeed shame or repentance, but certainly compunction, for behav-
ing as we do; we recognize, further, that it is our duty to make up somehow to
the promisee for the breaking of the promise.50

Few would quarrel with this, and yet it is a fact that my account of prima
facie obligation does not imply that it is true. Is this a problem with the
account?
I cannot see that it is. There might be a problem if the account implied
that Ross's view is false, but it doesn't. It is neutral with respect to the view,
and this seems perfectly acceptable, since the view is clearly not simply a
conceptual one; it is normative.51 To say that it is appropriate to feel such
compunction (as Ross obviously believes; it is not just that we are led to feel
it, we ought to feel it),52 and to say that there is a (presumably prima facie)
obligation to make amends — these are normative claims. However mis-

50 Ross (1930), p. 28.

51 "Clearly" may be a bit bold, given the remarks made in the penultimate paragraph of
Subsection 2.3.3.
52 This is a nonbinding sense of "ought," for reasons given in Subsection 3.1.7.

174
taken someone might be to deny them, the mistake would not be a concep-
tual one.
Indeed, although the view may be natural, it is clearly not uncontro-
versial. Some would claim that, as long as one does what is morally justifi-
able, then there is no cause for compunction at all.53 (Donagan's position,
cited in Section 5.1, would seem to suggest this.) Although I think that this
is false, nonetheless we would do well to note two additional points. First,
even if compunction (whatever that is, exactly) is called for in cases like
Steve's, certain other "negative" moral emotions (such as guilt or shame)
certainly are not; second, even if compunction is called for, still, as long as
one has clearly done what one overall ought, an overall sense of moral ease
would seem appropriate.54
Similarly, some would claim that, as long as one acts inculpably, then
even if, like Steve, one breaks a promise to someone, one does not owe that
person anything. Indeed, /would claim this (where it is understood that to
say that S owes S' something is to say that 5 has an obligation to S' to do
something. Here I am talking of what in Section 1.3 I called "obligation-
to" as opposed to "mere obligation." I shall say more about obligation-to
in Section 5.5). I acknowledge that this claim may well initially appear quite
implausible; my reasons for making it are too complicated to go into here,
however.55 Even if I am right, though, this still leaves untouched the close-
ly related claim that, for example, Steve has a prima facie obligation (even
if not a prima facie obligation-to-Dave) to make amends to Dave. This
seems to me likely to be true, although even here there are difficulties. First,
it must be acknowledged that the case could be filled in so that Steve clear-
ly doesn't have such a prima facie obligation; any scenario in which Steve is
incapable of making amends to Dave would have this implication. (Steve
might die; Dave might die or move away; and so on.) Second, it seems clear,
on reflection, that not every failure to satisfy a prima facie obligation gives
rise to a residual obligation to make amends, even when one is not inca-
pacitated in the relevant way. Suppose that Steve had a prima facie obliga-
tion to visit Dave on Monday, grounded not in any commitment to Dave
but simply in beneficence. Had he then helped Carl en route and thereby
rendered himself unable to visit Dave on Monday, certainly no explanation
or apology to Dave would seem called for.56 Perhaps a residual obligation

53 Gowans discusses this as an implication of what he calls the "Elimination Thesis" (as
opposed to the "Remainders Thesis"), in Gowans (1994), pp. 91-2.
54 Cf. Zimmerman (1993c) on the possibility of the "resolution" without "dissolution" of
ambivalence.
55 See Zimmerman (1994).
56 Cf. McConnell (1995); Pietroski (1993), p. 510, n. 27.

175
to visit Dave on Tuesday emerges from this, but perhaps not; for perhaps
the condition that gave rise to the desirability of Steve's visiting Dave on
Monday passes before Tuesday arrives.
Indeed, this seems to get to the heart of the matter. Is there a condition,
arising out of the failure to satisfy the original prima facie obligation, that
gives rise to the desirability of some further action? If so, then it is likely that
some residual prima facie obligation will have emerged. But not necessari-
ly; the agent might not have the relevant capacity. Indeed, it may not be
only the original agent who incurs such an obligation. If Steve was walk-
ing with Willis when he came upon Carl, then perhaps, while Steve was
busy tending to Carl, Willis had the obligation to call Dave and let him
know what was happening.57 Here an obligation emerges from Steve's
breaking his promise to Dave, but it is Willis's, not Steve's.
In general, then, we may say that, if residual obligations emerge from
the failure to satisfy prima facie obligations, this is perfectly consistent with
the account of prima facie obligation presented here. Whether or not they
do, however, is a complicated business, and it is certainly no defect in the
account that it fails to imply that they do.
Finally, it should be noted that residual obligations can arise not just from
the failure to satisfy prima facie obligations. They can arise also where an
overall obligation is not satisfied (whether or not this involves the failure to
satisfy a prima facie obligation). As discussed in Section 3.2, often, when an
overall obligation is not satisfied, a shift in obligations occurs, and a new
overall obligation emerges. Thus Mandy became obligated to babysit Bill's
child, and Victor became obligated to visit Bert. Also, as discussed in
Section 4.4, often, when an n-level overall obligation is not satisfied, an
tt+1-level overall obligation emerges. Thus you became obligated to vote
for Dick, and Alan became obligated to attend the meeting on the second
floor. These are all naturally classified as residual obligations; just like
Steve's residual obligation to apologize to Dave, they have emerged from
the failure to satisfy some initial obligation.

5.5 RIGHTS
5.5.1 The correlativity of rights and obligations
The term "right" is ambiguous, even when used as a noun. Wesley
Newcomb Hohfeld made a lasting contribution to our understanding
of rights when he distinguished among four species of them: "claims,"
57 Cf. McConnell (1995).

176
"liberties," "powers," and "immunities."58 According to him, the first two
are related in this way (to put the matter somewhat roughly):
(5.28) S' has a claim against S that S do A iff S has no liberty [is not at lib-
erty] as regards S' not to do A.
Hohfeld also held that a similar relationship binds powers and immunities.
My interest here is with how moral rights are related to moral obliga-
tions. This issue has already been briefly addressed in Section 1.3. There the
following four propositions (as formulated by Ross) were distinguished:
(1.4) a right of A against B implies a duty of B to A;
(1.5) a duty of B to A implies a right of A against B;
(1.6) a right of A against B implies a duty of A to B;
(1.7) a duty of A to B implies a right of A against B.
I gave reasons for rejecting (1.6) and (1.7) but said that I accept (1.4) and
(1.5). I wish now to elaborate a little on this treatment of (1.4) and (1.5)
The rights at issue in the foregoing statements are claims. Hohfeld, too,
accepts (1.4) and (1.5), which may now be formulated more fully as fol-
lows:
(1.4') if S' has a claim at T in W against S that S do A at T', then S has an
obligation at T in W to S' to do A at T';
(1.5') if S has an obligation at T in W to S' to do A at T\ then S' has a
claim at T in W against 5 that S do A at T'.
Combined, these statements yield the thesis that claims and obligations are
in a certain way correlative to one another. The following points should be
noted.
First, given the relationship between claims and liberties just noted in
(5.28), (1.40 and (1.50 imply the following:
(5.29) S has an obligation at T in W to S' to do A at T' iff 5 is not at lib-
erty at T in W as regards S' not to do A at T'.
Second, the correlativity of claims and obligations is compatible with
the normative priority of claims over obligations. It seems plausible to say
that obligations derive from claims, rather than vice versa. For example, it

58 Hohfeld (1919), p. 35ff. Hohfeld was explicitly concerned with giving a taxonomy of
legalrights,but his account is plausibly extended to moralrightsalso. Also, Hohfeld tend-
ed to prefer the term "privilege" to the term "liberty," although he acknowledged the
appropriateness of the latter (at Hohfeld, 1919, pp. 42-3). Most commentators appear to
prefer "liberty" to "privilege."

177
is a promisee's claim against a promiser that gives rise to the latter's duty;
the duty doesn't give rise to the claim. If this is so, and whatever exactly it
involves, it is perfectly consistent with the joint assertion of (1.4') and (1.5').
Third, (1.4') and (1.5') are intended to be consistent with the possibil-
ity that there be posthumous rights and also with the possibility that there
be rights held by persons not yet born or even conceived. In such cases it
would have to be true that S' has a claim at T even though S' does not exist
at T. I doubt whether this is indeed possible, but I don't wish to rule it out
here.59 Note that allowing for this possibility does not require allowing for
the possibility that there be post- or prevital obligations. Clearly it is impos-
sible for someone who has existed but will not exist to have obligations, for
such a person no longer has any choice among accessible worlds. I would
say, too, that it is impossible for a not-yet-existing person to have obliga-
tions (even remote ones), for I think that there being worlds accessible at
some time to someone requires that that person actually exist at that time.
But these are all difficult metaphysical issues that need not be resolved here.
Fourth, (1.4') and (1.5') serve to emphasize the distinction, introduced
in Section 1.3, between mere obligation and obligation-to. S's having an
obligation-to-S' is not the same as S's having an obligation towards or con-
cerning S\ Whether or not S has an obligation-to-5' to do A is a question
of whether or not S owes it to S' to do A; whether or not S has an obliga-
tion concerning S' is a question of whether or not the obligatory action,
A, itself somehow involves S'. Often, of course, the two will coincide. If
Pru has promised Zoe to take her to the zoo, then (let us assume) Pru has
an obligation-to-Zoe to take Zoe to the zoo. Here, Pru's obligation-to-
Zoe also concerns Zoe. But sometimes the one to whom an obligation is
owed is not the one whom the obligation concerns. If Pru has promised
Ollie to take Zoe to the zoo, then Pru's obligation still concerns Zoe, but
it is an obligation she owes to Ollie, not Zoe.60
Fifth, (1.4') and (1.5') serve to emphasize a related distinction, that
between mere claims and claims-against. For all that I have said so far, it
may be that someone can have a claim concerning something that is not a
claim-against-someone (or -something)61 to that thing. Thus (1.4') and
(1.5') do not assert that there is a correlativity between claims and obliga-

59 Cf. Feinberg (1980), pp. 173-6 and 180-2.

60 Cf. Hart (1979), p. 18.
61 It is sometimes asserted that we can, and do, have claims against entities that are not them-
selves individual persons. If there is to be correlativity in this sort of case, then it must be
that these entities can and do have obligations. This is an issue that I shall take up in
Chapter 9.

178
tions generally; they assert that there is a correlativity between claims-
against and obligations-to in particular. Objections that have been raised
against the correlativity thesis would appear to succeed, if they succeed at
all, only in impugning the general thesis; they leave the more particular
thesis untouched.
For example, David Lyons notes:
It is often said...that one should be charitable and generous — that one has a
duty to behave in that way. This appears to make good sense, even though no
other person could have a right to one's charity or generosity (for if the other
had a right, then what would be required of one would not be classifiable as
charity or generosity).

Of course, it is controversial whether or not one has an obligation to dis-

play charity or generosity. As noted in Section 1.3, it is arguable that all
obligation is obligation-to, and thus it might be argued that, if no one has
a claim to charity, then no one has an obligation to display charity. This,
however, serves only to confirm the correlativity of claims-against and
obligations-to. If there is an obligation to display charity but no claim to
charity, then the obligation simply isn't an obligation owed to anyone (or
anything)63 in particular, and so the more particular correlativity thesis is
not shown false by these considerations.64
Similarly, Joel Feinberg states:

The manifesto writers...who seem to identify needs, or at least basic needs,

with what they call "human rights," are more properly described...as urging
upon the world community the moral principle that all basic human needs
ought to be recognized as claims.. .worthy of sympathy and serious considera-
tion right now, even though, in many cases, they cannot plausibly be treat-
ed. . .as grounds of any other people's duties.65

Although Feinberg does not wholeheartedly endorse this way of talking,

he is nonetheless sympathetic to it. The point to be made here, though, is
simply that, even if we accept the propriety of talking of basic needs as
constituting claims in those cases where no one is in a position, and hence
no one has an obligation, to satisfy those needs, all this would mean is that

62 Lyons (1979a), p. 11.

63 Just as it might be (as stated in note 61 to this chapter) that entities that are not themselves
individual persons have obligations, so too it might be that such entities have claims.
64 Lyons would appear to agree; see Lyons (1979b), pp. 60-1. So too would Feinberg;
see Feinberg (1979), p. 79.
65 Feinberg (1979), p. 89.

179
it is possible for there to be claims that are not claims held against anyone
(or anything) in particular; and so, once again, the more particular correl-
ativity thesis is not threatened.
There is one sort of case, however, which I think poses a more serious
challenge to this correlativity thesis, in particular to the latter half of it,
(1.5'). It is commonly said that some moral duties are self-regarding. For
instance, Kant is famous for asserting that there is a duty not to commit sui-
cide and a duty to develop one's talents.66 Of course, this might simply be
interpreted as an assertion that we have duties concerning ourselves, in which
case they would not pose any challenge to (1.5'). And, even when so weak-
ly interpreted, some would dismiss the assertion on the grounds that moral-
ity has to do only with how we treat other people. This seems mistaken,
though. Considerations concerning self-respect would seem to indicate
that morality does indeed have in part to do with how we treat ourselves.67
If this is granted, then I don't think we can rest content simply with saying
that some of our duties concern ourselves. It would seem, rather, that we owe
it to ourselves to do certain things for ourselves. Yet it also seems odd to say
that we have claims against ourselves. This is the challenge to (1.5').68
Other than dismissing self-regarding duties altogether or simply rele-
gating them to the category of duties that merely concern the agent, there
are two possible responses to this challenge, short of capitulation. One is,
of course, to bite the bullet and insist that we do indeed have claims against
ourselves; it is just that we are not likely to press or stand on these claims if
they appear to be in danger of being flouted. I am not sure what to make
of this response. A second response is to try to plumb a middle ground
between mere obligation concerning oneself and obligation-to-oneself.
According to this response, self-regarding obligations, though not strictly
owed to oneself, nonetheless have an essential connection to oneself in that
they are grounded in considerations of self-respect; thus they do not mere-
ly concern oneself by accident, as it were.
I find the second response promising, although I concede that, unela-
borated, it remains pretty obscure. It is not my intention to try to clarify it
here, though. If it should ultimately fail, then perhaps we must grant that
(1.5') is false. Even then a closely related claim may be accepted, and that
is (1.5') conjoined with the stipulation that 5 is distinct from S'. ((1.4') will
still stand without qualification.)

66 Kant (1964), pp. 89-90.

67 Cf. Hill (1979).
68 Cf. Lyons (1979a), pp. 11-12.

180
 5.5.2 Absolute and prima facie rights
Like Ross, I am inclined to say that those obligations to which rights
(claims) are correlative are prima facie, and (1.4') and (1.5') are intended to
be so construed. This means that rights are themselves prima facie and so,
like the obligations to which they are correlative, subject to being overrid-
den.69 It means also that all those distinctions made earlier with respect to
prima facie obligations — concerning whether they are direct or indirect, in
principle or in practice, unconditional or conditional, and whether they are
permanent, universal, or necessary - apply also to rights. Consider the
(alleged, and suitably qualified) right of a promisee against a promiser to the
latter's fidelity. Corresponding, as it supposedly does, to a direct prima facie
obligation, it will be a direct right. As such, it can and will ground all sorts
of incidental rights; for example, if the promiser must drive to Boston to
ensure that he manifests fidelity, then the promisee has the right that the
promiser drive to Boston. Also, as was noted earlier, the prima facie obli-
gation of fidelity is a conditional one, the condition being that a commit-
ment is made; if and when the commitment is made, then the obligation
becomes unconditional (at least in practice, if not in principle). So too,
then, a (potential) promisee's right is conditional on the promise being
made; if and when it is made, the right becomes unconditional. Finally, if
a (suitably qualified) prima facie obligation of fidelity is a necessary one,
then the correlative right is necessary also.
Note that the failure to satisfy a right that is prima facie does not neces-
sarily constitute overall wrongdoing. If the right is merely prima facie and
one fails to satisfy it because one satisfies instead another right that overrides
it, then one does no overall wrong (unless there is a yet more stringent right
that one has failed to satisfy). Useful terminology to adopt here is Judith
Thomson's:70 we may say that, if a right (that is, a claim) is not satisfied, then
it is infringed; if such infringement is overall wrong, then the right is violat-
ed. We may also say that, if a right has been violated, then the violator has
not only done (overall) wrong, he has done a wrong; equivalently, he has
wronged the right-holder.
Some philosophers, however, conceive of rights as correlative not to
merely prima facie obligations but to overall obligations. This is to conceive
of rights as absolute. Let us now investigate this conception of rights.

69 Roughly, a right to X overrides a right to Yjust in case the obligation that corresponds
to X overrides the obligation that corresponds to Y. Cf. Thomson (1990), for a recent,
thorough treatment of rights understood as prima facie.
70 See Thomson (1986), p. 40, and (1990), p. 122.

181
 First, it will be helpful to distinguish degrees of the absoluteness of rights.
There are at least six. We may say, roughly, that someone has an absolute
right of the lowest degree - call it a grade-1 right -just in case someone
owes him a certain overall obligation; he has a grade-2 right just in case
someone always owes him such an obligation; he has a grade-3 right just in
case everyone always owes him such an obligation; there is a grade-4 right
just in case everyone of a certain sort is such that everyone always owes him
such an obligation; there is a grade-5 right just in case everyone is such that
everyone owes him such an obligation; and there is a grade-6 right just in
case it is necessarily the case that everyone is such that everyone always owes
him such an obligation. More precisely:

(5.30) S' has a grade-1 absolute claim at T in W against S that 5 do A at

T iff 5 has an overall obligation at T in W to S' to do A at T';
(5.31) S' has a grade-2 unconditional absolute claim in W against S that
S do A iff for all times T and T',
if (a) there is a world accessible to S from W at T in which S does
AztT', and
(b) there is a world accessible to 5 from W at T in which 5 does
not do A at T',
then (c) S' has a grade-1 absolute claim at T in W against S that S do
A at T\
(5.32) S' has a grade-3 unconditional absolute claim in W that A be done
iff for all persons S,
if (a) S exists in W,
then (b) S' has a grade-2 absolute claim in W against S that S do A;
(5.33) there is a grade-4 unconditional absolute claim in W that A be done
iff for all persons S',
if (a) S' exists in W and has a certain characteristic C,
then (b) S' has a grade-3 absolute claim in W that ^4 be done;
(5.34) there is a grade-5 unconditional absolute claim in W that A be done
iff for all persons S',
if (a) S' exists in W,
then (b) S' has a grade-3 absolute claim in W that A be done;
(5.35) there is a grade-6 unconditional absolute claim that A be done iff
for all worlds W, there is a grade-5 absolute claim in W that A be
done.

I think it is clear that other degrees of absoluteness might be distinguished,

but the foregoing should suffice to give an idea as to how absoluteness can
come in degrees. (Note that (5.33) can itself be "subdivided," depending

182
on what the characteristic, C, is taken to be. It might be a relatively simple
characteristic — innocence, for example — or a relatively complex charac-
teristic — innocence-cum-parenthood, say.) In addition, each of the fore-
going statements is supposed to be adequate to an account of so-called
negative rights (where, to be more explicit, reference to doing A would be
replaced by reference to not doing A).
The foregoing account of unconditional absolute rights can be straight-
forwardly supplemented with an account of conditional absolute rights.
We can say:
(5.36) S' has a grade-1 absolute claim at T in W against S that 5 do A at
T', on the condition that p is true iff S has an overall obligation at
T in W to S' to do A at T', on the condition thatp is true;
(5.37) S' has a grade-2 conditional absolute claim in W against S that 5 do
A iff for all times T and T', there is a condition^ such that,
if (a) there is a world accessible to 5 from W at T in which S does
A at T' andp is true, and
(b) there is a world accessible to 5 from W at T in which S does
not do A at T' and p is true,
then (c) S' has a grade-1 absolute claim at T in W against S that S do
A at T', on the condition that|> is true;
and so on through the various other grades. It might thus be claimed (truth-
fully or not; presumably not) that there is a grade-6 conditional absolute
right to fidelity.
Now, certainly no proponent of rights will quarrel with the claim that
some moral rights are grade-1 absolute. For even those who regard rights
as correlative to prima facie obligations will accept that sometimes the
obligation in question is not merely prima facie but overall. But problems
arise when it is claimed either that some moral rights are grade-2 absolute
(or grade-3, or higher) or that all moral rights are absolute (whether of grade
1 or higher). Let us examine these issues.
To claim that a moral right is grade-2 absolute (or higher) is to claim that
its unoverridden status holds on more than one occasion. The greater the
degree, the greater the generality of this status. What is problematic is that,
given the vast variety of considerations that may come into conflict with
that consideration which the right in question represents, the greater the
generality of "unoverriddenness" that one seeks to establish in the right,
the less likely it seems that one will succeed in establishing it. Still, some
have claimed success in such an endeavor. Alan Gewirth, for example, has
argued that a mother's right not to be tortured to death by her son is
183
absolute, 'where by "absolute right" he means a right that "cannot be over-
ridden in any circumstances."71 Here he appears to be contending that the
right in question is at least grade-4 absolute. (He claims that there are other
such rights, too, grounded in the general principle that there is an absolute
prohibition against degrading persons.)721 shall not dispute what Gewirth
says here, although it is surely open to debate. What can be said here is this:
if there are any such rights, there are surely not many of them. No one, for
example, is likely to wish to grant such an absolute status to the right to be
compensated for injury, or to the right that others not damage one's prop-
erty, and so on.
In general, then, it would appear that the greater the degree (that is, the
higher the grade) of absoluteness, the fewer absolute rights there are. But
this claim must now be qualified; for it can be argued that there are just as
many absolute rights as there are prima facie rights. One can argue this by
employing a rather obvious technical move (one that Gewirth himself
repudiates, though for reasons specific to his own particular agenda of iden-
tifying certain rights as being especially significant),73 and that is to build
into the description of the content of arighta list of conditions under which
it would be overridden if its content were not qualified in this way. For
example, it is often said that the right to life (understood as the right not to
be killed) is, though particularly weighty, not absolute (in a sense of
"absolute" that would seem to indicate at least grade 5, since the right is
referred to simply as "the" right to life, with no restriction to particular per-
sons or times), inasmuch as it is sometimes overall justifiable to kill people.
One response is to "downgrade" the right by restricting it to all innocent
people (thus understanding it to be a grade-4 absolute right, with C inter-
preted as innocence). But this response is frequently rejected on the basis
that it is sometimes overall justifiable to kill innocent people. To this one
might respond that it is never overall justifiable to kill innocent people
unless a certain (probably disjunctive) set of conditions Q through Cn
obtains74 and then declare that there is a grade-4 absolute conditional right
that innocent persons have not to be killed, where the condition in ques-
tion is that this set does not obtain. (Or, indeed, one might "upgrade" the
right back to its original status and claim that there is a grade-5 absolute con-
ditional right that all persons have not to be killed, where the condition in

71 Gewirth (1981), p. 2; the right is asserted on p. 8 and defended in ensuing pages.

72 Gewirth (1981), p. 16.
73 Gewirth (1981), p. 5.
74 Cf. Donagan (1977), p. 72.

184
question is that the set of Ct through Cn does not obtain and the persons in
question are innocent.) Indeed, this tactic of redescribing the content of
rights could be used to argue that there are in fact no prima facie rights at
all; rather, all rights are absolute, all (or certainly most) having this sort of
extensive, complicated, conditional content. But why think of rights in this
way?
The most prominent proponent of the view that all rights are absolute
is Joel Feinberg, at least in certain of his writings.75 According to Feinberg,
rights (of the sort that are correlative to obligations) are not just claims but
what he calls "valid claims." A claim is valid, he says, when "the balance of
reasons" favors "recognition" of the claim.76 Elsewhere, he says that a valid
claim is "a decisive case, invulnerable or conclusive." 77 It seems clear that
Feinberg here takes rights to be claims that are not overridden; for this rea-
son the obligations that they impose are not overridden either. Feinberg
reaches this conclusion through an examination of how legal rights func-
tion. He says:

"Having a claim".. .is an expression very much like the legal phrase "having a
prima facie case." A plaintiff establishes a prima facie case for the defendant's lia-
bility when he establishes grounds that will be sufficient for liability unless out-
weighed by reasons of a different sort that may be offered by the defendant...
In a parallel "prima facie sense" of "claim," having a claim to Xis not (yet) the
same as having a right to X, but is rather having a case of at least minimal plau-
sibility that one has a right to X, a case that does establish a right, not to X, but
to a fair hearing and consideration. Claims, so conceived, differ in degree: some
are stronger than others. Rights, on the other hand, do not differ in degree; no
one right is more of a right than another.

Now, I think that Feinberg may well be correct in what he says here, as
long as we restrict his account to legal rights. Suppose that Paul and Paula
have parted; their divorce was bitter, and now each is laying claim to a cer-
tain indivisible piece of property. A judge rules that the item belongs to
Paula. In this case (barring an appeal and overrule) it is Paula who has the

75 The passages from Feinberg that I am about to quote and that appear clearly to commit
him to the view that rights are not merely prima facie, would appear to be inconsistent
with what he says elsewhere (Feinberg, 1980, p. 225ff.) when he denies that all rights are
absolute and endorses Thomson's distinction between the violation of a right and the
mere infringement of a right.
76 Feinberg (1979), p. 88.
77 Feinberg (1980), p. 187.
78 Feinberg (1979), p. 88.

185
right to the item, not Paul; this is a right she holds against everyone, and so
everyone is obligated (overall, legally)79 not to steal it, damage it, and so on.
But even if Feinberg is correct in claiming that legal rights are not prima
facie but absolute, we may not automatically infer from this (as he appears
to do)80 that moral rights are similarly not prima facie but absolute. Notice
the implication. It would mean that our everyday discussions about moral
rights are to a very great extent ill formed. As noted earlier, we could not
say, without qualification, that someone who has been injured has a moral
right to compensation; we could not say that a property-owner has a moral
right that others not damage his property; and so on. We could not say these
things because it can happen, as we all know, that it is sometimes overall
morally right for an injurer not to give compensation, for someone to dam-
age another's property, and so on, due to overriding considerations. For the
same reason, many current debates would have to be reformulated. If a fetus
has a right not to be killed, and this is a (presumably at least grade-4) absolute
right, then no abortion is ever overall permissible; appeals to the rights of
the mother would be misguided, because either mistaken or irrelevant. If
a psychiatric patient, who has just expressed to his therapist a clear inten-
tion to commit murder, has a right to confidentiality, and this is an absolute
right, then no breach of such confidence is ever overall permissible; appeals
to the rights of his intended victim would be misguided, because either mis-
taken or irrelevant. But we don't normally find such appeals misguided, and
that is because we don't normally take moral rights to be absolute.
It is this implication of treating all moral rights as absolute — the impli-
cation that our everyday discussions of rights would need to be radically
revised — that seems to me the best reason to reject such treatment; for there
appears to be no compensating advantage. Others have given other reasons.
For example, it has been argued that the denial that certain rights are prima
facie (such as the right of property-owners that others not make unautho-
rized use of their property) cannot account for the obligation that others
have to compensate for certain actions (such as the overall justified, but
nonetheless unauthorized, use of property).81 This argument fails, I believe,
79 It is in fact not at all clear that legal obligations can be merely prima facie. When Feinberg
says that 5's having a (legal) claim to X presents a "prima facie case" that S has a right to
X, he is not saying that this imposes a prima facie obligation on others not to interfere
with S's use of X. He is, rather, using "prima facie" in a purely evidentiary sense; he is say-
ing, that is, that S's having a (legal) claim to X provides some reason (which is in princi-
ple open to conclusive rebuttal) to think that S has a (legal) right to X
80 See Feinberg (1979), p. 90.
81 See Thomson (1986), pp. 40-1, and Feinberg (1980), p. 230. (On Feinberg's use of this
argument, see note 75 to this chapter.)

186
in that a proponent of the view that all rights are absolute could certainly
seek to explain the (alleged)82 obligation of compensation without resort-
ing to the notion of the justifiable infringement of a right.83 The difficulty
in specifying the contents of absolute rights (in completing the "unless"-
clauses) is also often cited as a reason to reject the view that all moral rights
are absolute. However the significance of this difficulty should not be exag-
gerated, for there is a complementary difficulty that besets the view that
moral rights are prima facie, and that is the difficulty of specifying exactly
what the conditions are under which whatever moral rights there may be
are overridden. (Exactly when, for example, does fidelity trump gratitude?
Exactly when does gratitude trump fidelity?) Still, it may be that these dif-
ficulties are not completely parallel, for this reason. Suppose that rights are
prima facie and that the difficulties in specifying exactly, for all rights, the
conditions under which they are overridden, are simply insurmountable;
that is, no account of this sort can be given. Even if this is so, we can still
make explicit reference to such rights; we can name them, discuss them,
make certain determinations of their relative weights under certain cir-
cumstances, and so on. But on the view that all moral rights are absolute, if
we cannot fully specify their content, then we cannot make explicit refer-
ence to them at all. It is hard to see how, under such circumstances, we
could hope to justify our moral judgments by an appeal to rights. Yet we
commonly attempt to do just this.
In sum, we simply do not take all moral rights to be absolute, and I know
of no good reason why we should do so.

5.5.3 Rights and overall obligation

If moral rights are prima facie, what is their relation to overall moral obli-
gation? Here I have no precise account to offer. If I had given a precise
account of prima facie obligation-to, then, given the correlativity of prima
facie obligations-to and rights, I could have derived a precise account of
rights (or claim-rights and liberty-rights, at least). But I have not given such
an account of obligation-to; the account that I have offered in this chapter
is an account of mere prima facie obligation, and I confess myself unable to
say precisely what it is that transforms an obligation into an obligation-to.
In addition, if, as I believe, there are obligations that aren't obligations-to,
then it cannot be that a determination of what rights pertain in a given case

82 See Zimmerman (1994).

83 See Rainbolt (1993).

187
is in general all that is required for a determination of what obligations
pertain in that case.
Nonetheless, it is of course possible that an accounting of the pertinent
rights is of great help in determining what ought to be done. Rights, we
commonly think, are (frequently) especially weighty considerations. I have
already acknowledged that overall obligation may (on occasion) be episte-
mologically or explanatorily dependent on prima facie obligation; that is,
that determining what prima facie obligations pertain is (on occasion) help-
ful in, and perhaps even required for, determining what overall obligations
there are. If certain rights are correlative to certain particularly weighty (that
is, difficult to override) prima facie obligations, then it clearly can help,
when trying to determine what we ought overall to do, to determine
whether these rights are at issue. Still, for all that, the hope that a determi-
nation of the pertinent rights will by itself provide for a quick, uncompli-
cated determination of what we ought overall to do seems to me in many
cases to be a vain hope.

188
 Actualism and possibilism

6.1 THE DISPUTE

Rick has received an invitation to review a book. It would be deontically
best if he accepted the invitation and submitted his review on time. The
trouble is, if he accepted, he would not submit the review on time,
although he could. Perhaps it is just because he is lazy, perhaps it is because
he has taken on a great many other tasks, but the fact is that this task, though
accomplishable, would remain unaccomplished. It would be deontically
better for him to decline the invitation than to accept it but fail to submit
the review on time; for then, at least, someone else could be invited to do
it, someone less qualified but someone who would get the job done. What
ought (overall) Rick to do, accept the invitation or decline it?1
The situation may be pictured as in Figure 6.1 (where A stands for
"accept" and S for "submit"), and it may be represented as in Chart 6.1.
With the case so understood, it is clear what the present account of overall
obligation implies: Rick ought to accept the invitation. But a number of
philosophers believe that this is the wrong answer; they believe that Rick
ought to decline the invitation. This is not because they are unsympathe-
tic with the general idea that one ought to do one's deontic best; on the
contrary, they are impressed by the fact that Rick would do better if he
declined the invitation than if he accepted it, and it is for this reason that
they believe he ought to decline it.
This issue can be couched in terms of two broad doctrines, which have
come to be called "actualism" and "possibilism."2 What the actualist says is
roughly this:
(6.1) S ought (overall) to do A iff what would happen if S did A is deon-
tically superior to what would happen if S performed any alternative
action;
1 I borrow this illustration (in modified form) from Jackson and Pargetter (1986), p. 235,
who themselves borrow it (in modified form) from Goldman (1978), pp. 185—6.
2 These terms are introduced in Jackson and Pargetter (1986).

189
 7",

Figure 6.1
1: A & S ...
2:-A & ~S ...
3: A & ~S ...
Chart 6.1
whereas what the possibilist says is roughly this:
(6.2) 5 ought (overall) to do A iff what could happen if 5 did A is deon-
tically superior to what could happen if 5 performed any alternative
action,
where "could" expresses what in Chapter 2 I have called "personal possi-
bility."3 My account accords with possibilism; those who say that Rick
3 These formulations of actualism and possibilism are rough; for instance, there need not be
just one way that things would or could turn out if 5 did A. Still, they should suffice for
present purposes.
Note the following, however. My use of "would" in the statement of actualism pre-
supposes that, if 5 does A at T2 and an event E occurs at T3 as a consequence of this, then
it was true at T? that E would occur at T3 as a consequence of 5's doing A at T2. This is
so even if the probability (whether objective or otherwise) at Tt of E's occurring at T3 as
a consequence of 5's doing A at T2 was less than 1. In saying this, I don't mean to deny
that there is a sense of "would" according to which it would not be appropriate to say this
(in that "might" should be used instead — see Vallentyne, 1987b); I mean only that there
is a sense of "would" according to which it is appropriate to say this, and that it is this
sense that I'm employing here. This fact about how "would" is to be construed here also
affects how "could" is to be construed, for the two are intimately related. (See Subsections
2.2.2 and 2.2.3.) Suppose that S can at T, do A at T2, that there is a probability of .1 at
Tj that Ej will occur at T3 as a consequence of 5's doing A at T2, and that there is a prob-
ability of .9 at 7\ that E2 will occur at T3 as a consequence of 5's doing A at T2. Suppose
that, despite these probabilities, E1 would (in my sense) occur and E2 would not occur at
T3 if 5 did A at T2. Then Et could (in my sense) happen and E2 could not happen (unless
5 could make E2 happen by some other means). For more on the relevance of probabil-
ity to the present issue, see Section 6.4.

190
ought to decline the invitation (for the reason given above) are actualists.
The distinction between possibilism and actualism was brought to light, as
far as I know, only recently. At first actualism was favored;4 then possibil-
ism was touted;5 but since then actualism has been vigorously defended.6
In this chapter I shall investigate this dispute.

6.2 OBJECTIONS TO ACTUALISM

6.2.1 Certain theses denied
It might seem that possibilism is obviously preferable to actualism. If Rick
declines the invitation, then he takes Track 2. Clearly this is not the best he
can do, since Track 1 is better, and so anyone who is working within the
tradition that one ought to do one's deontic best (as the actualist avowed-
ly is) must surely accept that Rick does wrong in declining the invita-
tion. And yet the actualist says that it is not wrong for Rick to decline
the invitation, indeed that Rick ought to decline it. Isn't this plainly
unacceptable?
The actualist has a response here. He will point out that, according to
him, Rick does do wrong in not accepting the invitation. He doesn't do
wrong with respect to not-accepting as such, but he does do wrong with
respect to not-(accepting-and-submitting). Why? Because, even though it
is true that accepting would have been deontically inferior to its alterna-
tive, nonetheless accepting-and-submitting would have been deontically
superior to any of its alternatives.
But this response comes at a considerable cost, for it highlights the fact
that the actualist is committed to a denial of several very plausible theses.
First, the actualist must deny the following:
(2.40'd) if C(~£) & O(A&B), then O{B).
This is demonstrated, in the present case, by the fact that, according to actu-
alism, Rick is obligated to accept-and-submit but is not obligated to accept,
even though he can of course not-accept. Some actualists have explicitly
and boldly embraced this implication. For example, Frank Jackson and
Robert Pargetter, probably the most prominent proponents of actualism,
say this: "Perhaps an overweight Smith ought to stop smoking and eat less,

4 See Holly S. Goldman (1976) and Sobel (1976).

5 See Goldman (1978), Greenspan (1978), Thomason (1981), Humberstone (1983), and
Feldman (1986), Ch. 2.
6 See Jackson and Pargetter (1986) and Goble (1993).

191
but it may not be true that he ought to stop smoking. For it may be that
were he to stop smoking he would compensate by eating more."7
Notice that (2.40'd) is implied by the following (where <} signifies log-
ical possibility):

(6.3) if- O(A&~B) & C(~B), then, if O(A), then O(B).

(This can be seen by substituting A & B for A in (6.3), which yields:

(6.3') if- O((A&B) &~B) & C(~£), then, i£O(A&B), then O(B).

Clearly, - O((A&B) &~B) is necessarily true.) Thus actualism implies that

(6.3), also, is false. And (6.3) is itself implied by the following:

(2.43"') i£~C(A& ~B) & C(~B), then, if O(A), then O(£).

Thus actualism implies that (2.43'") is false, too.

The actualist must also deny this thesis:

(2.40'b) if O(A) & O(B), then O(A&B).

Let A be accepting-and-submitting and B be not-accepting. Rick cannot

both accept-and-submit and not-submit; hence, given

(2.46) ifO(^4),thenC(i4)

(a thesis to which the actualist is standardly committed), it is not the case

that Rick ought both to accept-and-submit and not-accept. Yet each,
according to the actualist, is such that Rick ought to do it. Similarly, the
actualist must also deny this:

(2.47) if O(A) & O(B), then C(A&B).

For, he says, Rick ought to accept-and-submit and ought also to not-

accept; yet Rick cannot do both. Now, some would deny (2.47) anyway,
claiming that moral dilemmas show it to be false. Although I reject this line
of argument (the issue will be discussed further in the next chapter), the
point to be made here is that the actualist does not reject (2.47) because he
accepts the possibility of moral dilemmas. Suppose that what would prompt
Rick not to submit the review on time, were he to accept the invitation to
do so, is sheer laziness; he just "couldn't" (more accurately: wouldnX) be
bothered. Even in this case the actualist would say that Rick ought to not-
accept. Yet he would also say that Rick ought to accept-and-submit. Now

7 Jackson and Pargetter (1986), p. 247. Cf. also Tannjo (1985), pp. 116-17.

192
Rick cannot do both, but it would surely be a gross distortion of his situa-
tion to portray it as a moral dilemma, for the actualist must admit that Rick
could do all that he ought (namely, accept-and-submit, the doing of which
would imply that it is not the case that Rick ought to not-accept).8

6.2.2 Sanctioning wrongdoing

This last point brings out what is perhaps oddest and, in my view, most
objectionable about the actualist's view: it sanctions wholly avoidable
wrongdoing. Let me not be misunderstood here. I do not mean that actu-
alism sanctions doing what the possibilist calls wrong (although this is true
on occasion). I mean that the actualist sanctions doing what the actualist calls
wrong. True, there is no flat-out inconsistency to the effect that in one
and the same breath the actualist says W(A) & P(A) or W(A) & O(A);10
nonetheless, the actualist is committed to claiming on occasion "P(A) (in
fact, O(A)), even though S cannot avoid doing wrong if he does A but can
avoid doing wrong if he doesn't do A" The case of Rick is just such an
occasion, for the actualist says that Rick ought to not-accept, even though
Rick cannot avoid doing wrong if he does not-accept — in that, if he does
not-accept, he cannot avoid not-(accepting-and-submitting), which is
wrong by the actualist's own lights — but can avoid doing wrong if he doesn't
not-accept (but rather accepts-and-submits). Why should Rick's obliga-
tions be thought to be thus tailored to his own easily avoidable moral fail-
ings? Of course, Rick's failure to submit the review on time might (in some
other version of the case) be due not to laziness but rather to his having
taken on a great many other tasks about which he is understandably con-
cerned, and here one might be inclined to accept what the actualist says
about Rick's obligations. This would be a mistake, though, for reasons that
I shall discuss in Section 6.4.
I have just claimed that no acceptable account of obligation sanctions
wholly avoidable wrongdoing, where this is understood to consist in the
conjunction of these propositions: S may, or ought, to do A; S cannot do
A without doing 23; it is wrong for S to do B; and S can act in a way that
includes his not doing A and his not doing B and his doing no wrong. Since
8 If Rick were to accept-and-submit, then it would not be true that, if he were to accept,
then he would not submit. Cf. Jackson and Pargetter (1986), pp. 242 and 249.
9 Goble apparently misunderstands me (in Goble, 1993, p. 161, n. 7), claiming that this
criticism (first presented in Zimmerman, 1990b) "begs the question, since the issue
between possibilism and actualism is what constitutes wrongdoing."
10 I say "inconsistency" rather than "contradiction," for reasons discussed in Section 7.1.

193
actualism allows for such a situation to arise, actualism must be rejected. But
here one might be tempted to invoke an argument of Derek Parfit's to the
effect that it can sometimes be right to cause oneself to do wrong. Parfit
gives a case of the following sort.11 Patty is a politician, and Enid is her
enemy. Enid tells Patty, truthfully, that she will kill all Patty's children if
and only if Patty refuses to be filmed performing some obscene act, the film
to be used later by Enid to persuade Patty to commit several minor crimes
(in that, if Patty refuses to commit these crimes, the film will be released to
the press, thereby ruining Patty's career). On the assumption that Patty
cannot save her children except by submitting to being filmed performing
the obscene act in question, Parfit claims that Patty ought to do this, even
though she knows that this will cause her to act wrongly later (by commit-
ting the crimes), since she knows that she will choose saving her career over
forgoing the crimes. Hence not only would it be right for Patty to cause
herself to do wrong, she in fact ought to cause herself to do wrong.
There are two main ways to construe this situation, and in neither case
do we have a sanctioning of wrongdoing of the sort that I am claiming to
be unacceptable. First, we can understand "cause" in such a way that a
cause's effects cannot be avoided, once the cause has occurred. In this case,
the situation may be represented as in Chart 6.2 (where 5 stands for "sub-
mit to being filmed" and C for "committing the later crimes"). Here the
submitting renders the later commission of the crimes inevitable. On this
understanding of the situation, the later commission is not wrong; hence,
contrary to Parfit's description of the situation, there is no wrongdoing
caused.
The second way to construe the situation - which is presumably Parfit's
way - is to understand "cause" in such a way that a cause's effects are avoid-
able, even when the cause has already occurred (but the effects haven't). In
this case, the situation may be represented as in Chart 6.3. Here, indeed,
the submitting is obligatory but the commission wrong. But this does not
constitute a sanctioning of wrongdoing in the foregoing sense. For
although it is being said that Patty ought to submit and that this would cause

1: S & C ...
2: ~S & ~C ...
3: ~S & C ...
C h a r t 6.2

11 Parfit (1984), p. 38.

194
 1: S & ~C ...
2: S & C ...
3: ~S & ~C ...
4: ~S & C ...
Chart 6.3

her to commit the crimes, which would be wrong, the fact is that it is open
to her to do what it is said she ought (namely, submit) while refraining from
doing the wrong in question.

6.2.3 A nondiachronic case

Recall from Section 4.3 the case of Charlie, captured in Chart 4.4, where
the actions in question, whichever track Charlie chooses, would be simul-
taneous. Suppose that, whether or not Charlie accelerates, he's going to
change lanes. Then actualism implies that it is not the case that he ought to
not-accelerate, even though it is the case that he ought to not-change-and-
not-accelerate. (Possibilism, of course, implies that he ought to not-accel-
erate.) Thus, as Jackson and Pargetter acknowledge,12 the diachronic nature
of the reviewing case is not essential to the present discussion. Indeed, the
case of Charlie brings out equally well the fact that actualism implies the
falsity of the theses just discussed.

6.2.4 A slightly amended version of actualism

Jackson and Pargetter subscribe to a version of actualism slightly but signi-
ficantly different from that with which we have been working so far. It is
this:13
(6.1') 5 ought to do A iff what would happen if S did A is deontically
superior to what would happen if S did not do A.
This version is no more successful than the first. Like (6.1), (6.1') also
implies that the theses just discussed are false, for it yields the same pres-
criptions as (6.1) for both Rick and Charlie. And like (6.1), (6.1') also sanc-
tions wholly avoidable wrongdoing. Indeed, (6.1') seems to me even more
problematic than (6.1), in that the former yields some counterintuitive pres-
criptions that the latter does not. For consider another case, that which was

12 Jackson and Pargetter (1986), p. 236.

13 Jackson and Pargetter (1986), p. 247.

195
first presented in Section 4.1 and captured in Chart 4.1, and with respect
to which we may suppose the following to be true: if you were to vote for
Tom, that would be better than if you were to perform any alternative
action (that is, better than if you were to vote for Dick, or Harry, or no
one). It follows that it is not the case that, if you were to vote for Dick, or
Harry, or no one, that would be better than if you were to perform any
alternative action (for it wouldn't be better than voting for Tom). But now
note this. It is consistent with the foregoing to suppose not only that, if you
were to vote for Tom, that would be better than if you did not votefor Tom,
but also that, if you were to vote for Dick, that would be better than if you
did not vote for Dick; for we may suppose that, if you didn't vote for Dick,
you'd vote for Harry. Hence, while both (6.1) and (6.1') imply that you
ought to vote for Tom, the latter (but not the former) also implies that you
ought to vote for Dick. This implication is surely to be rejected.

6.3 OBJECTIONS TO POSSIBILISM

Just how successful the foregoing criticisms of actualism are is debatable. I
argue: ifp then q (if actualism is true, then such-and-such a thesis is false);
not-q (the thesis is true);14 hence not-p (actualism is false). Adherents to
actualism may well respond: if p then q; p; hence q. This is not an uncom-
mon situation in philosophy, and in this case, as in others, an indisputable
resolution of it is probably not forthcoming. But there is room for some
maneuvering. For instance, the actualist might try to bolster his view by
pointing out that, although it implies the falsity of the theses just discussed,
it is nonetheless consistent with certain closely related theses. (For exam-
ple, Jackson and Pargetter, although they reject

(2.40'd) if C(~B) & O(A&B), then O(£),

in effect endorse this thesis:
(6.4) if C(~B) & O(A&B), and if S does (A&B), then O(£).15
Similarly, Lou Goble, although he rejects

(2.43'") if ~C(A&~B) & C(~£), then, if O(A), then O(B),

14 My account implies the truth of each of (2.40'd), (6.3), (2.43'"), (2.40'b), and (2.47). (See
Subsection 2.3.1 for a discussion of such implications.) It also implies that wrongdoing
is not to be sanctioned.
15 Jackson and Pargetter (1986), p. 248.

196
in effect endorses this thesis:
(6.5) if ~C(A&~B) & C(~B), then, if O(y4), then O(A&B))X6
While this might go some way toward mitigating the impact of the rejec-
tion of the original theses, it is surely a tack one is tempted to take only if
some good reason can be given for rejecting possibilism, which implies the
truth of the original theses without qualification. Here, though, the actu-
alist thinks that he can provide such reason; at least, Jackson and Pargetter
think that they can. I turn now to a consideration of certain arguments that
they have raised against possibilism.

6.3.1 Arbitrariness
The first objection that Jackson and Pargetter raise against possibilism is this.
They claim that possibilism is "arbitrary" in asserting that Rick ought to
accept the invitation, for although it is true that accepting is part of the
deontically best course of action open to him, it is also true that it is part of
the deontically worst course of action open to him. Surely, then, there must
be more to say before the question of what Rick ought to do is settled. In
themselves the fact that accepting is part of the best and the fact that it is part
of the worst show nothing. The first points to the answer that Rick ought
to accept, but the second points equally to the answer that he ought not to
accept. Possibilism errs in allowing the first fact to settle the matter.17
This is a puzzling objection. It rests on the assumption that acceptance's
being part both of what's deontically best and of what's deontically worst
points equally to its being obligatory and to its being wrong. No reason has
been given for this claim, and the possibilist of course rejects it. For while
the possibilist accepts the claim that, where a conjunction of acts (A&B) is
obligatory, then so is each conjunct (see (2.40'd) for a more precise rendi-
tion of this), he of course rejects the claim that, where a conjunction of acts
(B&C) is wrong, then so is each conjunct. There is no need, therefore, to
think that an act, B, cannot feature in each conjunction.
Consider an analogy. Suppose that a state of affairs, S, is part both of the
actual world, W\, and of some nonactual world, W2. Despite S's being part

16 Goble (1993), p. 144.

17 Jackson and Pargetter (1986), p. 237. Goble seems to have something similar in mind
when he says (in Goble, 1993, p. 137): "One problem with possibilism.. .is that when [it
says] that one ought to do the best one can..., [it disregards] the fact that the best one can
do may also be the very worst one can do... When that happens, the possibilist must
advocate that one ought to do the worst one can do."

197
of IV2, we have no hesitation in saying that S obtains. There is nothing per-
suasive in the following objection: "It is arbitrary to claim that S obtains,
for although it is true that S is part of Wu it is also true that it is part of W2.
Surely, then, there must be more to say before the question of whether 5
obtains is settled. In themselves the fact that S is part of W\ and the fact that
it is part of W2 show nothing. The first points to the answer that S obtains,
but the second points equally to the answer that it does not obtain."18

6.3.2 Advice
The second objection is this. Rick might turn to Addie for advice.
Knowing Rick's propensity for procrastination, Addie might well advise
Rick to decline the invitation. Is she advising him to act immorally? Is she
hoping for this? Clearly not. Her advice is the right advice, and this implies
that what she advises him to do is right.19
This objection cannot be accepted. It would prove too much, for it
could just as easily be applied to the voting case. Knowing your propensi-
ty for overlooking the best-qualified candidate, I might well advise you to
vote for Dick (rather than Harry). This advice would be just as appropriate
here as Addie's advice to Rick to decline the invitation. But, we all agree,
it is not the case that you ought to vote for Dick.
So where has the objection gone wrong? Presumably in its implicit
reliance on these two theses:

(6.6) if S ought to advise S' to do A, then S' ought to do A;

(6.7) if S advises S' to do A, and S' ought not to do A, then 5 advises S'
to act immorally.

With respect to the first thesis, all that needs to be said is this. Suppose that
(and that Addie is aware that) Rick will follow Addie's advice as to whether
or not to accept the invitation, but that nothing she can do would persuade
him to submit the review on time. Thus, although Rick can of course so act
that Rick submits the review on time, Addie cannot so act that Rick sub-
mits the review on time, and so the best that she can do does not match up
with the best that he can do. (Rick may be said to be intransigent relative to
18 Note, further, that the actualist, too, must acknowledge that, on occasion, "one ought
to do the worst one can do" (to use Goble's phrase - see the last note). After all, if Rick
would submit the review on time if he were to accept the invitation, then the actualist
would agree that he ought to accept, even though accepting is also (part of) the worst he
can do.
19 Jackson and Pargetter (1986), p. 237. Cf. Goble (1993), p. 139.

198
Addie concerning his submitting the review on time. This is an issue to be
taken up in some detail in Chapter 9.) Under the circumstances, then,
Addie has a choice between advising Rick to accept the invitation (and
thereby seeing to it that he does accept the invitation but does not submit
the review on time) and advising Rick to decline the invitation (and there-
by seeing to it that he does not accept the invitation and so, again, does not
submit the review on time). We may assume that the latter course of action
is deontically better. Here possibilism implies that Addie ought to advise
Rick to decline, while still implying that Rick ought to accept.20
As for the second thesis: there is a sense in which it is true and a sense in
which it is false. The sense in which it is true is this:

(6.7') if 5 advises S' to do A, and S' ought not to do A, then there is some
action such that:
(a) S advises that S' perform it; and
(b) it is wrong.

The sense in which the second thesis is false is this:

(6.7") if S advises S' to do A, and S' ought not to do A, then S advises that
S' perform a wrong action,

where this is understood to imply that the wrongness of the action forms
part of the content of the advice and thus part of the objective of the advi-
sor. After all, it is not Addie's objective (or hope), when advising Rick to
decline the invitation, that Rick should act immorally; that is something
which she correctly takes as given, that is, as beyond her remedy (though
not his). What she is aiming at is the minimization of /mmorality on his part,
consistent with the restrictions on her influence under which she is operating.
I have said that the possibilist rejects (6.6), but the actualist must reject
it too. For there is no necessity that what would happen if S' did A be deon-
tically better than what would happen if S' acted otherwise, even though
what would happen if S advised S' to do A would be deontically better than
what would happen if 5 acted otherwise. Suppose that Rick always acts
contrary to Addie's advice and that Addie can do nothing about this. Then,
on the actualist's view, Addie ought not to advise Rick to decline the invi-
tation, even though Rick ought to decline it. Similarly, both the possibilist
and the actualist must reject the converse thesis, namely:

(6.8) if 5 ' ought to do A, then 5 ought to advise S' to do A

20 Cf. Feldman (1986), pp. 55-7.

199
(where it is understood that 5 is in a position to give such advice). For both
the possibilist and the actualist agree that Rick ought to accept-and-sub-
mit, but both must acknowledge that it might happen that Addie ought not
to advise Rick so to act, in that such advice could easily backfire and thus
be deontically undesirable.
There is therefore no support for actualism to be found in considera-
tions concerning advice.

6.3.3 Detachment
The third objection that Jackson and Pargetter raise is this. They claim that
factual detachment for conditional obligation is valid; that is, that the
unconditional obligation to perform an action may be inferred from the
conditional obligation to perform it when the condition in question is sat-
isfied. Thus, given the fact that Rick ought to decline the invitation if he
won't submit the review on time and the fact that he won't submit the
review on time, it follows that he ought to decline it (period).21
In response, we need only note that, as remarked in Chapter 4, factual
detachment for conditional obligation is not valid. Consider the voting case
once again, as represented in Chart 4.1. You ought to vote for Dick if you
don't vote for Tom. Suppose that you won't vote for Tom. It doesn't fol-
low that you ought to vote for Dick. On the contrary, you still ought to
vote for Tom. Now, elsewhere, Jackson and Pargetter recognize this very
point.22 But then they try to save their original thesis by drawing a distinc-
tion between ways in which obligation may be subject to conditions. First
there is what they call "hypothetical obligation." A statement of hypothet-
ical obligation has the form, "If it were the case thatp, then S would be
(overall) obligated to do A " That is, what Jackson and Pargetter call hypo-
thetical obligation is obligation that involves a subjunctive conditional and
can be symbolized (as in Section 4.1) in this way:

(4.8) p>O(A).
As Jackson and Pargetter note, and as was noted in Section 4.1, factual
detachment is valid for statements of hypothetical obligation. But then
there is what Jackson and Pargetter call "restrictive obligation."23 A state-
ment of restrictive obligation has the form, "Given that p, S is (overall)
21 Jackson and Pargetter (1986), p. 238.
22 Jackson and Pargetter (1987), pp. 77-8. They also hint at it in note 20 on p. 253 of
Jackson and Pargetter (1986).
23 Jackson and Pargetter (1987), p. 77.

200
obligated to do A." By this they mean to give expression to what I have
called "conditional obligation proper." As Jackson and Pargetter note, and
as I have noted repeatedly, factual detachment is not valid for statements of
such obligation.
Making this perfectly correct distinction helps Jackson and Pargetter,
though, only if it can be shown that Rick's obligation to decline the invi-
tation, if he won't submit the review on time, is a hypothetical rather than
a restrictive one. This is of course what they contend, but their rationale is
problematic, for three reasons.
Consider, first, the voting case. There is an inconsistency in Jackson's
and Pargetter's position with respect to this case. They declare the obliga-
tion that you ought to vote for Dick, if you don't vote for Tom, to be a
restrictive one. This seems absolutely right. But then they say that, since
(factual) detachment for restrictive obligation is not valid, it is not the case
that you ought to vote for Dick. Yet we have seen that, on their version of
actualism (captured in (6.1')), you ought to vote for Dick (since your doing
so would be better than your not doing so).
Second, Jackson's and Pargetter's position overlooks the apparent iso-
morphism between the voting case, where it is granted by all concerned
that restrictive obligation is at issue, and the reviewing case. The mark of
restrictive obligation, as Jackson and Pargetter rightly note, is that it con-
cerns obligation relative to a restricted set of options. You ought to vote
for Tom, given all your options; but, with the option of voting for Tom
discounted, you ought to vote for Dick. Similarly, Rick ought to accept
the invitation, given all his options; but, with the option of submitting the
review on time discounted, he ought to decline the invitation. Or again:
you ought to vote for Tom; if you won't do this, then you ought at least to
vote for Dick. Similarly, Rick ought to submit the review on time; if he
won't do this, then he ought at least to (have the courtesy to) decline the
invitation. Jackson and Pargetter in fact admit that the propriety of the locu-
tion "at least" signals restrictive obligation. 5 Since it is proper to use it in
the reviewing case, again we find a tension in their position.
Third, their position is undermined by their own characterization of
hypothetical obligation. They say that what follows "if," in cases of hypo-
thetical obligation, involves not a restriction of options but a reordering of
them. This again seems quite right. Recall this example:

(4.10) if Tom is (or were) a crook, then you ought to vote for Dick.
24 Jackson and Pargetter (1987), p. 78.
25 Jackson and Pargetter (1986), p. 253, n. 20.

201
 1: ~A & ~S
2: A & ~S
3: A & S
Chart 6.4

As noted in Section 4.1, this is tantamount to saying that, if Tom were a

crook, then your options would not be ranked as in Chart 4.1 but in some
other way, such as in Chart 4.2. But notice how different this is from the
reviewing case. In saying that Rick ought to decline the invitation if he
won't submit the review on time, we are surely not saying that, since it is
true that he won't submit the review on time, his options are to be
reordered in some such way as in Chart 6.4.
I think, then, that we simply must accept that Rick's obligation to
decline the invitation, if he won't submit the review on time, is a restric-
tive one — that is, a conditional obligation proper. Since factual detach-
ment is acknowledged by all concerned to be invalid for such obligation,
we cannot argue that it is such detachment that warrants the claim that
the actualist wants to make, namely, that Rick ought to decline the
invitation.
Recall, however, that in saying that factual detachment is not valid for
conditional obligation proper, I am saying that we cannot derive a primary
unconditional obligation in this way. But we can, as noted in Section 4.4,
derive a subsidiary unconditional obligation in this way. If you don't vote
for Tom, you ought (at least) to vote for Dick; if you don't even do that,
you compound your wrongdoing, in that you violate a secondary uncondi-
tional obligation to vote for Dick. Similarly, if Rick won't submit the
review on time, then he ought to decline the invitation; if he doesn't even
do that, he compounds his wrongdoing by violating a secondary uncondi-
tional obligation to decline the invitation. Jackson and Pargetter claim that
there is a "positive advantage" to actualism.26 It can account for the fact that,
intuitively, if Rick declines the invitation, he meets some but not all of his
obligations. For, they say, he ought to decline (even though he ought also
to accept-and-submit). But my possibilist account also allows me to say that
Rick meets some but not all of his obligations in declining the invitation.
He meets the secondary obligation to decline; he fails to meet the primary
obligation to accept.27 Thus, contrary to Jackson's and Pargetter's claim,
actualism enjoys no advantage here.

26 Jackson and Pargetter (1986), p. 241.

27 This point is applicable to what Goble says in Goble (1993), p. 153.

202
 6.3.4 Act-individuation
The final objection that Jackson and Pargetter raise against possibilism is
this. It is clear that Rick ought not to delay the reviewing of the book; but,
by hypothesis, this is just what he would be doing if he accepted the invi-
tation to review it, for then the invitation would not be extended to any-
one else. That is, under the circumstances, his accepting the invitation
would be his delaying the review. Hence he ought not to accept.28
This objection rests on two assumptions. The first is that acts are to be
individuated coarsely and thus that Rick's accepting the invitation would
be identical with his delaying the review. The second is this:
(6.9) if S ought to do A, and his doing A is identical with his doing B,
then S ought to do B.
These assumptions are jointly unacceptable. Either Rick's accepting the
invitation is not identical with his delaying the review but is related to it in
some less intimate way (expressed by the term "by" in "Rick delays the
review by accepting the invitation"), or (6.9) is false, since, even if Rick's
accepting the invitation is his delaying the review, this identity holds only
contingently, and he ought not to allow it to hold. That is, the possibilist, work-
ing under the assumption that acts are to be individuated coarsely, would
accept that
(6.10) if S ought to do A, and he ought to allow his doing A also to be his
doing B, then S ought to do B,
but he would have no reason to accept (6.9). (Recall here the discussion in
Subsection 2.2.5 concerning Jones, who helped the old lady, as he ought
to have done, but who in doing so startled her, as he ought not to have
done. Someone who believes that Jones's helping the old lady and his star-
tling her are identical must reject (6.9) and resort to some such device as
saying that one and the same act can be obligatory under one description
but not under another.)

6.4 WHAT IT IS BEST TO DO

While it seems to me that the foregoing should suffice to put actualism to
rest, I grant that this thesis may still seem appealing. Even if it is conceded,
as surely it should be, that Rick ought to accept the invitation when it is
only laziness that would lead him to fail to submit the review on time, what
28 Jackson and Pargetter (1986), p. 239.

203
if instead he would fail to submit the review on time because he has taken
on a great many other tasks or for some other reason would find it very
difficult to do so, even though, strictly, he could do so? Isn't it plausible to
say what the actualist says in this case and declare that Rick is not obligated
to accept the invitation?
Consider the medicine case first presented in Section 4.2. There are two
medicines, A and B, that Audrey can administer. Let us suppose that, if she
administers A on Monday and A on Tuesday, then the patient will be fully
cured; if she administers B on Monday and B on Tuesday, the patient will
be pretty well off, but not as well off as on the first course of action. Any
other course of action would be disastrous for the patient. It seems that pos-
sibilism requires Audrey to administer A on Monday. But what if we now
say that, if Audrey were to administer A on Monday, she would not admin-
ister it on Tuesday, and this not because oflaziness but for some other whol-
ly understandable reason? For instance, the second dose of medicine A may
be very hard for Audrey to lay her hands on, although strictly she could do
so (it's in a safe, and she doesn't know the combination).29 Or perhaps it
wouldn't be difficult for Audrey to do this, but it's rather that she doesn't
know that two doses of A will cure the patient; on the contrary, she has
very good reason to believe that a second dose ofA would be fatal, although
she knows that with two doses of B the patient will be pretty well off. Under
these circumstances, don't we want to agree with the actualist that Audrey
ought not to administer A on Monday but should give the patient B
instead?
Or consider, finally, this case.30 Gary is playing a game. If he wins, great
good fortune will abound; if he loses, it will be disastrous; if the game ends
in a draw, there will be no significant consequences. He has in his hand a
token, which he can place in one of two boxes. The game consists simply
in the following. If he places the token in thefirstbox, then he either wins
or loses; if he places it in the second box, it's a draw. What determines
whether he wins or loses is whether he places the token in the first box with
his left hand or with his right hand. Although Gary is aware that this is so,
he has no idea which hand is correlated with great good fortune, which
with disaster. Ought Gary to place the token in the first box? Possibilism
appears to imply that he ought. But don't we want to say, with the actual-
ist, that whether or not Gary ought to do this depends on which hand he
would use and what would happen as a result?
29 See Goble (1993), pp. 140-1. I accept that Audrey can, strictly speaking, lay her hands
on the medicine. See the discussion in Subsection 2.2.3.
30 I borrow this illustration (in modified form) from Goble (1993), p. 145.

204
 It is not clear to me that it is correct to say that Rick ought to decline
the invitation (when he would fail to submit the review on time due to his
having so many other tasks to attend to), that Audrey ought to administer
medicine B on Monday (when she would have great difficulty in laying her
hands on a second dose ofA on Tuesday, or when she would fail to admin-
ister A on Tuesday in the belief that it would be disastrous to do so), or that
Gary ought not to place the token in the first box (if he would do so with
his right hand, say, and then disaster would ensue). After all, there is in prin-
ciple a distinction between what it would be wrong for someone to do and
what he would be blameworthy for doing, and while I readily grant that
the agents in these examples would not be blameworthy for failing to accept
the invitation, to administer A on Monday, and to place the token in the
first box (with the left hand), respectively, it does not follow that they
would do no wrong in failing to do these things.31 Still, it is also not clear
to me that it is incorrect to say that the agents ought to decline, administer
B, and put the token in the second box, respectively. But does this pose any
problem for possibilism?
I cannot see that it does. It was noted in Section 1.4 that the present
theory is not committed to any particular account of what deontic value
consists in. It is perfectly consistent with the present theory to say that, in
the relevant sense of "best," it would be best for Rick to decline, best for
Audrey to administer B on Monday, and best for Gary to place the token
in the second box. This would be because the relevant sense of "best"
(namely, that of "deontically best") is the same as that of "probably best"
or "most reasonably thought best" or something of the sort (where "best"
in these latter two phrases is understood in some other, substantive sense).
Thus the possibilist can accept the actualist's diagnosis of what the agent
ought to do in these cases, while of course rejecting the reason for this
diagnosis that the actualist himself gives.
But here the actualist may well protest.32 His point is not just that Rick
ought to decline the invitation, that Audrey ought to administer medicine
B on Monday, and that Gary ought to put the token in the second box
(given that he'd use his right hand if he put it in the first box). Rather,
it's also true that Rick ought to accept-and-submit, Audrey ought to
administer-A-on-Monday-and-administer-A-on-Tuesday, and Gary ought
to put-the-token-in-the-first-box-with-his-left-hand. Thus we do have
an objection to possibilism after all, in that in each case we have a
31 Indeed, it might well be that the agent would be blameworthy for accepting the invitation,
administering A on Monday, and so on. See Zimmerman (1988a), Ch. 3.
32 Goble would. See Goble (1993), p. 141.

205
counterexample to
(2.43'") if ~C(A&~B) & C(~B), then, i£O(A), then O(B).
But here I demur. If we grant that Rick ought to decline, why then also
grant that he ought to accept-and-submit? (Of course, actualism may well
yield this judgment, but actualism is unacceptable. It implies that Rick
ought to decline even when he would fail to submit the review on time out
of sheer laziness.) The reason why we are tempted to say that Rick ought
to decline, where his failure to submit the review on time would be due to
his having so many other tasks to attend to, is, I suggest, because of the great
difficulty he would have in submitting the review on time. But if this is so,
then there is equally good reason to say that it is not the case that he ought
to accept-and-submit. (This would be a matter of supererogation, an issue
to be discussed further in Chapter 8.) Or again, the great risk involved in
Audrey's case may lead us to say that it is not the case that she ought to
administer A on Monday; but this should lead us also to say that it is not the
case that she ought to administer-^4-on-Monday-and-administer-^4-on-
Tuesday. So too with Gary's putting-the-token-in-the-first-box-with-
his-left-hand. The risk, we may say, is too great — either (as in Audrey's
case) with respect to whether the complex action will be performed, or
(as in Gary's case) with respect to what will be achieved by means of the
complex action — and for that reason the complex action in question is not
obligatory.33
In sum, actualism has unacceptable implications, whereas possibilism
does not. It is clear which is preferable.
33 Goble's reason for saying that Rick ought to accept-and-submit, that Audrey ought to
administer-^4-on-Monday-and-administer-y4-on-Tuesday, and so on, is that it would be
"appropriate to regret" the agents' not performing these complex actions, although not
"appropriate to regret" their not performing the relevant component action. (See Goble,
1993, p. 141.) Perhaps we can agree that regret would be appropriate, or not, as Goble
indicates, but this is not determinative of what is obligatory, or not. Regret is appropri-
ate in many cases where no obligation is violated (I can quite appropriately regret some-
one's having died in an earthquake); and (overall) regret may be inappropriate even
where an obligation is violated, if things luckily turn out for the better (in some sense)
because of this violation.

206
 7

Dilemmas

7.1 THE NATURE OF MORAL DILEMMAS

The aloofness of the metaethicist may seem particularly objectionable
when the subject is moral dilemmas. A dispassionate inquiry into moral
dilemmas may appear especially inappropriate — their very nature demands
a passionate response — and those metaethicists who deny that such dilem-
mas can even arise may seem to have reserved for themselves a unique place
among out-of-touch academics.
To which I reply (as politely as possible): the criticism is misdirected. No
one is denying that life can present people with choices that are excruciat-
ing, both morally and personally. No one is denying that anyone who is
confronted with such a choice, whether as participant or simply as specta-
tor, should be deeply moved by it. The question is: just what is at stake in
such choices? What is the nature of the situation in which such choices
arise? To shun such inquiry is to court confusion, and that is what is
objectionable.
Moral dilemmas, as typically construed, are conflicts of moral obliga-
tion, but not all conflicts of obligation are dilemmas. A conflict of obliga-
tion involves there being two or more obligations which cannot all be
satisfied, although each can be. In earlier chapters I have distinguished two
main species of moral obligation- overall obligation and prima facie obliga-
tion — and from this emerge three main types of conflict: overall versus
overall, overall versus prima facie, and prima facie versus prima facie. Let
us consider the last type of conflict first.
Clearly, conflicts between prima facie obligations are possible. I began
Chapter 5 with a story about one such conflict: Steve had a prima facie
obligation to render first aid to Carl and also had a prima facie obligation
to keep his dinner date with Dave, but he couldn't satisfy both obligations.
Such conflicts are possible because the obligations are what Ross calls

207
"parti-resultant" rather than what he calls "toti-resultant":1 beneficence
required rendering the aid, but it was fidelity that required keeping the date.
The obligations stemmed from different considerations, and such consid-
erations may conflict. Even when conflicts between prima facie obligations
arise within one and the same "category" of obligation, the considerations
that conflict are distinct from one another. While gratitude required both
that Gertrude send Lucy lilies and that she send Chrissie chrysanthemums,
it was, more particularly, gratitude-to-Lucy that required the first and
gratitude-to-Chrissie that required the second.
There is nothing particularly mysterious about such conflicts of obliga-
tion, nor about conflicts between prima facie and overall obligations. (After
all, typically, and perhaps always,2 the most stringent prima facie obligation
will be an overall obligation.) Clearly not all such conflicts merit the label
"dilemma," however. This label should surely be reserved for occasions
when the choice between conflicting obligations is a hard one; and not all
conflicts of prima facie obligations (whether or not some of these are also
overall obligations) need involve hard choices.
Still, some clearly can. There are two principal ways in which a choice
between conflicting prima facie obligations might be hard. It might be hard
to tell which prima facie obligation is the most stringent, since more than
one might appear equally compelling. Or it might be hard to make the
choice for that prima facie obligation which is readily recognized as most
stringent, since the moral cost involved in failing to satisfy some opposing
prima facie obligation (or obligations) is so great.3 In both types of case, the
choice may be an especially wrenching one, and I have no quarrel with
calling such situations dilemmas.
Nonetheless, it is not such situations that are the focus of inquiry in this
chapter. Rather, I shall be looking into cases of conflict between overall
obligations. Such cases, too, appear to warrant the label "dilemma," since
they imply that overall wrongdoing cannot be avoided. I shall hence-
forth reserve the label "dilemma" for cases where overall wrongdoing
is inevitable, and it is only into dilemmas in this sense that I shall be
inquiring.
It is useful to distinguish between dilemmas and quandaries. A quandary
(in the present sense)4 is a situation in which the agent does not know what
it is overall right or obligatory for him to do. Clearly this can happen
1 Ross (1930), p. 28.
2 See the discussion of (5.25) and (5.27) in Subsection 5.4.3.
3 See Subsection 5.4.4.
4 Following Conee (1989), p. 133.

208
(despite what some philosophers say).5 Such a situation presents the agent
with a hard choice, but it doesn't involve inevitable wrongdoing; for the
agent may luckily make the right choice, despite his ignorance. Similarly,
dilemmas (in the present sense) don't arise simply in virtue of there being
a tie for what it is deontically best for the agent to do. Clearly, some cases
of such ties can be wrenching (Ruth Barcan Marcus describes what seems
to be such a case, where there are identical twins whose lives are in jeop-
ardy and only one can be saved),6 but they don't imply inevitable wrong-
doing. On the contrary, in cases of ties for what is deontically best, as the
present account of overall obligation implies (see (II) in Subsection 2.1.3),
the agent may, but is not obligated to, opt for any of the top-ranked worlds;
he would do wrong only if he opted for none of them. This point bears
repeating: such cases do not constitute dilemmas, in the present sense. This
bears repeating because at least one major recent work on dilemmas defines
dilemmas in just this way. Walter Sinnott-Armstrong says that a dilemma
arises when there is a conflict of two or more requirements (for all intents
and purposes, a conflict of prima facie obligations), none of which is over-
ridden.8 If there is an unoverridden prima facie obligation to do some act
A, then A is, as (5.18) implies, featured in some world accessible to the
agent to which there is none that is deontically superior. But this does not
imply that A is overall obligatory, merely that it is overall permissible; hence
it does not imply that refraining from doing A is overall wrong. Sinnott-
Armstrong grants this.10 Hence what he calls a moral dilemma is not a moral
dilemma in the present sense.
I noted in Subsection 2.2.6 that the present account of overall obliga-
tion rules in the possibility of certain dilemmas that I there called prohibition
dilemmas, as long as it is possible that each world accessible to the agent be
such that some other accessible world is deontically superior to it. In such
a situation the agent cannot avoid doing something wrong. But I there dis-
tinguished prohibition dilemmas from obligation dilemmas, which consist in
conflicts of overall obligation. Like prohibition dilemmas, obligation
5 See note 55 to Ch. 2.
6 Marcus (1987), p. 192.
7 Sinnott-Armstrong distances himself somewhat from what Ross says about prima facie
obligations, but nothing he says about requirements is incompatible with 'what I have
said in Chapter 5 about prima facie obligation. See Sinnott-Armstrong (1988),
pp. 97-102.
8 Sinnott-Armstrong (1988), pp. 19-20, 29.
9 As long as nothing is of deontic value unless it gives rise to a prima facie obligation; see
again the discussion of (5.25) and (5.27) in Subsection 5.4.3.
10 Sinnott-Armstrong (1988), p. 20.

209
dilemmas involve the inevitability of overall wrongdoing, but the two dif-
fer in that, whereas obligation dilemmas arise from a surfeit of obligations,
prohibition dilemmas arise from a dearth of obligations. Prohibition dilem-
mas involve inevitable overall wrongdoing because they involve the pos-
sibility that it would be overall wrong for 5 to do A even though there is
no act B that S ought overall to do instead; obligation dilemmas involve
inevitable overall wrongdoing because they involve the possibility that it
would be overall wrong for S to do A even though 5 ought overall to do
A. As I indicated in Subsection 2.2.6, the present account of overall oblig-
ation rules out the possibility of obligation dilemmas, barring certain com-
plications to be addressed in Section 7.4. It is this issue that is the focus of
this chapter. Henceforth, when I write of "dilemmas" without qualifica-
tion, I shall mean obligation dilemmas exclusively.
What renders (obligation) dilemmas especially problematic is not just
that they involve inevitable overall wrongdoing (after all, prohibition
dilemmas do too), but that they involve a certain moral inconsistency. The
inconsistency consists in the implication that one and the same act can be
both overall obligatory and overall wrong. This is an implication, at least,
if the following two propositions are granted:

(2.46) ifO(,4), thenC(y4);

(2.43'") if ~C(A&~B) & C(~£), then, if O{A), then O(B).

Suppose we have a dilemma, due to the fact that

(7.1) O(X) & O(Y) &~C(X&Y),

for some acts X and Y. Then, given (2.46), we may infer (substituting Y
for A)
(7.2) C(Y),

and hence also that

(7.3) ~C(X&Y) & C(Y) & O(X).

Then, given (2 A3'"), we may infer (substituting Xfor A and Yfor ~JB, and
so too ~Y for B)

(7.4) O(~Y),
which is, as I have said, inconsistent (in a way) with the second conjunct of
(7.1), namely,

(7.5) O(Y).
210
Of course, (7.4) and (7.5) do not flat-out contradict one another, as do (7.5)
and

(7.6) ~O(Y).

Nonetheless, there is clearly a troubling tension of sorts between (7.4) and

(7.5) to which anyone who wants to admit the possibility of moral dilem-
mas must attend. I shall argue that such tension is in fact not acceptable and
hence that moral dilemmas cannot arise.
More precisely, I shall argue that basic moral dilemmas are impossible,
where these are understood to consist in the following:

(7.7) (a) S ought overall at Tt to do A at T3;

(b) S ought overall at Tt to do B at T4; and
(c) S cannot at Tt do both A at T3 and B at T4\

where Tt is not later than either T3 or T4, but where the relation between
T3 and T4 is left open. ((7.7) of course concerns dilemmas that involve only
two actions that cannot both be performed, although basic dilemmas can
be envisaged that involve an indefinitely larger number of actions that can-
not all be performed.) It is basic dilemmas that appear to have been at the
center of much recent debate on moral dilemmas.

7.2 THE CASE AGAINST BASIC M O R A L DILEMMAS

One argument against the possibility of basic moral dilemmas is simply this:
my account of overall obligation implies that they are impossible. That this
is so is easily seen. If clauses (a) and (b) of (7.7) are both true, then the deon-
tically best worlds accessible to S at Tt are, on my account, worlds in which
Sboth does A at T3 and does B atT4. But this implies that clause (c) of (7.7)
is false. Of course, this might simply be taken as reason to reject my account;
but I don't think it should be, for there are independent reasons to reject
the possibility of basic moral dilemmas.
One reason is this: our own moral experience indicates that we do not
take ourselves to be, or to have been, in dilemmas in situations where many
people tend to diagnose a dilemma. I have in mind such cases as those that
proponents of dilemmas have (or at least appear to have - more on this in
the next section) themselves diagnosed as dilemmatic. A case presented by
Jean-Paul Sartre is probably the most celebrated of all.11 It is the case of a
student in France, during the Second World War, who must choose

11 Sartre (1973), pp. 35-7.

211
between joining the Free French Forces and taking care of his mother.
There seem to him to be equally good reasons to say of each option that he
ought overall to choose it, and yet he cannot choose both. Now certainly
such a situation can arise, for, as so far described, it amounts to "no more"
than a particularly excruciating quandary. What would you and I do in such
a case? Speaking for both of us, I am sure that we would seek advice as to
what to do, if we had someone to whom to turn for such advice. Indeed,
this is just what the student does; he consults Sartre (an unwise choice,
under the circumstances). But this indicates that we, and the student,
believe that we are not (or, at least, may well not be) in a genuine dilemma
after all; for if the advice we seek is moral advice (that is, advice as to what
we morally ought overall to do), then we believe that there is (or, at least,
may well be) a solution to our predicament which our advisor can point
out to us; and this indicates that we believe that at least one of the options
that we're faced with is not (or, at least, may well not be) overall obligato-
ry after all. In similar fashion, those of us who have been faced with such
choices frequently look back at the situation in question and wonder
whether we made the right choice; such wondering again indicates that we
believe that we were not (or, at least, may well not have been) in a genuine
dilemma after all. Of course, one reply to this is simply that our beliefs in
such cases are mistaken; another reply is that, while our beliefs in such cases
are not mistaken, there are other cases where we do not, or would not, hold
such beliefs. Perhaps so, but the present considerations surely indicate that
basic dilemmas aren't as common or as widespread as many of their propo-
nents appear to maintain.
There are other, more forceful reasons for rejecting the possibility of
basic moral dilemmas. One argument concerns these two theses, first
presented in Subsection 2.3.1:
(2.40'b) if O(A) & O(B), then O(A&B);
(2.46)
The first of these theses, often called the "principle of agglomeration,"
yields the following proposition (when times are explicitly stated):
(7.8) if (a) S ought overall at T, to do A at T3, and
(b) S ought overall at T? to do B at T4,
then (c) 5 ought overall at Tt to do both A at T3 and B at T4.
The second thesis yields the following proposition (when times are expli-
citly stated):

212
(7.9) if 5 ought overall at T, to do both A at T3 and B at T4, then 5 can
at T1 do both A at T3 and B at T4.
Jointly, (7.8) and (7.9) contradict (7.7), in that they imply that the con-
junction of clauses (a) and (b) of (7.7) is inconsistent with clause (c) of (7.7).
To preserve the possibility of dilemmas, then, one must reject either the
principle of agglomeration or the principle that "ought (overall)" implies
"can." I have defended the latter of these in Section 3.1, and I have no more
to say on the matter. But the principle of agglomeration deserves further
consideration.
Some proponents of dilemmas have explicitly rejected the principle of
agglomeration. In every case that I know of, the point seems to be this: it
may be reasonable for S to do A and reasonable for S to do B, but it doesn't
follow that it is reasonable for 5 to do both; and this applies to moral and
nonmoral reasons alike.12 For example, it may be reasonable (morally or
nonmorally) for Alex to accept Elvis's invitation to dinner and reasonable
for her to accept Irving's, but not reasonable for her to accept both. But we
must ask what "reasonable" means here. There are two principal senses: (a)
what reason requires, and (b) what reason permits;13 and there are two other
principal senses: (1) what is overall reasonable, and (2) what is prima facie
reasonable. We thus get four agglomeration principles for reasonableness,
namely, those corresponding to categories (al), (bl), (a2), and (b2).
Consider (a2) first. It is clear that agglomeration fails for prima facie
obligation; this explains how conflicts of prima facie obligation are possi-
ble. Steve has a beneficence-based reason to tend to Carl and a fidelity-
based reason to dine with Dave, but there is no X such that Steve has an
X-based reason both to tend to Carl and to dine with Dave (for he cannot
do both these things).14 Similar considerations pertain to (b2).
Consider (bl) next, where we now take this to concern what is overall
morally permissible. Again, it is clear that agglomeration fails here, that is,
that we must reject
(7.10) if P(A) &P(B), then P(A&B)9
and my account of overall obligation certainly doesn't imply that (7.10) is
true. It may well be overall permissible for Alex to accept Elvis's invitation
and overall permissible for her to accept Irving's, without its being overall
12 See Williams (1987), pp. 132-3; van Fraassen (1987), p. 146; Hudson (1986), pp. 16-18.
13 Cf. Gert (1988), pp. 23-4.
14 Cf. Swank (1985), pp. 113-14; Sinnott-Armstrong (1988), pp. 129-30; Pietroski (1993),
pp. 490, 504.

213
permissible for her to accept both. For it may be that the deontically best
worlds accessible to Alex are worlds in which she accepts one or other invi-
tation but not both.15
But the failure of agglomeration for categories (bl), (a2), and (b2) gives
us no reason to think that it fails for category (al); that is, we have been
given no reason by proponents of dilemmas to think that (2.40'b)/(7.8) is
false. Indeed, there seems to be very good reason to think that it is true,
independently of the fact that my account of obligation implies that it is
true. For, as Paul Pietroski notes,16 nondilemmatic cases seem clearly (bar-
ring the actualist's concerns, which I addressed in Section 6.2) to conform
to the principle of agglomeration, and it is difficult to explain this other than
by appealing to the principle itself.17 Of course, a proponent of dilemmas
could reply that the principle at issue is simply this:

(7.11) if O(A) & O(B) & C(A&B), then O(A&B);

but the addition of C(A&B) seems purely ad hoc.
A second argument against the possibility of basic dilemmas concerns
these four theses, first presented in Subsection 2.3.1:
(2.46) ifO(y4),thenC(/l);
(2.43 /// ) \£~C(A&~B) & C(~B), then, if O(A), then O(B);
(2.36') ifO(A),thenP(/l);
(2.37'a) if O{A)y then ~P(~A).
As noted in the last section, the first two of these jointly imply that, in cases
of dilemmas, there is an act A such that the following is true of it:

(7.12) O(A)&O(~A).
Let us now focus our attention on (2.36'), (2.37'a), and (7.12). The first of
these yields the following proposition (when times are explicitly stated):

(7.13) if S ought overall at Tt to do A at T3, then S may overall at Tt do

A at T3.
The contrapositive of the second yields this proposition (when A is substi-
tuted for ~A, and vice versa):

(7.14) if S may overall at Ti do A at T3, then it is not the case that S ought
overall at T7 not to do A at T3.

15 Cf. Sinnott-Armstrong (1988), pp. 133-4.

16 Pietroski (1993), pp. 503-4.
17 Cf. also Sinnott-Armstrong (1988), pp. 134-5.

214
And it is clear that (7.13) and (7.14) jointly contradict (7.12). To preserve
the possibility of dilemmas, then, one must reject at least one of the four
theses that function as premises in this argument.
Again, I shall not seek to defend (2.46) here; I have said all that I want
to say in Chapter 3. As to (2.43'"): the strong intuitive plausibility of this
proposition was noted in both Chapter 2 and Chapter 6; I have no more to
add here. And (2.36') and (2.37'a) are surely extremely plausible. A propo-
nent of basic dilemmas might nonetheless seek to undermine this second
argument by noting the following: I have already acknowledged that it is
possible for there to be the analogue of a basic dilemma when it is prima
facie, and not overall, obligation that is at issue and also when it is permis-
sibility (whether prima facie or overall), and not obligation, that is at issue;
hence I must admit that the analogues of this second argument break down
somewhere in these cases, and it may be that this would give us reason to
think that the second argument itself likewise breaks down. Let us now
examine this matter.
Consider prima facie obligation first. I of course grant that the analogue
of (7.12) is possible, that is, that it can happen that

(7.15) S ought prima facie at Tt to do A at T3 and S ought prima facie at

Tt not to do A at T3.
Moreover, the account of prima facie obligation that I have given in
Chapter 5 commits me to accepting this analogue of (2.46):

(7.16) if S ought prima facie at Tt to do A at T3, then S can at T1 do A at

And I grant also the truth of the relevant analogues of (2.43"') and (2.36').
But the analogue of (2.37'a)/(7.14) is to be rejected. That is, we should not
accept:

(7.17) if S may prima facie at Tt do A at T3, then it is not the case that S
ought prima facie at Tt not to do A at T3.

This is because prima facie obligations (and prima facie permissions) are, as
noted above, parti-resultant rather than toti-resultant. Thus, for example,
although Steve has a beneficence-based prima facie obligation (and prima
facie permission) to tend to Carl, he also has afidelity-basedprima facie
obligation not to tend to Carl but to have dinner with Dave; again, this
explains how conflicts of prima facie obligations are possible. But I cannot
see how this provides any support whatsoever for the proponent of basic
dilemmas.
215
 As to permissibility (whether prima facie or overall - let's focus on over-
all): it is clear that my account commits me to the analogue of (2.46),
namely,

(7.18) if P(A), then C(A);

and it also commits me to the analogue of (2A3"'). In addition, the
analogue of (2.36') is trivially true. But, once again, the analogue of
(2.37'a)/(7.14) is to be rejected. That is, we should not accept:

(7.19) if 5 may overall at T1 do A at T3, then it is not the case that 5 may
overall at Tt not do A at T3,
and this is for the obvious reason that ties for what is deontically best are
perfectly possible. Once again, though, I cannot see how this provides any
support whatsoever for the proponent of basic dilemmas.
I am thus insisting on the truth of (2.37'a)/(7.14), on the truth, that is,
of the claim that what it is overall obligatory to do it is not overall permis-
sible not to do, or, contrapositively, that what it is overall permissible to do
it is not overall obligatory not to do. I am not sure how to defend this claim,
beyond pointing to its being implied by the account of overall obligation
that I have provided. Perhaps one way to defend it stems from considera-
tions of supervenience. Many philosophers will agree that, if an act is oblig-
atory (whether prima facie or overall), as of a certain time, then this moral
fact supervenes on a certain set of nonmoral facts that obtains at that
time. This set of nonmoral facts grounds the obligatoriness of the act; it
renders the act obligatory. Similarly, if refraining from an act is permissible
(whether prima facie or overall), as of a certain time, then this moral fact
also supervenes on a certain set of nonmoral facts that obtains at that time.
This set of nonmoral facts grounds the permissibility of refraining from the
act; it does not, therefore, render the act obligatory. Now, as has been
repeatedly observed, it is of course quite possible that an act should be prima
facie obligatory and yet refraining from it be prima fade permissible. This is
because such moral facts are, as just noted, pdrfr'-resultant; they are, that is,
grounded in different sets of nonmoral facts, one set rendering the act oblig-
atory, the other not. But an act cannot be overall obligatory while refrain-
ing from it is overall permissible, for such moral facts arefofr'-resultant;they
are grounded in one and the same set of nonmoral facts (namely, the set of
all the nonmoral facts18 that obtain at the time in question), and it cannot
be that this set both renders the act obligatory and fails to render it so.

18 Or, at least, all the nonmoral facts that can serve as a supervenience base for moral facts.

216
 I find this argument persuasive. No doubt some will not find it so, either
because they are skeptical about the supervenience of moral on nonmoral
facts or because they believe that it begs the question to say that, if a set of
nonmoral facts grounds the permissibility of refraining from an act, then
that same set cannot ground the obligatoriness of the act. If so, I do not
know what else to say in defense of (2.37'a)/(7.14). Further defense of this
claim seems to me unnecessary, though, for it is the kind of proposition that
should serve as the touchstone of an ethical theory. If my account of over-
all obligation had contradicted it, then it is the account, not the claim, that
would have had to be rejected.

7.3 THE CASE FOR BASIC MORAL DILEMMAS

The ground covered in the last section is, for the most part, extremely well
trodden. Almost all the considerations that I have raised against the possi-
bility of basic dilemmas have been raised, many times, by others.19 But
they deserve to be stated again, because, despite what I take to be the
great strength of these considerations, it seems that nowadays proponents
of moral dilemmas far outnumber their opponents. This is a puzzling phe-
nomenon. Of course, it may be — may well be — that the appearance of dis-
agreement is merely an appearance, that what "proponents of dilemmas"
are proposing does not in fact conflict with what "opponents of dilemmas"
are opposing. This indeed seems frequently to be the case. Opponents of
dilemmas may, I think, in every case be taken to be opponents of what I
have called "basic moral dilemmas." But what is it that proponents of
dilemmas are proposing? Well, consider the article by E. J. Lemmon that
instigated the recent debate on dilemmas. There Lemmon distinguishes
various types of situations that he calls moral dilemmas, but none of them
seems to be a basic moral dilemma. In particular, what he calls "the
simplest variety of moral dilemma in the full sense" seems best understood
as simply a conflict of prima facie obligations, for the example he gives is
this:
A friend leaves me with his gun, saying that he will be back for it in the evening,
and I promise to return it when he calls. He arrives in a distraught condition,
demands his gun, and announces that he is going to shoot his wife because she
has been unfaithful. I ought to return the gun, since I promised to do so... And
yet I ought not to do so, since to do so would be to be indirectly responsible
19 See especially McConnell (1987), Conee (1987), and Feldman (1986), Section 9.1. Cf.
also Donagan (1987).

217
for a murder... I am in an extremely straightforward moral dilemma, evidently
resolved by not returning the gun.20
It could be that Lemmon regards this situation as what I have called a basic
moral dilemma, but his talk of resolution indicates otherwise. Conflicts of
prima facie obligation are resolved when the most stringent prima facie
obligation is acted on; no such resolution is available for conflicts of overall
obligation. Of course, conflicts of overall obligation might be subject to
some sort of nonmoral resolution (perhaps satisfying one of the obligations
would serve the agent better than satisfying the other), but it is apparent
that Lemmon is concerned with a moral resolution to the conflict that he
illustrates.
It is clear that other prominent proponents of dilemmas are also at times
concerned only with conflicts of prima facie obligation, so that there is in
reality no opposition on this score between them and opponents of dilem-
mas. For example, Bernard Williams allows for the (moral) solubility of at
least some of the conflicts of obligation that he is concerned with, even if
these are not "soluble without remainder,"21 that is, even if their resolution
leaves a certain sort of moral residue. Bas van Fraassen talks of "what ought
to be for one reason and what ought to be for another reason,"22 and this is
strongly reminiscent of the/wfr'-resultant nature of prima facie obligation.
Marcus declares that "it is a better fit with the moral facts that all dilemmas
are real, even where the reasons for doing x outweigh, and in whatever
degree, the reasons for doing y."23 And Christopher Gowans endorses an
approach to dilemmas according to which "even though S ought to do A
and S ought to do B, there is nonetheless some sense in which, in the final
analysis, (say) S ought to do B [and not ^4]."24 All such talk is most readily
understood in terms of conflicts of prima facie obligation and, as such,
seems wholly unobjectionable (except that Marcus's claim is surely too lib-
eral; as we saw earlier, not all conflicts of prima facie obligation warrant the
label "dilemma").
Nonetheless, it is also clear that proponents of dilemmas are at times con-
cerned with unresolvable conflicts of obligation, and this may seem to indi-
cate a concern with basic moral dilemmas. But let us not be hasty. There
can be situations that are (morally) unresolvable that don't constitute basic
dilemmas. For example, if more than one course of action is overall
20 Lemmon (1987), p. 105.
21 Williams (1987), p. 129.
22 Van Fraassen (1987), p. 141.
23 Marcus (1987), p. 193.
24 Gowans (1987a), p. 25.

218
permissible, then there is no moral resolution to this conflict between alter-
natives.25 Even if one chooses to call such a situation "dilemmatic," it is
certainly not a basic dilemma; for wrongdoing is not inevitable. Indeed, as
noted earlier, it is just such a situation that Sinnott-Armstrong calls "dilem-
matic," and so, although I am an opponent of dilemmas, I do not oppose
his view. And it may be that others have adopted his view. For example,
Williams at one point talks of "extreme cases of moral conflict, tragic
cases," where the following may be said:
The agonies that a man will experience after acting in full consciousness of such
a situation are not to be traced to a persistent doubt that he may not have cho-
sen the better thing; but, for instance, to a clear conviction that he has not done
the better thing because there was no better thing to be done.26
So described, a tragic case appears to be "no more" than a particularly
excruciating conflict between equally stringent prima facie obligations than
which there is none more stringent. Of course, I accept the possibility that
such cases may arise. Similarly, van Fraassen talks of conflicts "which can-
not be resolved in terms of one reason overriding another."27 This, too,
merely suggests a tie for top place.
Further evidence that proponents of dilemmas are often not concerned
with proposing the possibility of basic moral dilemmas may be adduced
simply from the fact that they frequently don't even bother to address the
formidable arguments presented in the last section against this possibility.
Such an attitude betrays a total lack of concern with the possibility of basic
dilemmas. And even when attention is apparently paid to these arguments,
we need to be careful what to infer from this. For example, Lemmon rejects
the principle that "ought" implies "can," and yet, as I have said, it is clear
that he is concerned with "prima facie ought." Of course, I believe that he
is mistaken to reject the principle even when so understood, but the point
here is that his apparent concern with the arguments presented in the last
section is, again, merely apparent. Others also may appear to have addressed
some of these arguments (Williams and Marcus both reject the principle of
agglomeration;29 van Fraassen rejects the principle that "ought" implies
"may"),30 but, again, it may be that their concern is not with these argu-
25 At least, none of the alternatives is preferable in terms of deontic value. There are other
morally relevant values, as noted in Ch. 2, n. 13.
26 Williams (1987), p. 123.
27 Van Fraassen (1987), p. 141.
28 Lemmon (1987), p. 114, n. 2.
29 Williams (1987), pp. 132-3; Marcus (1987), p. 200.
30 Van Fraassen (1987), pp. 145-6.

219
ments but with some closely related arguments. If so, there may not after
all be any conflict between their views and that of opponents of dilemmas.
Still, proponents of dilemmas do sometimes make remarks which sug-
gest that they accept the possibility of basic moral dilemmas, the possibili-
ty that an agent will inevitably do overall wrong. Williams talks of tragic
cases as cases where the agent accepts that "he has done something that he
ought not to have done, or not done something that he ought to have
done;"31 Marcus talks of being "damned if you do or ... damned if you
don't";32 and although such talk can be construed as advocating merely the
possibility that an agent will do prima facie wrong, it seems wise to grant
that it may be intended as an endorsement of the possibility that an agent
will inevitably do overall wrong and to examine it as such.
What, then, are the reasons that have (perhaps) been advanced by pro-
ponents of dilemmas for accepting the possibility of basic dilemmas? A
number of reasons appear to have been given. I shall now look into some
of these, in what I take roughly to be ascending order of plausibility.
First, it may be that some proponents of dilemmas believe that it is just
obvious that basic dilemmas can occur. Cases such as Sartre's, that of
Agamemnon at Aulis (cited by Williams),33 that of Sophie's choice34 (cited
by many), and others are often presented as being clearly dilemmatic. Well,
it is clear that a conflict of some sort is at issue in these cases, but is it clear
that what's at issue is a basic dilemma? How could this be, even from the
proponents' point of view? For they of course grant that there can be con-
flicts that are morally resolvable, and some of these may be very hard cases
(in the sense that it is very difficult to figure out just what the solution is).
If so, it would seem that it can never be obvious that a particular conflict con-
stitutes a basic dilemma rather than merely a very hard but resolvable case.
Another reason for accepting the possibility of basic dilemmas that may
have been proposed is that cases of symmetry imply that such dilemmas are
possible. One such case is that provided by Marcus of the twins whose lives
are in jeopardy. It is clear, however, that such cases do not imply that basic
dilemmas are possible; they simply involve equally overall permissible, not
equally overall obligatory, alternatives.
Dilemmas have often been diagnosed in cases not of symmetry but of
incommensurability.35 Such cases involve one course of action's being

31 Williams (1987), p. 123.

32 Marcus (1987), p. 194.
33 Williams (1987), p. 123.
34 Styron (1980).
35 See, e.g., Sinnott-Armstrong (1988), p. 58fF.; Lebus (1990).

220
neither superior nor inferior nor equal in deontic value to another. But, as
we saw in Subsection 2.2.6, what such cases, just like cases of symmetry,
imply is that each such alternative is equally overall permissible, not that
each is equally overall obligatory. It may be that such cases are or can be
especially wrenching, in that something of incomparable value is inevitably
forgone, but this doesn't give them the status of a basic dilemma. (Actually,
I don't see why they should be especially wrenching; Marcus's case of the
twins, which would appear to involve two comparable alternatives, seems
to me just about as wrenching as any situation can be.)
Some have diagnosed dilemmas when the agent cannot avoid having
"dirty hands." This may simply be an acknowledgment that some situa-
tions are, as Williams says, not "soluble without remainder," that is, that no
matter what one does, there will be a moral cost involved. But a moral cost
can be incurred without an overall obligation's being violated; when Steve
broke his dinner date with Dave, he incurred such a cost. Michael Stocker,
however, suggests that in certain cases, cases where one is wrongly coerced
by another to help implement some evil plan, there is a special reason to
talk of "dirty hands" and that, in these cases, wrongdoing is inevitable.36 An
example would be Williams's case of Jim and the Indians, cited in Section
2.1.37 But once again, while such cases may be especially wrenching, there
is no need to regard them as involving inevitable overall wrongdoing.
Obviously we may agree that it is usually wrong, overall, to implement evil
plans, but the question is what one should do under the actual circumstances.
After all, Jim's circumstances hardly merit being termed "usual." Perhaps,
under the circumstances, Jim ought overall to shoot the Indian, as the con-
sequentialist would presumably say. Perhaps he ought overall not to shoot
the Indian, as the Kantian would presumably say. Perhaps there is both a
strong reason to shoot and a strong reason not to shoot, as the Rossian
would presumably say. It may not be clear to Jim just what he ought
overall to do, in which case he is in a quandary. It may be that no option is
deontically better than the other, in which case he is in a situation involv-
ing either symmetry or incommensurability. But none of this implies
both that Jim ought overall to shoot and that he ought overall not to
shoot.
Remarks made by Marcus suggest the following argument:38
(7.20) one ought overall always to avoid conflicts of obligation;

36 Stocker (1990), p. 25.

37 Williams (1973), p. 98.
38 Marcus (1987), p. 199.

221
(7.21) unless such conflicts were possible and were basic dilemmas, it
would not be the case that one ought overall always to avoid them;

hence,

(7.22) basic dilemmas can occur.

Both premises are false. Sometimes it is not the case that one ought to avoid
conflicts of obligation of the sort that Marcus cites. After all, perhaps the
only way to have avoided being in a position to save one but not both twins
would have been to avoid the scene of the tragedy altogether, in which case
both lives would have been lost. And even when one ought to avoid such
conflicts of obligation, this can be explained without invoking basic dilem-
mas. For example, one ought (usually) to avoid making incompatible
promises. Why? Not because they put one in a basic dilemma, but simply
because they (usually) involve incurring unnecessary moral cost.
It might be argued, however, that such cost can give rise to further over-
all obligations and that this is, on these occasions, a sign that one was in a
basic dilemma. Perhaps one would have an overall obligation to explain to
the twins' next of kin why not both were saved; perhaps one would have
an overall obligation to apologize to, or even to compensate, the disap-
pointed promisee. But it is clear, as the discussion of residual obligation
in Subsection 5.4.4 attests, that such overall obligations can arise out
of the failure to meet prima facie obligations that are not themselves
overall obligations. There is no need to diagnose a basic dilemma in such
cases.
There is one type of moral residue in particular that is frequently cited
as a sign of moral dilemmas, and that concerns emotions. As noted in
Subsection 5.4.4, Ross believes that a certain feeling of compunction is
called for when a prima facie obligation is not satisfied. But in some cases
more than mere compunction is called for. Williams notes that it is some-
times appropriate to regret whatever it is that one has done, and he takes this
as evidence of dilemmas.39 If it is to be taken as evidence of basic dilemmas,
the argument would seem to be this:

(7.23) it is appropriate for S to regret doing A [or not doing A] only if it

was overall obligatory for 5 not to do A [or to do A];
(7.24) it is possible both that it would be appropriate for 5 to regret doing
A and that it would be appropriate for S to regret not doing A;
39 Williams (1987), p. 122ff.

222
hence,
(7.22') it is possible both that it was overall obligatory for 5 not to do A
and that it was overall obligatory for S to do A.
We must surely grant the second premise. Sartre's student had cause for
regret no matter which choice he made; Agamemnon likewise; Sophie too.
But the first premise is false. Regret is appropriate whenever a moral cost
is incurred, and this can happen without an overall moral obligation's being
violated.
This last point is widely accepted, but then proponents of dilemmas sug-
gest this modification to the argument: replace talk of regret with talk of
remorse or, equivalently, guilt feelings.40 Instead of (7.23) and (7.24) we would
then have, respectively:
(7.25) it is appropriate for 5 to feel remorse for (guilty about) doing A [or
not doing A] only if it was overall obligatory for S not to do A [or
to do A];
(7.26) it is possible both that it would be appropriate for 5 to feel remorse
for (guilty about) doing A and that it would be appropriate for S to
feel remorse for (guilty about) not doing A;
and we could then infer (7.22') afresh.
This is an interesting argument. I think we should agree that there is a
conceptual tie between remorse and overall wrongdoing that does not
obtain between regret and overall wrongdoing. Certainly being guilty is
intimately related to overall wrongdoing; moreover, it seems plausible to
say (although I shall express reservations about this shortly) that one feels
guilty about, or remorse for, doing something only if one believes that one
is guilty with respect to doing it. Given this, it seems plausible to say, along
with van Fraassen, that "it is appropriate to feel guilt if and only if one is
guilty";41 for then the belief that is part of the emotion will correspond with
the facts. Let us therefore (provisionally) accept the following:
(7.27) it is appropriate for S to feel remorse for (guilty about) doing A
[or not doing A] iff S is guilty with respect to doing A [or not
doing A].
The question is, when is it that one is guilty with respect to doing [or not
doing] something? One answer is simply this:
40 See, e.g., van Fraassen (1987), p. 147; Marcus (1987), pp. 196-7.
41 Van Fraassen (1987), p. 147.

223
(7.28) S is guilty with respect to doing A [or not doing A] iff 5 failed to
satisfy an overall obligation not to do A [or to do A].

That is, one is guiltyjust in case one does, or has done, overall moral wrong.
(7.27) and (7.28) jointly imply (7.25), the first premise of the argument
presently under consideration. But why, then, accept the second premise,
(7.26) (once we are clear that it is remorse in particular, rather than some
other negative emotion such as regret, that is at issue)? I know of no reason
having been advanced for accepting it, other than to insist that basic dilem-
mas are possible, and to cite this as a reason would of course be to beg the
question. (Actually, I'm not being quite honest here; I shall qualify this
remark in the next section.)
There is another answer to the question of when it is that one is guilty
with respect to doing [or not doing] something. It is this:

(7.29) 5 is guilty with respect to doing A [or not doing A] iff 5 believed
that he was failing to satisfy an overall obligation not to do A [or to
do^]. 4 2

I think we must agree that, on this understanding of what it is to be guilty,

the second premise of the argument, (7.26), is true. For it is, obviously,
quite possible for someone to believe that he ought overall to do some action
and also that he ought overall not to do it. But now it is the first premise,
(7.25), that is to be rejected. (7.27), conjoined with (7.29), does not war-
rant the claim that actual overall wrongdoing is necessary for actual guilt
and thus necessary for appropriate remorse; on the contrary, presumed
overall wrongdoing will suffice.43
Thus this, the strongest argument for basic dilemmas of which I am
aware, fails. It is worth noting, finally, that the account of the appropriate-
ness of remorse, (7.27), which I have provisionally conceded to the pro-
ponent of dilemmas, is in fact open to question. Suppose that Tracy, a truck
driver, has just run over Reggie, a small boy who darted out into the street
in front of Tracy's truck. Tracy did no wrong, let us assume; nor did she or
does she believe that she did wrong. She is therefore not guilty, according
to either (7.28) or (7.29), with respect to running over Reggie. Still, I think

42 This is roughly the answer given in Zimmerman (1988a), Chapter 3, with "guilty"
understood to mean the same as "culpable." The account given there actually supports
only the "only if," and not the "if," of (7.29), since more than such a belief is required
for culpability; freedom is also required.
43 Qualification: suffice along with other conditions, not among which is actual wrongdo-
ing. See the last note.

224
we would all be a little put out if she showed no remorse for what she had
done. True, there is, in a sense, nothing for her to feel guilty about, and yet
wouldn't such a reaction bespeak a coolness of character that would likely
prompt an absence of remorse in cases where it was clearly called for? I think
that what we should say here is that Tracy's reaction is intrinsically appro-
priate but (probably) extrinsically inappropriate, inasmuch as it (probably)
manifests a tendency to react in an intrinsically inappropriate way on other
occasions.44 It seems, then, that (J.21) is at best acceptable only as an
account of the conditions of the intrinsic appropriateness of remorse. But,
even when it is so understood, there is reason to question it. This is because
the claim concerning the connection between feeling guilty and being
guilty, on which I originally based (7.27), is itself dubious. Robert Roberts
has contended, with considerable plausibility, that feeling guilty doesn't
presuppose believing oneself to be guilty so much as construing oneself as
guilty.45 One example might be that of a lapsed Catholic, who believes the
lapse to be justifiable and yet feels guilty about not attending Mass. Roberts
likens such cases to optical illusions which are understood to be illusory and
yet persist. Given such cases, it is arguable that it is sometimes appropriate,
and not merely extrinsically so, for agents to feel guilty, even though they
have not done overall wrong and did not believe themselves to have done
overall wrong. If this is correct (and here I shall not seek to settle the ques-
tion), then (7.27) is to be rejected.

7.4 N O N B A S I C M O R A L DILEMMAS
Once again, the ground covered in the last section is, for the most part,
extremely well trodden.46 But once again, because proponents of dilemmas
are so numerous and so vocal, it deserves to be covered anew. Let me now
turn to some relatively fresh considerations, several of which are designed
to undermine the proponents' case even further.
I have so far been concerned with what I have called "basic moral dilem-
mas," that is, with situations where the following is true:

(7.7) (a) 5 ought overall at Tt to do A at T3;

(b) S ought overall at T1 to do B at T4; and
(c) S cannot at T1 do both A at T3 and B at T4.

44 Cf. Zimmerman (1988a), pp. 134-5.

45 Roberts (1988), pp. 183, 186-9, 195-6, and 201.
46 See especially McConnell (1987), (1993b), (1995); Conee (1987); Feldman (1986),
Section 9.1.

225
I have argued that such situations cannot arise. Notice that, inasmuch as
(2.46) (the claim that "ought" implies "can") is true, basic dilemmas may
in fact be more fully characterized as follows:
(7.7') (a) S ought overall at T, to do A at T3;
(b) 5 can at Tt do A at T3\
(c) S ought overall at T, to do B at T4;
(d) 5 can at T1 do B at T4; and
(e) S cannot at Tt do both A at T3 and B at T4.
Notice, also, that the term "basic moral dilemma" suggests that some moral
dilemmas are not basic. Consider, for example, this sort of situation (where
Tj is earlier than T2, and T2 is not later than either T3 or T4):

(7.30) (a) S ought overall at Tt to do A at T3;

(b) S can at T1 do ^4 at T3;
(c) 5 ought overall at T2 to do B at T4;
(d) S can at T2 do B at T4\ and
(e) S cannot at T2 do both A at T3 and B at T4.
Can such situations arise?
Indeed they can, and that they can seems to me of great significance. Let
us call the type of nonbasic dilemma captured in (7.30) a bind. Examples of
binds have in fact already been given. For instance, recall from Subsection
3.2.2 the case of Mandy, who on Friday tells Phil that she will be able to
help him move furniture on Sunday, thereby making a sort of informal
commitment to do so, but who on Saturday promises Bill that she will
babysit his child on Sunday, thereby making a formal commitment to do
so. Her situation may be pictured as in Figure 3.2. When Mandy makes her
promise to Bill on Saturday, she puts herself in a bind, for she can no longer
avoid doing overall wrong on Sunday. If she keeps her promise as she
ought, then she'll do the remote wrong of not helping Phil; if she helps Phil
as she (once) ought, then she'll do the immediate wrong of not babysitting
Bill's child. Here it is most important to note that the remote wrong, if it
were committed, would be a genuine wrong, as I argued in Subsection
3.2.2.
It seems to me perfectly appropriate to call binds a type of moral dilem-
ma, for they involve wrongdoing that is, in a way, inevitable. It is true that
the wrongdoing was not always inevitable, but it has become so (in
Mandy's case, it became so on Saturday). It is also true that there is no sin-
gle time at which the conflicting "oughts" apply (for then we would have
a basic dilemma); one applies at one time (from Friday to Saturday, in

226
Mandy's case), the other at another (from Saturday to Sunday). Nonetheless
overall wrong will have been done no matter what the agent chooses at the
time of action to do, for by then it is too late to extricate himself from the
dilemma.
Consider now once again the argument concerning remorse, but with
time indices explicitly provided as follows:
(7.25') it is appropriate for 5 to feel at T4 remorse for (guilty about) doing
A [or not doing A] at T3 only if there is a time Tsuch that it was
overall obligatory at Tfor 5 not to do A [or to do A] at T3;
(7.26') it is possible both that it would be appropriate for S to feel at T4
remorse for (guilty about) doing A at T3 and that it would be
appropriate for S to feel at T4 remorse for (guilty about) not doing
A at T3;
hence,
(7.22") it is possible both that there be a time Tsuch that it was overall
obligatory at T for S not to do A at T3 and that there be a time T
such that it was overall obligatory at Tfor S to do A at T3.
This argument may be sound. We should accept the first premise, if (7.27)
survives the doubts raised at the end of the last section and guilt is construed
as consisting in actual wrongdoing (see (7.28)). I think that we should also
accept the second premise. Now I said earlier that I knew of no reason hav-
ing been advanced to accept (7.26) other than to insist that basic dilemmas
are possible, but I also said that I wasn't being quite honest in making this
claim. I can now explain, for / now wish to advance a different reason for
accepting (7.26'). Look at (7.22"). It doesn't say that basic dilemmas are pos-
sible. That basic dilemmas are possible is expressed by the following
proposition:
(7.22'") it is possible that there be a time Tsuch that it was overall oblig-
atory at Tfor S not to do A at T3 and it was overall obligatory at
Tfor S to do A at T3.
Here the quantifier over times has broad scope, so that the time of "ought"
is said to be identical with the time of "ought not." But in (7.22") no such
identity is posited, and this is why it is acceptable. As long as binds are pos-
sible — and they are — we must accept (7.22"); for the following, which con-
stitutes a bind, implies (7.22"): it was overall obligatory at Tt for S not to
do A at T3 and it was overall obligatory at T2 for 5 to do A at T3. But this
doesn't commit us to accepting the possibility of basic dilemmas, and now
227
we can see that (7.26') is acceptable after all. Think of Mandy on Monday.
If she kept her promise to Bill on Sunday, then it would be appropriate for
her on Monday to feel remorse for not having helped Phil; and if she helped
Phil on Sunday, then it would be appropriate for her on Monday to feel
remorse for doing so, since she would thereby have failed to keep her
promise to Bill. In this way I think we can concede to proponents of
dilemmas their strongest argument and yet deny that basic dilemmas can
occur.
One objection to what I have just said is this.47 If it is to be appropriate
for Mandy to feel remorse on Monday for what she did (whatever it was)
on Sunday, then her belief concerning her guilt that renders her remorse
appropriate must itself be accurate. Thus she must either believe on
Monday that she was obligated from Friday to Saturday (but not beyond)
to help Phil on Sunday, or believe that she was obligated from Saturday to
Sunday to babysit Bill's child on Sunday, or both. But such time-specific
beliefs concerning remote obligations are surely extremely rare, and hence
it is extremely unlikely that Mandy's remorse will be appropriate after all.
While I agree (barring the doubts raised at the end of the last section)
that an agent's belief concerning his guilt must be accurate if his remorse is
to be appropriate, I don't think we need agree that it must be fully accurate.
If Mandy believes on Monday both that she ought to have helped Phil and
that she ought to have babysat Bill's child, then her remorse for having
failed to satisfy one of these obligations will be appropriate (given (7.28)),
whether or not the content of her belief was so specific as to include explic-
it time indices with respect to her obligations; at least, this is so as long as
she doesn't attach incorrect time indices to her obligations. Thus it need not
be at all rare that someone, who feels guilty about an action performed
when he was in a moral bind, does so in an appropriate way.
Binds are not the only type of nonbasic dilemma. Consider this situa-
tion (where "ought/' expresses primary overall obligation and "ought2"
secondary overall obligation, as discussed in Section 4.4):

(7.31) (a) S oughtj overall at T, to do A at T3;

(b) 5 can at T1 do A at T3;
(c) 5 ought2 overall at T1 to do B at T4;
(d) S can at Tt do B at T4; and
(e) S cannot at Tt do both A at T3 and B at T4.

47 What follows is borrowed from comments made by Earl Conee on an early version of
Zimmerman (1987) on the occasion of its presentation at the 1984 meeting of the Eastern
Division of the American Philosophical Association.

228
We may call this sort of situation a subsidiary dilemma. Such situations are
possible; indeed, examples have already been given. Recall Alan, for
instance, who oughtt to attend the 10:00 A.M. meeting on the first floor but
fails to do so and, as a result, ought2 to attend another 10:00 A.M. meeting
on the second floor. He can do each of these things, but he cannot do them
both.
Consider, finally, this situation (where S1 is one person and S2 another):
(7.32) (a) Sj ought overall at T, to do A at T3;
(b) Sj can at T, do A at T3;
(c) S2 ought overall at Tt to do B at T4;
(d) S2 can at Tt do B at T4; and
(e) S1 and S2 cannot jointly at Tt do both A at T3 and B at T4.

We may call this sort of situation an interpersonal dilemma. Whether or not

such situations can arise is an issue that I shall address in Section 9.4.
We find, then, that at least three types of nonbasic moral dilemma can
be distinguished. There are also variations on these types ("subsidiary
binds," "interpersonal binds," "interpersonal subsidiary dilemmas," and so
on). Moreover, the account of overall obligation that I have proposed,
while repudiating the possibility of basic dilemmas, accommodates the pos-
sibility of at least some nonbasic dilemmas. There is an aspect to this accom-
modation at which I have only hinted but which I wish now to
acknowledge explicitly. With respect to binds and subsidiary dilemmas, my
account implies that such dilemmas can arise only by way of some wrong-
doing. It thus gives detailed expression to the traditional Thomist distinc-
tion between a perplexity simpliciter and a perplexity secundum quid.4S The
distinction may be broadly characterized in this way: a perplexity simpliciter
arises if, through no wrongdoing on anyone's part, it happens that what-
ever can be done will be wrong; a perplexity secundum quid arises, on the
other hand, if it happens that whatever can be done will be wrong, but this
predicament has been induced by wrongdoing on someone's part. Alan
Donagan, one of the few philosophers to have entertained this distinction
in recent years, claims, as does Aquinas, that morality would be inconsis-
tent if it admitted perplexities simpliciter but not if it admitted perplexities
secundum quid; he claims further that morality is not inconsistent, and so does
not admit perplexities simpliciter, although it does admit perplexities secun-
dum quid.49 In this he echoes Georg von Wright, who puts the point pic-

48 Aquinas (1964-75), Ia2ae. 19,6 and 3a. 64,6.

49 Donagan (1977), pp. 144-5, and (1987), pp. 285-6.

229
turesquely as follows: "[P]redicament, though logically possible, can only
arise through antecedent sin... It is only as a consequence of a fall that a
man can come to be in a predicament."50 Von Wright talks of the agent's
situation before the "fall" as "prelapsarian." Let us say that (7.30) and (7.31)
represent two different types of predicament into which agents may enter
by way of a moral lapse.
It is no accident that the illustrations that have been given of nonbasic
dilemmas involve such lapses. (No illustration has been given of an inter-
personal dilemma, and you have no way yet to tell whether my account
accommodates such dilemmas or whether, if it does, it does so only by
virtue of some lapse. Again, I shall discuss this later, in Section 9.4.) The
illustration of a bind has it that Mandy wrongly promises Bill to babysit his
child. Such a lapse is required on my view, for only by way of such a lapse is
it possible that what formerly was deontically best (helping Phil move fur-
niture) no longer is so. Had Mandy not deviated from the strait and narrow
(departed from the path of righteousness, made a turn for the worse) by
making her promise to Bill, babysitting Bill's child would not have become
the best thing for her to do; it would have remained second-best.51 But
given the deviation onto a different path, helping Phil drops from first- to
second-best. Hence she becomes obligated to babysit Bill's child; but
hence, also, she is no longer obligated to help Phil. Her lapse does not
generate a basic dilemma.52
50 Von Wright (1968), p. 79.
51 It might be thought that Mandy's helping Phil move furniture could have been rendered
second-best not by Mandy but by something else (e.g., by Phil's not really caring very
much whether she helped him, or by Bill's child desperately needing to be babysat by
Mandy in particular), so that its being obligatory from Saturday on for Mandy to babysit
the child is not due to any wrongdoing on her part (and perhaps not on anyone's part).
But this is in fact not possible. Either Mandy could have prevented this other event - the
rendering second-best of her helping Phil - or she could not. If she could not, then her
helping Phil simply never was the best that she could achieve, contrary to the initial
hypothesis. If she could, then her failure to prevent the event either constituted a lapse
on her part or it didn't. If it did not, then the best that she could achieve involved allow-
ing her helping Phil to be rendered second-best; but in this case her helping Phil simply
never was the best that she could achieve, again contrary to the initial hypothesis.
Compare the remarks made in Subsection 3.2.4 concerning the cancellation of obliga-
tion.
52 Opponents of dilemmas have sometimes muddied the waters by claiming that perplex-
ities secundum quid are possible but without acknowledging that such perplexities cannot
take the form of basic dilemmas. (See von Wright, 1968, pp. 80—1; Donagan, 1977, pp.
144-5, and 1987, pp. 285-6; McConnell, 1987, p. 160.) Basic dilemmas cannot arise by
way of a lapse, because basic dilemmas are impossible, period, for the reasons given in
Section 7.2. Cf. Feldman (1986), pp. 204-5.

230
 It is clear, too, that subsidiary dilemmas can arise only by way of wrong-
doing. Alan's obligation to attend the meeting on the second floor arises
only out of his failure to attend the meeting on the first floor. Were he to
do what he ought! to do, it never would be the case that he ought2 to do
differently. I confess that here talk of "dilemma" is quite possibly mislead-
ing, for it seems to be at best an exaggeration to say that it is inevitable that
Alan do overall wrong. While it is true that he cannot avoid doing wrong
if he does what he ought2 to do, the fact is that he can avoid its being the
case that he ought2 to go to the meeting on the second floor; he can avoid
this by going to the meeting on the first floor, and in so doing he would do
no overall wrong at all. In this respect, subsidiary "dilemmas" are both like
and unlike binds. They are like binds in that the latter also do not involve
wrongdoing that can in no way be avoided by the agent; if Mandy had not
acted wrongly on Saturday by making her promise to Bill, she would have
avoided her bind altogether. But subsidiary "dilemmas" are unlike binds in
that the latter do involve there being a certain time (Saturday, in Mandy's
case) at which wrongdoing has become inevitable. So perhaps conflicts of
subsidiary obligations really should not be called a species of dilemma at all.
I include discussion of them here for the simple reason that their form
resembles that of other conflicts that are properly called dilemmatic (com-
pare (7.31) with (7.7), (7.30), and (7.32)) and because they can arise only
by way of a lapse.

231
 8

Supererogation

Steve, introduced in Chapter 5, stopped to render first aid to Carl, who had
had a car accident, thereby failing to keep his dinner date with Dave. This
case was used, following W. D. Ross, as an illustration of one prima facie
obligation overriding another and thereby constituting an overall obliga-
tion. But was Steve really overall obligated to do what he did? I am sure
that many would say that he was not. On the contrary, they would say that
he went well beyond the call of duty; his action was supererogatory.

8.1 THE NATURE OF SUPEREROGATION

8.1.1 Going beyond prima facie obligation
It may be that some writers consider supererogation simply to consist in
going beyond some prima facie duty or obligation. This is one way to inter-
pret, for instance, the following passage by Joel Feinberg:
A janitor has a duty to spend eight hours cleaning his employer's floors. He
works ten hours for eight hours' pay. Duty required eight hours; he gave duty
plus two, and thus in a perfectly intelligible sense, did "more than" his duty.1
This passage may seem to suggest that the janitor has a prima facie obliga-
tion (of fidelity, perhaps) to work eight hours; in working ten hours, he
goes beyond this duty and thus performs a supererogatory action. But even
if that is what this passage suggests, it is not, I am sure, what most people
have meant by the term "supererogatory," and it is not what I shall mean
by it. For it is clearly possible that, in going beyond some prima facie duty,
one is merely doing what one is overall obligated to do or, worse, failing
to do what one is overall obligated to do, and in neither case is one acting
supererogatorily.

1 Feinberg (1968), p. 397.

232
 8.1.2 Going beyond perfect obligation
Some writers conceive of supererogation in terms of going beyond perfect
overall duty or obligation. This leaves open the possibility that one never-
theless has an imperfect duty to do the act in question. In Subsection 2.1.4
several different senses of "(im)perfect duty" were distinguished, and some
of these have been proposed as appropriate for the definition of supereroga-
tion.
Kant claims that we have an imperfect duty of charity. One suggested
interpretation of this was that, while we have an overall obligation to act
charitably, we enjoy a certain leeway with respect to how to satisfy it, in
that we may choose between displaying charity to A and displaying it to B
and displaying it to C and so on. The duty is thus, in a certain sense, dis-
junctive (see (X) in Subsection 2.1.3). In this way, it might be said, when
1 act charitably towards a particular person, I am acting supererogatorily,
since I have no overall obligation to do that particular act; I could have sat-
isfied my overall obligation to display charity by acting charitably towards
someone else entirely.
This won't do. Suppose that I have an overall obligation either to act
charitably towards A or to act charitably towards B or to act charitably
towards C (there are no other candidate recipients of my charity), and that
therefore none of these individual acts is such that I have an overall oblig-
ation to perform it. Surely I do not "go beyond" duty, in the requisite sense,
simply by acting charitably towards A while bypassing B and C; I simply
fulfill my duty of charity. Of course, if I act charitably towards all three, I
might then do something supererogatory; but if so, this does not consist
simply in the performance of a duty that is imperfect, in the present sense.
Another suggested interpretation of "imperfect duty" was that the duty
in question is a duty not so much to act in a certain way as to adopt a cer-
tain maxim for action, and it has been proposed that a supererogatory action
is one that conforms with a maxim that satisfies such a duty.3 But while this
may be the best that can be done by way of squaring Kant's theory with
supererogation, it falls short of what is required for supererogation as this is
normally understood; for it does not accord the supererogatory act the
status of lying beyond duty altogether.4

2 Cf. Heyd (1982), pp. 121-2.

3 See Hill (1971), p. 71. Hill is speaking for Kant, and he suggests other conditions as well,
but not among these conditions is the agent's having no duty of any sort that applies to the
action in question.
4 Cf. Baron (1987), p. 244.

233
 Another suggested interpretation of "imperfect duty" was that of an
obligation that is not an obligation-to, that is, of an obligation that is not
correlative to some right. (See Subsection 5.5.1 for a discussion of this cor-
relativity.) I think many writers have understood supererogation in terms
of acts that are not owed to anyone in the way that an act to the performance
of which someone has a claim-right is owed to that person. Feinberg's
example of the janitor can certainly be construed in this way, as can his
example (cited in Section 1.3) of giving a match to a stranger.5 Similarly,
concerning cases such as Steve's Frances Myrna Kamm writes that only one
duty is involved (namely, Steve's duty-to-Dave to have dinner with him),
and she says that the alternative action (namely, Steve's helping Carl) thus
supersedes duty.6 But again this won't do, unless all overall obligations are
obligations-to. For otherwise one may have an overall obligation to do
something that goes beyond an obligation-to, and this can hardly rank as
supererogatory. For example, even if a lifeguard who is "off duty" goes
beyond any specific duty-to when saving a swimmer, whereas a lifeguard
who is "on duty" does not, if the former nonetheless has an overall obliga-
tion to save the swimmer — if, that is, he would do overall wrong in not
saving the swimmer — then he can hardly be said to be acting supereroga-
torily in saving the swimmer.

8.1.3 Going beyond overall obligation

No, for an act to be supererogatory, it must be such that the agent has no
overall obligation to perform it, period. That is how I shall understand
"supererogatory," and that is, I believe, how most people understand it.8
Of course, that is not all that's required. The act must also be such that the
agent has no overall obligation not to perform it. Thus it must be neither
obligatory nor wrong; it must be (not just personally optional but) morally
optional. And it must also be especially valuable, in some way. After all, not
all optional acts are supererogatory; some merely serve to fulfill a disjunc-
tive obligation rather than to "go beyond" obligation.

5 Feinberg (1968), pp. 392-3. Actually, Feinberg would not call giving the stranger a match
supererogatory, since it does not involve any great sacrifice or risk. More on this later.
6 Kamm (1985), pp. 119-20.
7 See New (1974), p. 179ff.; Heyd (1982), pp. 119-20.
8 The two most extensive treatments of supererogation to date concur. See Heyd (1982),
pp. 120, 125; Mellema (1991), p. 11 and passim.
9 On the distinction between personal and moral optionality, see the discussion in
Subsection 2.1.2 immediately following the presentation of (I).

234
 There is a flip-side to supererogation that some writers have recently
remarked on. We may call it "suberogation."10 Whereas a supererogatory
act goes beyond duty, a suberogatory act, we may say, falls short of decen-
cy. Perhaps Steve acted in a supererogatory manner when he stopped and
helped Carl; let us now assume so. Let us assume, that is, that he would have
done no overall wrong if he had passed Carl by and kept his date with Dave.
But what if, in passing Carl by, he had taunted Carl, mocking his misfor-
tune? This would have been a nasty thing to do, but perhaps it wouldn't
have been wrong. If it wouldn't have been wrong, it would have been
suberogatory: optional but especially disvaluable, in some way. (Here and
henceforth in this chapter, when I use the term "optional," I mean that
which is morally optional.)
How to render more precise the sense of "valuable" that pertains to
supererogation (and that of "disvaluable" that pertains to suberogation) is
a very tricky question that I shall address later. (Be forewarned, though: I
will not attempt to resolve this question.) Nonetheless, this much should
be said now: a supererogatory act is more valuable than any merely option-
al — that is, optional but not supererogatory — alternative; for a supereroga-
tory act "goes beyond" what duty requires and, where there are moral
options, duty requires only that one not do what is not optional. (Similarly,
a suberogatory act is less valuable than some merely optional - that is,
optional but not suberogatory - alternative.) This condition must not be
misunderstood. In saying that supererogatory acts go beyond duty by being
more valuable than any merely optional alternative, I am not saying that all
supererogatory acts are more valuable than all nonsupererogatory (whether
obligatory or merely optional or wrong) acts. Suppose that in Situation 1
John, a janitor, has two alternatives, that of working ten hours (act A) and
that of working eight (act B); and let us suppose that A would be
supererogatory and B not. In Situation 2 John has two other alternatives,
that of working twelve hours (act C), as he has promised to do, and that of
working eleven hours (act D); and let us suppose that C would be obliga-
tory and D therefore wrong. Whatever it is that makes A more valuable
than B, it seems plausible to say that both C and D are therefore more valu-
able still, even though A is supererogatory while neither C nor D is
supererogatory (indeed, the latter is wrong). The point is, however, that A
is more valuable than any nonsupererogatory alternative, that is, more valu-
able than any nonsupererogatory option that John has in Situation 1. This

10 Chisholm (1968) was perhaps the first to emphasize this phenomenon. It is there called
"offence." The term "suberogatory" is used in Driver (1992) and Stump (1992).

235
point has sometimes been missed. For example, Roderick Chisholm says
the following:
[I]f we contrast the man who performs an heroic act of duty with the man who
performs a trifling act of supererogation,.. .we can say that the obligatory act of
the former is better.. .than.. .the supererogatory act of the latter... [Hence we]
cannot say that the supererogatory must be better than the obligatory.11
This observation leads Chisholm to give an account of supererogation that
does not acknowledge any sense in which the supererogatory "goes
beyond" duty.12 In doing so, while he allows for the superiority of C to A,
he neglects the superiority of A to B.
To repeat, then: a supererogatory act is one that is both optional and
more valuable, in some way, than any nonsupererogatory alternative.
Insofar as it involves optionality, supererogation is of course agent- and
time-relative, in that an act's being optional for one agent at one time is no
guarantee that it is optional for that same agent at another time, let alone
for another agent at that same or another time. It can happen, therefore,13
that it is supererogatory for S at Tt to do A at T3 but not supererogatory for
5 at T2 to do A at T3; and it can also happen that it is not supererogatory for
S at T1 to do A at T3 but supererogatory for S at T2 to do A at T3. As to the
first possibility: perhaps at T2 S makes a morally optional promise to do A
at T3, thereby rendering the act obligatory; or perhaps at T2 S makes an
optional promise not to do A at T3, thereby rendering the act wrong. As to
the second possibility: perhaps at T2 S commits an act that converts his
doing A at T3 from obligatory or wrong to optional. (Such an act would
itself have to be wrong. See the discussion of lapses in Section 7.4.)
It is worth noting that talk of the optionality of a supererogatory act is,
if left unqualified, most naturally associated with acts not being construed
as act-tokens. As noted in Subsection 2.2.5, the claim that Jones ought to
help that old lady on Monday is probably not to be understood as imply-
ing that there is a particular act-token that is obligatory; rather, it is best
understood either as implying that a certain act-type (or some other abstract
entity)14 is obligatory or as implying that a certain disjunction of act-tokens
is obligatory. Similarly, the claim that it is optional, indeed supererogato-
ry, for Jones to help that old lady on Monday is probably not to be under-

11 Chisholm (1968), pp. 415-16 and 421. Cf. Chisholm and Sosa (1966), p. 331.
12 Chisholm (1968), p. 424ff.
13 For all that has been said so far; but see the discussion of the possibility of supererogation
in the next section.
14 See notes 63 and 68 to Ch. 2.

236
stood as implying that there is a particular act-token that is optional
(supererogatory); rather, it is best understood either as implying that a
certain act-type is optional (supererogatory), or as implying that a certain
conjunction of act-tokens is such that each token is optional (supererogatory).
Similar remarks pertain to the suberogatory. Insofar as the suberogato-
ry is optional, it can happen that an act is suberogatory for an agent at one
time but not at another. Furthermore, if "suberogatory" is to be applied to
act-tokens, it is probably best applied to conjunctions of them.

8.2 THE POSSIBILITY OF S U P E R E R O G A T I O N

The foregoing is intended to provide a rough account of the nature of
supererogation and suberogation. Even if it is accurate as far as it goes, how-
ever, this of course does not imply that there are, or even can be, any
supererogatory or suberogatory acts; still less does it imply that those acts
that we normally call supererogatory are in fact so. The philosophical liter-
ature is full of reasons for thinking that supererogation is impossible. Many
of the most influential normative theories seem to rule it out. It appears, for
instance, to be inconsistent with certain strains of Christianity,16 with
Kantianism,17 with utilitarianism and, more generally, consequentialism,18
and with Ross's theory;19 in every case the problem seems (roughly) to be
either that, according to the theory, no action that is nonobligatory can be
valuable in the requisite way, or that, according to the theory, no action
that is valuable in the requisite way can be nonobligatory.
In addition, there may be good reason to deny the possibility of subero-
gation even if that of supererogation is accepted. This is because there is a
certain asymmetry between these concepts that might not at first be appar-
ent.21 In particular, the greater the extent to which an act goes beyond duty,
the more supererogatory it is; its status as optional is not thereby threatened.
But the same cannot be said of suberogatory acts. That is, it is not the case
that the greater the extent to which an act falls short of decency, the more
suberogatory it is; on the contrary, its status as optional is thereby threat-

15 Cf. Mellema (1991), p. 29ff.

16 Cf. Heyd (1982), p. 26ff.; Mellema (1991), Ch. 3.
17 Pace Hill (1971).
18 Cf. Feldman (1978), pp. 48-50; Heyd (1982), p. 76ff.; Mellema on Feldman's theory
in Mellema (1991), p. 79ff.; New (1974); Kagan (1989).
19 See the discussion of (5.25) and (5.27) in Subsection 5.4.3.
20 Cf. Heyd (1982), p. 73.
21 See especially Heyd (1982), p. 128ff., and Mellema (1991), p. 185ff.

237
ened, in that at some point it presumably becomes wrong. And it may be
that this point is reached immediately, as it were, so that the status ofsubero-
gation is never attained. It seems to me not accidental that the most plau-
sible of the relatively few examples of allegedly suberogatory acts that exist
in the literature (including, if it is plausible, the example given earlier in this
chapter of Steve's taunting Carl) are most readily understood as examples
where the agent has managed, in some morally repugnant way, not to vio-
late an obligation- to. For instance, Chisholm and Ernest Sosa talk of minor
acts of discourtesy (such as taking too long in a restaurant when others are
waiting) and of not-so-minor acts committed while competing with
others (such as advertising in a way calculated to drive a competitor out of
business), and in such cases they explicitly say that the act is permissible
because the agent is acting "within his rights."22 Similarly, Julia Driver gives
several illustrations of what she takes to be suberogatory actions (a single
person takes a seat on a train, thereby preventing two other people from
sitting together, although another seat is available; a man refuses to donate
a kidney to his brother, although the latter will die without it), and she too
explicitly talks of the agent as acting "within his rights."23 But if the realm
of obligation is not exhausted by what there is an obligation-to to do, then
these examples lose some of their force. Perhaps we can agree that, typi-
cally, one acts within one's (liberty-)rights, that is, that one violates no one
else's (claim-)rights, when one is discourteous to another or "plays hard-
ball" in competitive contexts or refuses to donate a bodily part, but this does
not mean that we should agree that in such cases one does no overall wrong.
And if one does do overall wrong in such cases, then clearly one's action is
not optional, and hence not suberogatory.
There would seem to be powerful reasons, therefore, for thinking that
supererogation is impossible, and further, independent reasons for think-
ing that suberogation is impossible. However, determination of whether or
not these moral phenomena are in fact impossible would require a detailed
investigation of the theories to which I have just alluded (something that I
shall not attempt) as well as a detailed account of the pertinent senses of
"valuable" and "disvaluable" that apply to supererogation and suberoga-
tion (something that I have already said I shall not attempt). In the absence
of such investigation, it would seem premature to rule out the possibility
of supererogation and suberogation. Yet it may seem that this is just what
my account of obligation does. For it is a maximizing account, and certainly

22 Chisholm and Sosa (1966), p. 326.

23 Driver (1992), p. 287.

238
many philosophers appear to believe that maximization itself precludes
supererogation.24 If this belief is correct, then, unless cogent arguments can
be given, independently of considerations about maximization, to support
such preclusion (as arguments were given in Chapter 7 against the possi-
bility of basic moral dilemmas), such preclusion may be taken as a sign of
the inadequacy of the proposed account of obligation. But is the belief
correct?

8.2.1 Supererogation and discretion

Supererogatory acts, it has been said, are at once both optional and espe-
cially valuable. But a maximizing theory links value with obligation, and
thereby hangs the problem. As Michael Slote puts it, when discussing con-
sequentialism in particular: "[C]onsequentialism, precisely because it insists
that one do the best one can, leaves no room for moral supererogation: bar-
ring ties for first place, everything it morally permits it morally requires."25
And yet such ties are possible, and thus acts can be optional. So why say that
maximization precludes supererogation altogether?
It might be replied that, while a maximizing theory does admittedly
imply that some acts can be optional, it doesn't allow for the possibility that
acts be fully optional, whereas supererogatory acts are not just optional but
fully optional. An act is fully optional only if it is not just not obligatory but
also such that its performance does not constitute a way to fulfill some
obligation; it is, as it were, wholly beyond the reaches of obligation. Now
it is true that a maximizing theory is hardly likely to classify an optional act
as fully optional. Consider a case that has the oversimplified form depicted
in Figure 8.1. Here the agent has four and only four alternative courses of
action, the first two equally valuable, the others less so. In this case, acts A
and B are optional, whereas C is wrong. But neither A nor B is fully option-
al, since there is still an obligation lurking not too far in the background,
namely, the obligation to do either A or B (see (X) in Subsection 2.1.3).
Thus, although the agent is not obligated to do A and is not obligated to
do B, performance of either will fulfill an obligation.

24 Included in this number are some who accept and also some who reject a maximizing
theory. See, for example, the following, all of whom appear essentially to hold the opin-
ion in question: Moore (1965), p. 78; Baier (1958), pp. 203-4; Smart (1973), p. 55; New
(1974), p. 179ff.; Heyd (1982), p. 78; Schemer (1984), pp. 21-2; Feldman (1986), pp.
48-50; Slote (1989), pp. 3, 171; Kagan (1989), pp. xi-xii, 1-2, 7-8, and elsewhere;
Stocker(1990),p.311.
25 Slote (1989), p. 3.

239
 Figure 8.1

But here two points are to be noted, one minor, the other not. The
minor point is this: a maximizing theory can in principle allow for the pos-
sibility of fully optional acts, as is evident in the case depicted in Figure 8.2.
The much more important point is this: it is not true that supererogatory
acts must be fully optional. Consider the case in Figure 8.1. Why is it that
it is obligatory to do either A or B? The answer is, of course, that any course
of action involving neither A nor B is wrong. In doing either A or B, the
agent is thus fulfilling the obligation not to do something else. It cannot be
accepted that this alone renders both A and B nonsupererogatory. After all,
it is surely possible that a supererogatory act have some alternative that is

Figure 8.2
240
wrong (even if not every alternative can be wrong — for then it would be
obligatory), so that, if the supererogatory act is performed, that very per-
formance, though of course not itself obligatory, will nonetheless consti-
tute the fulfillment of an obligation, to wit, the obligation not to perform
the alternative act.
Why is it, then, that maximizing theories have been thought not to
accommodate supererogation? Well, consider the classic maximizing
theory of act utilitarianism, which is traditionally seen to be an amalgama-
tion of hedonism (the view that all and only states of pleasure are intrinsi-
cally good and all and only states of pain are intrinsically bad) and
consequentialism (the view that one ought to produce the most favorable
balance of intrinsic goodness over intrinsic badness that one can). It is often
said that act utilitarianism rules out supererogatory action,26 but this is ques-
tionable. What it certainly does do is imply that many of the acts that we
would normally call supererogatory are in fact not so, since it implies that
these acts are either obligatory or wrong. But this is not the same as imply-
ing that there are no supererogatory acts, let alone the same as implying that
there can be no such acts. Indeed, the consequentialist aspect of act utilitar-
ianism (which is what makes the latter a maximizing theory) rules in the pos-
sibility of optional acts (as long as ties in sum intrinsic value are in principle
possible, and there seems to be no reason to deny this). This fact might sug-
gest to some that it is act utilitarianism's hedonistic, rather than conse-
quentialist, aspect that renders it inimical to supererogation, and that
another version of consequentialism would accommodate supererogation
more readily.
But actually it is not that easy. If we turn to the nonhedonistic conse-
quentialism of G. E. Moore, we find that Moore's theory, too, has a hard
time accounting for supererogation. While this theory (which allies conse-
quentialism with a theory ofvalue according to which not just states ofplea-
sure and pain but also other states — knowledge, moral qualities, beauty27 —
have non-neutral intrinsic value) of course allows for the possibility of acts'
being optional — as does any consequentialist or, more broadly, maximiz-
ing theory - it seems not to square with our common intuitions concern-
ing supererogation any better than act utilitarianism does. For example, just
as act utilitarianism implies that many instances of self-sacrifice that we
would normally call supererogatory are in fact either obligatory or wrong,
so too does Moore's theory imply this. If one can maximize intrinsic value
26 See, for example: Baier (1958), pp. 203-4; Smart (1973), p. 55; New (1974), p. 179ff.;
Feldman (1978), pp. 49-50; McConnell (1980), p. 36ff.
27 Moore (1965), p. 102.

241
only by way of extreme self-sacrifice, then Moore's theory implies that one
must, one is obligated to, undertake such sacrifice.28 This observation applies
equally well to the neoconsequentialist view of Fred Feldman, couched in
terms of possible worlds rather than in terms of consequences,29 and it is this
view that inspired the account of obligation that I have proposed.30 Must we
therefore conclude that maximization does indeed preclude supererogation?
Not yet. It could be that the problem lies with (neo)consequentialism
in particular rather than maximizing theories in general. Although
(neo)consequentialism allows for acts' being optional, its focus on the pro-
motion of intrinsic value neglects the fact that it is in the area of personal
interest that we commonly look for a moral theory to give the agent some
"slack," some latitude as to how to proceed. What, then, if a maximizing
theory were to admit into the final determination of right and wrong cer-
tain agent-relative values? What if, in the final computation of what is best
and what not, an agent's own projects and commitments were given a
weight disproportionate to that which their intrinsic value alone supplies?
This possibility has been explicitly admitted in the account of obligation
that I have proposed; recall that in this account S's being obligated at T to
perform some act is analyzed in terms of the value for S at T of accessible
worlds in which he does and does not perform the act. Does this feature
of the account provide the means of reconciling maximization with
supererogation?
Again, it is not that easy. Once more, there are two points to be made,
one minor, the other not. The minor point is that it would be a mistake to
think that supererogation essentially concerns agent-relative values. Some
people might claim, for instance, that, although it is obligatory to refrain
from killing babies whenever possible, it is supererogatory to refrain from
killing fetuses. Whether or not this is correct, it is certainly intelligible, and
no tie to agent-relative values is, or need be, involved. The much more
important point is this: even though a maximizing theory that incorporates
agent-relative values in the way indicated allows, as do all maximizing
theories, for certain acts to be optional, it does not appear to allow for cer-
tain acts to be discretionary in the way that supererogation requires. Consider
a case where A and B are each optional, according to the maximizing the-
ory in question, and where their values (and for this reason the values of the
top-ranked worlds that include them) are, respectively, predominantly
agent-neutral and predominantly agent-relative. For example, A might be
28 Cf. Heyd (1982), pp. 77-8.
29 Cf.Mellema(1991),p. 87.
30 See Ch. 2, n. 7.

242
a charitable act of some sort (in which the agent has no vested personal
interest), while B is the purchase of some high-priced compact-disk play-
er (in which the agent has considerable vested personal interest). Now con-
sider a very slightly modified case, where B's value marginally overrides A's
(and for this reason no accessible v4-world is quite as valuable as any acces-
sible B-world), because the agent's vested personal interest is marginally
increased. The maximizing theory would imply that in this modified case
neither A nor B is optional; on the contrary, A is wrong and B obligatory!
Similarly, if ^4's value were marginally to override B's, then the maximiz-
ing theory would once again imply that neither A nor B is optional; on the
contrary, A would be obligatory and B wrong. In such cases, engaging in
the sort of self-sacrifice that A would entail is clearly not left to the agent's
discretion.31
It is probably at this point that many will diagnose a fatalflawin all max-
imizing theories. It is not, as it may at first have seemed, that such theories
cannot admit that certain acts are optional; clearly they can. It is rather that
such theories cannot admit that certain acts are discretionary, in the sense
that it is left to the agent's discretion whether or not to perform the act when
its performance and nonperformance are of different values. This, it might
be claimed, is something that is essential to supererogation, but something
that maximizing theories cannot accommodate.
Somewhat more precisely, and applied specifically to the account of
obligation that I have proposed, the argument is this:
(8.1) necessarily, for any act A, if the performance of A is supererogatory
and its nonperformance merely optional, then all accessible ~A-
worlds are less valuable than some accessible /1-world;
(8.2) if my account of obligation is correct, then, necessarily, for any act
A, if all accessible ~/l-worlds are less valuable than some accessible
^4-world, then the performance of A is obligatory;33
(8.3) necessarily, for any act A, if the performance of A is obligatory, then
it is not supererogatory;
hence,
(8.4) if my account of obligation is correct, then, necessarily, there is no
act A such that the performance of A is supererogatory and its non-
performance merely optional.
31 Cf. Bennett (1989), pp. 56-7.
32 Cf. Raz (1975), p. 164. Cf. also Stocker (1990), p. 311. See also Heyd (1982), pp. 167-8.
33 This is a bit rough. It ignores subclause (c4) of (I), which is rendered necessary by con-
siderations laid out in Subsection 2.2.6.

243
An analogous argument can be made concerning suberogation:
(8.5) necessarily, for any act A, if the performance ofA is suberogatory and
its nonperformance merely optional, then all accessible ^4-worlds are
less valuable than some accessible ~;4-world;
(8.6) if my account of obligation is correct, then, necessarily, for any act
A, if all accessible ^4-worlds are less valuable than some accessible
~y4-world, then the performance of A is wrong;
(8.7) necessarily, for any act A, if the performance of A is wrong, then it
is not suberogatory;

hence,
(8.8) if my account of obligation is correct, then, necessarily, there is no
act A such that the performance of A is suberogatory and its non-
performance merely optional.

8.2.2 The division of values

There is but one possible way for me to resist the conclusions of these argu-
ments while hanging on to maximization. Like any maximizer, I must
endeavor both to have my cake and to eat it - that is, endeavor both to say
that, where A is supererogatory, its performance is superior to its nonper-
formance and, where A is suberogatory, its performance is inferior to its
nonperformance, and yet also to say that, since A is optional, its perfor-
mance and its nonperformance are of equal value. This can be achieved by,
and only by, declaring that there is more than one set of values pertinent to
the moral evaluation of an act.34 One set of values must be said to be perti-
nent to the determination of right and wrong and obligation, the other not.
Where A is supererogatory, and so it is left to the agent's discretion whether
or not to perform it, a maximizing theory must declare its performance to
be nondeontically superior but deontically equivalent to its nonperfor-
mance; and where A is suberogatory, a maximizing theory must declare its
performance to be nondeontically inferior but deontically equivalent to its
nonperformance. In this way, the first two premises of each argument will
be seen to be true, but only because of equivocation on the term "less valu-
able," and so, despite the evident truth of each of the third premises, the

34 Note: moral evaluation. It is in some morally relevant way that supererogatory acts are
superior to their nonsupererogatory alternatives. Simon seems to miss this in Simon
(forthcoming).

244
arguments are to be declared unsound and their conclusions thereby
resisted.
This solution should come as no surprise. In previous chapters I stressed
that my concern, when giving my account of obligation, was with the deon-
tic superiority of one action, or world, over another. This was intended to
leave open the possibility that one act, or world, be superior, even moral-
ly, to another in a nondeontic way.35 Let me now elaborate.
Suppose that consequentialists are correct and that 'what determines
whether or not an act is morally obligatory has to do with and only with
how it and its alternatives actually promote intrinsic value. In this case,
deontic superiority consists in superiority concerning such promotion.
Nonetheless, it is surely possible that a consequentialist should accept that
acts are open to a sort of moral evaluation that has nothing to do with their
deontic status. For example, some consequentialist might say that an act is
morally meritorious if and only if its agent intends by means of it to achieve
what is intrinsically best. Since intended results need not match actual
results, it would follow that an act can be morally obligatory without being
morally meritorious, and vice versa. A morally meritorious act would be
superior in a morally relevant way to a nonmeritorious one, but this supe-
riority would be nondeontic.
Note that, on this consequentialist view, there is a common core to an
act's deontic and its nondeontic moral status. Both have to do with the pro-
motion of intrinsic value. What distinguishes them is that the one has to
do with the actual promotion of such value, the other with its intended
promotion.
A nonconsequentialist can similarly distinguish between an act's deon-
tic and its nondeontic moral status. For example, someone might say that
whether or not an act is morally obligatory has to do with and only with
whether it actually respects people's rights. In this case, deontic superiori-
ty consists in superiority concerning such respect. Nonetheless, it is surely
possible that a nonconsequentialist of this sort should accept that an act is
morally meritorious if and only if its agent intends by means of it to respect
people's rights. Since, as before, intended results need not match actual
results, it would follow that an act can be morally obligatory without being
morally meritorious, and vice versa. Also, as before, there would in this case
be a common core to an act's deontic and its nondeontic moral status.
There would seem, then, to be good reason to accept the possibility that
an act be subject to a variety of moral evaluations, only some of which have

35 See note 13 to Ch. 2.

245
to do with its deontic status. A maximizing theory of obligation can exploit
this possibility in order to accommodate supererogation. Unless there are
further restrictions on supererogation that I have so far overlooked (an issue
to which I shall return in the next subsection), I therefore conclude that
maximization does not preclude supererogation.
Even if this conclusion is accepted, though, it may be thought that it
comes at too high a price; for some may find such a division of values to be
a deus ex machina with no rationale other than to save maximization. But
this is not so. First, it seems to me that appeal to such division is routine.
When someone is told that he should or ought to do some act, although he
doesn't have to,36 what's apparently meant is that it is valuable in a way that
doesn't render it obligatory. (Thus "ought" is here used in a nonbinding
sense. See Section 1.2.) Why should this common appeal to a double stan-
dard37 be deemed unavailable to a maximizing theory? The problem with
most maximizing theories is that they don't make such an appeal, not that
they can't, and this is why they cannot (as they stand) account for
supererogation. Second, and more importantly, there is no such thing as a
nonmaximizing theory of obligation, as I pointed out in Chapters 1 and 2,
unless such a theory does not imply that an obligatory act is "unique in
respect of value" (as Moore puts it)38 - that is, it does not in some way rank
an obligatory act higher than its nonobligatory alternatives - in which case
I would say that it completely misconstrues the notion of obligation.39 If so,
then the strategy of dividing values is the strategy that any theory of obliga-
tion must adopt if it is to admit supererogation and suberogation.
The argument underlying all this is very simple. It is as follows:
(8.9) if the performance of A is not deontically equivalent to its nonper-
formance, then one is obligatory and the other wrong;
(8.10) if one is obligatory and the other wrong, then neither is optional;
(8.11) if neither is optional, then neither is supererogatory or suberoga-
tory;
hence,
(8.12) if either the performance of A or its nonperformance is either
supererogatory or suberogatory, then they are deontically equi-
valent;
36 Cf. Feinberg (1968), pp. 392-3. Cf. also Chisholm (1968), p. 417.
37 Cf. Attfield (1979). Cf. also Raz (1975).
38 See note 37 to Ch. 1.
39 This is not to say that every theory of obligation exactly fits the form of (I), as I acknowl-
edged in Subsection 2.1.2.

246
hence,
(8.13) any superiority (due to being supererogatory) or inferiority (due to
being suberogatory) of the performance of A to its nonperfor-
mance, or vice versa, is nondeontic.
And so, if the strategy of dividing values comes at too high a price, then
supererogation and suberogation must simply be declared impossible.
But is the price too high? I don't see why. Surely it is plausible to say
that there is more to morality than the deontic. An act, or its agent, may be
praiseworthy or blameworthy, virtuous or vicious, and so on,40 and only if
such matters are directly reducible to the act's being obligatory, permissi-
ble, or wrong - which is surely not the case - would the act's deontic sta-
tus exhaust its moral status. If this is so, then supererogation and
suberogation can involve nondeontic matters such as these. The question
is which.
Well, this is the question that I said I would not try to resolve: precise-
ly what senses of "valuable" and "disvaluable" are at issue in supereroga-
tion and suberogation. Many different answers have been proposed.
Almost all accounts of supererogation claim that, if an act is supererogato-
ry, then either it or its agent is in some way praiseworthy, but just what goes
into praiseworthiness is controversial. Some claim that, if an act or agent is
to be praiseworthy, the act must be unusually valuable, and hence that there
are no trifling acts of supererogation;41 others allow for trifling acts of
supererogation that are, because trifling, only minimally praiseworthy.42
Some philosophers claim that, for an act to be supererogatory, its omission
must be such that either it or its agent is not blameworthy.43 Some philoso-
phers claim that a supererogatory act, even though it is not obligatory, is
such that it ought (in a nonbinding sense) to be done.44 Some philosophers
40 I would prefer to say that agents, rather than acts, are praiseworthy or blameworthy, vir-
tuous or vicious, and so on, but this might seem to put some strain on my contention
that there is a variety of ways to evaluate acts, morally. Recall, however, from Section 1.5
that, even when it comes to obligation, I am in fact not at all sure that it is best to cate-
gorize the evaluation in question as drt-evaluation. Perhaps it is better categorized as a
matter of a^enf-evaluation, in which case the strategy of dividing values is under no strain
at all.
41 E.g., Feinberg (1968), p. 400.
42 E.g., Chisholm (1968), p. 417, although he resists talk of praiseworthiness as such; Heyd
(1982), p. 127; Montague (1989), p. 107; Mellema (1991), p. 111.
43 E.g., Chisholm (1968), p. 424, although he resists talk of blameworthiness as such; Heyd
(1982), p. 115; Mellema (1991), p. 5. If this condition is not met, Mellema says that the
act can at best be "quasi-supererogatory." See Mellema (1991), Ch. 5.
44 E.g., Feinberg (1968), p. 392; Chisholm (1968), p. 417; Forrester (1975), pp. 225-6.

247
say that a supererogatory act, though not obligatory, is nonetheless "rec-
ommended," in a quasi-obligatory way, by morality.45 Some philosophers
claim that an act is supererogatory only if it has good consequences, or is
intended to have good consequences.46 Some claim that an act is
supererogatory only if it is performed for someone else's sake. And so on.
A similar variety of conditions may be proposed for suberogation. The mat-
ter is clearly both complex and controversial.
It is perhaps easier to envisage a division of morally relevant values when
the nondeontic values are not said to supply a reason for acting, harder
when they are said to supply such a reason. Suppose we agree that an act is
supererogatory only if it or its agent is praiseworthy. On some accounts of
praiseworthiness, an act or agent is rendered praiseworthy by the motive
underlying the act; on some accounts of motivation, the motive underly-
ing an act is beyond the agent's control; and, on some accounts of reasons,
what is beyond the agent's control cannot feature in a reason to perform an
act. Thus some accounts can declare the performance of a supererogatory
act to be wholly on a par with its nonperformance with respect to reasons
for action; the nondeontic superiority of the performance will be explained
without appeal to reasons for action.
On the other hand, some accounts of supererogation will declare the
superiority of the performance of a supererogatory act to constitute a spe-
cial reason for performing it. If this is essential to supererogation, then, if
supererogation is possible, there must be not only two sets of morally rel-
evant values but also two sets of moral reasons for acting. But, while it may
be difficult to see just how such an account should go, it is clearly not with-
out precedent. On the contrary, the common invocation of what is ideal
rather than obligatory seems, often at least, to constitute an appeal to a non-
deontic set of reasons for acting.

8.2.3 Continuity
It may seem that something has gone wrong in the foregoing defense of the
compatibility of maximization with supererogation and suberogation, in
that I have overlooked a condition of supererogation and suberogation
which blocks the strategy of dividing values. This is the condition that

45 E.g., Velazco y Trianosky (forthcoming).

46 E.g., Heyd(1982),p. 115.
47 E.g., Heyd(1982),p. 115.
48 Cf. Heyd (1982), p. 181.

248
David Heyd calls the condition of "continuity"49 and which Gregory
Mellema characterizes as follows: "There is a common and continuous scale
of value shared by supererogation and duty, and the difference between
supererogation and duty is reflected in the degree to which the value is real-
ized in each. The value is realized to a greater degree in supererogation than
in obligation."50 How can the strategy of dividing values be reconciled with
this talk of a common and continuous scale of value?
Whether or not such reconciliation is possible depends on just what is
involved in this condition of continuity that (allegedly) applies to
supererogation. It may seem that it is simply being claimed that the value
exhibited by supererogatory action, like that exhibited by obligatory
action, is a morally relevant value. Clearly I agree with this. But this can-
not be all that is meant; for, although this gives sense to the talk of there
being a common scale ofvalue, it does not give sense to the talk of there being
a continuous scale of value.
It may seem that it is simply being claimed that all supererogation
involves the sort of "oversubscription" (to use Feinberg's term) that is dis-
played by John when he works a supererogatory ten hours instead of the
merely optional eight, for clearly there is a sort of continuity operative here.
But this, too, cannot be what is meant, for two reasons. First, like Feinberg,
both Heyd and Mellema acknowledge that there are cases of supereroga-
tion that do not involve such oversubscription, as when, for instance, one
gives a stranger a match.51 Second, the sort of continuity that cases of
oversubscription display and cases of nonoversubscription do not concerns,
not a type of value, but a type of action (in John's case, the type of action
being that of working).
Perhaps what is being claimed is that, if A is supererogatory and some
alternative B merely optional, then both A and B must exhibit a common
type of morally relevant value, A exhibiting more of this value than B. This
is compatible with what I have said (although I doubt that it is true); for
even if B is merely optional, it may nonetheless exhibit a certain nondeon-
tic value, just not as much of this value as is needed to render it, like A,
supererogatory. For example, if no act of supererogation can be trifling, as
some claim, then a trifling favor will not be supererogatory; nonetheless, as
a favor it may be good in a certain way that is irrelevant to its deontic sta-
tus and thus share a type of nondeontic value with its less trifling,
supererogatory alternatives.
49 Heyd (1982), p. 5.
50 Mellema (1991), p. 26.
51 Feinberg (1968), p. 397; Heyd (1982), p. 135; Mellema (1991), p. 27.

249
 Perhaps, though, what is being claimed is that, ifA is supererogatory and
B merely optional, then both A and B owe their respective statuses wholly
to a certain single type of morally relevant value that they both exhibit, their
statuses being different solely due to the fact that the former exhibits more
of this value than the latter. Given (8.12) and (8.13), I cannot accept this;
but I know of no reason to accept it.
Now, some may respond to this dismissal of the condition of continu-
ity (understood according to the latest interpretation) as follows: certain
accounts of supererogation clearly include it, and thus, even if the condi-
tion hasn't been established as necessary for supererogation, still it is per-
fectly intelligible, and hence something must have gone wrong with the
argument for (8.12) and (8.13).
Perhaps the sort of account of obligation and supererogation that might
be thought most obviously to include the condition of continuity (under-
stood according to the latest interpretation) is a "satisficing" account. Slote,
who is favorably disposed towards such an account, says the following:
From the standpoint of common-sense it can be acceptable, permissible, to be
less than optimifically benevolent or beneficent, to give others less than one is
in a position to give them, so long as one is sufficiently beneficent, that is, gives
others a good deal or enough.52
If one does well enough, one's act is "satisficing," even if not maximizing,
with respect to the good that one can achieve; and an act need only be sat-
isficing to avoid being wrong. Then, with respect to supererogation, Slote
says this:
[S]atisficing forms of.. .consequentialism.. .are consequentialistic because they
treat overall consequences as the only thing relevant to act-evaluation and
because, more particularly, they regard any act with overall consequences less
good than those of some alternative as morally less good than that alternative.
But they allow for supererogation and for the moral permissibility of less than
optimal behavior.53
Note: "the only thing relevant." Even if such a satisficing theory is to be
rejected for some reason, doesn't its form demonstrate that a theory can sat-
isfy the condition of continuity (understood according to the latest inter-
pretation), and hence that my contention, that supererogation mwsdnvolve
a division of values, is false?
Well, recall that my contention that supererogation requires a division
of values rests on the dual claims that maximizing theories of obligation can
52 Slote (1989), p. 26, italics deleted.
53 Slote (1989), p. 142.

250
accommodate supererogation only in this way and that there is no such
thing as a nonmaximizing theory of obligation. Perhaps the former claim
will be granted, but what about the latter? What, in particular, about satis-
ficing theories of obligation?
There is of course a tension between the claim that obligation involves
maximization and the claim that it involves satisficing; but it is only a sur-
face tension. To resolve it, one need only recognize that I am saying that
obligation involves maximization of some sort, and this is clearly compati-
ble with its involving satisficing ofanother sort. In particular, if two acts, one
of which optimizes a certain value and the other of which merely satisfices
that value, are both equally permissible, then clearly they are to be ranked as
deontic equals; their deontic status is not determined solely by the degree to
which they realize the value in question. For instance, a theory might say
that an act is permissible if and only if it displays beneficence to at least
degree 5, supererogatory if and only if it displays beneficence to a degree
greater than 5. This is intelligible, but what it implies is that it is not simply
beneficence that is pertinent to the determination of right, wrong, and
obligation. Rather, it is beneficence-to-at-least-degree-5 that is at issue;
any acts that display this characteristic are deontically equivalent, whether
or not one displays a greater degree of beneficence than another. And it is
not beneficence-to-at-least-degree-5 that determines whether and to what
extent an act goes beyond duty; on the contrary, it is beneficence-to-a-
degree-greater-than-5. Thus the condition of continuity (understood
according to the latest interpretation) is not satisfied by this theory. Despite
there being a common core (namely, beneficence) to that characteristic
which determines an act's deontic status and that which determines its non-
deontic status, the characteristics are distinct.
(One would, of course, like a rationale for making the cutoff point
between mere optionality and supererogation be beneficence to degree 5.
One such rationale might be that other people's needs are satisfied when
and only when they are treated with beneficence to at least degree 5, and
that one is obligated to satisfy other people's needs but not obligated to do
more than this. Another rationale might give a different story. Once the
rationale, whatever it is, is given, it will be easier, I think, to see that on any
theory that accommodates supererogation, even a satisficing theory, there
is indeed more than one consideration relevant to the evaluation of an act,
contrary to what Slote says. For instance, if the rationale were the one just
illustrated, then an act's deontic value would turn on whether it meets other
people's needs, whereas its nondeontic value would turn on something else
entirely.)
251
 Perhaps this will have served to show how modest the strategy of divid-
ing values really is. Drawing a distinction between deontic and nondeon-
tic values merely reflects the fact that supererogatory acts and their merely
optional alternatives are both alike (with respect to optionality) and unalike
(with respect to being beyond duty), and, as such, it seems to me wholly
beyond reproach.
It should be noted, finally, that I have not been arguing that supereroga-
tion and suberogation are possible. For all that I have said, they may yet be
impossible. Certainly not every theory of obligation can, as it stands,
accommodate them. Those theories that deny any division of moral values
cannot, and others that do allow for such division may nonetheless under-
mine themselves. For example, any theory that says that something's being
good from the ideal point of view is itself a deontic reason for trying to
achieve it would appear to have undercut any advantage that this division
of values initially afforded it with respect to the accommodation of
supererogation. Ross's theory seems to have just this difficulty. In distanc-
ing himself from consequentialism by divorcing the right from the good,
Ross may at first seem to have proposed a theory that can account for
supererogation in a way that Moore's theory (unaccompanied by any divi-
sion of values) cannot. For the special prima facie duties of which Ross
speaks that vie for the status of duty proper do not owe their deontic status
to the fact (if it is a fact) that the acts in question, or what they bring about,
are in themselves desirable. Their deontic ground lies elsewhere, and this
would appear to allow Ross to declare that the desirability of an act, or of
what the act brings about, provides at most a nondeoritic reason for its per-
formance. But he does not say this. Rather, he declares the desirability of
an act, or of what it brings about, itself to be a deontic reason for perform-
ing the act; there is, he says, a general duty to maximize intrinsic value.54
Thus, as long as it is always possible for an agent to promote intrinsic value,
Ross seems no better able than Moore to accommodate supererogation.55
Even if a proponent of supererogation avoids such pitfalls, however,
there may be other reasons to find fault with what he says. Perhaps an act's
being supererogatory does (or would) constitute a moral reason to perform
it, and perhaps no such reason can be nondeontic after all.56 Or perhaps,
once it is clearly recognized that what is especially valuable about
supererogation is nondeontic, any inclination to say that what is (appar-
ently) supererogatory is in fact neither obligatory nor wrong is to that extent
54 Ross (1930), p. 27.
55 Contrast this assessment of Ross's theory with that given in Mellema (1991), pp. 72-3.
56 Cf.Kagan(1989),p.381.

252
undermined. (For example, some people may be moved to call an act
supererogatory because they wish to ensure that the agent is not blamed for
its nonperformance. But if an agent can avoid blame and yet do wrong,
then there may be reason to doubt that no "supererogatory" act is morally
wrong.)57 Whether or not any of this is so, I shall not seek here to deter-
mine. My aim has been merely to argue that maximization does not, as
such, preclude either supererogation or suberogation.
57 Cf. Moore (1965), pp. 79-80, with respect to consequentialism. Cf. also Baron (1987),
p. 262, with respect to Kant's theory.

253
 Cooperation

You and I never act in isolation. Like it or not, we are all members of
groups. Groups can achieve things that individuals alone cannot. This is an
obvious fact that has so far been neglected in this book, but it is a fact that
poses a serious problem for the account of obligation that I have proposed.

9.1 THE PROBLEM

The account that I have proposed is presented in Chapter 2. Its main claim
is captured in (I). For the time being, let us work with the following sim-
pler paraphrase of (I):
(9.1) for any person 5 and times T and T", S ought at T to do at T" the
best that he can at T do at T'.
The problem that besets this account may be illustrated as follows.1 Suppose
that there are two voters, Vincent and Virgil, and two candidates for elec-
tion, Smith and Jones, and suppose that the voters' options may be ranked
according to the matrix given in Chart 9.1. That is, it would be (deonti-
cally) best if both Vincent and Virgil voted for Smith; next best if one voted
for Smith and the other abstained; next best if both voted for Jones; next
best if one voted for Jones and the other abstained; and worst if either both
abstained or one voted for Smith and the other for Jones. The question is:
given (9.1), what should Vincent and Virgil do?
Before we attempt to answer this question, we should make note of
what the illustration presupposes. In particular, we must recognize that the
numbers represent the values of certain outcomes achievable by the group
of Vincent and Virgil, that is, achievable by them, jointly. It is thus being
presupposed that there are certain worlds accessible to this group, and also
1 The problem has received extensive treatment in Regan (1980). It is anticipated in
Gibbard (1965), Sobel (1968), and Barnes (1971). My illustration borrows from Regan
(1980), p. 18, and Feldman (1986), p. 155.

254
 Virgil
Vote for Vote for
Smith Jones Abstain
Vote for Smith 10 0 7
Vincent Vote for Jones 0 5 3
Abstain 7 3 0
Chart 9.1
that these worlds may be ranked according to their relative values. Let us
look at each presupposition in turn.
There can be no doubt that groups can sometimes achieve things that
subgroups or individuals cannot. Perhaps a group of four people can lift a
heavy piece of furniture but no fewer than four can do so. In general, then,
we must reject the claim that, whatever a group can do, some proper sub-
group or individual member can do also.
What about the converse? Should we accept or reject the claim that,
whatever a proper subgroup or individual member of a group can do, the
group itself can do? It may seem that we should reject this too. Perhaps one-
quarter of the assembled company can squeeze into the back of a
Volkswagen, but the whole group cannot; perhaps Paul can marry Paula,
whereas the group of Peter and Paul cannot; and so on. But these facts
should not blind us to the truth that whatever is accessible to a proper sub-
group or individual member of a group is also accessible to the group. If
one-quarter of the assembled company can squeeze into the car, then cer-
tainly the group as a whole can so act that one-quarter of its number
squeezes into it; if Paul can marry Paula, then certainly the group of Peter
and Paul can so act that Paul marries Paula.
In general, then, we should reject the first and accept the second of the
following two claims (where "{S^,..., SH}" signifies the group of persons
Sj through SM, and where an individual member S( of a group is, for the sake
of convenience, treated as the subgroup {5,}):
(9.2) if Wf is accessible to {Su ..., Sn} from W at T, then W is accessible
to some proper subgroup, if any, of {Su ..., Sn} from W at T;
(9.3) if W' is accessible to some proper subgroup of {Su ..., Sn} from W
at T, then W* is accessible to {Sl9 ..., Sn} from W at T.3
It is the falsity of (9.2) that underlies the problem with which I am present-
ly concerned.
2 This illustration is borrowed from Humberstone (1983), p. 29.
3 Cf. Humberstone (1983), p. 29; Feldman (1986), pp. 158-9.

255
 The second presupposition of the illustration is that worlds that are
accessible to a group have values (and may be ranked accordingly); that is,
they have deontic values relative to the group. Some philosophers will reject
this presupposition. If a group (as opposed to individuals in the group) can-
not comply with the Categorical Imperative, then presumably Kant would
reject the presupposition, since according to him deontic value relative to
an agent is rooted entirely in that agent's ability to comply with the
Categorical Imperative. Similarly, if a group cannot display gratitude, then
Ross would presumably find the presupposition suspect, at least in this
regard. But other philosophers would not hesitate to accept the presuppo-
sition. Consequentialists, for example, will accept that worlds that are
accessible to groups have, like all worlds, an intrinsic value. And Ross, inso-
far as he accepts that intrinsic value is relevant to the determination of
obligation and hence counts as a source of deontic value, will also accept
the presupposition, at least in this regard.
There is a twist to this second presupposition. If Chart 9.1 is to be used
in the determination of Vincent's and Virgil's individual obligations, it must
be that the numbers that serve to indicate the relative desirability of out-
comes do so in an agent-neutral way. That is, the unanimous election of
Smith must be seen to be superior relative to all perspectives — to Vincent's,
to Virgil's, and to {Vincent, Virgil}'s - and thus the agent-relativization
that is built into (I) must be seen to be insignificant.
By conjoining (9.3) with the second presupposition, we may assert the
following:
(9.4) if a world of deontic value Vis accessible to some proper subgroup
of { St, ..., Sn} from W at T, then a world of deontic value Fis acces-
sible to {Su ..., Sn} from W at T.
That is, whatever deontic good (or bad) a subgroup can achieve, the group
can achieve also. But, as with (9.2), the converse does not hold. That is, we
should reject this:
(9.5) if a world of deontic value Vis accessible to {S2, ..., Sn} from W at
T, then a world of deontic value Vis accessible to some proper sub-
group, if any, of {Su ...,£„} from H^at T.
The problem with (9.1) that I am about to expose has, therefore, only
limited application; it applies only to those theories that satisfy the form of
(I) and subscribe to both the presuppositions noted. This is not to say that
theories that avoid this problem are superior. First, if they avoid it by fail-
ing to satisfy the form of (I), they are inferior; for (I), as I have argued
256
throughout, would (so far) appear to be the form of any acceptable theory
of obligation. Second, if they avoid it by avoiding the presuppositions in
question, they face a problem of their own. The first presupposition is sure-
ly undeniable, and denial of the second comes at considerable cost.4 Third,
the problem is soluble, as I shall seek to show.
Having made note of these presuppositions, let us ask again: given (9.1),
what should Vincent and Virgil do? We cannot yet say; more information
is needed. Can Vincent act in such a way that Virgil will vote for Smith? If
so, (9.1) directs Vincent to act in that way and to vote for Smith too.
Similarly, mutatis mutandis, for Virgil. Thus there is a scenario in which (9.1)
is satisfied and in which the best possible outcome (ten units of deontic
value) is achieved.5 This of course poses no problem for (9.1). Nor does this
scenario: Vincent can act in such a way that Virgil will vote for Smith (so
that, we may say, Virgil is not intransigent relative to Vincent concerning
Smith's being unanimously elected), but Virgil cannot act in such a way that
Vincent will vote for Smith (because Vincent is intransigent relative to
Virgil concerning Smith's being unanimously elected; Vincent insists on
voting for Jones). Here (9.1) directs Vincent to act in such a way that Virgil
will vote for Smith and also directs Vincent himself to vote for Smith; but
it directs Virgil to vote for Jones (since this is the best he can do, given
Vincent's intransigence). It may at first seem odd that (9.1) should direct an
agent to perform an action which, if performed, guarantees that less than
the best possible outcome will be achieved (in the sense that ten units of
deontic value will be forgone in favor of five), but on reflection this is sure-
ly right. After all, it's not Virgil's fault that voting for Jones is the best he
can do, it's Vincent's; it is Vincent who does not do the best he can.
But now notice this. There is a sense in which (9.1) seems to allow two
wrongs to make a right. Suppose that Virgil is just as intransigently in favor
of Jones as Vincent is. Then not only does (9.1) direct Virgil to vote for
Jones (since this is the best that he can do under the circumstances), it directs
Vincent to do this too (since this is now also the best that he can do). Thus,
given (9.1), neither Vincent nor Virgil does wrong in voting for Jones, even
though this yields only five units of deontic value rather than ten; even
though, that is, the best that both could do is not done. Somehow, Vincent's
intransigently voting for Jones is transformed from being wrong to being
right when it is balanced by a matching intransigence on Virgil's part. This
4 The cost in question is this: they are subject to analogues of the Prisoner's Dilemma and
are thus "directly collectively self-defeating," as Parfit puts it. See Parfit (1984), pp. 54 and
95ff. Cf. Regan (1983), p. 108ff.
5 Cf. Regan (1980), pp. 52-5.

257
implication of (9.1) should surely trouble anybody inclined to the view that
one ought to do the best one can, inasmuch as it demonstrates that there is
a sense in which universal satisfaction of (9.1) is compatible with the best
that can be done not being done.

9.2 NONSOLUTIONS
9.2.1 Unrequited cooperation
It may seem that the problem raised in the last section has brought to light
a limitation to (9.1) that is easily remedied. (9.1) was drawn up with the
individual alone in mind; it ignored the possibility of cooperation. The
problem brings this out starkly: through a mutual failure to cooperate,
Vincent and Virgil, though doing the best they can individually,6 fail to do
the best they can jointly. To ensure that they do the best they can jointly,
each need only do his part in the mutually cooperative action open to both.
Each, that is, need only vote for Smith and the best possible outcome will
be assured.
This suggests the following view:
(9.6) for any person S and times T and T\ S ought at T to do at T" what
he would do at T" if his group did at T' the best that it could at T
do at T ' (i.e., S ought to "do his part").7
Apart from the fact that (9.6) implies that the best is done if everyone acts
according to it, it has what may seem to be another attractive feature.
Notwithstanding the discussion of supererogation in Chapter 8, it might
still be thought that theories of the sort captured by (9.1) are too demand-
ing, that they require of each of us that we exhaust ourselves in the pursuit
of the good (for example, by giving and giving to charity), since many
others will in fact not pursue the good and we cannot get them to do so
(that is, they are intransigent relative to us concerning the pursuit of the
good) — although, if everyone were to pursue the good, none of us would
have to extend himself all that much (if everyone were to give to charity,
none of us would have to give very much). (9.6), on the other hand, seems
6 Here and henceforth, when I say that two (or more) people do the best they can individ-
ually, I mean simply that each does the best he can in fact do under the circumstances. I
don't mean that each does the best he can do "all on his own," i.e., exclusive of any coop-
eration from others. And so, if full cooperation is in fact forthcoming, the best that two
(or more) people can do individually will in fact be to do whatever it is that they would
do if they were to do the best they could jointly. At least, this is so given (9.4) above.
7 Cf. Regan (1980), p. 85, on what Regan calls "COP."

258
to imply that we each ought to pursue the good only to that extent which
would be required of us under the counterfactual condition where every-
one does his part. And this might appear to be just what should be said.
Why, after all, should one person be obligated to make up for the moral
failures of another?
But this won't do. (9.6) faces a number of problems.
First, a relatively minor problem: (9.6) overlooks the fact that a person
may belong to several groups at once.8
Second, a far more serious problem: (9.6) violates the thesis that "ought"
implies "can." This is illustrated by any case in which one person's doing
his part depends on another's doing his, and where the latter is intransigent
relative to the former concerning the group's generation of the best possi-
ble outcome. For example, my part in generating the best possible outcome
may be to enter a room whose door only you can unlock. If you adamant-
ly refuse to do your part, then I cannot do mine.
Third, whether or not the discussion of supererogation in Chapter 8 is
acknowledged to have dealt successfully with the apparent overdemand-
ingness of (9.1), certainly (9.6) doesn't deal with it. The claim, unqualified,
that no one should be said to be obligated to make up for the moral failures
of another is surely far too sweeping. Just consider a case where all that is
needed to prevent a child from dying is the donation of one dollar. There
are one hundred people present who can each give one cent; you can give
one dollar without any hardship whatsoever; none of the others gives
money to the child; and you give one cent, smugly assuring yourself that,
in doing your part, you've done your duty. The child dies, of course.
Similarly, with reference to the voting case, surely it is in (9.1)'s favor that
it directs Virgil to vote for Jones, given Vincent's intransigence; for had he
voted for Smith, zero rather than five units of deontic value would have
been achieved. Such unrequited cooperation does no good at all.10

9.2.2 Procedural cooperation

What it seems we want, then, is a theory that will direct Virgil to vote for
Jones, given Vincent's intransigence, but which will not permit a similar
intransigence on Virgil's part. Donald Regan has produced a very elabo-

8 Cf. Feldman (1986), p. 153.

9 Cf. Feldman (1986), p. 165.
10 Cf. Humberstone (1983), p. 29. "Cooperation" ought perhaps to be put in quotes here,
since it must fail. (Cf. Regan, 1980, p. 126ff., on what goes into complete cooperation.)

259
rate theory designed to meet this goal. Cutting through the elaboration, we
may give this rough version of Regan's proposal:

(9.7) for any person 5 and times T and T", S ought at T to:
(a) hold himself ready at T (and up to T') to cooperate with others at T';
(b) identify prior to T ' other would-be cooperators at T'; and
(c) do his part at T ' in the best action that the group of actual cooper-
ators can at T do at T'.n

This is an attractive proposal. If everyone follows the prescribed procedure,

the best that can be done by all will be done; anyone not prepared to coop-
erate does wrong; anyone prepared to cooperate but frustrated by other
noncooperators, who nonetheless does his best under these less than ideal
circumstances, does what he ought. Doesn't this capture just what we want
to capture?
Unfortunately not. There are two principal objections to be made, based
on the fact that the proposal prescribes a certain procedure. 12 First, Regan's
theory violates the thesis that "ought" implies "can": S may be unable to
do what (9.7) requires. (In particular, S may be unable to identify other
would-be cooperators.) Second, Regan's theory ignores the possible deon-
tic cost involved in undertaking the specified procedure. 13 It could be that
identifying other would-be cooperators is itself a very costly activity, and
to overlook this is surely inconsistent with the view that one ought to do
the best one can.

9.2.3 Group obligation

A proposal made prior to Regan's has been resurrected by several others in
response to the difficulties that Regan's faces.14 It is, in effect, that (9.1) be
generalized thus:
(9.8) for any persons St...Sn and times T and T", {Su ..., Sn} ought at T
to do at T ' the best that {Si9 ..., Sn} can at T do at T'. 15
11 Regan (1980), pp. 135-6.
12 See Barley (1984), pp. 157-9; Conee (1983), pp. 422-3; and Feldman (1986), pp. 171-
3. Regan responds to such criticism (in Regan, 1980, p. 169ff.), but the response seems
unsatisfactory.
13 Regan deliberately ignores this (Regan, 1980, p. 176ff.).
14 Originally: Postow (1977), p. 51 (and see also McKinsey, 1981, p. 317). Later: Conee
(1983), pp. 422-3; Parfit (1984), p. 73; Feldman (1986), p. 161; Jackson (1987), p. lOOff.
15 As noted in Subsection 2.1.3, "at T'" is supposed to cover "throughout T'," "during
T'," and so on; for it is clear that the individual components of a group action may be
temporally scattered.

260
(Where n equals 1, (9.8) is equivalent to (9.1), insofar as we are treating each
individual S{ as {S,}.) This proposal takes advantage of the presuppositions
that give rise to the problem in the first place. If we acknowledge that
groups can sometimes achieve outcomes that individuals alone cannot, and
if we acknowledge that these outcomes are themselves of deontic value,
then why not accept that groups can be the agents of right and wrong and
can thus have obligations of their own? If we do accept this and accept (9.8)
as the correct account of it, we can then point to a certain lack of corre-
spondence between individual and group obligation. For example, given
the falsity of (9.5), we must reject the following:

(9.9) if{S ? ,..., Sn} ought at T to do A at T", then some proper subgroup,
if any, of { Si9 ..., Sn} ought at T to do A at T.
For the best achievable by the group may not be achievable by any proper
subgroup, as in the case of Vincent and Virgil. And we must also reject the
following for the same reason:

(9.10) if some proper subgroup of {Sl9 ..., Sn} ought at T to do A at T",

then {Sl9 ..., Sn} ought at T to do A at T'.
In addition, while we must of course accept that, where there is full coop-
eration - that is, everyone does his part in the best action performable by the
group - then neither the group nor any subgroup does wrong, we should
not accept that, if all its members act as they ought, then neither the group
nor any subgroup does wrong. That is, we must accept the following:
(9.11) i f S j d o e s ^ a t T\ and...and Sn does An at Ttt, and if {Si9 ..., Sn}
thereby does at T' the best that {Su ..., Sn} can at T do at T', then
it is not the case that {Sl9 ..., Sn} or any subgroup of {Su ..., Sn}
goes wrong at or after T in so acting at T';
but, as the present case of Vincent and Virgil shows, we must reject the fol-
lowing:
(9.12) if 5^ ought at T to do At at T\ and.. .and Sn ought at T to do An at
T'n9 and if St does At at T\ and.. .and Sn does An at T'n9 then it is
not the case that {Sl9 ..., Sn} or any subgroup of {Sh ..., Sn} goes
wrong at or after T in so acting at T".
It is on the falsity of (9.12) in particular that proponents of (9.8) have wished
to capitalize in the attempt to solve the problem before us. Their sugges-
tion is this:16 it is not that, where Vincent's intransigence is matched by
16 See Feldman (1986), p. 177; Jackson (1987), p. 100.

261
Virgil's, two wrongs have made a right. On the contrary, although (9.8) of
course yields the same story about obligation and wrongdoing in the indi-
vidual case as (9.1) does — each of Vincent and Virgil ought to vote for Jones
and hence does no wrong in doing so — unlike (9.1), (9.8) still diagnoses a
wrongdoing in this case: Vincent and Virgil jointly do wrong in not elect-
ing Smith. Thus we need not accept the disturbing and unsatisfactory con-
tention that nothing goes wrong in the election ofJones.
This is an intriguing and elegant proposal, but I think it must be reject-
ed. First, it is not clear that groups of more than one member can have
obligations. I have acknowledged that, if the problem with which we're
concerned is genuine, then wre must grant the presupposition that groups
can achieve outcomes that have a certain deontic value. But this does not
mean that we must accept that groups can have obligations. When I pre-
sented (I) in Chapter 2,1 noted17 that it might be that we should say that an
agent, to be the subject of obligations, must be a moral agent; if not, we may
have to accept that nonhuman animals have obligations (for they, too, may
be able to achieve outcomes that have a certain deontic value). What is
involved in an agent's being a moral agent is unclear, but arguably it must
have the capacity to conceive of its action in moral terms,18 and it seems
safe to say that no group with more than one member (as opposed to indi-
viduals in the group) has this capacity. Furthermore, some have contend-
ed that, for something to be obligatory, it must be something about which
decisions can be made, and that joint outcomes achievable by groups with
more than one member (as opposed to the individual component outcomes
achievable by individual members of the group) are not entities of this
sort.19 If either of these points is correct, then, despite the fact (if it is a fact)
that groups with more than one member can achieve an outcome that has
deontic value and, often, has a degree of deontic value that no outcome
achievable by some proper subgroup has, they cannot be obligated to
achieve this outcome.
My rejection of the proposal is not based on these points, however.
Indeed, I am prepared to accept (9.8) as an interesting and fruitful exten-
sion of (9.1) and to accept, therefore, that Vincent and Virgil jointly do
wrong when they unanimously elect Jones. What I wish to deny is that this
is all that goes wrong in the election ofJones. The problem is simply this:
on the present proposal, the group wrongdoing is not attributable to any of
the members of the group; neither Vincent nor Virgil can be faulted for his
17 In note 9 to Ch. 2.
18 Cf. Thomson (1990), p. 215.
19 Cf. Smith (1986), p. 342; McConnell (1988a), p. 27.

262
intransigence.20 Suppose that Vincent and Virgil are friends. Virgil sees that
Vincent is intransigently in favor ofJones; not wanting to have his friend
do wrong, Virgil decides to be intransigently in favor ofJones too, believ-
ing that Vincent will thereby be absolved. (He resolves not to listen to
entreaties even from Vincent himself; he believes that Vincent, being his
friend but subscribing to an account of obligation other than (9.8), may
attempt to dissuade him from acting in this way.) Now, if Virgil acts in this
way, (9.8) - as Virgil recognizes - implies only that Vincent and Virgil
jointly do wrong and that neither does wrong singly. But this is too much
to swallow; the scenario still smacks of two (individual) wrongs making a
right.
The proponent of (9.8) might reply in this way. What's wrong with
intransigence, when it's met with matching intransigence? Given Vincent's
intransigence, Virgil's intransigence makes no difference; Vincent's intran-
sigence alone rules out attainment of the group's best. So too, mutatis mutan-
dis, for Vincent's intransigence making no difference. Thus, under the
circumstances, there is nothing wrong with the attitude of either.21
But there is something wrong with this attitude. True, there is a sort of
overdetermination here, such that each agent's being a party to achieving
less than the best outcome in a way renders the other's being a party to it
redundant. Nonetheless, each is a. party to it, by virtue of his intransigence;
each contributes to the failure to do what is best; each does wrong. Neither
(9.1) nor (9.8) appears adequate to this fact, and the problem remains.

9.2.4 Counterfactual cooperation

The solution, I believe, is in outline simply (and unsurprisingly) this: one's
moral obligation is to do the best one can while avoiding intransigence; that
is (to coin a term), one ought transigently to do the best one can. Doing the
best one can must be accompanied by the adoption of a certain attitude. This
is in essence Regan's proposal shorn of its objectionable aspect: one must
do the best one can (see clause (c) of (9.7)) while holding oneself ready to
cooperate (clause (a)), but there is no obligation to undertake something

20 Here "faulting" someone means ascribing wrongdoing to that person. Thus "attribut-
ing" a group wrongdoing to an individual involves finding that that individual has done
wrong with respect to the group outcome in question. It is natural to use the phrase "S
is responsible for X" instead of the phrase "Xis attributable to S," but one should be wary
of this. At least, one should not take wrongdoing itself to suffice for culpability. Cf.
Zimmerman (1988a), pp. 41-3.
21 Cf. Feldman (1986), p. 177.

263
(identification of cooperators — clause (b)) that may prove either impossi-
ble or overly costly.
But what is it to "hold oneself ready to cooperate"? One answer is this:
to hold oneself ready (that is, to be prepared) to cooperate is simply to be
such that one would do one's part if all others did theirs. When conjoined
with the injunction to do one's best, this yields the following thesis:

(9.13) for any person S and times T and T', S ought at T to:
(a) do at T' the best that he can at T do at T'; and
(b) be at T such that he would do his part at T' if all others did theirs.

One problem with this proposal is that it is too narrow. It overlooks the
hypothetical case where not all others cooperate but where a sufficient
number do so and their doing so would generate an outcome superior to
that generated by 5 if S were to do the best he could in the actual situation.
Surely, if we wish to insist on transigence, we wish to have 5 prepared to
cooperate in the generation of any superior outcome, not just the best pos-
sible, so long as this outcome would in fact be the best available (given the
behavior of any noncooperators). Suppose, for example, that the best that
Bob, Carol, and Ted can jointly do is to raise and push aside a tree that has
fallen on Alice. Suppose also that, if Bob refuses to cooperate, the best that
Carol and Ted can jointly do is to raise the tree so that Alice can crawl out
from under it. Suppose finally that, if both Bob and Carol refuse to coop-
erate, the best that Ted can do alone is go for help. What we want is for
Ted to be prepared to help raise and push aside the tree if both Bob and
Carol are prepared to cooperate, and (9.13) gives us that. But what we also
want, and what (9.13) does not give us, is for Ted to be prepared to help
raise the tree if Carol is prepared to cooperate but Bob is not.
One way to try to provide what's missing is to amend (9.13) as follows:

(9.13') for any person S and times T and T', 5 ought at T to:
(a) do at T' the best that he can at T do at T'; and
(b) be at T such that he would do at T" the best that he could at T do
at T' no matter what the circumstances might be at T'. 22

There are of course difficulties of interpretation here, but notice how

(9.13') would deal with the problem afflicting (9.13). Suppose that neither
Bob nor Carol is prepared to cooperate with Ted. Then (9.13') is designed
to direct Ted to (a) go for help but also (b) be prepared to cooperate with
both Bob and Carol if they are prepared to cooperate (something that (9.13)
22 Cf. Schwartz (1985), p. 85, for a similar proposal concerning compliance with a policy.

264
also says) and to be prepared to cooperate just with Carol if Bob is not pre-
pared to cooperate (something that (9.13) doesn't say).
Nonetheless (9.13') is problematic. The requirement of counterfactual
cooperation stipulated in clause (b) is too strong. Suppose that Fred were
force-fed some mind-altering drug. Would he do the best he could under
these circumstances? Quite conceivably not (especially if the drug altered
his moral perception). Would this show that he was in fact unprepared to
cooperate? It would not.
A second problem with (9.13') is that it appears not to conform with the
thesis that "ought" implies "can." While it is of course always in one's
power to do the best one can, is it always in one's power to be such that
one would do the best one could if circumstances were different? Fred
appears to provide a counterexample. Let us assume that, having ingested
the drug, he would not (although of course he could) do the best that he
could do. There seems to be no reason to assume that whether or not this
is true of him is presently up to him.

9.2.5 Mere openness

One attractive suggestion that does conform with the thesis that "ought"
implies "can" is to view transigence as openness, where this is understood
as follows:
(9.14) S is open at T to G concerningp at T ' iff
(a) {S, G} can at T so act at T ' thatp is true; and
(b) G can at T so act at T' that p is true.
(Here 5 is distinct from G, and G is a group of persons, possibly with just
one member.) W e might then say:

(9.15) for any person 5 and times T and T\ S ought at T to:

(a) do at T' the best that he can at T do at T'; and
(b) for any group of persons G and proposition p, be open at T to G
concerning p at T" iff {5, G} so acting at T ' that p is true is the
best that {5, G} can at T do at T'.

The reason why this conforms with the thesis that "ought" implies "can"
is this. If {S, G} can so act thatp is true, then all that S has to do to see to it
that G can so act that p is true is do his part in the group action in question.
N o w it may be (as noted before in the discussion ofyou, me, and the locked
door) that S cannot do his part because G prevents him from doing so;
nonetheless G needn't do this and so can act in such a way that S does his

265
part - as long, that is, as S doesn't prevent G from so acting. Each of 5 and
G can, then, perform his or its respective part if the other does; otherwise
it would not be the case that they could jointly perform the action in ques-
tion. Thus each of S and G can be open to the other concerning p; each of
them can so act that the other can so act that p is true. If either S or G
cannot so act that p is true, it is because the other has avoidably prevented
this.
I think that we should accept that transigence is openness, as this is
defined in (9.14). (9.15) might therefore appear to give expression to the
solution to which I have alluded. Indeed, (9.15) is notably superior to any
of the other proposals discussed. For example, it avoids the severity of
(9.13') in that it requires simply that others can so act that 5 cooperates and
not that S would cooperate no matter what the circumstances. Like (9.13'),
though, (9.15) aims to satisfy, as (9.13) did not, the requirement that S's
preparedness to cooperate be general.
The fact is, however, that (9.15) faces serious problems; the claim that
one ought transigently to do the best one can is only roughly accurate.
First, (9.15) on occasion issues in a basic moral dilemma, namely, that 5
remain transigent while performing an action that precludes such transi-
gence. Suppose, for example, that it is time for Vincent to vote, that Virgil
has not yet voted, but that Virgil remains intransigently in favor ofJones.
Then clause (a) of (9.15) requires that Vincent now vote for Jones. In doing
this, Vincent cannot of course remain transigent concerning Smith's unan-
imous election; yet clause (b) requires that he still be transigent in this
respect, since it is still possible for {Vincent, Virgil} to so act that Smith is
unanimously elected, and this can be achieved only by Vincent's voting for
Smith.
It may seem that the obvious solution to this first problem is simply to
stipulate in clause (b) that T is earlier than T", to stipulate, that is, that tran-
sigence concerning the best outcome achievable by the group is required
only up until the time at which 5 must act in order to do what is in fact the
best that he can do. But a second problem is that, even with this amend-
ment, (9.15) would still on occasion issue in a basic moral dilemma. This is
because it can sometimes happen that the only way in which 5 can be open
to G concerning some group-achievable outcome is by doing his part in
the achievement, or what would be the achievement, of that outcome.
Suppose, for example, that it is not yet time to vote, that both Vincent and
Virgil are going to vote for Jones, and that each is closed (that is, not open)
to the other concerning Smith's being unanimously elected; but suppose
now that this mutual intransigence is due simply to the fact that Vincent
266
and Virgil live at different ends of the country and cannot interact. Here it
is the mere fact that Vincent is going to vote for Jones that makes it the case
that he is closed to Virgil concerning Smith's being unanimously elected;
so too, mutatis mutandis, for Virgil's being closed to Vincent. Their being
mutually intransigent in this regard is due, therefore, not to any orneriness
on anyone's part, but simply to their mutual disconnectedness. Their
inability to interact renders each inflexible to, unmanipulable by, the other;
neither can affect the way the other votes, and hence, however the one
votes, it is perforce afait accompli for the other. Now, (9.15) condemns even
such closedness, but it ought not to. For, under the circumstances, even if
T is earlier than T", the only way in which Vincent can at T be open to
Virgil concerning Smith's being unanimously elected at T" is by actually
voting for Smith at T". Thus clause (b) of (9.15) implies that Vincent ought
at T to vote for Smith at T', even though, given the fact that Virgil will
vote for Jones, clause (a) implies that Vincent ought at T not to vote for
Smith at T" but to vote for Jones instead.
Perhaps this result should not surprise us. If Vincent and Virgil are dis-
connected in this way, then, even though there is a sense in which they can
jointly, or as a group, achieve Smith's unanimous election, such action
hardly seems paradigmatic of what we normally think of as joint, or group,
action. It is certainly not concerted action, action that is undertaken by a
group of interacting individuals. For this reason, even if we accept (in the
spirit of (9.8)) that Vincent and Virgil have, as a group, done wrong, there
would seem to be no reason to think of such wrongdoing as attributable
even in part to any member(s) of the group. If anything, it is "attributable
to nature."23
A third problem with (9.15) is that it requires that S be open to any
group's best group-achievable outcome. This is surely too sweeping; once
again, it can issue in a basic dilemma. For it can happen that what S must
do to remain open to one group's best achievable outcome conflicts with
what he must do to remain open to another group's best achievable out-
come. For example, it could happen that Bob's and Carol's and Ted's joint-
ly doing their best involves Ted's taking ten paces to the right, while Carol's
and Ted's jointly doing their best involves Ted's taking ten paces to the left.
A final problem with (9.15) is that, like (9.7), it ignores the possible cost-
liness of doing what it requires. Even openness can be costly, as when (per-
haps - it depends on just what deontic value consists in) S's openness to G,
having become public knowledge, prompts someone to perform a harm-
ful act that would otherwise not have been performed.
23 This is an ironic use of "attributable." Nature can do no wrong.
267
 9.3 T H E S O L U T I O N : U N I N T R U S I V E T R A N S I G E N C E
All four problems with (9.15) can, I believe, be handled by a simple stipu-
lation. The requirement that S be open to G must be conditional on such
openness not preventing S from doing what is in fact the best that he can
do under the circumstances. One ought transigently to do one's best, but
the transigence must be unintrusive.
I propose, therefore, this revision of (9.1) as a way of resolving the prob-
lem concerning cooperation that besets that proposition.

(9.1/) for any person 5 and times T and T", S ought at T to:
(a) do at T 7 the best that he can at T do at T7; and
(b) for any group of persons G and proposition p, be open at T to G
concerning p at T ' iff
(1) {S, G} so acting at T ' thatp is true is the best that {5, G} can
at T do at T', and
(2) for any act A' and time T " such that T 7/ is no earlier than T
and no later than T 7 , \£A' is the best that S can at T do at T " ,
then S can at T both be open at T to G concerning p at T ' and
do ,4'at T "

Roughly: clause (a) requires that 5 do the best he can; and subclause (bl)
requires that he do it transigently, while subclause (b2) requires that this
transigence be unintrusive.
Just as (9.1) only gives rough expression to (I), so (9.1') only gives rough
expression to what I take to be the proper revision of (I). But before I set
out this revision more precisely, let me address certain issues. In particular,
let me explain how it is that the requirement that transigence be unintru-
sive is designed to solve the four difficulties that (9.15) faces, and then let
me indicate certain other important features of (9.17).
In stipulating that transigence is required only when it is unintrusive,
(9.1') clearly avoids the possibility that 5 can meet his requirement to do
what is best only if he fails to meet his requirement to be transigent. Hence
the sort of basic dilemma outlined in the first problem with (9.15) cannot
arise on (9.17). Nor can the second sort of basic dilemma outlined in the
second problem. If the best that Vincent can in fact do is vote for Jones, and
the only way in which he can be transigent concerning Smith's unanimous
election is to vote for Smith, then he is not required to be transigent in this
respect. Hence (9.17) requires transigence only where S is connected to G,
that is, only where whether or not S does his part in contributing to the
best achievable by { S, G} is not of necessity zfait accompli for G. Where there

268
is connectedness, then, S can (although he might, of course, refuse to do
so) leave it up to G whether or not S does his part, and it is in virtue of this
fact that S ought to be open to G concerning what is the best achievable by
{ S, G}. As an illustration, consider Vincent and Virgil once again, but now
think of them as mutually connected. Vincent's being connected to Virgil
concerning Smith's being unanimously elected consists in this: Vincent can
act in such a way that it is left up to Virgil whether both of them vote for
Smith. (How? This will of course depend on other details of the case. (9.1')
itself includes no stipulation as to just how this might be achieved. Perhaps
Vincent can simply inform Virgil — truthfully — that he will vote as Virgil
wishes.) That Vincent can act in this way is, of course, no guarantee that he
will. While disconnectedness ensures lack of influence, connectedness
doesn't ensure influence. If Vincent opts for intransigence concerning
Smith's being unanimously elected, then, even though he is connected to
Virgil in this respect, Virgil cannot influence Vincent in this regard, and so
Virgil's access to the best that is jointly achievable is effectively obstructed
by Vincent. This is Vincent's fault; because of his being connected to Virgil,
his intransigence is not "attributable to nature."
As to the third problem with (9.15): the best that anyone can ever do
(given, as (9.4) implies, that an individual can never achieve a good that is
not achievable by some group of which he is a member, and also, as the fal-
sity of (9.5) implies, that a group can sometimes achieve a good that no
member can) is to do his part in the best group action that those who can
in fact be moved to cooperate can perform. If S's holding himself ready to
cooperate in another group action in fact conflicts with his holding himself
ready to cooperate in this group action, then it conflicts with his doing what
is in fact best, and (9.17) therefore implies that he ought not to hold himself
ready to cooperate in the other group action.
Finally, if S's holding himself ready to cooperate is unduly costly, then
it prevents his doing the best he can, and, if that is the case, then once again
(9.1') implies that he ought not to hold himself ready to cooperate.
Let me now indicate certain other important features of (9.1').
First, (9.17) of course conforms with the thesis that "ought" implies
"can." One can always do the best one can; one can (as noted in the dis-
cussion of (9.15)) always be transigent; when one cannot achieve both,
clause (b2) kicks in and says that the requirement of transigence is nullified.
Still, it is important to note that (9.1') differs from (9.1) in this respect,
namely, that, whereas the latter implies that it is only what one does that
counts (with respect to acting as one ought), the former implies that how
one does it counts as well. In this way, (9.1') appears to accord with Regan's
269
claim that a satisfactory solution to our problem must involve a theory that
is not "exclusively act-oriented," where (it seems) a theory is exclusively
act-oriented just in case one's satisfying it depends solely on whether (and
not on how) one performs some act.24 Sometimes this difference between
(9.1) and (9.1') will issue in a difference in what action is required. This is
so in the case where at least one of Vincent and Virgil can both do his best
and be transigent concerning Smith's unanimous election. In this case,
whereas (9.1) implies that both Vincent and Virgil do as they ought if they
both intransigently vote for Jones, (9.1') implies that both do as they ought
only if they both transigently vote for Smith. For, suppose that Vincent can
both do his best and be transigent concerning Smith's being unanimously
elected. Then (9.1') implies that he ought to be transigent in this respect,
and, if he is, then Virgil can so act that Smith is unanimously elected, and
so Virgil would not do his best by voting for Jones; and the same holds,
mutatis mutandis, for Virgil's being transigent in this respect.
Nonetheless, and this is the second important feature of (9.1'), if neither
Vincent nor Virgil can both do his best and be transigent concerning
Smith's being unanimously elected, then (9.1') does not differ from (9.1) in
its implication that each of them ought to vote for Jones. This is so even
though they would and could jointly do their best by unanimously electing
Smith. Consider, for instance, the case where Vincent and Virgil are mutu-
ally disconnected with respect to Smith's being unanimously elected. Here
the group does not do its best (and, if groups of more than one member can
be obligated, the group does wrong), but no individual does wrong. But
here, I submit, this result is not anomalous, precisely because of the dis-
connectedness. It is when transigence can be unintrusive that a group's fail-
ure to achieve its best is attributable to some member's doing wrong;
otherwise it is "attributable to nature" and must be accepted as such.
Thirdly, if everyone other than S does his part in producing the best that
is jointly achievable, then (9.1') certainly implies that S does no wrong if
he does his part too, for he can do no better (given (9.4)). But does (9.1')
imply that 5 ought to do his part? If we construe "do his part" liberally, so
that an agent can on occasion do his part simply by letting others get on
with what they're doing, the answer is yes. But it is not the case that S must

24 Regan (1980), 113-15. This description of (9.1') may be misleading, however. The atti-
tude of (in) transigence is always within the agent's control — it is an optional attitude — and
clearly its adoption will sometimes involve action (as when Vincent tells Virgil that he will
vote as Virgil wishes); and if, as I suspect (and as noted in Subsection 2.2.1), nothing is in
one's control except by way of a decision, and if a decision can be said to constitute a min-
imal type of action, then in fact adopting (in) transigence will always involve action.

270
always actively contribute to the achievement of the best group-achievable
outcome. There are three sorts of cases where his remaining idle in this
regard is permissible. First, it may be that he cannot actively contribute to
the outcome. Second, it may be that, although he can actively contribute
to some outcome, he ought nonetheless to remain idle, because only in this
way can the group achieve its best. (If Rachel is a superior swimmer to
Roberta and Roberta's joining in the rescue of a drowning person would
only mess things up, then Roberta ought to stand by while Rachel carries
out the rescue.) Third, it may be that S is permitted a choice as to whether
to remain idle, because there are more than enough people actively con-
tributing to the group-best and his active contribution, though not coun-
terproductive, would not help either. In this last sort of case, if S does not
actively contribute to the outcome, he gets a kind of free ride; the best is
achieved without any effort on his part. Such free riding seems to me unob-
jectionable, though, if accompanied by the transigence that (9.1/) requires.
Fourthly, the illustrations used so far (of Vincent and Virgil; of Bob and
Carol and Ted and Alice; of Rachel and Roberta) might be most natural-
ly understood to concern group actions whose individual components
occur simultaneously, but such simultaneity is not essential to the illustra-
tions. Indeed, I have already mentioned the possibility that Vincent's vote
precede Virgil's, and it may be that Bob and Carol and Ted must perform
actions that occur at (slightly) different times in order to save Alice. Now,
it might seem that the temporal order of the individual component actions
in a group action can make a difference as to just what the members' indi-
vidual obligations are - that whether or not to participate is optional for the
earlier links in the chain in a way in which it is not for the last link; for by
the time the last link has a chance to contribute, all the others have con-
tributed, their participation is settled, and so it is now clearly up to him
alone whether the outcome is achieved. But in fact to single out the last
link in this way would be a mistake. For each link there is the question of
whether he ought to participate, and the answer is contingent on others'
actions, whether or not these actions have already taken place. Thus, where
ten are needed to deliver some vital message, what the first link, Archie,
ought to do is to be determined in just the same way as what the last link,
Jack, ought to do. If all nine of Archie, Barry,..., and Ian have cooperated,
then Jack ought to pass the message along too. Similarly, if all nine of Barry,
Chip,..., and Jack will cooperate, then Archie ought to pass the message
along too. On the other hand, if at least one of Archie, Barry,..., and Ian
has not cooperated, then it is not the case that Jack ought to cooperate (for
he cannot; he has no message to pass on). Similarly, if one of Barry,

271
Chip,..., and Jack will not cooperate, then (presumably) it is not the case
that Archie ought to cooperate (for his effort will be wasted; he could pre-
sumably do something better — unless, of course, he could prevail upon the
noncooperators to cooperate). In each instance, then, the timing of others'
actions is irrelevant to the determination of an agent's individual obliga-
tion, except insofar as such timing renders the others insusceptible to the
agent's influence. 5
In summary, then, (9.1') condemns, as (9.1) does not, one's obstructing
another's access to the best possible outcome, when avoidance of such
obstruction is compatible with one's doing one's best. In this way, the
problem facing (9.1) is solved. However, (9.17) only gives rough expres-
sion to the revision of (I) that I think is called for. At this point I offer the
following more precise account.
Recall (I). It says this:

(I) 5 ought, at T in W, to do A at time T ' iff

(a) there is a world W' such that W' is accessible to 5 from W atT and 5
does v4 at T ' i n Wf;
(b) there is a world W" such that W" is accessible to S from W at T and
S does not do A at T ' in W"; and
(c) for all worlds W" such that W" is accessible to S from W' at T and S
does not do A at T' in W'\ there is a world W' such that
(1) W' is accessible to S from W at T,
(2) S does A at T ' in W\
(3) the deontic value for 5 at T of W* is greater than the deontic value
for S at T of W", and
(4) there is no world W'" such that
(i) W is accessible to S from W at T,
(ii) 5 does not do A at T' in W"\ and
(iii) the deontic value for 5 at T of W"' is greater than the deon-
tic value for S at T of W'.
Let us say that any act and time that meet the conditions laid out in (I) for
A and T ' satisfy (I). Then, I think, the revision of (I) that is adumbrated in
(9.1') may be put this way:

(I") 5 ought, at T in W, to act as follows:

(a) for any act A and time T', do A at T ' iff A and T ' satisfy (I); and
(b) for any act A and time T' that satisfy (I), and for any group of persons
G and proposition^?, be open at T to G concerning p at T' iff

25 Cf. Lyons (1965), pp. 79-80; Parfit (1984), p. 30.

272
 (1) for all worlds W" such that W" is accessible to {S, G} from W
at T and S does A at T' in W", there is a world W' such that
(i) W' is accessible to {S, G} from W at T,
(ii) {5, G} so acts at T ' in W' that p is true,
(iii) the deontic value26 of W' is greater than the deontic value
of W\ and
(iv) there is no world W'" such that
(a) W'" is accessible to {S, G} from W at T,
(j8) 5 does ,4 at T ' in W'", and
(y) the deontic value of W'" is greater than the deontic
value of W'\ and
(2) for any act A' and time T " such that T " is no earlier than T and
no later than T', if ^4' and T " satisfy (I), then there is a world W'
such that
(i) W' is accessible to S from ^ at T,
(ii) S does ,4' at T " in P^', and
(iii) S is open at T in W' to G concerning p at T'.
Similar revision is also required for (II)—(XII), but I shall show mercy and
forgo the details. Furthermore, if a group of more than one individual can
be obligated to do something, as (9.8) contends, then (I") can be extended
to account for this, in just the same manner as (9.1) is transformed into (9.8);
but, again, I shall forgo the details.
I shall end with two observations. The first concerns the puzzle posed
by Bernard Williams and first presented in Section 1.2. It concerns these
three statements:
(1.1) someone ought to help that old lady:
(1.2) Jones is the only person who can help her;
(1.3) Jones ought to help her.
Williams contends that (1.1) does not express an obligation but that (1.3)
does, and yet that (1.3) follows from (1.1) and (1.2). I disputed this. I wish
now to point to a possible reading of (1.1) which Williams doesn't consi-
der and which may account for the inclination, if there is one, to accept the
inference in question. It is this: as Williams says, (1.1) does not express an
individual obligation, but it does express a group obligation. That is, one
might take (1.1) to mean this:
(1.1") that group ought to help that old lady (by virtue of some member (s)
of it doing so).
26 Recall the presupposition in Section 9.1 that deontic value is agent-neutral.

273
Taken by itself, this does not ascribe an obligation to any particular indi-
vidual in the group to help the old lady. But if we now say, with (1.2), that
only Jones is so situated that he can carry out the group's obligation, then
we may indeed infer, with (1.3), that he, as an individual, is obligated to
help the old lady. For the best that he can achieve can be no better than the
best that the group can achieve, and if the group can achieve its best only
by his acting in this way, then he can achieve his best only by acting in this
way. This of course does not help Williams's cause, which is to establish
some important link between binding and nonbinding senses of "ought,"
but it does at least provide a way to grant him his claim that the inference
is to be accepted even though the first premise does not express an indi-
vidual obligation while the conclusion does.
Finally, it may well have occurred to you that the move from (I) to (I")
is unnecessary. For, you may say, (I) can itself accommodate the require-
ment concerning transigence, as long as it is acknowledged that, in any sit-
uation, the best that one can do is to perform some act transigently. In other
words, transigence is just one of the features of (and thus just one of the fea-
tures that contributes to the deontic value of) what one does and is simply
to be included in the calculus of values along with all the other value-con-
tributory properties of actions (whatever they may be).
I have no real quarrel with this. As long as transigence is acknowledged
as being deontically significant, my main point in this chapter has been
accepted. Just how this acknowledgment is to be made is not so important.
If we grant that transigence itself has deontic value, then we can revert from
(I") to (I). This may seem simpler and hence preferable, but we should not
be misled. We can revert to (I) only if we stipulate that transigence, as
spelled out in clause (b) of (I"), is a necessary feature of the best that an
agent can do. There is, therefore, no gain in simplicity. In fact, this way of
proceeding strikes me as pretty messy. We would need to be able to talk of
what would otherwise be best if transigence were bracketed, in order to be
able to identify just what it is concerning which one ought to be transigent.
I don't know quite how this should go. But if it can be done, so be it. The
result would be an account of obligation extensionally equivalent to the
one that I have proposed.

9.4 INTERPERSONAL MORAL DILEMMAS

If groups of more than one individual can be obligated, as (9.8) contends,
then, as I have noted, (I") can be extended to account for this. Given this
extension, the question of group moral dilemmas could be treated in a man-
274
ner analogous to that employed in Chapter 7 when individual moral dilem-
mas were discussed. That is, basic group moral dilemmas would be impos-
sible - being ruled out by analogues to (2.46) (the thesis that "ought
(singly)" implies "can (singly)"), (2.40'b) (the principle of agglomeration
for "ought (singly)"), and so on27 - but certain nonbasic group moral
dilemmas (group binds, group subsidiary dilemmas) would be possible.
However, the fact that individuals never act in isolation gives rise to the
possibility of another sort of dilemma that I have called an "interpersonal
moral dilemma." As stated in Section 7.4, the form of a (two-person) inter-
personal moral dilemma is this:

(7.32) (a) St ought overall at T1 to do A at T3;

(b) Sj can at Ti do A at T3;
(c) S2 ought overall at T1 to do B at T4;
(d) S2 can at T1 do B at T4; and
(e) St and S2 cannot jointly (that is, {Su S2} cannot) at Ti do both
A at T3 and B at T4.
I postponed comment on such dilemmas to this chapter.
One instance of an interpersonal dilemma is this.28 Donna ought to buy
her daughter a Cabbage Patch doll for her birthday tomorrow; Donald
ought to buy his daughter a Cabbage Patch doll for her birthday tomor-
row; and the only store in town that carries these dolls has just one left (and,
I suppose we should add, Donna's daughter is distinct from Donald's). How
might this case arise?
If Donna and Donald are mutually disconnected, then this case could
arise only by way of some lapse; hence the dilemma would constitute a per-
plexity secundum quid, as explained in Section 7.4. This is so for the follow-
ing reason. Suppose that Donna cannot see to it whether Donald buys the
doll, and suppose that Donald does as he ought and buys the doll. Then
every world accessible to Donna is one in which Donald buys the doll; hence
no world accessible to her is one in which she buys it; hence it is not the case
after all that she ought to buy it. Thus it can happen that Donna ought to
buy the doll only if Donald fails to act as he ought; her obligation presup-
poses his lapse. The same applies, mutatis mutandis, where Donald cannot
see to it whether Donna buys the doll.
The point here is not simply the obvious one that a conflict of obliga-
tions implies that some wrong is done; it's that the sort of conflict present-

27 Cf. McConnell (1988a), p. 27.

28 This illustration is borrowed, in modified form, from McConnell (1988a), p. 28.

275
ly at issue, in the circumstances presently at issue, presupposes that some
wrong is done. It is not simply that, if one person satisfies his obligation,
then the other fails to satisfy his; it's that the one has an obligation only if
the other fails to satisfy his obligation.
If Donna and Donald are connected, however, the story is different. But
if we make the presuppositions operative throughout this chapter - in par-
ticular, the presupposition that deontic value is agent-neutral - the result is
still that interpersonal moral dilemmas can arise only by way of some lapse.
This result is obvious if we admit the possibility of group obligation; for a
group can always achieve its best and, if it does so, then, as (9.11) says, no
member does wrong. If a situation arises, then, in which some member can-
not avoid some member's doing wrong, this can only be because the group
does wrong; the members' having conflicting obligations presupposes the
group's lapse. But even if we do not admit the possibility of group obliga-
tion, still such conflicting obligations presuppose a lapse on someone's part.
There are three possibilities. Either Donna's buying the doll is deontically
superior to Donald's doing so (more precisely, every world accessible to
{Donna, Donald} in which Donald buys the doll is deontically inferior to
some world accessible to {Donna, Donald} in which Donna buys the doll),
or Donald's buying the doll is deontically superior, or they are of equal
deontic value. Suppose that the first is the case. Given the agent-neutrali-
ty of deontic value, it must be that, if Donna is transigent concerning her
buying the doll, then Donald can and ought to ensure that she buys the doll
and that he does not buy it himself; hence, if in fact he has the obligation to
buy it, this presupposes both that Donna does not do her best and that she
is intransigent concerning her (and thus the group's) best, and so it presup-
poses a double lapse on her part. So too, mutatis mutandis, for the case where
Donald's buying the doll is deontically superior to Donna's doing so. As to
the final possibility: if the group-best is for one or the other of Donna and
Donald to buy the doll, then it could not be individually better for Donna
or Donald to buy it unless either Donna or Donald acts so as to preclude
achievement of what is group-best; this, too, requires a lapse on someone's
part.
Finally, if Donna and Donald are connected but deontic value is not
agent-neutral, it appears that interpersonal moral dilemmas can arise with-
out any lapse on someone's part. Their dilemma would be rooted simply
in the agent-relativity of deontic value. If it is thought that no moral dilem-
ma of any sort can arise unless it constitutes a perplexity secundum quid, this
would constitute a reason for rejecting the possibility that deontic value be
significantly agent-relative.
276
 Postscript

I have endeavored in this book to develop and defend a certain analysis of

the concept of moral obligation, an analysis that can be put to good use in
the resolution of a number of philosophical issues. In the Preface, I made
mention of several questions and claimed that they would receive answers
in the ensuing pages. And so they have, although some of the answers have
been somewhat tentative. My hope is that these answers will have served
collectively to set the concept of moral obligation in high relief Such clar-
ification is valuable not just for its own sake. Although the account of oblig-
ation that I have presented is, appropriately, neutral regarding competing
substantive theories of obligation, it should provide a useful framework
within which these theories can be developed and assessed. Thus the exer-
cise in metaethics undertaken in the foregoing pages has significant impli-
cations for normative ethics. Of course, it has not been my purpose to
expound on these implications. Undoubtedly, too, my account has impli-
cations in the area of applied ethics, but tracing these is a very complex mat-
ter, and I offer no suggestions on how best to carry out this task. To repeat
what was said in the Preface, then, anyone seeking solutions to such prob-
lems as those besetting the unfortunate Indianapolis would be ill advised to
look for them in this book. I trust that by now that much is clear.
Nonetheless, if it can be agreed that some progress has been made here
regarding the questions that have been addressed, I shall be content.

277
 Appendix: List ofpropositions

(A) 11 (1.2)4 (2.14) 40

(B)ll (1.3)4 (2.15) 45
(C)ll (1.30 5 (2.16) 45
(D)12 (1.4)6 (2.17) 47
(E)12 (1.40 177 (2.18) 47
(MO) 75 (1.5)6 (2.19) 48
(1.50 177 (2.20) 48
(1)26 (1.6)6 (2.21) 51
(I') 133 (1.7)6 (2.22) 51
(I") 272 (1.8) 11 (2.23) 51
(II) 32 (1.9)11 (2.24) 52
(III) 33 (1.10) 11 (2.25) 52
(IV) 33 (1.11)11 (2.26) 52
(V)34 (2.27) 53
(VI) 35 (2.1)21 (2.28) 54
(VII) 35 (2.2) 21 (2.29) 54
(VIII) 35 (2.3) 21 (2.30) 54
(IX) 35 (2.4) 22 (2.31) 54
(X)36 (2.5) 22 (2.32) 54
(XI) 36 (2.50 24 (2.33) 56
(XII) 36 (2.6) 22 (2.34) 56
(XIII) 119 (2.7) 22 (2.35) 57
(XIII') 133 (2.70 24 (2.36) 63
(XIV) 120 (2.8) 22 (2.360 64
(XV) 121 (2.80 24 (2.37) 63
(XVI) 133 (2.9) 23 (2.370 64
(2.10) 23 (2.37'a) 65
(1.1)4 (2.11)39 (2.37'b) 65
(1.1')5 (2.12) 40 (2.37'c) 65
(1.1") 273 (2.13) 40 (2.37") 65
278
(2.38) 63 (2.56) 77 (3.22) 90
(2.38') 65 (2.57) 77 (3.23) 90
(2.39) 63 (2.58) 78 (3.24) 92
(2.39') 65 (3.25) 92
(2.39") 66 (3.1) 80 (3.26) 96
(2.39'") 66 (3.1') 80 (3.27) 96
(2.40) 63 (3.1") 80 (3.28) 100
(2.40') 66 (3.2) 80 (3.29) 101
(2.40'a) 66 (3.2') 80 (3.29') 103
(2.40'b) 66 (3.2") 80 (3.29") 104
(2.40'c) 68 (3.3) 81 (3.30) 106
(2.40'd) 68 (3.4) 81 (3.31) 106
(2.40'e) 68 (3.4') 82 (3.32) 111
(2.41) 63 (3.5) 81 (3.33) 113
(2.41') 68 (3.6) 81
(2.41'a) 68 (3.7) 81 (4.1) 114
(2.41'b) 69 (3.8) 81 (4.1') 115
(2.42) 63 (3.9) 83 (4.2) 114
(2.42') 69 (3.9') 83 (4.3) 115
(2.43) 63 (3.9") 83 (4.3') 122
(2.43') 69 (3.9'") 83 (4.4) 115
(2.43") 69 (3.10) 83 (4.4') 115
(2.43'") 70 (3.10') 83 (4.5) 115
(2.44) 63 (3.10") 83 (4.5') 115
(2.44') 70 (3.10'") 84 (4.6) 115
(2.45) 63 (3.11)85 (4.7) 116
(2.45') 70 (3.11a) 85 (4.8) 116
(2.46) 71 (3.12) 85 (4.9) 116
(2.46') 85 (3.13) 86 (4.10) 116
(2.47) 71 (3.13') 86 (4.11) 117
(2.48) 72 (3.14) 86 (4.12) 117
(2.49) 72 (3.14') 86 (4.13) 117
(2.49') 72 (3.15) 87 (4.13') 122
(2.50) 72 (3.16) 87 (4.14) 118
(2.50') 73 (3.17) 87 (4.15) 118
(2.51) 76 (3.17') 89 (4.16) 118
(2.52) 76 (3.18) 87 (4.17) 120
(2.53) 76 (3.19) 88 (4.17') 121
(2.54) 76 (3.20) 90 (4.18) 121
(2.55) 77 (3.21) 90 (4.18a) 121
279
 (4.18b) 121 (4.52) 130 (5.23) 165
(4.19) 121 (4.53) 130 (5.24) 166
(4.20) 121 (4.54) 130 (5.25) 170
(4.21) 121 (4.55) 135 (5.26) 171
(4.22) 122 (4.56) 135 (5.27) 172
(4.23) 122 (4.57) 137 (5.28) 177
(4.24) 123 (4.58) 138 (5.29) 177
(4.25) 123 (4.59) 138 (5.30) 182
(4.25') 138 (4.60) 138 (5.31) 182
(4.26) 123 (4.61) 140 (5.32) 182
(4.26') 138 (4.62) 140 (5.33) 182
(4.27) 123 (4.63) 140 (5.34) 182
(4.27') 124 (4.64) 140 (5.35) 182
(4.28) 123 (5.36) 183
(4.28') 124 (5.1) 143 (5.37) 183
(4.28") 138 (5.2) 144
(4.29) 124 (5.3) 144 (6.1) 189
(4.30) 124 (5.4) 144 (6.1') 195
(4.31) 124 (5.5) 144 (6.2) 190
(4.32) 124 (5.6) 145 (6.3) 192
(4.33) 125 (5.7) 148 (6.3') 192
(4.34) 126 (5.7') 167 (6.4) 196
(4.35) 126 (5.8) 149 (6.5) 197
(4.36) 126 (5.9) 149 (6.6) 198
(4.37) 126 (5.9') 158 (6.7) 198
(4.38) 126 (5.10) 149 (6.7') 199
(4.39) 126 (5.10') 168 (6.7") 199
(4.40) 127 (5.11) 151 (6.8) 199
(4.41) 127 (5.12) 151 (6.9) 203
(4.42) 127 (5.13) 151 (6.10) 203
(4.43) 127 (5.14) 152
(4.44) 127 (5.15) 152 (7.1)210
(4.45) 127 (5.16) 152 (7.2) 210
(4.46) 127 (5.17) 158 (7.3) 210
(4.47) 127 (5.18) 162 (7.4) 210
(4.48) 127 (5.19) 165 (7.5) 210
(4.49) 128 (5.20) 165 (7.6)211
(4.50) 128 (5.20) 165 (7.7)211
(4.50') 134 (5.21) 165 (7.7') 226
(4.51) 130 (5.22) 165 (7.8) 212
280
(7.9) 213 (7.26) 223 (8.13)247
(7.10)213 (7.26') 227
(7.11)214 (7.27) 223 (9.1) 254
(7.12)214 (7.28) 224 (9.1') 268
(7.13)214 (7.29) 224 (9.2) 255
(7.14)214 (7.30) 226 (9.3) 255
(7.15)215 (7.31) 228 (9.4) 256
(7.16)215 (7.32) 229 (9.5) 256
(7.17) 215 (9.6) 258
(7.18)216 (8.1)243 (9.7) 260
(7.19) 216 (8.2) 243 (9.8) 260
(7.20) 221 (8.3) 243 (9.9) 261
(7.21) 222 (8.4) 243 (9.10)261
(7.22) 222 (8.5) 244 (9.11)261
(7.22') 223 (8.6) 244 (9.12)261
(7.22") 227 (8.7) 244 (9.13)264
(7.22'") 227 (8.8) 244 (9.13') 264
(7.23) 222 (8.9) 246 (9.14)265
(7.24) 222 (8.10) 246 (9.15)265
(7.25) 223 (8.11)246
(7.25') 227 (8.12)246

281
 List of works cited

Anscombe, G. E. M. (1969). Intention. Ithaca: Cornell University Press.

Aquinas, St. Thomas (1964—75). Summa Theologiae, translated by Thomas
Gilby et al. New York: McGraw-Hill
Aqvist, Lennart (1967). "Good Samaritans, Contrary-to-Duty Imperatives,
andEpistemic Obligations." Nous, 1: 361—79.
Attfield, Robin (1979). "Supererogation and Double Standards." Mind,
88: 481-99.
Atwell, John (1978). "Ross and Prima Facie Duties." Ethics, 88: 240-9.
Austin, J. L. (1968). "A Plea for Excuses." In The Philosophy of Action,
edited by Alan R. White. Oxford: Oxford University Press, 1968,
19-42.
Baier, Kurt (1958). The Moral Point of View. Ithaca: Cornell University
Press.
Barley, Blake (1984). Review of Regan (1980). Nous, 18: 152-9.
Barnes, Gerald (1971). "Utilitarianisms." Ethics, 82: 56-64.
Baron, Marcia (1987). "Kantian Ethics and Supererogation." Journal of
Philosophy, 84: 237-62.
Bennett, Jonathan (1989). "Two Departures from Consequentialism."
Ethics, 100: 54-66.
Berlin, Isaiah (1979). Concepts and Categories. New York: Viking Press.
Brand, Myles (1971). "The Language of Not Doing." American Philosophical
Quarterly, 8: 45-53.
Broad, C. D. (1946). "Some of the Main Problems in Ethics." Philosophy,
21:99-117.
Broad, C. D. (1985). Ethics. Dordrecht: Martinus Nijhoff.
Castafieda, Hector-Neri (1968). "A Problem for Utilitarianism." Analysis,
28: 141-2.
Castafieda, Hector-Neri (1975). Thinking and Doing. Dordrecht: D.
Reidel.

282
Castaneda, Hector-Neri (1977). "Ought, Time, andDeontic Paradoxes."
Journal of Philosophy, 74: 775-91.
Castaneda, Hector-Neri (1981). "The Paradoxes of Deontic Logic: The
Simplest Solution to All of Them in One Fell Swoop." In Hilpinen
(1981): 37-85.
Castaneda, Hector-Neri (1989a). "Paradoxes of Moral Reparation:
Deontic Foci vs. Circumstances." Philosophical Studies, 57: 1—21.
Castaneda, Hector-Neri (1989b). "Moral Obligation, Circumstances, and
Deontic Foci (A Rejoinder to Fred Feldman)." Philosophical Studies,
57: 157-74.
Chisholm, Roderick M. (1963). "Contrary-to-Duty Imperatives and
Deontic Logic." Analysis, 24: 33—6.
Chisholm, Roderick M. (1968). "Supererogation and Offence." In
Thomson and Dworkin (1968): 412-29.
Chisholm, Roderick M. (1974). "Practical Reason and the Logic of
Requirement." In Practical Reason, edited by Stephan Korner. New
Haven: Yale University Press, 1974, 1-17.
Chisholm, Roderick M. (1976). Person and Object. La Salle, 111.: Open
Court.
Chisholm, Roderick M. (1989). Theory of Knowledge, 3rd edition.
Englewood Cliffs: Prentice-Hall.
Chisholm, Roderick M. and Sosa, Ernest (1966). "Intrinsic Preferability
and the Problem of Supererogation." Synthese, 16: 321-31.
Clark, Michael (1993). "On Wanting to Be Morally Perfect." Analysis, 53:
54-6.
Conee, Earl (1983). Review of Regan (1980). Journal of Philosophy, 80:
415-24.
Conee, Earl (1987). "Against Moral Dilemmas." In Gowans (1987b):
239-49.
Conee, Earl (1989). "Why Moral Dilemmas Are Impossible." American
Philosophical Quarterly, 26: 133-41.
Davidson, Donald (1980). Essays on Actions and Events. Oxford: Clarendon
Press.
Davis, Lawrence H. (1979). Theory of Action. Englewood Cliffs: Prentice-
Hall.
Donagan, Alan (1977). The Theory of Morality. Chicago: University of
Chicago Press.
Donagan, Alan (1987). "Consistency in Rationalist Moral Systems." In
Gowans (1987b): 271-90.

283
Dreier, James (1993). "Structures of Normative Theories." Monist, 76:
22-40.
Driver, Julia (1992). "The Suberogatory." AustralasianJournal ofPhilosophy,
70: 286-95.
Feinberg, Joel (1968). "Supererogation and Rules." In Thomson and
Dworkin (1968): 391-411.
Feinberg, Joel (1973). Social Philosophy. Englewood Cliffs: Prentice-Hall.
Feinberg, Joel (1979). "The Nature and Value of Rights." In Lyons
(1979c): 78-91.
Feinberg, Joel (1980). Rights, Justice, and the Bounds of Liberty. Princeton:
Princeton University Press.
Feldman, Fred (1978). Introductory Ethics. Englewood Cliffs: Prentice-Hall.
Feldman, Fred (1986). Doing the Best We Can. Dordrecht: D. Reidel.
Feldman, Fred (1989a). "The Paradox of the Knower." Philosophical
Studies, 55: 93-100.
Feldman, Fred (1989b). "Concerning the Paradox of Moral Reparation
and Other Matters." Philosophical Studies, 57: 23-39.
Feldman, Fred (1990). "A Simpler Solution to the Paradoxes of Deontic
L o g i c . " Philosophical Perspectives, 4: 309—41.
Feldman, Fred (1993). "On the Consistency of Act- and Motive-
Utilitarianism." Philosophical Studies, 70: 201-12.
Follesdal, Dagfinn and Hilpinen, Risto (1970). "Deontic Logic: An
Introduction." In Hilpinen (1970): 1-35.
Foot, Philippa (1980). "The Problem of Abortion and the Doctrine of
Double Effect." In Killing and Letting Die, edited by Bonnie Stein-
bock. Englewood Cliffs: Prentice-Hall, 1980, 156-65.
Forrester, James (1984). "Gentle Murder, or the Adverbial Samaritan."
Journal of Philosophy, 86: 193-7.
Forrester, Mary (1975). "Some Remarks on Obligation, Permission, and
Supererogation." Ethics, 85: 219—26.
Frankena, William K. (1963). "Obligation and Ability." In Philosophical
Analysis, edited by Max Black. Englewood Cliffs: Prentice-Hall, 1963,
148-65.
Frankfurt, Harry G. (1969). "Alternate Possibilities and Moral
Responsibility." Jowratf/o/P/n/osop/*)/, 66: 829-39.
Frankfurt, Harry G. (1988). The Importance of What We Care About. Cam-
bridge: Cambridge University Press.
Fried, Charles (1978). Right and Wrong. Cambridge, Mass.: Harvard
University Press.

284
Gert, Bernard (1988). Morality: A New Justification of the Moral Rules. New
York: Oxford University Press.
Gewirth, Alan (1981). "Are There Any Absolute Rights?" Philosophical
Quarterly, 31: 1-16.
Gibbard, Allan (1965). "Rule-Utilitarianism: Merely an Illusory Alter-
native?" Australasian Journal of Philosophy, 43: 211-20.
Goble, Lou (1991). "Murder Most Gentle: The Paradox Deepens."
Philosophical Studies, 64: 217-27.
Goble, Lou (1993). "The Logic of Obligation, 'Better' and 'Worse.'"
Philosophical Studies, 70: 133-63.
Goldman, Alvin I. (1976). A Theory ofHuman Action. Princeton: Princeton
University Press.
Goldman, Holly S. (1976). "Dated Rightness and Moral Imperfection."
Philosophical Review, 85: 449-87.
Goldman, Holly S. (1978). "Doing the Best One Can." In Values and
Morals, edited by Alvin I. Goldman and Jaegwon Kim. Dordrecht:
D. Reidel, 1978, 185-14.
Gorr, Michael (1979). "Omissions." Tulane Studies in Philosophy, 28:93—102.
Gowans, Christopher (1987a). "Introduction: The Debate on Moral
Dilemmas." In Gowans (1987b): 3-33.
Gowans, Christopher (1994). Innocence Lost. New York: Oxford
University Press.
Gowans, Christopher, ed. (1987b). Moral Dilemmas. New York: Oxford
University Press.
Greenspan, Patricia S. (1975). "Conditional Oughts and Hypothetical
Imperatives." Journal of Philosophy, 72: 259-76.
Greenspan, Patricia S. (1978). "Oughts and Determinism: A Response to
Goldman." Philosophical Review, 87: 77—83.
Hart, H. L. A. (1979). "Are There Any Natural Rights?" In Lyons (1979c):
14-25.
Henderson, G. P. (1966). "'Ought' Implies 'Can.'" Philosophy, 41:101-12.
Heyd, David (1982). Supererogation. Cambridge: Cambridge University
Press.
Hill, Thomas E., Jr. (1971). "Kant on Imperfect Duty and Super-
erogation." Kant-Studien, 72: 55-76.
Hill, ThomasE.Jr. (1979). "Servility and Self-Respect." InLyons (1979c):
111-24.
Hilpinen, Risto, ed. (1970). Deontic Logic: Introductory and Systematic Readings.
Dordrecht: D. Reidel.

285
Hintikka, Jaakko (1970). "Some Main Problems in Deontic Logic." In
Hilpinen (1970), 59-104.
Hohfeld, Wesley Newcomb (1919). Fundamental Legal Conceptions. New
Haven: Yale University Press.
Hudson, Stephen D. (1986). Human Character and Morality. London:
Routledge and Kegan Paul.
Humberstone, I. L. (1983). "The Background of Circumstances." Pacific
Philosophical Quarterly, 64: 19-34.
Humberstone, I. L. (1991). "Two Kinds ofAgent-Relativity." Philosophical
Quarterly, 41: 144-66.
Hume, David (1888). Treatise of Human Nature, edited by L. A. Selby-
Bigge. Oxford: Clarendon Press.
Jackson, Frank (1985). "On the Semantics and Logic of Obligation." Mind,
94: 177-95.
Jackson, Frank (1987). "Group Morality." In Metaphysics and Morality, edit-
ed by Philip Pettit, Richard Sylvan, and Jean Norman. Oxford: Basil
Blackwell, 1987, 91-110.
Jackson, Frank (1991). "Decision-theoretic Consequentialism and the
Nearest and Dearest Objection." Ethics, 101: 461-82.
Jackson, Frank and Pargetter, Robert (1986). "Oughts, Options, and
Actualism." Philosophical Review, 95: 233—55.
Jackson, Frank and Pargetter, Robert (1987). "Two Puzzles about
Conditional Obligation." Philosophical Papers, 16: 75—83.
Kagan, Shelly (1989). The Limits of Morality. Oxford: Clarendon Press.
Kamm, Frances Myrna (1985). "Supererogation and Obligation." Journal
of Philosophy, 82: 118-38.
Kant, Immanuel (1964). Groundwork of the Metaphysic of Morals, translated
by H. J. Paton. New York: Harper and Row.
Kim, Jaegwon (1976). "Events as Property Exemplifications." In Action
Theory, edited by Myles Brand and Douglas Walton. Dordrecht: D.
Reidel, 1976, 159-77.
Lebus, Bruce (1990). "Moral Dilemmas: Why They are Hard to Solve."
Philosophical Investigations, 13: 110—25.
Lemmon, E. J. (1987). "Moral Dilemmas." In Gowans (1987b): 101-14.
Lemos, Ramon M. (1980). "Duty and Ignorance." Southern Journal of
Philosophy, 18: 301-12.
Lewis, David (1973). Counterfactuals. Cambridge, Mass.: Harvard University
Press.
Loewer, Barry and Belzer, Marvin (1991). "Prima Facie Obligation: Its
Deconstruction and Reconstruction." In John Searle and His Critics,
286
 edited by Ernest Lepore and Robert van Gulick. Cambridge, Mass.:
Basil Blackwell, 1991, 359-70.
Lyons, David (1965). Forms and Limits of Utilitarianism. Oxford: Clarendon
Press.
Lyons, David (1979a). "Introduction." In Lyons (1979c): 1-13.
Lyons, David (1979b). "Rights, Claimants, and Beneficiaries." In Lyons
(1979c): 58-77.
Lyons, David, ed. (1979c). Rights. Belmont, Calif: Wadsworth.
Marcus, Ruth Barcan (1966). "Iterated Deontic Modalities." Mind, 75:
580-2.
Marcus, Ruth Barcan (1987). "Moral Dilemmas and Consistency." In
Gowans (1987b): 188-204.
Martinich, A. P. (1987). "Obligation, Ability and Prima Facie Promising."
Philosophia, 17: 323-30.
McCloskey, H. J. (1963). "Ross and the Concept of Prima Facie Duty."
Australasian Journal of Philosophy, 41: 336-45.
McCloskey, H. J. (1973). "Utilitarianism: Two Difficulties." Philosophical
Studies, 24: 62-3.
McConnell, Terrance C. (1980). "Utilitarianism and Supererogatory
Acts," Ratio, 22: 36-8.
McConnell, Terrance C. (1987). "Moral Dilemmas and Consistency in
Ethics." In Gowans (1987b): 154-73.
McConnell, Terrance C. (1988a). "Interpersonal Moral Conflicts."
American Philosophical Quarterly, 25: 25-35.
McConnell, Terrance C. (1988b). "Ross on Duty and Ignorance." History
of Philosophy Quarterly, 5: 79-95.
McConnell, Terrance C. (1989). "'Ought' Implies 'Can' and the Scope of
Moral Requirements." Philosophia, 19: 437-54.
McConnell, Terrance C. (1993a). Gratitude. Philadelphia: Temple
University Press.
McConnell, Terrance C. (1993b). "Dilemmas and Incommensurateness."
Journal of Value Inquiry, 27: 247—52.
McConnell, Terrance C. (1995). "Moral Residue and Dilemmas." In
Moral Dilemmas and Moral Theory, edited by H. E. Mason. New York:
Oxford University Press, forthcoming.
McGinn, Colin (1992). "Must I Be Morally Perfect?" Analysis, 52: 32-4.
McKinsey, Michael (1975). "Levels of Obligation." Philosophical Studies,
35: 385-95.
McKinsey, Michael (1981). "Obligations to the Starving." Nous, 15:
309-23.
287
Mellema, Gregory (1991). Beyond the Call ofDuty. Albany: State University
of New York Press.
Mellema, Gregory (1993). "Quasi-Obligation and the Failure to Be
Virtuous." Journal of Social Philosophy, 24: 176-85.
Mendola, Joseph (1987). "The Indeterminacy of Options." American
Philosophical Quarterly, 24: 125-36.
Mill, John Stuart (1957). Utilitarianism. Indianapolis: Bobbs-Merrill.
Montague, Phillip (1989). "Acts, Agents, and Supererogation." American
Philosophical Quarterly, 26: 101-11.
Montefiore, Alan (1958). "'Ought' and 'Can.'" Philosophical Quarterly, 8:
24-40.
Moore, G. E. (1903). Principia Ethica. Cambridge: Cambridge University
Press.
Moore, G. E. (1922). Philosophical Studies. London: Routledge and Kegan
Paul.
Moore, G. E. (1965). Ethics. Oxford: Oxford University Press.
Moore, Robert E. (1979). "Refraining." Philosophical Studies, 36: 407-24.
Nagel, Thomas (1986). The View from Nowhere. New York: Oxford
University Press.
New, Christopher (1974). "Saints, Heroes, and Utilitarians." Philosophy,
49: 179-89.
Nozick, Robert (1974). Anarchy, State, and Utopia. New York: Basic
Books.
Parfit, Derek (1984). Reasons and Persons. Oxford: Clarendon Press.
Pietroski, Paul M. (1993). "Prima Facie Obligations, Ceteris Paribus Laws
in Moral Theory." Ethics, 103: 489-515.
Pollock, John L. (1976). Subjunctive Reasoning. Dordrecht: D. Reidel.
Postow, B. C. (1977). "Generalized Act Utilitarianism." Analysis, 37:
49-52.
Powers, Lawrence (1967). "Some Deontic Logicians." Nous, 1: 381-400.
Prichard, H. A. (1949). Moral Obligation. Oxford: Clarendon Press.
Purtill, Richard L. (1973). "Deontically Perfect Worlds and Prima Facie
Obligations." Philosophia, 3: 429-38.
Rainbolt, George W. (1993). "Moral Rights Conflict and Compensation:
An Objection to Thomson." Presented at the 1993 Central Division
Meeting of the American Philosophical Association.
Raz, Joseph (1975). "Permissions and Supererogation," American
Philosophical Quarterly, 12: 161-8.
Regan, Donald H. (1980). Utilitarianism and Co-operation. Oxford:
Clarendon Press.
288
Regan, Donald H. (1983). "Against Evaluator Relativity: A Response to
Sen." Philosophy and Public Affairs, 12: 93-112.
Richman, Robert J. (1983). God, Free Will, and Morality. Dordrecht: D.
Reidel.
Roberts, Robert C. (1988). "What an Emotion Is: A Sketch." Philosophical
Review, 97: 183-209.
Robinson, Richard (1971). "Ought and Ought Not." Philosophy, 46:
193-202.
Ross, Alf (1941). "Imperatives and Logic." Theoria, 7: 53-71.
Ross, W. D. (1930). The Right and the Good. Oxford: Clarendon Press.
Ross, W. D. (1939). Foundations of Ethics. Oxford: Clarendon Press.
Rousseau, Jean Jacques (1992). Discourse on the Origins of Inequality, edited
by Roger W. Masters and Christopher Kelly. Hanover, N.H.:
University Press of New England.
Russell, Bertrand (1970). "The Elements of Ethics." In Readings in Ethical
Theory, edited by Wilfrid Sellars and John Hospers, 2nd edition.
Englewood Cliffs: Prentice-Hall, 1970, 3-28.
Sartre, Jean-Paul (1973). "Existentialism Is a Humanism." In Existentialist
Philosophy, edited by James A. Gould and Willis H. Truitt. Encino,
Calif: Dickenson, 1973, 27-40.
Sayre-McCord, Geoffrey (1986). "Deontic Logic and the Priority of Moral
Theory." Nous, 20: 179-97.
Scheffler, Samuel (1984). The Rejection of Consequentialism. Oxford:
Clarendon Press.
Schwartz, Thomas (1985). "Dr. Krankheit and the Concept of Comp-
liance." American Philosophical Quarterly, 22: 81—7.
Sen, Amartya (1983). "Evaluator Relativity and Consequentialist
Evaluation." Philosophy and Public Affairs, 12: 113-32.
Seung, T. K. and Bonevac, Daniel (1992). "Plural Values and
Indeterminate Rankings." Ethics, 102: 799-813.
Shaw, P. D. (1965). "'Ought' and 'Can.'" Analysis, 23: 118-20.
Simon, Caroline J. (forthcoming). "Supererogation: Puzzle or Pseudo-
Problem?"
Sinnott-Armstrong, Walter (1985). "'Ought to Have' and 'Could Have.'"
Analysis, 45: 44-8.
Sinnott-Armstrong, Walter (1988). Moral Dilemmas. Oxford: Basil Blackwell.
Slote, Michael (1989). Beyond Optimizing. Cambridge, Mass.: Harvard
University Press.
Smart, J. J. C. (1973). "An Outline of a System of Utilitarian Ethics." In
Smart and Williams (1973): 3-74.
289
Smart, J. J. C. and Williams, Bernard (1973). Utilitarianism: For and Against.
Cambridge: Cambridge University Press.
Smith, Holly M. (1986). "Moral Realism, Moral Conflict, and Compound
Acts." Journal ofPhilosophy, 83: 341-5.
Snare, Frank (1974). "The Definition of Prima Facie Duties." Philosophical
Quarterly, 24: 235-44.
Sobel, Jordan Howard (1968). "Rule-Utilitarianism." Australasian Journal
of Philosophy, 46: 146-65.
Sobel, Jordan Howard (1976). "Utilitarianism and Past and Future
Mistakes." Nous, 10: 195-219.
Sorensen, Roy A. (1991). "Moral Dilemmas, Thought Experiments, and
Conflict Vagueness." Philosophical Studies, 63: 291-308.
Stocker, Michael (1971). "'Ought' and 'Can.'" Australasian Journal of
Philosophy, 49: 303-16.
Stocker, Michael (1990). Plural and Conflicting Values. Oxford: Clarendon
Press.
Strawson, Galen (1986). Freedom and Belief. Oxford: Clarendon Press.
Stump, Eleonore (1992). "God's Obligations." Philosophical Perspectives, 6:
475-91.
Styron, William (1980). Sophie's Choice. New York: Bantam.
Swank, Casey (1985). "Reasons, Dilemmas and the Logic of'Ought.'"
Analysis, 45: 111-16.
Tannsjo, Torbjorn (1985). "Moral Conflict and Moral Realism." Journal of
Philosophy, 82: 113-17.
Thomason, Richmond H. (1981). "Deontic Logic and the Role of
Freedom in Moral Deliberation." In New Studies in Deontic Logic,
edited by Risto Hilpinen. Dordrecht: D. Reidel, 1981, 177-86.
Thomson, Judith Jarvis (1986). Rights, Restitution, and Risk. Cambridge,
Mass.: Harvard University Press.
Thomson, Judith Jarvis (1990). The Realm of Rights. Cambridge, Mass.:
Harvard University Press.
Thomson, Judith Jarvis and Dworkin, Gerald, eds. (1968). Ethics. New
York: Harper and Row.
Unger, Peter (1977). "Impotence and Causal Determinism." Philosophical
Studies, 31:289-305.
Vallentyne, Peter (1987a). "Prohibition Dilemmas and Deontic Logic."
Logique et Analyse, 117-18: 113-21.
Vallentyne, Peter (1987b). "Utilitarianism and the Outcomes of Actions."
Pacific Philosophical Quarterly, 68: 57-70.

290
Vallentyne, Peter (1988). "Gimmicky Representations and Moral
Theories." Metaphilosophy, 19: 253—63.
Vallentyne, Peter (1989). "Two Types of Moral Dilemmas." Erkenntnis,
30: 301-18.
Vallentyne, Peter (1992). "Moral Dilemmas and Comparative Concep-
tions of Morality." Southern Journal of Philosophy, 30: 117—24.
van Fraassen, Bas (1987). "Values and the Heart's Command." In Gowans
(1987b): 138-53.
van Inwagen, Peter (1983). An Essay on Free Will. Oxford: Clarendon Press.
van Rijen, Jeroen (1993). Review of Feldman (1986). Nous, 27: 264-6.
Velazco y Trianosky, Gregory (forthcoming). "On Defining the
Supererogatory."
von Wright, Georg Henrik (1968). "An Essay in Deontic Logic and the
General Theory of Action." Ada Philosophica Fennica, 21.
Vorobej, Mark (1986). "Conditional Obligation and Detachment."
Canadian Journal of Philosophy, 16: 11—26.
Walton, Douglas N. (1980). "Omitting, Refraining and Letting Happen."
American Philosophical Quarterly, 17: 319-26.
White, Alan R. (1975). Modal Thinking. Oxford: Basil Blackwell.
Widerker, David (1991). "Frankfurt on 'Ought Implies Can' and
Alternative Possibilities." Analysis, 51: 222—4.
Williams, Bernard (1973). "A Critique of Utilitarianism." In Smart and
Williams (1973), 77-150.
Williams, Bernard (1981). Moral Luck. Cambridge: Cambridge University
Press.
Williams, Bernard (1987). "Ethical Consistency." In Gowans (1987b):
115-37.
Zimmerman, Michael J. (1984). An Essay on Human Action. New York:
Peter Lang.
Zimmerman, MichaelJ. (1986). "Subsidiary Obligation." Philosophical Studies,
50: 65-75.
Zimmerman, Michael J. (1987). "Remote Obligation." American Philoso-
phical Quarterly, 24: 199-205.
Zimmerman, Michael J. (1988a). An Essay on Moral Responsibility. Totowa,
N.J.: Rowman and Littlefield.
Zimmerman, Michael J. (1988b). "Lapses and Dilemmas." Philosophical
Papers, 17: 103-12.
Zimmerman, Michael J. (1990a). "The Range of Options." American Philoso-
phical Quarterly, 27: 345-55.

291
Zimmerman, MichaelJ. (1990b). "Where Did I Go Wrong?" Philosophical
Studies, 58: 83-106.
Zimmerman, Michael J. (1992). "Cooperation and Doing the Best One
Can." Philosophical Studies, 65: 283-304.
Zimmerman, Michael J. (1993a). "Obligation, Responsibility, and
Alternate Possibilities." Analysis, 53: 51-3.
Zimmerman, Michael J. (1993b). "Supererogation and Doing the Best
One Can." American Philosophical Quarterly, 30: 373-80.
Zimmerman, Michael J. (1993c). "A Plea for Ambivalence." Meta-
philosophy, 24: 382-9.
Zimmerman, MichaelJ. (1994). "Rights, Compensation, and Culpability."
Law and Philosophy, 13: 1—32.
Zimmerman, MichaelJ. (1995a). "Prima Facie Obligation and Doing the
Best One Can." Philosophical Studies, 78: 87-123.
Zimmerman, Michael J. (1995b). "Responsibility Regarding the
Unthinkable." Midwest Studies in Philosophy, 20: forthcoming.

292
 Index of names
Anscombe, G. E. M , 55, 55n., 56, 278
Aquinas, St. Thomas, 229, 229n., 278
Aqvist, Lennart, 73n., 74n., 278
Attfield, Robin, 246n., 278
Atwelljohn, 171n., 278
Austin, J. L., 93, 93n., 94, 94n., 278
Badhwar, Neera, xiii
Baier, Kurt, 239n., 241n., 278
Barley, Blake, 260n., 278
Barnes, Gerald, 254n., 278
Baron, Marcia, 233n., 253n., 278
Belzer, Marvin, 173n., 282
Bennett, Jonathan, 243n., 278
Berlin, Isaiah, 58n., 278
Bonevac, Daniel, 58n., 285
Brand, Myles, 38n., 278
Broad, C. D., 2n., 3n., 6n., 12, 12n.,
13n., 37n., 44n., 56n., 141n.,
278
Castaiieda, Hector-Neri, 2n., 3n., 21,
21n., 36n., 37n., 44, 45n., 73n.,
97n., 124, 124n., 128, 128n.,
134, 278, 279
Chisholm, Roderick M., xiv, 2n., 3,
3n., 4n., 39n., 47n., 77n., 127,
127n., 128, 134, 173n., 235n.,
236, 236n., 238, 238n., 246n.,
247n., 279
Clark, Michael, 71n., 279
Clarke, Randolph, xiii
Clinton, Bill, 45
Coley, James, xiii
Conee, Earl, xiii, 208n., 217n., 225n.,
228n., 260n., 279
Copp, David, xiii
293
Davidson, Donald, 43n., 55, 55n.,
56n., 279
Davis, Lawrence H., 4In., 279
Donagan, Alan, 17n., 40n., 141, 142,
142n., 143, 166n., 175, 184n.,
217n., 229, 229n., 230n., 279
Dreier James, 18n., 280
Driver, Julia, 235n., 238, 238n., 280
Dworkin, Gerald, 286
Feinberg, Joel, 6n., 7n., 8, 8n., 9, 9n.,
178n., 179, 179n., 185, 185n.,
186, 186n., 232, 232n., 234n.,
246n., 247n., 249, 249n., 280
Feldman, Fred, xiv, In., 2n., 4n., 12,
12n., 18n., 21n., 24n., 31, 31n.,
37n., 42n., 45n., 46n., 49n.,
50n., 51n., 60n., 61, 61n., 64n.,
74n., 75, 75n., 76, 76n., 77n.,
78,97n., 114, 114n., 115,
n., 116, 116n., 118n., 121,
n., 122, 122n., 125n., 128,
128n., 143, 143n., 145, 145n.,
199n., 217n., 225n., 230n.,
237n., 239n., 241n., 242, 254n.,
255n., 259n., 260n., 261n.,
263n., 279, 280, 287
Follesdal, Dagfinn, 73n., 280
Foot, Philippa, 38n., 280
Forrester, James, 127n., 280
Forrester, Mary, 247n., 280
Frankena, William K., 80n., 90n.,
91n., 280
Frankfurt, Harry G., 52, 52n., 66, 67n.,
85, 85n., 86, 87, 88, 89, 280,
287
Fried, Charles, 17n., 40n., 280
Gert, Bernard, 213n., 281
Gewirth, Alan, 183, 184, 184n., 281
Gibbard, Allan, 254n., 281
Goble, Lou, 127n., 191n., 193n., 196,
197n., 198n., 202n., 204n.,
205n., 206n., 281
Goldman, Alvin I., 55n., 97n., 281
Goldman, Holly S., 31, 31n., 97n.,
117n., 136n., 189n., 191n., 281
Gorr, Michael, 38n., 281
Gowans, Christopher, 175n., 218,
218n.,281
Greenspan, Patricia S., 114n., 121n.,
122n., 136n., 191n., 281
Haji, Ishtiyaque, xiii
Hart, H. L. A., 7n., 178n., 281
Henderson, G. P., 95n., 281
Heyd, David, xiii, 233n., 234n.,
237n., 239n., 242n., 243n.,
247n., 248n., 249, 249n., 281
Hill, Thomas E., Jr., xiii, 7n., 39n.,
180n., 233n., 237n., 281
Hill, Thomas E., Sr., xiii
Hilpinen, Risto, 73n., 280, 281
Hintikka, Jaakko, 31, 31n., 144n.,
145n., 282
Hoffman, Joshua, xiii
Hohfeld, Wesley Newcomb, 176, 177,
177n., 282
Hudson, Stephen D., 213n., 282
Humberstone, I. L., 3n., 191n., 255n.,
259n., 282
Hume, David, 17n., 282
Jackson, Frank, 4n., 16, 16n., 17, 17n.,
116n., 189n., 191, 191n.,
192n., 193n., 195, 195n., 196,
196n., 197, 197n., 198n., 200,
200n.,201,201n.,202,202n.,
203, 203n., 260n., 261n., 282
Kagan, SheUy, 5n., 237n., 239n.,
252n., 282
Kamm,. Frances Myrna, 234, 234n.,
282
294
Kant, Immanuel, 17, 17n., 18, 19, 28,
29, 30, 31, 38, 38n., 39, 40, 42,
180, 180n., 233, 233n., 253n.,
256,281,282
Kim, Jaegwon, 55n., 281, 282
King, John, xiii
Landers, Ann, ix .
Lebus, Bruce, 220n., 282
Lemmon, E. J., 217, 218, 218n., 219,
219n., 282
Lemos, Ramon M., 17n., 50n., 282
Leplin, Jarrett, xiii
Lewis, Carl, 46
Lewis, David, 18n., 60n., 172n., 282
Loewer, Barry, 173n., 282
Lyons, David, 7n., 45n., 179, 179n.,
180n., 272n., 283
Marcus, Ruth Barcan, 2n., 70n., 209,
209n., 218, 218n., 219, 219n.,
220,220n.,221,221n.,222,
222n., 223n., 283
Martinich, A. P., 166n., 283
McCloskey, H. J., 50n., 173n., 283
McConnell, Terrance C , xiv, 16n.,
44n., 59n., 93, 94, 94n., 108n.,
142n., 175n., 176n., 217n.,
225n., 230n., 241n., 262n.,
275n., 283
McGinn, Colin, 7In., 283
McKinsey, Michael, 31, 3In., 97n.,
131, 131n., 132, 132n., 136,
136n., 137, 137n., 138, 138n.,
260n., 283
Mele, Alfred, xiii
Mellema, Gregory, xiii, 3n., 55n.,
234n., 237n., 242n., 247n.,
249, 249n., 252n., 284
Mendola, Joseph, 60n., 284
Mill, John Stuart, 39n., 284
Montague, Phillip, 247n., 284
Montefiore, Alan, 93, 93n., 94, 284
Moore, G. E., 3n., 15n., 90n., 91,
91n.,92n.,239n.,241,241n.,
242, 246, 252, 253n., 284
Moore, Robert E., 38n., 284
Nagel, Thomas, 18n., 284
New, Christopher, 234n., 237n.,
239n., 241n., 284
Nozick, Robert, 18n., 19n., 284
Parfit, Derek, 194, 194n., 257n.,
260n., 272n., 284
Pargetter, Robert, 116n., 189n., 191,
191n., 192n., 193n., 195,
195n., 196, 196n., 197, 197n.,
198n.,200, 200n.,201,201n.,
202, 202n., 203, 203n., 282
Pietroski, Paul M., 59n., 145n., 173n.,
175n.,213n., 214, 214n., 284
Pollock, John L., xiv, 147, 147n., 148,
148n., 172n., 284
Postema, Gerald, xiv
Postow,B. C.,260n.,284
Powers, Lawrence, 95n., 284
Prichard, H. A., 3n., 11, lln., 12, 13,
13n., 14, 16, 16n., 17n., 20n.,
41n., 43, 43n., 44, 44n., 50n.,
56n., 92n., 97n., 284
Purtill, Richard L., 145n., 284
Quinn, Philip, xiv
Rainbolt, George W., 187n., 284
Raz, Joseph, 243n., 246n., 284
Reath, Andrews, xiv
Regan, Donald H., 254n., 257n., 258n.,
259, 259n., 260, 260n., 263,
269, 270n., 278, 279, 284, 285
Richman, Robert J., 95n., 285
Roberts, Robert C , 225, 225n., 285
Robinson, Richard, 95n., 285
Rosenkrantz, Gary, xiv
Ross, Alf, 72, 72n., 118n., 122, 285
Ross, W. D., 3n., 5, 5n., 6, 6n., 7, 7n.,
11, lln., 12, 12n., 13, 16, 16n.,
17n., 18, 20n., 43, 43n., 44,
44n., 50n., 92, 92n., 93, 109,
109n., 141, 141n., 142, 143,
143n., 146, 150, 150n., 159,
159n., 160n., 163, 163n., 164,
164n., 166, 166n., 167, 168,
169, 169n., 170, 172, 172n.,
295
173, 173n., 174, 174n., 177,
181,207,208n.,209n.,220,
232, 237, 252, 252n., 256, 285
Rousseau, Jean Jacques, 17n., 285
Russell, Bertrand, 11, lln., 12n., 285
Sartre, Jean-Paul, 211, 211n., 212,
220, 223, 285
Sayre-McCord, Geoffrey, xiv, 75n.,
285
Scheffler, Samuel, 29n., 239n., 285
Schwartz, Thomas, 264n., 285
Sen, Amartya, 18n., 285
Seung, T. K., 58n., 285
Shaw, P. D., 81n., 285
Sider, Ted, xiv
Simon, Caroline J., 244n., 285
Sinnott-Armstrong, Walter, 58, 58n.,
82, 82n., 83, 83n., 84, 84n.,
90n.,91,91n.,93,93n.,94,
95n.,98n., 110, 111, llln.,
209, 209n., 213n., 214n., 219,
220n., 285
Slote, Michael, 18n., 61n., 239, 239n.,
250, 250n.,251,285
Smart, J. J. C , 239n., 241n., 285, 286
Smith, Holly M., 262n., 286
Snare, Frank, 170, 171, 171n., 172,
173, 173n., 286
Sobel, Jordan Howard, 191 n., 254n.,
286
Sorensen, Roy A., 60n., 286
Sosa, Ernest, xiv, 236n., 238, 238n.,
279
Stocker, Michael, 19, 19n., 95n., 98n.,
221,221n.,239n.,243n.,286
Strawson, Galen, 14n., 286
Stump, Eleonore, 235n., 286
Styron, William, 220n., 286
Swank, Casey, 213n., 286
Tannsjo, Torbjorn, 192n., 286
Thomason, Richmond H., 108n.,
191n.,286
Thomson, Judith Jarvis, xi, 13, 13n.,
17n.,23n.,27n., 181,
186n., 262n., 284, 286
Tirrell, Lynne, xiv
Twain, Mark, 90
Unger, Peter, 66n., 286
Vallentyne, Peter, xiv, 19n., 28n.,
55n., 59n., 62, 62n., 190n.,
286, 287
van Fraassen, Bas, 213n., 218, 218n.,
219, 219n., 223, 223n., 287
van Inwagen, Peter, 90n., 287
van Rijen, Jeroen, 89n., 287
Velazco y Trianosky, Gregory, xiv,
248n., 287
296
von Wright, Georg Henrik, 229, 230,
230n., 287
Vorobej, Mark, 130, 130n., 287
Walton, Douglas N., 38n., 287
Washington, George, 89, 90
White, Alan R., 82, 82n., 90n., 91,
91n.,92n.,93,93n.,95n., 115,
287
Widerker, David, 85, 85n., 86, 94, 287
Williams, Bernard, 4, 4n., 5, 29, 29n.,
213n.,218,218n.,219,219n.,
220,220n.,221,221n.,222,
222n., 273, 274, 286, 287
 Index of subjects
absolutism, 40
accessibility, 45-50, 60-2, 79, 85, 86,
89, 108n., 109n., 167, 254-7,
269
and choice, 25-6, 27n., 41-2,
46-8, 59, 61n., 103, 178, 199,
208-9, 212, 223, 271
of possible worlds, 25-6, 48, 276
see also "can"; options
act-categories, 11-18
act-evaluation, 20, 244-6, 247n., 250-1
act-individuation, 203
see also actions, types and tokens
act-oriented theories, 270
actions
compound, 22, 23n., 34-5, 37, 63,
66-8, 237
courses of, 24, 80, 240
group, Ch. 9 passim
types and tokens, 53-7, 164n.,
171-2, 236-7
actualism, Ch. 6 passim
objections to, 191-6
advice, 1,3, 13, 198-200,212
agent-evaluation, 20, 247n.
agents, moral, 27n., 261-2
agglomeration, 68, 71, 212-14, 219,
275
alternatives, 14-15, 18-19, 21-2,
85-9, 189-91, 195-6, 218-21,
235-6, 239-41, 244n., 245-6,
249-50, 252
see also options, personal
analysis, 75-8, 170, 172-3
attitudes, 40, 263, 270n.
augmentation, 121, 129, 144-5
297
beneficence, 7, 141, 143, 146, 150-7,
161-4, 167, 171, 175, 208, 213,
215, 217-8, 250-1
"best," see value, maximization of
binds, 226-31, 275
blame, 13, 16-17, 44n., 85-8, 93,
108-10, 131, 132n., 205, 224n.,
247, 253, 263n.
and "can avoid," 81-2, 86-90,
95-6
see also options, personal
and control, 40-6, 48-53, 92-3,
96-7, 248, 270n.
and freedom, 41, 46, 50-3, 66, 79,
85-9, 90n., 224n.
and knowledge, 49-50, 204
see also accessibility; "ought,"
and "can"
Categorical Imperative, 18, 28-31,
116,256
character, 20, 71, 90, 225
charity, 38-9, 116, 179, 233, 243,
258-9
choice, see accessibility, and choice
commensurability
intramoral, 6, 58-9, 61n., 132,
134-5, 220-1
moral and nonmoral, 1—2, 6
comparability, see commensurability
compunction, 174-5, 222
conflicts of obligation, see dilemmas;
obligation, conflicts of
conjunction, see actions, compound
connectedness, see disconnectedness
conscientiousness, 16-17, 95
consequentialism, 18-19, 27, 30-1,
221, 237, 239, 241-2, 245, 250,
252, 253n., 256
continuity, see value, continuity of
control, see, "can," and control
cooperation, 42-5, Ch. 9 passim
counterfactual, 263—5
procedural, 259-60
unrequited, 258-9
culpability, see blame
deontic status, 14-15, 221, 233,
237-8, 245-7, 249-52
deontic value, see value, deontic
deontology, 17n., 18-19, 27, 30-1
detachment
deontic, 122, 128-30, 146-8,
160-2, 168-9
factual, 114-17, 120-1, 124-31,
133, 138, 144-6, 148, 160, 171,
200-2
necessity, 121-2, 129, 145-8, 160,
168-9
determinism, 45n., 46, 51, 66
dilemmas, 117, 192-3, Ch. 7 passim
avoidance of, 221—2
basic, 211-27, 229-30, 239, 266-8,
275
group,275
interpersonal, 229-30, 274-6
nonbasic, 225-31, 275
obligation-, 62-3, 209-10
prohibition-, 62, 209-10
subsidiary, 228-9, 231, 275
disconnectedness, 267—70, 275—6
discretion, 239-44
disjunction, 35-7, 62-6, 68-9, 72-3,
144n., 233, 236, 239-40
divine command theory, 18—19
division of values, see value, division of
duty
absolute, 15n.
ceteris paribus, 5
conditional, 5, 143, 146, 163, 169
correlativity to rights, 6—9, 39
see also rights, correlativity to
obligations
298
perfect and imperfect, 38-40, 57,
233-4
positive and negative, 38
proper, 5, 141
sans phrase, 5, 141
self-regarding, 180
see also obligation
excuses, see justification, and excuse
experience, 211—12
fault, 257, 262, 263n., 269
fidelity, 141-3, 146-7, 150-1, 155-64,
166-7, 170-1, 174, 181, 183,
187, 208, 213, 215, 232
free-riding, 271
freedom, 58, 174
see also "can," and freedom
gratitude, 59-60, 90, 92, 112-13,
154-9, 163, 167, 170, 187, 208,
256
guilt, 175, 223-5, 227-8
hedonism, 241
ideals, 2-4, 6n., 8-9, 71, 90-4, 248,
252
see also "ought," binding and non-
binding
intransigence, 198-9, 257-9, 261,
266-7, 269-70, 276
justice, 18, 30, 163-4, 174
justification
and excuse, 86, 93-5
lapses, 104, 225, 230-1, 236, 275-6
see also wrongness and wrong-
doing
laziness, 189, 192-3, 203-4, 206
liberty, see freedom; rights, liberties
luck, 44, 49, 110, 206n., 209
lying, 89-90
making amends, see reparation
maximization, see value, maximization
of
maxims, 19, 23, 31, 39, 42, 233
minimal difference, 147-50, 153-4,
156, 158-60, 166-9
motives, 20, 42, 71, 92-3, 159n., 248
nonmaleficence, 7, 151—3, 156n., 157,
161-4, 167-8
obligation
absolute, 5, 57, 141
actual, 5, 141
all else being equal, 145-60
all things considered, 5-6
cancellation of, 107-11, 230n.
component, 141n.
compound, 152, 161-2, 166n.
conditional, 20, 72, 109, Ch. 4
passim, 145-7, 149-53, 156-7,
159-60, 162-9, 170n., 181, 183,
200-2
conflicts of, 141-4
see also dilemmas
defeasibility of, see augmentation
and descriptions, 56—7, 203
direct and indirect, 145, 147,
149-64, 166-8, 181
dynamics of, 20, 75, Ch. 3 passim,
125, 227-8, 236
extinction of, 113
formal, 12
group, 178n., 179n., 260-3, 270,
273-6
hypothetical, 200-1
immediate, 52, 96-113, 226
in practice, 169, 170n., 174, 181
in principle, 168-9, 170n., 174, 181
incidental, 150-3, 155, 157, 161-3
and indefiniteness, 57—63
and intention, 17
and knowledge, 17, 50, 108-10,
164-9, 173-4
levels of, 131-40, 152n., 176
material, 12
mere, 10-11, 13, 20-1, 39, 175,
178, 180, 187
moral and nonmoral, 1-2, 10
necessary, 165-70, 181-2
objective and subjective, 10—20
overall, 5-10, 27, 39, 59, 95,
299
111-12,114,123,131,135,
141, 143-7, 149-50, 152, 154,
169-76, 181-4, 186-90, 200-1,
207-37, 275
overriding of, 111-13, 135-6, 141,
144, 155, 162-3, 170, 172, 181,
188, 209, 219, 232
permanent, 165, 170n., 181
prima facie, 5-10, 18, 20, 94, 111,
123, 136, Ch. 5 passim, 207-9,
213, 215-20, 222, 232, 252
primacy of, 99-100, 102
remote, 52, 96-113, 178, 226, 228
residual, 142, 174-6, 222
restoration of, 112-13
restrictive, 200-2
resultant, 141n.
subsidiary, 20, 128-40, 202, 228-9,
231
supersession of, 97, 111—13
and timing, 105-6, 271-2
see also obligation, dynamics of
-to, 7-10, 39, 175, 178-80, 187,
234, 238
universal, 165-6, 170n., 181
see also "ought"; paradoxes of
obligation
openness, 265—7
see also transigence
options
indeterminacy of, 60
infinity of, 60-2
moral, 27, 33, 59, 234-44, 246,
249-52, 271
personal, 1, 15, 26-7, 31, 33, 40,
52, 60, 62, 65-8, 70, 89, 106,
119-20,123,125,130,139,150-1,
168,212,221,234-5,254
reordering of, 116-17, 201-2
restriction of, 97, 119-20, 201
"ought"
binding and nonbinding, 2—5, 9—10,
64, 91-2, 94, 174n., 246-7, 274
and "can," 1, 26-7, 32, 40, 46, 71,
75, 77-100, 111-13, 136, 155-6,
166n., 167, 179-80, 210-12,
214-16, 219, 225-6, 259-60,
265-6, 269-70, 275
"ought" (Contd.)
ideal, see "ought," binding and
nonbinding
iteration of, 63, 70-1
and knowledge, 49-50
and material implication, 63,
69-70, 118n.,123n., 144
and "may," see permissibility
passive-transformable statements of,
5,91
and tense, 37, 82-4
see also obligation
oversubscription, 249
paradoxes of obligation, 72-5, 191,
122-8, 131
Chisholm's, 127-8, 134
Gentle Murder, 127
Good Samaritan, 73-5, 122, 151n.
Knower, 74-5, 122, 126
Reparation, 126-7
Ross's, 72-3, 118n., 122
Second Best Plan, 124-5
SuzyMae, 95n., 122n.
perfection, 70-1
permissibility, 32-7, 55, 59, 63-5,
68-70, 81-2, 152-4, 164, 186,
209,213-21,238-9,247,
250-1,271
perplexities
secundum quid, 229, 230n., 275-6
simplidter, 229
personal possibility, see accessibility
possibilism, Ch. 6 passim
objections to, 196-203
possible worlds, 31, 45, 147, 242
see also accessibility, of possible
worlds
praise, 87-8, 247-8
preparedness to act, 42, 260, 264-6,
269
probability, 190n.
see also value, probable
promises, 18, 38, 59-60, 110-11,
141-5, 166n., 174-6, 178, 222
see also fidelity
punishment, 84
300
quandaries, 208-9, 212, 221
reasons, 150, 153, 155, 157, 168, 170,
185, 213-14, 218-19, 221, 248,
252
regret, 16, 206n., 222-4
remorse, see guilt
reparation, 126-7, 154, 156-9, 163-4,
184, 186-7
respect, 17-18, 174, 180,245
responsibility, 13-14, 16, 44n., 52-3,
84-8, 93, 217-18, 263n.
tightness and doing right, 12, 15, 16n.,
29, 33-4, 53, 143, 172, 194,
198, 208, 242, 244, 251-2,
257-8, 261, 263
rights, 17-18, 23, 33n., 50, 156n.,
176-88, 245
absolute and prima facie, 6-7,
181-8
claims, 6-9, 176-88, 234, 238
conditional, 183-5
correlativity to obligations, 6—9,
176-80, 183, 187-8, 234
see also duty, correlativity to
rights
incidental, 181
infringement of, 181, 187
legal, 177n., 185-6
liberties, 177, 187, 238
to life, 184-6
overriding of, 181, 183-7
violation of, 181
risk, 206, 234n.
satisficing, 250-1
self-imposed impossibility, 72,
95-101,110,113,117,122,
175
self-imposed necessity, 95-7, 105,
113
Simple Assumption, 132, 133n., 134-8
suberogation, 235, 237-8, 244, 246-8,
252-3
supererogation, 3, 206, Ch. 8 passim,
258-9
supervenience, 216-17
transigence, 263—6, 276
unintrusive, 268-74
utilitarianism, 18, 50, 115, 143, 237,
241
value
actual, 11,15-17, 109
agent-relative, 18, 27-30, 159-60,
242-3, 256,276
continuity of, 248-53
deontic, 14-15, 19-20, Ch. 2
passim, 98n., 108, 118-20, 122,
132-6, 139, 145-7, 149-50,
153-4, 156-9, 162, 164, 167-8,
170, 189-91, 195, 197,
199-200,205,209,211,214,
216, 219n., 221, 230, 244-9,
251-2, 254, 256-7, 259-62,
267-8, 272-4,276
division of, 244-53
instrumental, 15, 159n.
intrinsic, 18, 23, 27-8, 31, 75, 78,
159, 167, 241-2, 245, 252,
256
limits of, 3In.
maximization of, 18-19, Ch. 2
passim, 238-44, 246, 248,
250-3, Ch. 9 passim
moral, 14-15
301
morally relevant, 15, 18, 30, 58-9,
248-51
see also value, division of
perceived, 11, 15-17, 31, 79-81
probable, 11, 15-17, 79-81, 109,
205
substantive views of, 15-19, 23,
27-8,30-1,50,139,277
time-relative, 27-30, 159
vice, 30n., 247
virtue, 30n., 159n., 247
wrongness and wrongdoing, 9, 12,
14n., 15-16, 28-9, 33-6, 53,
57, 61-2, 65, 70-1, 74, 86-90,
93-4,98-105,107-11,126,
152-3, 166, 172, 181, 191, 197,
199,205,208-10,219-21,
223-7, 229-31, 234-6, 238-44,
246-7, 250-3, 260-3, 267, 270,
275-6
causing of, 194—5
compounding of, 128, 132-5, 202
levels of, 132, 134, 138-40
minimization of, 128, 134, 199
sanctioning of, 193—5, 196n.
seriousness of, 17, 29, 99, 138-40
timing of, 100-5
wrongs, 33n., 181
two making a right, 257-8, 262