You are on page 1of 20

Journal of Philosophy, Inc.

Some Ordinalist-Utilitarian Notes on Rawls's Theory of Justice by John Rawls

Review by: Kenneth J. Arrow
The Journal of Philosophy, Vol. 70, No. 9 (May 10, 1973), pp. 245-263
Published by: Journal of Philosophy, Inc.
Stable URL: .
Accessed: 06/04/2014 16:33

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact

Journal of Philosophy, Inc. is collaborating with JSTOR to digitize, preserve and extend access to The Journal
of Philosophy.

This content downloaded from on Sun, 6 Apr 2014 16:33:21 PM

All use subject to JSTOR Terms and Conditions



AWLS'S major work has been widely and correctlyac-
of the notion
claimedas the mostsearchinginvestigation
of justice in modern times. It combines a genuine and
fruitfuloriginalityof viewpoint with an extraordinarysystematic
evaluation of foundations,implicationsfor action, and connections
withotheraspectsof moral choice.The specificpostulatesforjustice
that Rawls enunciates are quite novel, and yet, once stated, they
clearlyhave a strongclaim on our attentionas at least plausible
candidatesforthe foundationsof a theoryof justice. The arguments
foracceptingthesepostulatesare part of the contractariantradition,
but have been developed in many new and interestingways. The
implicationsof these postulatesfor specificaspects of the institu-
tions of liberty,particularlycivil liberty,and for the operationsof
the economic order are spelled out in considerableand thoughtful
detail (as an economistaccustomedto much elementarymisunder-
standingof the nature of an economyon the part of philosophers
and social scientists,I mustexpressmygratitudefor the sophistica-
tion and knowledgewhich Rawls displayshere). Finally, the rela-
tionsbetweenjusticeof social institutionsand the notion of morally
rightbehavioron the part of individualsis analyzedat considerable
and intelligentlength.
It will become clear in the sequel that I have a numberof ques-
tions and objections to Rawls's theory.Indeed, it is not surprising
that no theoryof justice can be so compellingas to forestallsome
* Cambridge, Mass.: The Belknap Press of the Harvard UniversityPress, 1971.
xv, 607 p. Cloth $15.00, paper $3.95.
t This note was prepared with assistance fromGrant N.S.F. No. GS28626X.

This content downloaded from on Sun, 6 Apr 2014 16:33:21 PM

All use subject to JSTOR Terms and Conditions

objections; indeed, that veryfact is disturbingto the quest for the

concept of justice, as I shall brieflynote in the last section of this
paper. These questionsare a tributeto the breadthand fruitfulness
of Rawls's work.
My critical stance is derived from a particular tradition of
thought:thatof welfareeconomics.In the prescriptionof economic
policy,questionsof distributivejustice inevitablyarise (not all such
questions arise, only some; in particular,justice in the allocation
of freedomsratherthan goods is not part of the formalanalysisof
welfareeconomics,though some economistshave made strongin-
formaland unanalyzed commitmentsto some aspects of freedom).
The implicit ethical basis of economic policy judgment is some
versionof utilitarianism.At the same time, descriptiveeconomics
has relied heavily on a utilitarian psychologyin explaining the
choices made by consumersand other economic agents. The basic
theoremof welfareeconomics: that, under certain conditions,the
competitiveeconomic systemyields an outcome that is optimal or
efficient(in a sense which requires careful definition),depends on
the identificationof the utilitystructuresthat motivatethe choices
made by economic agents with the utilitystructuresused in judg-
ing the optimalityof the outcome of the competitivesystem.As a
result,the utilityconceptswhich,in one formor another,underlie
welfarejudgmentsin economicsas well as elsewhere(accordingto
Rawls's and many other theoriesof justice) have been subjected to
an intensivescrutinyby economists.There has been more emphasis
on their operational meaning, but perhaps less on their specific
content;philosophershave been more prone to analyze what indi-
viduals should want,whereeconomistshave been contentto identify
"should" with"is" fortheindividual (not forsociety).
I do not mean that all economistsor even those who have con-
cerned themselveswith welfarejudgmentswill agree with the fol-
lowing remarks,but I do want to suggestthe backgroundout of
In section I, I will highlightthe basic assumptionsof Rawls's
theoryand stressthose aspects which especially intersectmy inter-
ests.I will be brief,since by now the theoryis doubtlessreasonably
familiarto the reader. In section II, I raise some specificquestions
about differentaspects of the theory,in particular, the logic by
which Rawls proceedsfromthe general point of view of the theory
(the "original position," the "differenceprinciple" in its general
form)to more specificimplications,such as the priorityof liberty
and the maximin principle for distributionof goods. Section iII is

This content downloaded from on Sun, 6 Apr 2014 16:33:21 PM

All use subject to JSTOR Terms and Conditions

the central section of this paper; in it I raise a number of the

epistemologicalissues that seem to me to be crucial in the develop-
ment of most kinds of ethical theoryand in particular Rawls's:
How do we knowotherpeoples' welfareenough to apply a principle
of justice? What knowledgeis assumed to be possessedby those in
Rawls's original position when theyagree to a set of principles?In
sectioniv, I statemore explicitlywhat may be termedan ordinalist
(i.e., epistemologicallymodest) versionof utilitarianismand argue
that,in theseterms,Rawls's positiondoes not differsharply.A brief
sectionv discussesthe role of majorityand otherkinds of votingin
a theoryof justice,especiallyin lightof the discussionin sectionIv.
Section vi turnsto a differentline, an examinationof the implica-
tion of Rawls's theoryfor economic policy. Finally, in section VII,
some of the precedingdiscussionsare applied and extended to raise
some questions about the possibilityof any theoryof justice; the
criterionof universalizabilitymay be impossible to achieve when
people are really different,particularlywhen differentlife experi-
encesmean thattheycan neverhave the same information.
The centralpart of Rawls's theoryis a statementof fundamental
propositionsabout the natureof a just society,what may be thought
of as a systemof axioms. On the one side, it is soughtto justifythese
axioms as derivingfroma contractmade among rational potential
membersof society; on the other side, the implications of these
axiomsforthedeterminationof social institutionsare drawn.
The axioms themselvescan be thoughtof as divided into two
parts: one is a general statementof the notion of justice,the second
a moredetailed elaborationof morespecificforms.
The generalpoint of view is a stronglyaffirmed egalitarianism,to
be departedfromonlywhen it is in the interestof all to do so. "All
social values-liberty and opportunity,income and wealth,and the
bases of self-respect-areto be distributedequally unless unequal
distributionof any,or all, of thesevalues is to everyone'sdisadvan-
tage" (p. 62; parentheticalpage references are to Rawls's book). This
generalizeddifference principle,as Rawls termsit, is no tautology.
In particular,it implies that even natural advantages,superiorities
of intelligenceor strength, do not in themselvescreateany claims to
greaterrewards.The principlesof justice are "an agreementto re-
gard the distributionof natural talentsas a common asset and to
sharein thebenefitsof thisdistribution"(101).
Personally,I share fully this value judgment; and, indeed, it is
implied by almost all attemptsat full formalizationof welfareeco-

This content downloaded from on Sun, 6 Apr 2014 16:33:21 PM

All use subject to JSTOR Terms and Conditions

nomics.' But a contradictoryproposition: that an individual is

entitledto what he creates,is widelyand unreflectively held; when
teaching elementaryeconomics,I have had considerable difficulty
in persuadingthe studentsthat thisproductivityprinciplewas not
It may be worth stressingthat the assumptionof what may be
termedasset egalitarianism:that all the assetsof society,including
personal skills,are available as a commonpool forwhateverdistri-
bution justice calls for,is so much takenforgrantedthatit is hardly
argued for. All the alternativesto his principles of justice that
Rawls considersimply asset egalitarianism(though some of them
are veryinegalitarianin result,since more goods are to be assigned
to thosemostcapable of using them).The productivityprincipleis
not even considered.It must be said, on the otherhand, that asset
egalitarianismis certainlyan implicationof the "original position"
contract. (The practical implications of asset egalitarianismare,
however,severelymodifiedin the directionof the productivity prin-
ciple by incentiveconsiderations;see sectionvi below).
But Rawls's theoryis a much more specificstatementof the con-
cept of justice. This consistsof two parts. First,among the goods
distributedby the social order, libertyhas a priorityover others;
no amount of materialgoods is consideredto compensatefor a loss
of liberty.Second,among goods of a givenpriorityclass,inequalities
should be permittedonly if theyincrease the lot of the least well
off.The firstprinciplewill be referredto as the priorityof liberty,
the second as the maximinprinciple (maximizingthe welfareat its
minimumlevel; Rawls himselfrefersto this as the difference prin-
Rawls arguesforthesetwo principlesas being thosewhichwould
be agreed to by rational individuals in a hypotheticaloriginal
position,where theyhave full general knowledgeof the world,but
do not know which individual theywill be. The idea of this "veil
of ignorance" is that principlesof justice must be universalizable;
theymust be such as to command assent by anyone who does not
take account of his individual circumstances.If it is assumed that
rational individualsunder thesecircumstanceshave some degree of
aversion to uncertainty,then they will find it desirable to enter
into an insurance agreement:that the more successfulwill share
1 See A. Bergson,Essays in Normative Economics (Cambridge, Mass.: Harvard,
1966), ch. I; P. A. Samuelson, The Foundations of Economic Analysis (Cam-
bridge, Mass.: Harvard, 1947), pp. 230-248; or F. Y. Edgeworth (London: Kegan
Paul, 1881), pp. 56-82.

This content downloaded from on Sun, 6 Apr 2014 16:33:21 PM

All use subject to JSTOR Terms and Conditions

with the less, thoughnot so much as to make themboth worse off.

Thus, the original-positionargumentdoes lead to a generalized
view of justice. Rawls then furtherargues that his more specific
principles(priorityof libertyand the maximin principle) also fol-
low fromthe original-positionargument,at least in the sense of
being preferableto other principlesadvanced in the philosophical
literature,such as classicalutilitarianism.
Two final remarks on the general nature of Rawls's system:
(1) The principlesof justice are intendedto apply to the choice of
social institutions,not to the actual allocative decisions of society
separately.(2) The -principlesare supposed to characterizean ideal
state of justice. If the ideal state is not achieved, they do not in
themselvessupply any basis for deciding that one non-ideal state
is more or less just than another."Questions of strategyare not to
be confusedwiththoseof justice.... The forceof opposingattitudes
has no bearing on the question of rightbut only on the feasibility
of arrangementsof liberty"(231). It is intended of course that a
characterizationof ideal or optimal statesof justice is a firststep in
a complete ordering of alternativeinstitutionalarrangementsas
moreor less just.
From the viewpointof the logical structureof the theory,a central
question is the extentto which the assumptionof the original posi-
tion really implies the highlyspecificformsof Rawls's two rules.
Let me take the priorityof libertyfirst.This is given a centralplace
in presentation,and at a numberof points the fact that the theory
puts such emphasison libertyis used to distinguishit favorablyfrom
utilitarianism;the latter,it is argued, might easily lead to sacri-
ficingthe libertyof a few for the benefitof many. "Each person
possessesan inviolabilityfounded on justice that even the welfare
of societyas a whole cannot override.For thisreason justice denies
that the loss of freedomfor some is made rightby a greatergood
sharedby others"(3/4).
Despite its importance,the definitiveargumentfor the priority
of libertyis postponed to verylate in the book (541-548). The key
argumentis thatthepriorityof libertyis desiredby everyindividual.
In technicalterms,each individual has a lexicographical(or "lexi-
cal" in Rawls's simplification)orderingof goods of all kinds,with
libertycoming first;of any two possible states,an individual will
always preferthat with the most liberty,regardlessof other goods
(such as income), and will choose accordingto income only among
stateswith equal liberty."The suppositionis that . . . the persons

This content downloaded from on Sun, 6 Apr 2014 16:33:21 PM

All use subject to JSTOR Terms and Conditions

. . . will not exchange a lesserlibertyfor an improvementin their

economicwell-being,at least not once a certainlevel of wealth has
been attained. . . As the conditionsof civilization improve,the
marginalsignificancefor our good of furthereconomic and social
advantagesdiminishesrelativeto theinterestsof liberty"(542).
The argumentis clearlyan empiricaljudgment,and the reader
can decide forhimselfhow muchweightit will bear. I want to bring
out another aspect, the relation to utilitarianism.If in fact each
individual assigns priorityto libertyin the lexicographicalsense,
then the most classical sum-of-utilities criterionwill do the same
forsocial choice; the rule will be for societyto maximize the sum
of individuals' libertiesand then,among those stateswhich accom-
plish this,choosethatwhichmaximizesthe sum of satisfactions from
Let me now turnto the maximinrule (this is to be applied sepa-
ratelyto libertyand to the nonprioritygoods). The justificationap-
pears mostexplicitlyon pages 155-158; it is mainlyan argumentfor
maximin as against the sum-of-utilities criterion.It should firstbe
noted that the original-positionassumption had also been put
forthby the economistsW. S. Vickrey2 and J. C. Harsanyi3; but
theyuse it to supply a contractarianfoundationto a formof utili-
tarianism(discussedat considerablelengthby Rawls, 161-175).They
startfromthe position,due to F. P. Ramsey,and J. von Neumann
and 0. Morgenstern,that choice under riskyconditionscan be de-
scribed as the maximization of expected utility. In the original
position,each individualmaywithequal probabilitybe anymember
of the society.If thereare n membersof the societyand if the ith
memberwill have utilityui under some given allocation decision,
then the value of thatallocation to any individual is xui(l/n), since
I/n is the probabilityof being individual i. Thus, in choosing
among alternativeallocationsof goods,each individual in the origi-
nal positionwill want to maximizethisexpectation,or, what is the
same thingfora givenpopulation,maximizethe sum of utilities.
2 "Measuring Marginal Utility by Reactions to Risk," Econometrica, xiii
(1945): 319-333, p. 329; "Utility, Strategy,and Social Decision Rules," Quarterly
Journal of Economics,LXXIV (1960): 507-535, pp. 523f.
Vickrey's 1945 statementhas been overlooked by all subsequent writers,not
surprisingly,since it received relatively little emphasis in a paper overtly de-
voted to a seeminglydifferentsubject. I read the paper before I was concerned
with the theoryof social choice; the implications for that theorywere so easy
to overlook that they did not occur to me at all when they would have been
3 "Cardinal Utility in Welfare Economics and the Theory of Risk-taking,"
Journal of Political Economy, LXI (1953): 434/5; "Cardinal Welfare, Individual-
istic Ethics, and Personal Comparisonsof Utility,"ibid., LXIII (1955): 309-321.

This content downloaded from on Sun, 6 Apr 2014 16:33:21 PM

All use subject to JSTOR Terms and Conditions

Rawls,however,startingfromthe same premises,derivesthe state-

ment that societyshould maximizemin ui. The argumentseems to
have two parts: first,that in an original position,where the quality
of an entirelife is at stake,it is reasonable to have a high degreeof
aversion to risk,and being concernedwith the worstpossible out-
come is an extreme form of risk aversion; and, second, that the
probabilitiesare in fact ill definedand should not be employedin
such a calculation. The firstpoint raises some questions about the
meaning of the utilitiesand does not do justice to the fact that,at
least in Vickreyand Harsanyi,the utilitiesare already so measured
as to reflectriskaversion(see some furtherdiscussionin sectionIv).
The second point is a versionof a recurrentand unresolvedcontro-
versyin the theoryof behaviorunder uncertainty;are all uncertain-
ties expressibleby probabilities?The view that theyare has a long
historyand has been given an axiomatic justificationby Ramsey 4
and by L. J. Savage.5The contraryview has been upheld by F. H.
Knight6 and by manywriterswho have held to an objectiveview of
probability;the maximin theoryof rational decision-making under
uncertaintywas set forthby A. Wald T specificallyin the lattercon-
text.Among economists,G. L. S. Shackle8 has been a noted advo-
cate of a more general theorywhich indudes maximin as a special
case. L. Hurwicz and 19 have givena set of axioms whichimplythat
choice will be based on some functionof the maximum and the
It has, however,long been remarkedthatthe maximintheoryhas
some implicationsthat seem hardlyacceptable. It implies that any
benefit,no matterhow small, to the worst-off member of society,
will outweighany loss to a better-off individual, provided it does
not reduce the second below the level of the first.Thus, therecan
easily exist medical procedures which serve to keep people barely
alive but with little satisfactionand which are yet so expensive as
to reduce the rest of the population to poverty.A maximin prin-
ciple would apparentlyimplythatsuch proceduresbe adopted.
4 F. P. Ramsey, "Truth and Probability," in The Foundations of Mathe-
matics and Other Logical Essays (London: K. Paul, Trench, Trubner, 1931),
p. 156-198.
5 XThe Foundations of Statistics(New York: Wiley, 1954).
6 Risk, Uncertainty,and Profit(New York: Houghton Mifflin,1921).
7 "Contributions to the Theory of StatisticalEstimation and Testing Hypothe-
ses," Annals of Mathematical Statistics,x (1939):299-326.
8 Expectations in Economics (Cambridge: UniversityPress, 1949) and subse-
quent works.
9"An Optimality Criterion for Decision-making under Ignorance," in C. F.
Carter and J. L. Ford, eds., Uncertaintyand Expectation in Economics (Oxford:
Basil Blackwell, 1972), pp. 1-11.

This content downloaded from on Sun, 6 Apr 2014 16:33:21 PM

All use subject to JSTOR Terms and Conditions

Rawls considersthis argument,but rejectsit on the ground that

it will not occur in practice.He fairlyconsistently assumesthat the
actual societyhas the propertyhe calls close-knittedness: "As we
raise the expectationsof the more advantaged the situationof the
worstoffis continuouslyimproved. . For the greaterexpections
of the more favoredpresumablycover the costsof trainingand en-
courage betterperformance"(158). It is hard to analyze this argu-
mentfairlyin shortcompass.On the face of it, it seemsclearlyfalse;
there is nothingeasier than to point out changes that benefitthe
well-offat the expense of the poor, including the least advantaged,
e.g.,simultaneousreductionof the income tax forhigh bracketsand
of welfarepayments.Rawls holds that one must considerhis prin-
ciples in their totality,in particular,a stronglyexpresseddemand
for open access to all positions.But, even with perfectequality of
opportunity,therewill presumablyremain inequalities due to bio-
logical and cultural inheritance(Rawls nowhere advocates aboli-
tion of thefamily)and chanceevents,and, once inequalitiesdo exist,
the harmonyof interestsseems to be less than all-pervasive.In any
case, the assumptionof close-knittedness underminesall the distinc-
tions that Rawls is so carefulto make. For, if it holds, thereis no
difference in policy implicationbetweenthe maximinprincipleand
the sum of utilities; if all satisfactionsgo up together,the conflict
betweentheindividual and thesocietydisappears.
Many theoriesof justice,includingboth Rawls's and utilitarianism,
imply that the social institutionsor their creatorshave access to
some kindsof knowledge.This raises the question whethertheycan
in fact or even in principle have such knowledge.In this section,
two epistemologicalquestions are raised, though there are others:
(1) How can interpersonalcomparisonsof satisfactionbe made?
and (2) What knowledgeis available in theoriginalposition?

1. The problem of interpersonalcomparisonof utilitiesseems to

bother economistsmore than philosophers.As already indicated,
utilityor satisfactionor any other similar concept appears in eco-
nomic theoryas an explanationof individual behavior,forexample,
as a consumer.Specifically,it is hypothesizedthat the individual
chooseshis consumptionso as to maximizehis utility,subject to the
constraintsimposed by his budget. But, for this purpose,a quanti-
tativelymeasurable utility is a superfluousconcept. All that is
needed is an ordering,thatis, a statementforeach pair of consump-
tion patternsas to which is preferred.Any numericalfunctionover

This content downloaded from on Sun, 6 Apr 2014 16:33:21 PM

All use subject to JSTOR Terms and Conditions

the possible consumptionpatternshaving the propertythat it as-

signslargernumbersto preferredbundles could be thoughtof as a
utilityfunction.Clearly,then, any monotonic transformation of a
utilityfunctionis also a utilityfunction.
To turnthe matteraround, it mightbe asked, How can we have
any evidence about the magnitudeof utility?The only evidenceon
an individual's utilityfunctionis supplied by his observable be-
havior, specificallythe choices he makes in the course of maximiz-
ing the function.But such choices are definedby the preference
orderingand must thereforebe the same for all utilityfunctions
compatiblewiththatordering.Hence, thereis no quantitativemean-
ing forutilityforan individual. (This ordinalistpositionwas intro-
duced into economicsby V. Pareto and I. Fisher and has become
fairlyorthodoxin thelast thirtyyears.)
If the utilityof an individual is not measurable,then a fortiori
the comparisonof utilitiesof different individuals is not meaning-
ful. In particular,the sum-of-utilities
as it stands. Rawls's maximin criterionalso implies interpersonal
comparison,for we must pick out the least advantaged individual,
and thatrequiresstatementsof the form,"individual A is worseoff
than individual B." Unlike the sum-of-utilities approach, however,
this does not require that the units in which different individuals'
utilitiesare measuredbe comparable,only that we be able to rank
different individuals accordingto some scale of satisfaction.But we
do not have any underlyingnumerical magnitude to use for this
purpose, and the question still remains,What is the operational
meaningof the interpersonalcomparison?
If one is to take the sum-of-utilitiescriterionseriously,then it
would have to be consideredpossible forindividuals to have differ-
ent utility functions; in particular, they might derive different
amounts of satisfactionfromthe same incrementsto their wealth.
Then, the utilitarianwould have to agree that the sum of utilities
would be increasedby shiftingwealth to the more sensitiveindi-
viduals. This does not occur in Rawls's theory,but somethingparal-
lel to it does. Consider an individual who is incapable of deriving
much pleasure fromanything,whetherbecause of psychologicalor
physical limitations.He may well be the worst-off individual and,
therefore,be the touchstoneof distributionpolicy,even thoughhe
deriveslittlesatisfactionfromthe additional income.
In the usual applications of the sum-of-utilities approach, the
problem of differingutilitiesis dodged by assumingit away; it is
postulated that everyonehas the same utilityfunction.This avoids

This content downloaded from on Sun, 6 Apr 2014 16:33:21 PM

All use subject to JSTOR Terms and Conditions

not only what may be thoughtof as the injustice of distributing

incomein favorof the more sensitive,but also the problemof ascer-
tainingin detail what the utilityfunctionsare, a task which might
be thoughtimpossible,as argued above, or at least verydifficult in
practice,if the ordinalistposition is not accepted. Rawls criticizes
this utilitarian evasion, though cautiously; he does not wish to
reject interpersonalcomparisons(90/1). But in fact he winds up
with a somewhatsimilar approach. He introducesthe interesting
conceptofprimarygoods,thosegoods whichare needed whateveran
individual's preferencerelation ("rational plan of life," in Rawls's
terms)is. These mightbe liberties,opportunities,and income and
wealth. Then, even though individuals might have very different
uses for these primarygoods, we need consider only some simple
index of themforpurposesof interpersonalcomparison.Thus, the
factthatone individual was satisfiedwithwaterand soy flour,while
another was desperatewithout pre-phylloxeraclarets and plovers'
eggs, would have no bearing on the interpersonalcomparison; if
theyhad thesame income,theywould be equally well off.
If this comparisonappears facetious,consider the haemophiliac
who needs about $4000 worthper annum of coagulant therapyto
arriveat a stateof securityfrombleeding at all comparable to that
of the normal person. Does equal income mean equality? If not,
then,to be consistent,Rawls would have to add health to the list of
primarygoods; but then there is a trade-off between health and
wealth which involves all the conceptual problems of differing
The restrictionto some list of primarygoods is probablyessential.
I have but two comments:(1) so long as there is more than one
primarygood, thereis an index-numberproblem in commensurat-
ing the differentgoods, which is in principle as difficultas the
problem of interpersonalcomparabilitywith which we started;
(2) if we could resolve the problem of interpersonalcomparability
in Rawls's systemby reducingeverythingin effectto a single pri-
marygood, we could do the same in the sum-of-utilities approach.
To thelast statement, however,thereis a qualification:the maximin
criterionrequires only interpersonalordinality,whereas the classi-
cal view requiresinterpersonallycomparable units; to that extent,
the Rawls systemis epistemologically less demanding.

2. Let us turnfromthe epistemologicalproblemsof the current

decision-makerfor societyto those in the original position. Indi-
viduals are supposed to know the laws of the physical and the

This content downloaded from on Sun, 6 Apr 2014 16:33:21 PM

All use subject to JSTOR Terms and Conditions

social worlds,but not to know who they are or will be. But em-
pirical knowledge is afterall uncertain,and even in the original
position individuals may disagree about the factsand laws of the
universe.For example, Rawls arguesforreligioustolerationon the
grounds that one doesn't know what religion one will have, and
thereforeone wantssocietyto tolerateall religions.Operationally,a
Catholic would have to recognizethat in the original position he
wouldn't know he would be a Catholic and would thereforehave to
tolerateProtestantsor Jews or whatever,since he mightwell have
been one. But suppose he repliesthatin factCatholicismis the true
religion,that it is part of the knowledgewhich all sensible people
are supposed to have in the original position,and that he insistson
it for the salvationof all mankind.How could this be refuted?
Indeed, just thissortof argumentis raisedby writerslike Marcuse,
not to mentionany totalitarianstate and, withinwider limits,any
state. Only those who correctlyunderstand the laws of society
should be allowed to expresstheirpolitical opinions. I feel I know
that Marxism (or laissez-faire)is the truth;therefore,in the origi-
nal position, I would have supported suppressingother positions.
Even Rawls permitssuppressionof those who do not believe in
I hope it is needlessto say that I am in favorof verywide tolera-
tion. But I am not convincedthatthe originalpositionis a sufficient
basis for this argument,for it transfersthe problem to the area of
There is another kind of knowledge problem in the original
position: that about social preferences.Rawls assumes that indi-
viduals are egoistic,theirsocial preferencesbeing derivedfromthe
veil of ignorance.But whyshould therenot be viewsof benevolence
(or envy) even in the original position?All that is required is that
theynot referto named individuals.But if theseare admitted,then
therecan be disagreementover the degreeof benevolenceor malevo-
lence, and the happy assumption,thatthereare no disagreementsin
the originalposition,disappears.
It will alreadyhave been seen thatmyattitudetowardutilitarianism
is ambivalent.On the one hand, I findit difficultto ascribe opera-
tional meaning to the utilitiesto be added. On the other hand, I
have suggestedthat the practical differencesbetween the maximin
and the sum-of-utilitiescriteriaare not great,and indeed that the
maximin principle would lead to unacceptable consequencesif the
worldweresuch thattheyreallydiffered.

This content downloaded from on Sun, 6 Apr 2014 16:33:21 PM

All use subject to JSTOR Terms and Conditions

In this section,I will take up several differentpoints raised by

Rawls, and tryto defendutilitarianismagainstthem.
First,let me extenda littlethe discussionof the Vickrey-Harsanyi
position,which Rawls calls average utilitarianism.In part, this dis-
cussion continues the epistemological considerationsof the last
section. As Ramsey and von Newmann and Morgensternhave
shown,if one considerschoice among riskyalternatives,there is a
sense in which a quantitativeutilitycan be given meaning.Specifi-
cally, if choice among probabilitydistributionssatisfiescertainap-
parentlynatural rationalityconditions,then it can be shown that
thereis a utilityfunction(unique up to a positive linear transfor-
mation) on the outcomessuch thatprobabilitydistributionsof out-
comes are orderedin accordancewith the mathematicalexpectation
of theutilityof theoutcome.
By itself,thistheoremdoes not establishany welfareimplications
forthisutilityfunction;afterall, the choice among probabilitydis-
tributionsof outcomes could equally well be described by any
monotonictransformation of theexpectedutility.When I firstwrote
on this matter,10I thereforedenied the welfare relevance of ex-
pected-utility theory.But the Vickrey-Harsanyi argumentputs mat-
tersin a different perspective;if an individual assumeshe may with
equal probability be any member of society, then indeed he
evaluates any policy by his expected utility,where the utilityfunc-
tion is specificallythat definedby the von Neumann-Morgenstern
theorem.Rawls thereforeerrswhen he argues thataverageutilitari-
anism assumesrisk neutrality(165); on the contrary,the degree of
riskaversionof the individualsis alreadyincorporatedin the utility
function.This point may be given furtherstrengthby noting that
the maximin criterion,far from being opposed to average utili-
tarianism,can be regardedas a limitingcase of it. For let U be any
utilityfunction,in the sense of a functionthat representsprefer-
ences withoutuncertainty.Then, forany a > 0, -U-a is an increas-
ing functionof U and so also is a utilityfunction.Any memberof
thisfamilycould be thevon Neumann-Morgenstern utilityfunction,
i.e., that utilityfunctionfor which it is true that the individual
seeks to maximizeexpected utility.It is easy to see that,the larger
the value of a, the higherthe degreeof riskaversion.Then, accord-
ing to Vickrey,the value of a policyto an individual with a random
stakein societywould be
V = 2(-U,)-_a =

10 Social Choice and Individual Values (New York: Wiley, 1951), firsted.,
pp. 9/10.

This content downloaded from on Sun, 6 Apr 2014 16:33:21 PM

All use subject to JSTOR Terms and Conditions

But a social-welfarefunctionis only an index of choice and can

itselfbe subject to monotonic transformation;
hence, another cri-
terionthatwould yieldthesame choice is

W = (- V)-" = (2U7-a)-l'

It can, however,easilybe proved that,as a approachesinfinity, rep-

resentingincreasingdegreesof riskaversion,W approachesmin U4.
I do not wish to argue that average utilitarianismmeets all the
problems that can be raised. Rawls veryproperlypoints out that
each individual may have a differentutilityfunction,so that, al-
thougheach wishes to maximizea sum of utilities,each individual
has a different utilityfunctionin his maximand (173); in addition,
the use of equiprobabilityin thiscase is certainlynot beyondcavil.
A second of Rawls's objectionsto utilitarianismis that it may re-
quire that some individuals sacrificefor the benefitof others,so
that othermen appear to be means,not only ends (181, 183). But I
don't follow this argumentat all. A maximin principle certainly
seems to imply that the betteroffshould sacrificefor the less well
off,if that will in fact help. The talentsof the more able are, in
Rawls's system(and in my value judgments),to be used on behalf
of the less able; is thisnot usingsomepeople as means?
A thirdcriticismof classical utilitarianismis that it makes an il-
legitimateanalogy between individuals and society."The classical
view results,then,in impersonality, in the conflationof all desires
into one systemof desire" (188). But it would appear to me a purely
formalrequirementof any theoryof justice that it act as such a
conflation.A theoryof justice is presumablyan orderingof alterna-
tive social states,and thereforeis formallyanalogous to the indi-
vidual's orderingof alternativesocial states. Further,Rawls and
Bentham and I would certainlyall agree that justice should reflect
individual satisfactions;hence, the social choice made in accord-
ance with any of these theoriesof justice is "a conflationof all de-
sires." No doubt perfectionisttheoriesor those based on religious
considerationswould not be so characterized;but Rawls is not de-
The expressionand aggregationof individual preferencesthrough
voting does not have a high place in Rawls's system:"There is
nothing to the view, then, that what the majoritywills is right"
(356). The legislatorsor votersare thoughtof as expertsin justice
and are not to vote in self-interest.The assumptionseems to be
simplythat the workingsof justice will not alwaysbe clear and that

This content downloaded from on Sun, 6 Apr 2014 16:33:21 PM

All use subject to JSTOR Terms and Conditions

a pooling of opinions is worthwhile; a majoritymakes more sense

fromthispoint of view than a minority.
Clearly,thereis somethingto Rawls's position,which indeed he
shares with many political philosophers,as he notes. A political
systemin which there is no other-regardingness will not function
at all. Further,Rawls is rightin sayingthat the analogy with the
marketis imperfect.In the market,he agrees that selfishbehavior
is socially correct,but holds that the political process can never
lead to perfectjustice if based on self-seeking
behavior.But I would
argue that the analogy,though imperfect,is not completelywrong
either.Political competitiondoes servesome of the same functions
in its sphere as economic competition.Further,the expressionof
one's own interestsin voting seems to me an essential part of the
informationprocess needed for voting.Unless votersexpress their
interests,how is anyone going to know if the ends of justice are in
fact being carried out? "If I am not for myself,then who is for
me?," said Hillel, though he continued in more Rawlsian terms,
"and if I am not forothers,thenwho am I?"
To put the matter more emotionally,I would hold that the
notion of votingaccordingto one's own beliefsand thensubmitting
to the will of the majorityrepresentsa recognitionof the essential
autonomyand freedomof others.It recognizesthatjustice is a pool-
ing of irreduciblydifferent individuals,not the carryingout of poli-
cies alreadyknownin advance.
Rawls's views have implicationsmostdirectlyfor the redistribution
of income,both among contemporariesand acrossgenerations.The
maximinrule would seem on the face of it to lead to radical equali-
zation of income. Indeed, so would the sum-of-utilities
rule, if it is
assumed that all individuals have the same utilityfunctionwhich
displays decreasing marginal satisfactionsfrom additional incre-
mentsof income.Rawls, however,holds that the close-knittedness of
membersof the societymeans thatperfectequality of income is not
to the advantage of the least well-off,
but that typicallytheywill
benefitby an increasein income to some higherup in the income
scale. Rawls is ratherbriefon why one mightexpect this kind of
relation, but economistshave laid considerable stresson the in-
centiveeffectsof taxation.Assumethateach individual can produce
a certainamountper hour worked,but that this productivityvaries
fromindividual to individual. In the absence of taxation,the least
productiveindividual will be the worstoff.Therefore,a Rawlsian
(or even an old-fashionedutilitarian) may advocate a tax on the

This content downloaded from on Sun, 6 Apr 2014 16:33:21 PM

All use subject to JSTOR Terms and Conditions

income of the more able to be paid out to the less able. This is, in
fact,essentiallythe widespreadproposal for a negativeincome tax.
But since the effortto produce may in itselfdetractfromsatisfac-
tion, an income tax will lead individuals to reduce the number of
hours theyworkand thereforethe amount theyproduce. If the tax
rate on the more able is high enough, the amount of work will go
down so much that the amount collectedin taxes forredistribution
to the worst offwill actually decrease. It is at this stage that the
economybecomes close knit.
The conflictbetween incentiveand equity occurs in a utilitarian
frameworkand was already noted by Edgeworth(who was really
veryconservativeand was glad to escape from the rigorousegali-
tarianismto which his utilitarianismled). The mathematicalprob-
lem of choosing a tax schedule to maximize the sum of utilities,
taking account of the adverse incentiveeffects,is a very difficult
one; it was broached by Vickreyin his 1945 paper (op. cit.) and
analyzedby Mirrlees,"1 Fair,12and Sheshinski,13 among others.More
recently,the tax implications of the Rawls criterionhave been
analyzed along similar lines in forthcomingpapers by Atkinson,
Phelps, and Sheshinski.The practical implicationsof this research
are as yet dubious, primarilybecause too little is known about the
magnitude of the incentive effects,particularly in the upper
As I have indicated, Rawls is inexplicit about the incentive
effectsand so does not give clear guidance to the determinationof
tax rates. On pages 277-279 he argues for progressiveincome and
inheritancetaxes to achieve justice,but thereis no indication how
the rates should be chosen. Clearly, the philosophy of justice is
under no obligationto tell us what the ratesshould be in a numeri-
cal sense; but it is supposed to definethe rule that translatesany
given set of factsinto a tax schedule.The maximin rule would, on
the face of it, lead to perfectequalization, i.e., 100 per cent taxation
above a certainlevel, with correspondingsubsidiesbelow it. As far
as I can see, it is only the incentivequestion that preventsus from
The incentivequestion raises another issue with regard to the
obligation of an individual to performjustice (Rawls has much to
11J. A. Mirrlees, "An Exploration in the Theory of Optimal Income Tax-
ation," Review of Economic Studies,xxxviii (1971): 175-208.
12 R. C. Fair, "The Optimal Distribution of Income," Quarterly Journal of
Economics, LXXXV (1971): 551-579.
18 E. Sheshinski, "The Optimum Linear Income Tax," Review of Economic
Studies,xxxix (1972): 297-302.

This content downloaded from on Sun, 6 Apr 2014 16:33:21 PM

All use subject to JSTOR Terms and Conditions

say on the notion of duties and obligationson individuals,though

I have slighted this discussion in this review). If each individual
revealedhis productivity(the amount he could produce per unit of
time), it would be possible to achieve a perfectreconciliationof
justice and incentives;namely,tax each individual accordingto his
ability, not according to his actual output. Then he could not
escape taxes by workingless, and so the tax systemwould have no
adverse incentive effects.Practical economists would reject this
solution,because it would be taken for grantedthat no individual
would be truthfulif the consequencesof truth-telling were so pain-
ful. But Rawls, like most social philosophers,takes it for granted
that individuals are supposed to act justly,at least in certain con-
texts. For example, as legislatorsor voters,it is an obligation or
duty to judge accordingto the principlesof justice, not according
If, then,an individual is supposed to assesshis own
to self-interest.
potential forearningincome,is therean obligation to be truthful?
One of the most difficultquestionsin allocative justice is the dis-
tributionof wealth over generations.To what extentis one genera-
tion obligated to save, so as to increase the welfare of the next
generation?The traditionaleconomicproblemhas been the general
act of investmentin productive land, machines, and buildings
whichproduce goods in the future;more recently,we have become
especially concerned with preservationof undisturbed environ-
ments and natural resources.The most straightforward utilitarian
answeris that the utilitiesof futuregenerationsenterequally with
those of the present.But since the present generation is a very
small part of the total numberof individuals over a horizon easily
measurable in thousandsof years,the policy conclusion would be
that virtuallyeverythingshould be saved and verylittle consumed,
a conclusionwhichseemsoffensive to commonsense.The mostusual
formulationhas been to asserta criterionof maximizinga sum of
discountedutilities,in which the utilitiesof futuregenerationsare
given successivelysmallerweights.The implicationsof such policies
seem to be more in accordancewithcommonsense and practice,but
thefoundationsof such a criterionseemarbitrary.
Rawls argues that the maximin criterion,properlyinterpreted,
can be applied to the determinationof a just rate of savings(284-
292). In the original position,individuals do not know which gen-
eration theybelong to and should thereforejudge of a just rate on
that basis. That is, theyagree to leave a fixedfractionof theirin-
come to the next generationin returnfor receivingan equal frac-

This content downloaded from on Sun, 6 Apr 2014 16:33:21 PM

All use subject to JSTOR Terms and Conditions

tion of the previousgeneration'sincome. There are two difficulties

with thisargument:(1) Why should theyagree on a fractionrather
than some more complicatedrule, for example, an increasingfrac-
tion as wealth increases?(2) More serious,it would appear that the
maximin rule would most likely lead to zero as the agreed-onsav-
ings rate; for the firstgenerationwould lose under any positive
savings rate, whereas the welfare of all futuregenerationswould
increase.This point is reinforcedstronglyif one adds the empirical
factof technologicalprogress,so thateven in the absence of savings
the successivegenerationsare gettingbetteroff.Then a maximin
policy would call for improvingthe lot of the earlier generations,
which can only be done by negativesaving (runningdown existing
capital equipment) if at all possible. (To be precise,the above argu-
ment is valid only in the absence of population growth.If popula-
tion is growing,thenzero savingwould mean less capital per person
and thereforea falling income per capita. Hence, a maximin rule
in the absence of technologicalprogresswould call for positive
saving; it can easily be shown that the rule would be that the rate
of savingsequals the rate of population growthmultiplied by the
Rawls, however,modifiesthe motivationsin the original position
at this point in the argument."The parties are regardedas repre-
sentingfamilylines, say, with ties of sentimentbetween successive
generations"(292). This is a major departurefromthe egoisticas-
sumptionsheld up to this point about behavior and choice in the
original position. It should be noted that so long as fathersthink
more highlyof themselvesthan of theirsons or even more highlyof
theirsons than of subsequent generations,the effectof this modi-
ficationis verymuch the same as thatof discountingfutureutilities.
Although my guess is that any justificationfor provision for the
futurewill run somewhatalong these lines, it cannot be said that
the solutionfullyescapes all difficulties.(1) It introducesan element
of altruisminto the original position; if we introducefamilysenti-
ments,why not others (nation, tribal)? And why not elementsof
envy?(2) One mightlike a theoryof justice in which the role of
the familywas derivedratherthan primitive.In a reexaminationof
social institutions,why should the familyremain above scrutiny,
its role being locked into the original assumptions?(3) Anyway,the
familyargumentfor saving has an implicationthat should be dis-
played and mightbe questioned. Presumablythe burden of saving
should fall only on those with childrenand perhaps in proportion

This content downloaded from on Sun, 6 Apr 2014 16:33:21 PM

All use subject to JSTOR Terms and Conditions

to the numberof children.Since education and public construction

are essentiallyformsof saving, taxes to support them should fall
only on thosewithchildren.In the originalposition,thisis just the
sort of contractthat would be arrived at if the concern for the
futurewerebased solelyon familyties.
Rawls's workis based on the hypothesisthat thereis a meaningful
universalconcept of justice. If thereis, it surelymust,as he says,
be universalizablein some sense, that is, based on principles that
are symmetricamong the particularaccidents that distinguishone
individual's position from another. But as I look around at the
many conflictsthat plague our humanity,I findmany for which I
can imagine no argumentof a symmetric nature which would con-
One problem is that any actual individual must necessarilyhave
limitedinformationabout the world,and different individualshave
differentinformation.Hence, theycannot possiblyargue themselves
back into an original position with common information,even if
theysucceed in "forgetting"who theyare. Indeed, one of the most
brilliantpassages in Rawls's book is that on what he calls "social
union" (520-530). He argues that no human life is enough to en-
countermore than a small fractionof the experiencesneeded for
completeness,so that individuals have a natural complementarity
with each other (a more mundane version of this idea is Adam
Smith'sstresson the importanceof the divisionof labor). The social
natureof man springsfromthisvariegationof experience.But pre-
cisely the same differentiations imply differingand incompletely
communicablelife experiencesand therewiththe possible impossi-
bilityof agreeingon thejust action in any concretesituation.
Indeed, the thrustof Rawls's work,particularlyin its latterpas-
sages,is highlyharmonistic;the principlesof justice are stable, ac-
cording to Rawls, because the moral education they induce rein-
forces them. But if the specificapplication of the principles is
judged to be differentaccording to differentlife experiences(and
of coursedifferent geneticexperiences),even as betweenparent and
child,thentheneeded concordanceof viewsmayevaporate.
To put the mattersomewhatdifferently, many sociologistswould
hold that, in a world of limited information,conflictunresolved
by appeal to commonlyaccepted principles may have a positive
value; it is the means by which informationabout others is con-
veyed.In its own sphere,thisis the role assignedto competitionby
economists;if everyoneattemptedto act justly at everymomentin

This content downloaded from on Sun, 6 Apr 2014 16:33:21 PM

All use subject to JSTOR Terms and Conditions

his economic life, it might be difficultever to find out what the

trueinterestsof anyonewere.
To the extent that individuals are really individual, each an
autonomous end in himself,to that extenttheymust be somewhat
mysteriouisand inaccessibleto each other.There cannot be any rule
thatis completelyacceptable to all. There must,or so it now seems
to me, be the possibilityof unadjudicable conflict,which may show
itselflogicallyas paradoxes in the processof social decision-making.

Economics,Harvard University


TEHEREis no needto summarizetheargumentof thisphilo-
sophical epic. In itsbasic outline it is sufficiently
well known
to the readersof thisjournal fromRawls's articlesover the
last twentyyears.In this book Rawls has filledin gaps in the argu-
ment,answerednumerouscriticalobjections,applied his theoryto
problems of justice in politics, economics, education, and other
importantareas,and buttressedit witha theoryof moral psychology
and otherargumentativereinforcement. The resultis a remarkably
thoroughtreatisewhich well deservesto be called a philosophical
Rawls's primaryaim, he tells us, is to provide a "workable and
systematicmoral conception" (viii) to oppose utilitarianism.Until
now, the opponentsof utilitarianismhave been unable to provide
an equally systematicalternativeof theirown, and have contented
themselveswith a series of ad hoc amendmentsand restrictionsto
utilitarianismdesigned to bring it into closer harmonywith our
spontaneousmoral sentiments,at whatevercost in theoreticaltidi-
ness. They are likely to concede that one of the prime duties of
social policymakersis to promotesocial utility,but then insistthat
one may not properlypursue that commendablegoal by grinding
the faces of the poor, framingand punishing the innocent,falsify-
ing history,and so on. On the level of personal ethics,such moral-
ists as W. D. Ross admit utilitarianduties of beneficenceand non-
maleficence,but supplementthemwith quite nonutilitarianduties
of veracity,fidelity,and the like, and thereis no way of tellingin
advance which duty must trump the others when circumstances

This content downloaded from on Sun, 6 Apr 2014 16:33:21 PM

All use subject to JSTOR Terms and Conditions