Professional Documents
Culture Documents
Many social scientists other than economists call it, the theory of con-
psychologists try to account for the sumer's choice) has become exceed-
behavior of individuals. Economists ingly elaborate, mathematical, and
and a few psychologists have pro- voluminous. This literature is almost
duced a large body of theory and a unknown to psychologists, in spite of
few experiments that deal with indi- sporadic pleas in both psychological
vidual decision making. The kind of (40, 84, 103, 104) and economic
decision making with which this body (101, 102, 123, 128, 199, 202) litera-
of theory deals is as follows: given ture for greater communication be-
two states, A and B, into either one of tween the disciplines.
which an individual may put himself, The purpose of this paper is to re-
the individual chooses A in prefer- view this theoretical literature, and
ence to B (or vice versa). For in- also the rapidly increasing number of
stance, a child standing in front of a psychological experiments (performed
candy counter may be considering by both psychologists and econo-
two states. In state A the child has mists) that are relevant to it. The
$0.25 and no candy. In state B the review will be divided into five sec-
child has $0.15 and a ten-cent candy tions: the theory of riskless choices,
bar. The economic theory of decision the application of the theory of risk-
making is a theory about how to pre- less choices to welfare economics, the
dict such decisions. theory of risky choices, transitivity in
Economic theorists have been con- decision making, and the theory of
cerned with this problem since the games and of statistical decision
days of Jeremy Bentham (1748- functions. Since this literature is un-
1832). In recent years the develop- familiar and relatively inaccessible to
ment of the economic theory of con- most psychologists, and since I could
sumer's decision making (or, as the not find any thorough bibliography
1 on the theory of choice in the eco-
This work was supported by Contract nomic literature, this paper includes
N5ori-166, Task Order I, between the Office
of Naval Research and The Johns Hopkins a rather extensive bibliography of the
University. This is Report No. 166-1-182, literature since 1930.
Project Designation No. NR 145-089, under
that contract. I am grateful to the Depart- THE THEORY OF RISKLESS CHOICES"
ment of Political Economy, The Johns Hop-
kins University, for providing me with an Economic man. The method of
office adjacent to the Economics Library those theorists who have been con-
while I was writing this paper. M. Allais,
2
M. M. Flood, N. Georgescu-Roegen, K. O. No complete review of this literature is
May, A. Papandreou, L. J. Savage, and es- available. Kauder (105, 106) has reviewed the
pecially C. H. Coombs have kindly made very early history of utility theory. Stigler
much unpublished material available to me. (180) and Viner (194) have reviewed the
A number of psychologists, economists, and literature up to approximately 1930. Samuel-
mathematicians have given me excellent, but son's book (164) contains an illuminating
sometimes unheeded, criticism. Especially mathematical exposition of some of the con-
helpful were C. Christ, C. H. Coombs, F. tent of this theory. Allen (6) explains the con-
Mosteller, and L. J. Savage. cept of indifference curves. Schultz (172) re-
380
THEORY OF DECISION MAKING 381
cerned with the theory of decision continuous and differentiable. Stone
making is essentially an armchair (182) has recently shown that they
method. They make assumptions, can be abandoned with no serious
and from these assumptions they de- changes in the theory of choice.
duce theorems which presumably can Rationality. The crucial fact about
be tested, though it sometimes seems economic man is that he is rational.
unlikely that the testing will ever This means two things: He can
occur. The most important set of weakly order the states into which he
assumptions made in the theory of can get, and he makes his choices so
riskless choices may be summarized as to maximize something.
by saying that it is assumed that the Two things are required in order
person who makes any decision to for economic man to be able to put all
which the theory is applied is an available states into a weak ordering.
economic man. First, given any two states into which
What is an economic man like? He he can get, A and B, he must always
has three properties, (a) He is com- be able to tell either that he prefers
pletely informed. (6) He is infinitely A to B, or that he prefers B to A, or
sensitive, (c) He is rational. that he is indifferent between them.
Complete information. Economic If preference is operationally defined
man is assumed to know not only as choice, then it seems unthinkable
what all the courses of action open to that this requirement can ever be
him are, but also what the outcome of empirically violated. The second
any action will be. Later on, in the requirement for weak ordering, a
sections on the theory of risky choices more severe one, is that all prefer-
and on the theory of games, this as- ences must be transitive. If economic
sumption will be relaxed somewhat. man prefers A to B and B to C, then
(For the results of attempts to in- he prefers A to C. Similarly, if he is
troduce the possibility of learning indifferent between A and B and
into this picture, see 51, 77.) between B and C, then he is in-
Infinite sensitivity. In most of the different between A and C. It is not
older work on choice, it is assumed obvious that transitivity will always
that the alternatives available to an hold for human choices, and experi-
individual are continuous, infinitely ments designed to find out whether
divisible functions, that prices are or not it does will be described in the
infinitely divisible, and that economic section on testing transitivity.
man is infinitely sensitive. The only The second requirement of ra-
purpose of these assumptions is to tionality, and in some ways the more
make the functions that they lead to, important one, is that economic man
must make his choices in such a way
views the developments up to but not includ- as to maximize something. This is
ing the Hicks-Allen revolution from the point the central principle of the theory of
of view of demand theory. Hicks's book (87) choice. In the theory of riskless
is a complete and detailed exposition of most
of the mathematical and economic content of choices, economic man has usually
the theory up to 1939. Samuelson (167) has been assumed to maximize utility. In
reviewed the integrability problem and the re- the theory of risky choices, he is as-
vealed preference approach. And Wold (204, sumed to maximize expected utility.
205, 206) has summed up the mathematical In the literature on statistical de-
content of the whole field for anyone who is
comfortably at home with axiom systems and cision making and the theory of
differential equations. games, various other fundamental
382 WARD EDWARDS
principles of decision making are gists have tended to reject out of
considered, but they are all maximi- hand the theories that result from
zation principles of one sort or an- these assumptions. This isn't fair.
other. Surely the assumptions contained in
The fundamental content of the Hullian behavior theory (91) or in
notion of maximization is that eco- the Estes (60) or Bush-Mosteller
nomic man always chooses the best (36, 37) learning theories are no more
alternative from among those open realistic than these. The most useful
to him, as he sees it. In more techni- thing to do with a theory is not to
cal language, the fact that economic criticize its assumptions but rather
man prefers A to B implies and is to test its theorems. If the theorems
implied by the fact that A is higher fit the data, then the theory has at
than B in the weakly ordered set least heuristic merit. Of course, one
mentioned above. (Some theories in- trivial theorem deducible from the
troduce probabilities into the above assumptions embodied in the concept
statement, so that if A is higher than of economic man is that in any
B in the weak ordering, then eco- specific case of choice these assump-
nomic man is more likely to choose A tions will be satisfied. For instance,
than B, but not certain to choose A.) if economic man is a model for real
This notion of maximization is men, then real men should always
mathematically useful, since it makes exhibit transitivity of real choices.
it possible for a theory to specify a Transitivity is an assumption, but it
unique point or a unique subset of is directly testable. So are the other
points among those available to the properties of economic man as a
decider. It seems to me psychologi- model for real men.
cally unobjectionable. So many differ- Economists themselves are some-
ent kinds of functions can be maxi- what distrustful of economic man
mized that almost any point actually (119, 156), and we will see in subse-
available in an experimental situation quent sections the results of a num-
can be regarded as a maximum of ber of attempts to relax these as-
some sort. Assumptions about maxi- sumptions.
mization only become specific, and Early utility maximization theory.
therefore possibly wrong, when they The school of philosopher-economists
specify what is being maximized. started by Jeremy Bentham and
There has, incidentally, been al- popularized by James Mill and others
most no discussion of the possibility held that the goal of human action is
that the two parts of the concept of to seek pleasure and avoid pain.
rationality might conflict. It is con- Every object or action may be con-
ceivable, for example, that it might sidered from the point of view of
be costly in effort (and therefore in pleasure- or pain-giving properties.
negative utility) to maintain a weakly These properties are called the utility
ordered preference field. Under such of the object, and pleasure is given
circumstances, would it be "rational" by positive utility and pain by nega-
to have such a field? tive utility. The goal of action, then,
It is easy for a psychologist to point is to seek the maximum utility. This
out that an economic man who has simple hedonism of the future is
the properties discussed above is very easily translated into a theory of
unlike a real man. In fact, it is so choice. People choose the alternative,
easy to point this out that psycholo- from among those open to them, that
THEORY OP DECISION MAKING 383
leads to the greatest excess of positive completing goods, like right and left
over negative utility. This notion of shoes, which obviously do not have
utility maximization is the essence of independent utilities). Edgeworth
the utility theory of choice. It will (53), who was concerned with such
reappear in various forms through- nonindependent utilities, pointed out
out this paper. (Bohnert [30] dis- that total utility was not necessarily
cusses the logical structure of the an additive function of the utilities
utility concept.) attributable to separate commodities.
This theory of choice was embodied In the process he introduced the no-
in the formal economic analyses of all tion of indifference curves, and thus
the early great names in economics. began the gradual destruction of the
In the hands of Jevons, Walras, and classical utility theory. We shall re-
Menger it reached increasingly so- turn to this point shortly.
phisticated mathematical expression Although the forces of parsimony
and it was embodied in the thinking have gradually resulted in the elimi-
of Marshall, who published the first nation of the classical concept of
edition of his great Principles of utility from the economic theory of
Economics in 1890, and revised it at riskless choices, there have been a
intervals for more than 30 years few attempts to use essentially the
thereafter (137). classical theory in an empirical way.
The use to which utility theory was Fisher (63) and Frisch (75) have de-
put by these theorists was to estab- veloped methods of measuring margi-
lish the nature of the demand for nal utility (the change in utility [u]
various goods. On the assumption with an infinitesimal change in
that the utility of any good is a amount possessed [Q], i.e., du/dQ)
monotonically increasing negatively from market data, by making assump-
accelerated function of the amount of tions about the interpersonal simi-
that good, it is easy to show that the larity of consumer tastes. Recently
amounts of most goods which a con- Morgan (141) has used several vari-
sumer will buy are decreasing func- ants of these techniques, has dis-
tions of price, functions which are cussed mathematical and logical flaws
precisely specified once the shapes of in them, and has concluded on the
the utility curves are known. This is basis of his empirical results that the
the result the economists needed and techniques require too unrealistic
is, of course, a testable theorem. (For assumptions to be workable. The
more on this, see 87, 159.) crux of the problem is that, for these
Complexities arise in this theory techniques to be useful, the com-
when the relations between the modities used must be independent
utilities of different goods are con- (rather than competing or complet-
sidered. Jevons, Walras, Menger, ing), and the broad commodity clas-
and even Marshall had assumed that sifications necessary for adequate
the utilities of different commodities market data are not independent.
can be combined into a total utility Samuelson (164) has shown that the
by simple addition; this amounts to assumption of independent utilities,
assuming that the utilities of different while it does guatantee interval scale
goods are independent (in spite of utility measures, puts unwarrantably
the fact that Marshall elsewhere dis- severe restrictions on the nature of
cussed the notions of competing the resulting demand function. Else-
goods, like soap and detergents, and where Samuelson (158) presented,
384 WARD EDWARDS
primarily as a logical and mathe- the same amount of utility from
matical exercise, a method of measur- 10-apples-and-l-banana as you do
ing marginal utility by assuming from 6-apples-and-4-bananas. Then
some time-discount function. Since these are two points on an indiffer-
no reasonable grounds can be found ence curve, and of course there are
for assuming one such function rather an infinite number of other points on
than another, this procedure holds no the same curve. Naturally, this is not
promise of empirical success. Mar- the only indifference curve you may
shall suggested (in his notion of "con- have between apples and bananas. It
sumer's surplus") a method of utility may also be true that you are in-
measurement that turns out to be different between 13-apples-and-5-
dependent on the assumption of con- bananas and 5-apples-and-15-banan-
stant marginal utility of money, and as. These are two points on another,
which is therefore quite unworkable. higher indifference curve. A whole
Marshall's prestige led to extensive family of such curves is called an in-
discussion and debunking of this difference map. Figure 1 presents
notion (e.g., 28), but little positive such a map. One particularly useful
comes out of this literature. Thur- kind of indifference map has amounts
stone (186) is currently attempting of a commodity on one axis and
to determine utility functions for amounts of money on the other.
commodities experimentally, but has Money is a commodity, too.
reported no results as yet. The notion of an indifference map
Indifference curves. Edgeworth's can be derived, as Edge worth derived
introduction of the notion of in- it, from the notion of measurable
difference curves to deal with the utility. But it does not have to be.
utilities of nonindependent goods was Pareto (146, see also 151) was seri-
mentioned above. An indifference ously concerned about the assump-
curve is, in Edgeworth's formula- tion that utility was measurable up
tion, a constant-utility curve. Sup- to a linear transformation. He felt
pose that we consider apples and that people could tell whether they
bananas, and suppose that you get preferred to be in state A or state B,
but could not tell how much they
25 preferred one state over the other. In
other words, he hypothesized a utility
_J 20 function measurable only on an ordi-
0_ nal scale. Let us follow the usual
QL economic language, and call utility
< 15 measured on an ordinal scale ordinal
utility, and utility measured on an
10 interval scale, cardinal utility. It is
Ld meaningless to speak of the slope, or
CO marginal utility, of an ordinal utility
function; such a function cannot be
differentiated. However, Pareto saw
that the same conclusions which had
0 5 10 15 20 25 been drawn from marginal utilities
could be drawn from indifference
NUMBER OF BANANAS curves. An indifference map can be
FIG. 1. A HYPOTHETICAL INDIFFERENCE MAP drawn simply by finding all the com-
THEORY OF DECISION MAKING 385
presents no evidence for this asser- con and eggs. By assuming that all
tion. The crux of the experiment students had the same indifference
was the attempt to predict the in- curves, they were able to derive a com-
difference curves between shoes and posite indifference map for bacon and
overcoats from the other indifference eggs. No mathematical assumptions
curves. This was done by using the were necessary, and the indifference
other two indifference curves to infer map is not given mathematical form.
utility functions for shoes and for Some judgments were partly or com-
overcoats separately, and then using pletely inconsistent with the final map,
these two utility functions to predict but not too many. The only conclu-
the total utility of various amounts sion which this experiment justifies is
of shoes and overcoats jointly. The that it is possible to derive such a
prediction worked rather well. The composite indifference map.
judgments of the one subject used are The final attempt to measure an
extraordinarily orderly; there is very indifference curve is a very recent one
little of the inconsistency and vari- by the psychologists Coombs and
ability that others working in this Milholland (49). The indifference
area have found. Thurstone says, curve involved is one between risk
"The subject . . . was entirely naive and value of an object, and so will be
as regards the psychophysical prob- discussed below in the section on the
lem involved and had no knowledge theory of risky decisions. It is men-
whatever of the nature of the curves tioned here because the same meth-
that we expected to find" (18S, p. ods (which show only that the in-
154). He adds, "I selected as subject difference curve is convex to the
a research assistant in my laboratory origin, and so perhaps should not be
who knew nothing about psycho- called measurement) could equally
physics. Her work was largely well be applied to the determination
clerical in nature. She had a very of indifference curves in riskless
even disposition, and I instructed her situations.
to take an even motivational attitude Mention should be made of the
on the successive occasions . . . I was extensive economic work on statisti-
surprised at the consistency of the cal demand curves. For some reason
judgments that I obtained, but I am the most distinguished statistical de-
pretty sure that they were the result mand curve derivers feel it necessary
of careful instruction to assume a uni- to give an account of consumer's
form motivational attitude."3 From choice theory as a preliminary to the
the economist's point of view, the derivation of their empirical demand
main criticism of this experiment is curves. The result is that the two
that it involved imaginary rather best books in the area (172, 182) are
than real transactions (200). each divided into two parts; the first
The second experimental measure- is a general discussion of the theory
ment of indifference curves is reported of consumer's choice and the second
by the economists Rousseas and Hart a quite unrelated report of statistical
(157). They required large numbers economic work. Stigler (179) has
of students to rank sets of three com- given good reasons why the statistical
binations of different amounts of ba- demand curves are so little related to
the demand curves of economic
8
Thurstone, L. L. Personal communication, theory, and Wallis and Friedman
December 7, 1953. (200) argue plausibly that this state
388 WARD EDWARDS
of affairs is inevitable. At any rate, C may be more than one j.n.d. apart,
there seems to be little prospect of and so one may be preferred to the
using large-scale economic data to fill other. This argument is, of course,
in the empirical content of the theory wrong. If A has slightly more utility
of individual decision making. than B, then the individual will
Psychological comments. There are choose A in preference to B slightly
several commonplace observations more than SO per cent of the time,
that are likely to occur to psycholo- even though A and B are less than
gists as soon as they try to apply the one j.n.d. apart in utility. The 50 per
theory of riskless choices to actual cent point is in theory a precisely
experimental work. The first is that defined point, not a region. It may in
human beings are neither perfectly fact be difficult to determine because
consistent nor perfectly sensitive. of inconsistencies in judgments and
This means that indifference curves because of changes in taste with time.
are likely to be observable as in- The second psychological observa-
difference regions, or as probability tion is that it seems impossible even
distributions of choice around a to dream of getting experimentally
central locus. It would be easy to an indifference map in w-dimensional
assume that each indifference curve space where n is greater than 3. Even
represents the modal value of a nor- the case of w = 3 presents formidable
mal sensitivity curve, and that choices experimental problems. This is less
should have statistical properties important to the psychologist who
predictable from that hypothesis as wants to use the theory of choice to
the amounts of the commodities rationalize experimental data than
(locations in product space) are to the economist who wants to de-
changed. This implies that the defi- rive a theory of general static equilib-
nition of indifference between two rium.
collections of commodities should be Experiments like Thurstone's (185)
that each collection is preferred over involve so many assumptions that it
the other 50 per cent of the time. is difficult to know what their empiri-
Such a definition has been proposed cal meaning might be if these assump-
by an economist (108), and used in tions were not made. Presumably,
experimental work by psychologists the best thing to do with such ex-
(142). Of course, SO per cent choice periments is to consider them as tests
has been a standard psychological of the assumption with the least face
definition of indifference since the validity. Thurstone was willing to
days of Fechner. assume utility maximization and in-
Incidentally, failure on the part of dependence of the commodities in-
an economist to understand that a volved (incidentally, his choice of
just noticeable difference (j.n.d.) is a commodities seems singularly un-
statistical concept has led him to fortunate for justifying an assump-
argue that the indifference relation is tion of independent utilities), and so
intransitive, that is, that if A is in- used his data to construct a utility
different to B and B is indifferent to function. Of course, if only ordinal
C, then A need not be indifferent to C utility is assumed, then experimental
(8, 9, 10). He argues that if A and B indifference curves cannot be used
are less than one j.n.d. apart, then A this way. In fact, in an ordinal-
will be indifferent to B; the same of utility universe neither of the prin-
course is true of B and C; but A and cipal assumptions made by Thurstone
THEORY OF DECISION MAKING 389
can be tested by means of experi- point has not been universally recog-
mental indifference curves. So the nized, and certain economists (e.g.,
assumption of cardinal utility, though 82, 154) still defend cardinal (but not
not necessary, seems to lead to con- interpersonally comparable) utility
siderably more specific uses for ex- on grounds of its necessity for wel-
perimental data. fare economics.
At any rate, from the experimental Pareto's principle. The abandon-
point of view the most interesting ment of interpersonal comparability
question is: What is the observed and then of cardinal utility produced
shape of indifference curves between a search for some other principle to
independent commodities? This ques- justify economic policy. Pareto
tion awaits an experimental answer. (146), who first abandoned cardinal
The notion of utility is very similar utility, provided a partial solution.
to the Lewinian notion of valence He suggested that a change should
(120, 121). Lewin conceives of be considered desirable if it left
valence as the attractiveness of an everyone at least as well off as he
object or activity to a person (121). was before, and made at least one
Thus, psychologists might consider person better off.
the experimental study of utilities to Compensation principle. Pareto's
be the experimental study of valences, principle is fine as far as it goes, but
and therefore an attempt at quantify- it obviously does not go very far.
ing parts of the Lewinian theoretical The economic decisions which can be
schema. made on so simple a principle are few
and insignificant. So welfare eco-
APPLICATION OF THE THEORY OF nomics languished until Kaldor (98)
RISKLESS CHOICES TO WEL- proposed the compensation prin-
FARE ECONOMICS4 ciple. This principle is that if it is
The classical utility theorists as- possible for those who gain from an
sumed the existence of interpersonally economic change to compensate the
comparable cardinal utility. They losers for their losses and still have
were thus able to find a simple an- something left over from their gains,
swer to the question of how to de- then the change is desirable. Of
termine the best economic policy: course, if the compensation is actually
That economic policy is best which paid, then this is simply a case of
results in the maximum total utility, Pareto's principle.
summed over all members of the But Kaldor asserted that the
economy. compensation need not actually be
The abandonment of interpersonal made; all that was necessary was
comparability makes this answer use- that it could be made. The fact that
less. A sum is meaningless if the it could be made, according to
units being summed are of varying Kaldor, is evidence that the change
sizes and there is no way of reducing produces an excess of good over harm,
them to some common size. This and so is desirable. Scitovsky (173)
observed an inconsistency in Kaldor's
4
The discussion of welfare economics given position: Some cases could arise in
in this paper is exceedingly sketchy. For a which, when a change from A to B
picture of what the complexities of modern has been made because of Kaldor's
welfare economics are really like (see 11, 13,
14, 86, 118, 124, 127, 139, 140, 148, 154, 155, criterion, then a change back from B
166, 174). to A would also satisfy Kaldor's
390 WARD EDWARDS
that one case might have come about the idea that perhaps only two sub-
by chance. No evidence was found jective probability functions are nec-
for the existence of concave indiffer- essary.
ence curves, which are certainly in- Santa Monica Seminar. In the
consistent with the theory of risky summer of 1952 at Santa Monica,
decisions. This experiment is a fine California, a group of scientists con-
example of the strength and weak- ferred on problems of decision mak-
ness of the Coombs approach. It ing. They met in a two-month semi-
makes almost no assumptions, takes nar sponsored by the University of
very little for granted, and avoids Michigan and the Office of Naval
the concept of error of judgment; as Research. The dittoed reports of
a result, much of the potential in- these meetings are a gold mine of
formation in the data is unused and ideas for the student of this problem.
rarely can any strong conclusions be Some of the work done at this semi-
drawn. nar is now being prepared for a book
A most disturbing possibility is on Decision Processes edited by R. M.
raised by experiments by Marks (133) Thrall, C. H. Coombs, and R. L.
and Irwin (94) which suggest that the Davis, of the University of Michigan.
shape of the subjective probability Several minor exploratory experi-
function is influenced by the utilities ments were done at this seminar.
involved in the bets. If utilities and Vail (190) did an experiment in which
subjective probabilities are not inde- he gave four children the choice of
pendent, then there is no hope of pre- which side of various bets they
dicting risky decisions unless their wanted to be on. On the assumption
law of combination is known, and it of linear utilities, he was able to com-
seems very difficult to design an ex- pute subjective probabilities for these
periment to discover that law of com- children. The same children, how-
bination. However, the main dif- ever, were used as subjects for a
ferences that Marks and Irwin found number of other experiments; so,
were between probabilities attached when Vail later tried them out on
to desirable and undesirable alterna- some other bets, he found that they
tives. It is perfectly possible that consistently chose the bet with the
there is one subjective probability highest probability of winning, re-
function for bets with positive ex- gardless of the amounts of money in-
pected values and a different one for volved. When 50-50 bets were in-
bets with negative expected values, volved, one subject consistently chose
just as the negative branch of the the bet with the lowest expected
Markowitz utility function is likely value. No generalizable conclusions
to be different from the positive can be drawn from these experiments.
branch. The results of my probabil- Kaplan and Radner (100) tried out
ity preference experiments showed a questionnaire somewhat like
very great differences between the Coombs's method of measuring sub-
probability preference patterns for jective probability. Subjects were
positive and for negative expected - asked to assign numbers to various
value bets (57), but little difference statements. The numbers could be
between probability preferences at anything from 0 to 100 and were to
different expected-value levels so represent the likelihood that the
long as zero expected value was not statement was true. The hypotheses
crossed (59). This evidence supports to be tested were: (a) for sets of ex-
THEORY OF DECISION MAKING 401
haustive and mutually exclusive ples of this type. However, from a
statements in which the numbers as- simple-minded psychological point of
signed (estimates of degree of belief) view, these examples are irrelevant.
were nearly equal, the sums of these It is enough if the theory of choice
numbers over a set would increase can predict choices involving familiar
with the number of alternatives (be- amounts of money and familiar
cause low probabilities would be over- probability differences—choices such
estimated) ; (b) for sets with the same as those which people are accustomed
numbers of alternatives, those with to making. It may be necessary for
one high number assigned would have economic theory that the theory of
a lower set sum than those with no choice be universal and exceptionless,
high numbers. The first prediction but experimental psychologists need
was verified; the second was not. not be so ambitious. This is fortu-
Any judgments of this sort are so nate, because the introduction of the
much more likely to be made on the variance and higher moments of the
basis of number preferences and utility distribution makes the prob-
similar variables than on subjective lem of applying the theory experi-
probabilities that they offer very mentally seem totally insoluble. It is
little hope as a method of measuring difficult enough to derive reasonable
subjective probabilities. methods of measuring utility alone
Variance preferences. Allais (2, 3, from risky choices; when it also be-
4) and Georgescu-Roegen (78) have comes necessary to measure subjec-
argued that it is not enough to apply tive probability and to take the
a transform on objective value and on higher moments of the utility dis-
objective probability in order to pre- tribution into account, the problem
dict risky decisions from expected seems hopeless. Allais apparently
utility (see also 188); it is also neces- hopes to defeat this problem by using
sary to take into account at least the psychophysical methods to measure
variance, and possibly the higher utility (and presumably subjective
moments, of the utility distribution. probability also). This is essentially
There are instances in which this what Coombs has done, but Coombs
argument seems convincing. You has recognized that such procedures
would probably prefer the certainty are unlikely to yield satisfactory
of a million dollars to a 50-50 chance interval scales. The dollar scale of
of getting either four million or noth- the value of money is so thoroughly
ing. I do not think that this prefer- taught to us that it seems almost im-
ence is due to the fact that the ex- possible to devise a psychophysical
pected utility of the 50-50 bet is less situation in which subjects would
than the utility of one million dollars judge the utility, rather than the dol-
to you, although this is possible. A lar value, of dollars. They might
more likely explanation is simply judge the utility of other valuable
that the variances of the two propo- objects, but since dollars are the
sitions are different. Evidence in usual measure of value, such judg-
favor of this is the fact that if you ments would be less useful, and even
knew you would be offered this choice these judgments would be likely to be
20 times in succession, you would contaminated by the dollar values of
probably take the 50-50 bet each the objects. I would get more utility
time. Allais (5) has constructed a from a new electric shaver than I
number of more sophisticated exam- would from a new washing machine,
402 WARD EDWARDS
but because of my knowledge of the criticized conventional utility the-
relative money values of these ob- ory's attack on this problem. Since
jects, I would certainly choose the the usual frequency theory of prob-
washing machine if given a choice ability conceives of the probability as
between them. Somewhat similar the limit of the outcomes of a large
arguments can be applied against number of similar trials, it is ques-
using psychophysical methods to tionable that notions which use prob-
measure subjective probability. A ability in the ordinary sense (like the
final point is that, since these subjec- notion of maximizing expected util-
tive scales are to be used to predict ity) are applicable to unique choices.
choices, it would be best if they could However, this seems to be an experi-
be derived from similar choices. mental problem. If notions which use
Other approaches. Shackle (175) ordinary probability are incapable of
has proposed a theory of decision predicting actual unique choices,
making under risk and uncertainty. then it will be necessary to seek other
This theory is unique in that it does theoretical tools. But so long as a
not assume any kind of maximizing generally acceptable probability can
behavior. For every possible out- be defined (e.g., as in the unique toss
come of a decision made in a risky or of a coin), it is not necessary to as-
uncertain situation, Shackle assumes sume a priori that theories based on
that there is a degree of potential conventional probabilities will be in-
surprise that this, rather than some adequate. When no generally ac-
other, outcome would occur. Every ceptable probability can be defined,
outcome-potential surprise pair is then the problem becomes very dif-
ranked in accordance with its ability ferent.
to stimulate the mind (stimulation in- Cartwright and Festinger (38, 41)
creases with increasing outcome and have proposed a theory about the
decreases with increasing potential time it takes to make decisions which
surprise). The highest-ranking posi- is in some ways similar to those dis-
tive outcome-potential surprise pair cussed in this section. The main dif-
and the highest-ranking negative pair ference is that they add the concept
are found, and these two possibilities of restraining forces, and that they
alone determine what the individual conceive of all subjective magnitudes
will do. Semi-mathematical methods as fluctuating randomly around a
are used to predict the outcome of mean value. From this they deduce
consideration of possible lines of ac- various propositions about decision
tion. Although attempts have been times and the degree of certainty
made to relate it to Wald's minimax which subjects will feel about their
principle for statistical decision func- decisions, and apparently these prop-
tions (see below), the fact remains ositions work out experimentally
that most critics of the Shackle point pretty well (38, 3Q, 61, 62). The
of view have judged it to be either too Lewinian theoretical orientation
vague to be useful, or, if specified in seems to lead to this kind of model;
detail, too conducive to patently ab- Lewin, Dembo, Festinger, and Sears
surd predictiona (e.g., 201). (122) present a formally similar
Shackle's point of view was de- theory about level of aspiration. Of
veloped primarily to deal with unique course, the notion of utility is very
choices—choices which can be made similar to the Lewinian notion of
only once. Allais (3) has similarly valence.
THEORY OF DECISION MAKING 403
imputation F if one or more of the when you consider that the other
players has separately greater gains player is out to get you, and so will
(or smaller losses) in X than in F and do his best to make the worst possible
can, by acting together (in the case of outcome for you occur. If this rule is
more than*one player), enforce the expressed geometrically, it asserts
occurrence of X, or of some other im- that the point you should seek is a
putation at least as good. The rela- saddle-point, like the highest point in
tionship of domination is not transi- a mountain pass (the best rule for
tive. crossing mountains is to minimize the
A solution is a set of imputations, maximum height, so explorers seek
none of which dominates another, out such saddle-points).
such that every imputation outside Before we go any further, we need
the solution is dominated by at least a few more definitions. Games may
one imputation within the solution. be among any number of players, but
Von Neumann and Morgenstern as- the simplest game is a two-person
sert that the task of the theory of game, and it is this kind of game
games is to find solutions. For any which has been most extensively and
game, there may be one or more than most successfully analyzed. Funda-
one. One bad feature of the theory of mentally, two kinds of payoff ar-
games is that it frequently gives a rangements are possible. The sim-
large, or even infinite, number of solu- plest and most common is the one in
tions for a game. which one player wins what the other
The above definitions make clear player loses, or, more generally, the
that the only determiner of behavior one for which the sum of all the pay-
in games, according to this theory, is ments made as a result of the game is
the amounts of money which may be zero. This is called a zero-sum game.
won or lost, or the expected amounts In nonzero-sum games, analytical
in games with random elements. The complexities arise. These can be di-
fun of playing, if any, is irrelevant. minished by assuming the existence
The minimax loss principle. The of a fictitious extra player, who wins
notions of domination and of solution or loses enough to bring the sum of
imply a new fundamental rule for payments back to zero. Such a ficti-
decision making—a rule sharply dif- tious player cannot be assumed to
ferent from the rule of maximizing have a strategy and cannot, of course,
utility or expected utility with which interact with any of the other players.
this paper has been concerned up to In zero-sum two-person games,
this section. This rule is the rule of what will happen? Each player, ac-
minimizing the maximum loss, or, cording to the theory, should pick his
more briefly, minimax loss. In other minimax strategy. But will this re-
words, the rule is to consider, for each sult in a stable solution? Not always.
possible strategy that you could Sometimes the surface representing
adopt, what the worst possible out- the possible outcomes of the game
come is, and then to select that strat- does not have a saddle-point. In this
egy which would have the least ill- case, if player A chooses his minimax
effects if the worst possible outcome strategy, then player B will have an
happened. Another way of putting incentive not to use his own minimax
the same idea is to call it the principle strategy, because having found out
of maximizing the minimum gain, or his opponent's strategy, he can gain
maximin gain. This rule makes con- more by some other strategy. Thus
siderable sense in two-person games the game has no solution.
408 WARD EDWARDS
Various resolutions of this problem agree with this), nor is it likely to be
are possible. Von Neumann and of any practical use in telling you
Morgenstern chose to introduce the how to play a complicated game; the
notion of a mixed strategy, which is a crux of the theory of games is the
probability distribution of two or principle of choosing tfre strategy
more pure strategies. The fundamen- which minimizes the maximum ex-
tal theorem of the theory of games is pected financial loss; and the theory
that if both players in a zero-sum defines a solution of a game as a set
two-person game adopt mixed strat- of imputations which satisfies this
egies which minimize the maximum principle for all players.
expected loss, then the game will al- Assumptions, In their book von
ways have a saddle-point. Thus each Neumann and Morgenstern say "We
person will get, in the long run, his have . . . assumed that [utility] is
expected loss, and will have no in- numerical . . . substitutable and un-
centive to change his behavior even restrictedly transferable between the
if he should discover what his op- various players." (197, p. 604.) Game
ponent's mixed strategy is. Since A is theorists disagree about what this
already getting the minimum possible and other similar sentences mean.
under the strategy he chose, any One likely interpretation is that they
change in strategy by B will only in- assume utility to be linear with the
crease A's payoff, and therefore cause physical value of money involved in
B to gain less or lose more than he a game and to be interpersonally
would by his own minimax strategy. comparable. The linear utility curves
The same is true of B. seem to be necessary for solving two-
Games involving more than two person games; the interpersonal com-
people introduce a new element—the parability is used for the extension to
possibility that two or more players n persons. Attempts are being made
will cooperate to beat the rest. Such to develop solutions free of these as-
a cooperative agreement is called a sumptions (176).
coalition, and it frequently involves Statistical decision functions. Von
side-payments among members of the Neumann (195) first used the mini-
coalition. The method of analysis for max principle in his first publication
three-or-more-person games is to con- on game theory in 1928. Neyman
sider all possible coalitions and to and Pearson mentioned its applic-
solve the game for each coalition on ability to statistical decision prob-
the principles of a two-person game. lems in 1933 (144). Wald (198), who
This works fairly well for three-per- prior to his recent death was the
son games, but gets more complicated central figure in the statistical deci-
and less satisfactory for still more sion-function literature, first seriously
people. applied the minimax principle to sta-
This is the end of this exposition of tistical problems in 1939. Appar-
the content of von Neumann and ently, all these uses of the principle
Morgenstern's book. It is of course were completely independent of one
impossible to condense a tremendous another.
and difficult book into one page. The After Theory of Games and Eco-
major points to be emphasized are nomic Behavior appeared in 1944,
these: the theory of games is not a Wald (198) reformulated the problem
model of how people actually play of statistical decision making as one
games (some game theorists will dis- of playing a game against Nature,
THEORY OF DECISION MAKING 409
The statistician must decide, on the confined their discussions to cases in
basis of observations which cost which the concepts of minimax loss
something to make, between policies, and minimax regret amount to the
each of which has a possible gain or same thing. Other suggested princi-
loss. In some cases, all of these gains ples are: maximizing the maximum
and losses and the cost of observing expected gain, and maximizing some
can be exactly calculated, as in in- weighted average of the maximum
dustrial quality control. In other and minimum expected gains (93).
cases, as in theoretical research, it is None of these principles commands
necessary to make some assumption general acceptance; each can be
about the cost of being wrong and the made to show peculiar consequences
gain of being right. At any rate, when under some conditions (see 170).
they are put in this form, it is obvious Experimental games. The concepts
that the ingredients of the problem of of the theory of games suggest a new
statistical decision making have a field of experimentation: How do
gamelike sound. Wald applied the people behave in game situations?
minimax principle to them in a way Such experimentation would center
essentially identical with game the- on the development of strategies, par-
ory. ticularly mixed strategies, and, in
A very frequent criticism of the three-or-more-person games, on the
minimax approach to games against development of coalitions and on the
Nature is that Nature is not hostile, bargaining process. You should* re-
as is the opponent in a two-person member that the theory of games
game. Nature will not, in general, does not offer a mathematical model
use a minimax strategy. For this predicting the outcomes of such
reason, other principles of decision games (except in a few special cases);
making have been suggested. The all it does is offer useful concepts and
simple principle of maximizing ex- language for talking about them, and
pected utility (which is the essence of predict that certain outcomes will
the Bayes's theorem [15, 198] solution not occur.
of the problem) is not always applica- A few minor experiments of this
ble because, even though Nature is kind have been conducted by Flood,
not hostile, she does not offer any a mathematician, while he was at
way of assigning a probability to each Rand Corporation. He usually used
possible outcome. In other words, colleagues, many of whom were ex-
statistical decision making is a prob- perts in game theory, and secretaries
lem of uncertainty, rather than of as subjects. The general design of
risk. Savage has suggested the prin- his experiments was that a group of
ciple of minimaxing regret, where re- subjects were shown a group of de-
gret is defined as the difference be- sirable objects on a table, and told
tween the maximum which can be that they, as a group, could have the
gained under any strategy given a first object they removed from the
certain state of the world and the table, and that they should decide
amount gained under the strategy among themselves which object to
adopted. Savage believes (170, also choose and how to allocate it. In the
personal communication) that neither first experiment (64) the allocation
von Neumann and Morgenstern nor problem did not arise because enough
Wald actually intended to propose duplicate objects were provided so
the principle of minimaxing loss; they that each subject could have one of
410 WARD EDWARDS
the kind of object the group selected. better. Physical isolation of one sub-
The subjects were Harvard under- ject from another would make it pos-
graduates, and the final selection was sible to match each subject against a
made by negotiation and voting. In standard bargainer, the experimenter
the second experiment (65), in which or a stooge, who bargains by a fixed
the subjects were colleagues and sec- set of rules that are unknown to the
retaries, a long negotiation process subject. Flood (personal communica-
eliminated some of the objects, but a tion) is conducting experiments of
time limit forced a selection by lot this sort. For three-or-more-person
from among the rest. Further negoti- games, Asch's (16) technique of using
ations to solve the allocation problem a group consisting of only one real
were terminated by a secretary, who subject and all the rest stooges might
snatched the object, announced that well be used. It would be interesting,
it was hers, and then tried to sell it. for instance, to see how the probabil-
No one was willing to buy, so the ex- ity of a coalition between two players
periment terminated. Other experi- changes as the number and power of
ments (66, 67) showed that coalitions players united against them increase.
sometimes form, that a sophisticated The theory of games is the area
subject could blackmail the group for among those described in this paper
an extra side-payment by threatening in which the uncontrolled and casu-
to change his vote, and that the ally planned "pilot experiment" is
larcenous secretary, having succeeded most likely to occur. Such experi-
once, had to be physically restrained ments are at least as dangerous here
in subsequent sessions to prevent as they are in the area of risky de-
more larceny. The general conclusion cision making. Flood's results sug-
suggested by all these experiments is gest that it is especially important to
that even experts on game theory are use naive subjects and to use them
less rational and more conventional only once, unless the effects of expert-
than game theory might lead experi- ness and experience are the major
menters to expect. concern of the experiment.
Psychological comments. The most
nutritive research problems in this SUMMARY
area seem to be the social problems of For a long time, economists and
how bargaining takes place. Flood's others have been developing mathe-
experiments left bargainers free and matical theories about how people
used physical objects, whose utilities make choices among desirable alter-
probably vary widely from subject to natives. These theories center on the
subject, as stimuli to bargain over. notion of the subjective value, or
This is naturalistic, but produces utility, of the alternatives among
data too complex and too nonnumeri- which the decider must choose. They
cal for easy analysis. A simpler situa- assume that people behave rationally,
tion in which the possible communi- that is, that they have transitive
cations from one bargainer to an- preferences and that they choose in
other are limited (perhaps by means such a way as to maximize utility or
of an artificial vocabulary), in which expected utility.
the subjects do not see one another, The traditional theory of riskless
and in which the object bargained choices, a straightforward theory of
over is simple, preferably being utility maximization, was challenged
merely a sum of money, would be by the demonstration that the mathe-
THEORY OF DECISION MAKING 411
matical tool of indifference curves for utility and subjective probability
made it possible to account for risk- measured on an ordered metric scale,
less choices without assuming that and did some experiments to test im-
utility could be measured on an in- plications of the model.
terval scale. The theory of riskless Economists have become worried
choices predicted from indifference about the assumption that choices
curves has been worked out in detail. are transitive. Experiments have
Experimental determination of indif- shown that intransitive patterns of
ference curves is possible, and has choice do occur, and so stochastic
been attempted. But utility meas- models have been developed which
ured on an interval scale is necessary permit occasional intransitivities.
(though not sufficient) for welfare The theory of games presents an
economics. elaborate mathematical analysis of
Attention was turned to risky the problem of choosing from among
choices by von Neumann and Mor- alternative strategies in games of
genstern's demonstration that com- strategy. This paper summarizes the
plete weak ordering of risky choices main concepts of this analysis. The
implies the existence of utility meas- theory of games has stimulated in-
urable on an interval scale. Hosteller terest in experimental games, and a
and Nogee experimentally deter- few bargaining experiments which
mined utility curves for money from can be thought of in game-theoretical
gambling decisions, and used them to terms have been performed.
predict other gambling decisions. All these topics represent a new
Edwards demonstrated the existence and rich field for psychologists, in
of preferences among probabilities in which a theoretical structure has al-
gambling situations, which compli- ready been elaborately worked out
cates the experimental measurement and in which many experiments need
of utility. Coombs developed a model to be performed.
REFERENCES
1. ALCHIAN, A, The meaning of utility 6. ALLEN, R. G. D. The nature of indiffer-
measurement. Amer. econ. Rev., 1953, ence curves. Rev. econ. Stud., 1933, 1,
43, 26-50. 110-121.
2. ALLAIS, M. Fondements d'une thebrie 7. ALLEN, R. G. D. A note on the determi-
positive des choix comportant un nateness of the utility function. Rev.
risque et critique des postulats et econ. Stud., 1934, 2, 155-158.
axiomes de 1'ecole americaine. Col- 8. ARMSTRONG, W. E. The determinateness
logue Internationale du Centre Na- of the utility function. Econ. J., 1939,
tional de la Recherche scientifique, 1952, 49, 453-467.
No. 36. 9. ARMSTRONG, W. E. Uncertainty and the
3. ALLAIS, M. Le comportement de 1'hom- utility function. Econ. J., 1948, 58,
rae ratlonnel devant le risque: critique 1-10.
des postulats et axiomes de 1'ecole 10. ARMSTRONG, W. E. A note on the theory
americaine. Econometrica, 1953, 21, of consumer's behavior. Ox}, econ.
503-546. Pap., 1950, 2, 119-122.
4. ALLAIS, M. L'Extension des theories de 11. ARMSTRONG, W. E, Utility and the
1'equilibre economique gen6ral et du theory of welfare. Ox}, econ. Pap.,
rendement social au cas du risque. 1951, 3, 259-271.
Econometrica, 1953, 21, 269-290. 12. ARROW, K. J. Alternative approaches to
5. ALLAIS, M. La psychologic de 1'homme the theory of choice in risk-taking
rationnel devant le risque: La theorie situations. Econometrica, 1951, 19,
et I'exp6rience. J, soc. Statist,, Paris, 404-437.
1953, 94, 47-73. 13. ARROW, K. J. An extension of the basic
412 WARD EDWARDS
theorems of classical welfare econom- 31. BOREL, E. La thforie du jeu et les equa-
ics. In J. Neyman (Ed.), Proceedings tions integrates & noyau symfitrique.
of the second Berkeley symposium on C. R. Acad. Set., Paris, 1921, 173,
mathematical statistics and probability. 1304-1308. (Trans, by L. J. Savage in
Berkeley: Univ. of Calif. Press, 1951. Econometrica, 1953, 21, 97-100.)
Pp. S07-S32. 32. BOREL, E. Sur les jeux oil interviennent
14. ARROW, K. J. Social choice and indi- 1'hasard et I'habilitfi des joueurs. In
vidual values. New York: Wiley, 1951. E. Borel, Theorie des probability.
15. ARROW, K. J., BLACKWELL, p., & Paris: Librairie Scientifique, J. Her-
GIRSHICK, M. A. Bayes and minimax mann, 1924. Pp. 204-224. (Trans, by
solutions of sequential decision prob- L. J. Savage in Econometrica, 1953, 21,
lems. Econometrica, 1949, 17, 213-244. 101-115.)
16. ASCH, S. E. Social psychology. New 33. BOREL, E. Algebre et calcul des proba-
York: Prentice-Hall, 1952. bilitfe. C. R. Acad. Sci., Paris, 1927,
17. ATTNEAVE, F. Psychological probability 184, 52-53. (Trans, by L. J. Savage in
as a function of experienced frequency. Econometrica, 1953,21, 116-117.)
/. exp. Psychol, 1953, 46, 81-86. 34. BOREL, E. Traitt du calcul des probabil-
18. BAUMOL, W. J. Community indifference. ites et de ses applications, applications
Rev. econ. Stud., 1946, 14, 44-48. des jeux de hasard. Vol. IV, No. 2.
19. BAUMOL, W. J. The Neumann-Morgen- Paris: Gauthier-Villars, 1938.
stern utility index—an ordinalist 35. BROSS, I. Design for decision. New
view. /. polit. Econ., 1951, 59, 61-66. York: Macmillan, 1953.
20. BAUMOL, W. J. Discussion. Amer. econ. 36. BUSH, R. R., & MOSTELLER, F. A
Rev. Suppl, 1953, 43, 415-416. mathematical model for simple learn-
21. BERGSON (BURK), A. Reformulation of ing. Psychol. Rev., 1951, 58, 313-323.
certain aspects of welfare economics. 37. BUSH, R. R., & MOSTELLER, F. A model
Quart. J. Econ., 1938, 52, 310-334. for stimulus generalization and dis-
22. BERNARDELLI, H. Note on the determi- crimination. Psychol. Rev., 1951, 58,
nateness of the utility function. Rev. 413-423.
econ. Stud., 1934, 2, 69-75. 38. CARTWRIGHT, D. Decision-time in rela-
23. BERNARDELLI, H. The end of marginal tion to differentiation of the phenome-
utility theory? Economica, 1938, 5, nal field. Psychol. Rev., 1941, 48, 425-
192-212. 442.
24. BERNARDELLI, H. A reply to Mr. 39. CARTWRIGHT, D. The relation of de-
Samuelson's note. Economica, 1939, 6, cision-time to the categories of re-
88-89. sponse. Amer. J. Psychol., 1941, 54,
174-196.
25. BERNARDELLI, H. A rehabilitation of the
40. CARTWRIGHT, D. Survey research: psy-
classical theory of marginal utility.
Economica, 1952, 19, 254-268. chological economics. In J. G. Miller
(Ed.), Experiments in social process.
26. BERNOULLI, D. Specimen theoriae New York: McGraw-Hill, 1950. Pp.
novae de mensura sortis. Comentarii 47-64.
Academiae Scientiarum Imperiales Pe- 41. CARTWRIGHT, D., & FESTINGER, L. A
tropolitanae, 1738, 5, 175-192. (Trans, quantitative theory of decision. Psy-
by L. Sommer in Econometrica, 1954, chol. Rev., 1943, 50, 595-621.
22, 23-36.) 42. CLARK, J. M. Realism and relevance in
27. BILODEAU, E. A. Statistical versus In- the theory of demand. J. polit. Econ.,
tuitive confidence. Amer. J. Psychol., 1946, 54, 347-353.
1952, 65, 271-277. 43. COOMBS, C. H. Psychological scaling
28. BISHOP, R. L. Consumer's surplus and without a unit of measurement. Psy-
cardinal utility. Quart. J. Econ., 1943, chol. Rev., 1950, 57, 145-158.
57, 421-449. 44. COOMBS, C. H. Mathematical models in
29. BISHOP, R. L. Professor Knight and the psychological scaling. /. Amer.
theory of demand. /. polit. Econ., statist. Ass., 1951, 46, 480-489.
1946, 54, 141-169. 45. COOMBS, C. H. A theory of psychologi-
30. BOHNERT, H. G. The logical structure of cal scaling. Bull. Engng Res. Inst.
the utility concept. In R. M. Thrall, Univer. Mich., 1952, No. 34.
C. H. Coombs, & R. L. Davis (Eds.), 46. COOMBS, C. H. Theory and methods of
Decision Processes. New York: Wiley, social measurement. In L. Festinger &
in press. D. Katz (Eds.), Research methods in
THEORY OF DECISION MAKING 413
the behavioral sciences. New York: An empirical test of a quantitative
Dryden, 1953. Pp. 471-S3S. theory of decision. J. exp. Psychol.,
47. COOMBS, C. H. A method for the study 1943,32,411^23.
of interstimulus similarity. Psycho- 63. FISHER, I. A statistical method for
metrika, in press. measuring "marginal utility" and
48. COOMBS, C. H., & BEARDSLEE, D. C. testing the justice of a progressive in-
Decision making under uncertainty. come tax. In J. Hollander (Ed.),
In R. M. Thrall, C. H. Coombs, & Economic essays contributed in honor of
R. L. Davis (Eds.), Decision processes. John Bates Clark. New York: Macmil-
New York: Wiley, in press. lan, 1927. Pp. 157-193.
49. COOMBS, C. H., & MILHOLLAND, J. E. 64. FLOOD, M. M. A preference experiment.
Testing the "rationality" of an indi- Rand Corp. Memo., November 1951,
vidual's decision making under un- No. P-256.
certainty. Psychometrika, in press. 65. FLOOD, M. M. A preference experiment
50. CORLETT, W. J., & NEWMAN, P. K. A (Series 2, Trial 1). Rand Corp. Memo.,
note on revealed preference and the December 1951, No. P-258.
transitivity conditions. Rev. econ. 66. FLOOD, M. M. A preference experiment
Stud., 1952, 20, 156-158. (Series 2, Trials 2, 3, 4). Rand Corp.
51. DE FINETTI, B. La prevision: ses lois Memo., January 1952, No. P-263.
logiques, ses sources subjectives. Ann. 67. FLOOD, M. M. A preference experiment
Inst. Poincare, 1937, 7, 1-68. (Series 3). Unpublished memorandum,
52. DE FINETTI, B. Recent suggestions for Rand Corporation. February 25, 1952.
the reconciliation of theories of prob- 68. FLOOD, M. M. Some experimental
ability. In J. Neyman (Ed.), Proceed- games. Rand Corp. Memo., March
ings of the second Berkeley symposium 1952, No. RM-789-1. (Revised June
on mathematical statistics and prob- 1952.)
ability. Berkeley: Univer. of Calif. 69. FLOOD, M. M. Testing organization
Press, 1951. theories. Rand Corp. Memo., Novem-
53. EDGEWORTH, F. Y. Mathematical ber 1952, No. P-312.
psychics. London: Kegan Paul, 1881. 70. FLOOD, M. M. An experimental multi-
54. EDWARDS, W. Discussion. Econometrica, ple-choice situation. Rand Corp.
1953, 21, 477. (Abstract) Memo., November 1952, No, P-313.
55. EDWARDS, W. Experiments on economic 71. FR£CHET, M. Emile Borel, initiator of
decision-making in gambling situa- the theory of psychological games and
tions. Econometrica, 1953, 21, 349- its application. Econometrica, 1953,
350. (Abstract) 21, 95-96.
56. EDWARDS, W. Information, repetition, 72. PRICKET, M.,& VON NEUMANN, J. Com-
and reinforcement as determiners of mentary on the three notes of Emile
two-alternative decisions. Amer. Borel. Economelrica, 1953, 21, 118-
Psychologist, 1953, 8, 345. (Abstract) 126.
57. EDWARDS, W. Probability-preferences 73. FRIEDMAN, M., & SAVAGE, L. J. The
in gambling. Amer. J. Psychol., 1953, utility analysis of choices involving
66, 349-364. risk. J. polit. Econ., 1948, 56, 279-304.
58. EDWARDS, W. Probability preferences (Reprinted with minor changes in
among bets with differing expected G. J. Stigler & K. E. Boulding [Eds.],
values. Amer. J. Psychol., 1954, 67, Readings in price theory. Chicago:
56-67. Richard D. Irwin, 1952. Pp. 57-
59. EDWARDS, W. The reliability of prob- 96.)
ability preferences. Amer. J. Psychol., 74. FRIEDMAN, M., & SAVAGE, L. J. The
1954, 67, 68-95. expected-utility hypothesis and the
60. ESTES, W. K. Toward a statistical measurability of utility. /. polit.
theory of learning. Psychol. Rev., Econ., 1952, 60, 463-475.
1950, 57, 94-107. 75. FRISCH, R. New methods of measuring
61. FESTINGER, L. Studies in decision: I. marginal utility. In R. Frisch, Beitrage
Decision-time, relative frequency of zur okonomischen'Theorie. Tubingen:
judgment and subjective confidence Mohr, 1932.
as related to physical stimulus differ- 76. GEORGESCU-ROEGEN, N. The pure
ences. J. exp. Psychol., 1943, 32, 291- theory of consumer's behavior. Quart.
306. J. Econ., 1936, 50, 545-593.
62. FES,TINGE;R, L. Studies in decision: II. 77- GE.QRGESCU-RQEGEN, N- The theory of
414 WARD EDWARDS
choice and the constancy of economic functions of probability and desirabil-
laws. Quart. J. Econ., 1950, 64, 125- ity of outcomes. /. Pers., 1953, 21,
138. 329-335.
78. GEORGESCU-ROEGEN, N. Utility, ex- 95. JARRETT, JACQUELINE M. Strategies in
pectations, measurability, prediction. risk-taking situations. Unpublished
Paper read at Econometric Soc., Ph.D. thesis, Harvard Univer., 1951.
Kingston, September, 1953. 96. JENKINS, W. O., & STANLEY, J. C., JR.
79. GIRSHICK, M. A., & BLACKWELL, D. Partial reinforcement: a review and
Theory of games and statistical de- critique. Psychol. Bull., 1950, 47, 193-
cisions. New York: Wiley, 1954. 234.
80. GOOD, I. J. Probability and the weighing 97. JOHNSON, W. E. The pure theory of
of evidence. London: Griffin, 1950. utility curves. Econ. J., 1913, 23, 483-
81. GRIFFITH, R. M. Odds adjustments by 513.
American horse-race bettors. Amer. 98. KALDOR, N. Welfare propositions and
J. Psychol., 1949, 62, 290-294. inter-personal comparisons of utility.
82. HARSANYI, J. C. Cardinal utility in wel- Econ. J., 1939, 49, 549-552.
fare economics and in the theory of 99. KALDOR, N. A comment. Rev. econ.
risk-taking. /. polit. Econ., 1953, 61, Stud., 1946, 14, 49.
434-435. 100. KAPLAN, A., & RADNER, R. A question-
83. HART, A. G. Risk, uncertainty, and the naire approach to subjective prob-
unprofitability of compounding prob- ability—some experimental results.
abilities. In 0. Lange, F. Mclntyre, Working Memorandum 41, Santa
& T. O. Yntema (Eds.), Studies in Monica Conference on Decision Prob-
mathematical economics and econo- lems, August 15, 1952.
metrics. Chicago: Univer. of Chicago 101. KATONA, G. Psychological analysis of
Press, 1942. Pp. 110-118. business decisions and expectations.
84. HAYES, S, P., JR. Some psychological Amer. econ. Rev., 1946, 36, 44-62.
problems of economics. Psychol. Bull., 102. KATONA, G. Contributions of psycho-
1950, 47, 289-330. logical data to economic analysis. /.
85. HERSTEIN, I. N., & MILNOR, J. An Amer. statist. Ass., 1947, 42, 449-459.
axiomatic approach to measurable 103. KATONA, G. Psychological analysis of
utility. Econometrica, 1953, 21, 291- economic behavior. New York: Mc-
297. Graw-Hill, 1951.
86. HICKS, J. R. The foundations of welfare 104. KATONA, G. Rational behavior and eco-
economics. Econ. J., 1939, 49, 696- nomic behavior. Psychol. Rev., 1953,
712. 60, 307-318.
87. HICKS, J. R. Value and capital. Oxford: 105. KAUDER, E. Genesis of the marginal
Clarendon Press, 1939. utility theory from Aristotle to the
88. HICKS, J. R., & ALLEN, R. G. D. A re- end of the eighteenth century. Econ.
consideration of the theory of value. J., 1953, 63, 638-650.
Economica, 1934, 14, 52-76, 196-219. 106. KAUDER, E. The retarded acceptance of
89. HILDRETH, C. Alternative conditions for the marginal utility theory. Quart. J.
social orderings. Econometrica, 1953, Econ., 1953, 67, 564-575.
21, 81-94. 107. KAYSEN, C. A revolution in economic
90. HOUTHAKKER, H. S. Revealed prefer- theory? Rev. econ. Stud., 1946, 14, 1-
ence and the utility function. Eco- 15.
nomica, 1950, 17, 159-174. 108. KENNEDY, C. The common sense of in-
91. HULL, C. L. Principles of behavior, an difference curves. Oxf. econ. Pap.,
introduction to behavior theory. New 1950, 2, 123-131.
York: D. Appleton-Century, 1943. 109. KNIGHT, F. H. Risk, uncertainty, and
92. HURWICZ, L. The theory of economic be- profit. Boston: Houghton Mifflin,
havior. Amer. econ. Rev., 1945, 35, 1921.
909-925. (Reprinted in G. J. Stigler & 110. KNIGHT, F. H. Realism and relevance
K. E. Boulding [Eds.], Readings in in the theory of demand. /. polit.
price theory. Chicago: Richard D. Econ., 1944, 52, 289-318.
Irwin, 1952. Pp. 505-526.) 111. KNIGHT, F. H. Comment on Mr.
93. HURWICZ, L. What has happened to the Bishop's article. /. polit. Econ., 1946,
theory of games? Amer. econ. Rev. 54, 170-176.
Suppl., 1953, 43, 398^05. 112. KUHN, H. W., & TUCKER, A. W. (Eds.)
94. IRWIN, F. W. Stated expectations as Contributions to the theory of games.
THEORY OF DECISION MAKING 415
Vol. I. Ann. Math. Stud., No. 24. theory of games. New York: McGraw-
Princeton: Princeton Univer. Press, Hill, 1952.
1950. 130. MALINVAUD, E. Note on von Neumann-
113. KUHN, H. W., & TUCKER, A. W. (Eds.) Morgenstern's strong independence
Contributions to the theory of games. axiom. Econometrica, 1952, 20, 679.
Vol. II. Ann. Math. Stud., No. 28. 131. MANNE, A. S. The strong independence
Princeton: Princeton Univer. Press, assumption—gasolene blends and
1953. probability mixtures. Econometrica,
114. LANCASTER, K. A refutation of Mr. 1952, 20, 665-669.
Bernardelli. Economica, 1953, 20, 132. MARKOWITZ, H. The utility of wealth.
259-262. J. polit. Econ., 1952, 60, 151-158.
115. LANDAHL, H. D. A neurobiophysical in- 133. MARKS, ROSE W. The effect of prob-
terpretation of certain aspects of the ability, desirability, and "privilege"
problem of risks. Bull. Math. Bio- on the stated expectations of children.
physics, 1951, 13, 323-335. J. Pers., 1951, 19, 332-351.
116. LANGE, O. The determinateness of the 134. MARSCHAK, J. Neumann's and Morgen-
utility function. Rev. econ. Stud., 1933, stern's new approach to static eco-
1, 218-225. nomics. J. polit. Econ., 1946, 54, 97-
117. LANGE, O. Note on the determinateness 115.
of the utility function. Rev. econ. 135. MARSCHAK, J. Rational behavior, un-
Stud., 1934, 2, 75-77. certain prospects, and measurable
118. LANGE, O. The foundations of welfare utility. Econometrica, 1950, 18, 111-
economics. Econometrica, 1942, 10, 141.
215-228. 136. MARSCHAK, J. Why "should" statisti-
119. LANGE, O. The scope and methods of cians and businessmen maximize
economics. Rev. econ. Stud., 1945, 13, "moral expectation"? In J. Neyman
19-32. (Ed.), Proceedings of the second
120. LEWIN, K. Principles of topological psy- Berkeley symposium on mathematical
chology. New York: McGraw-Hill, statistics and probability. Berkeley:
1936. Univer. of Calif. Press, 1951. Pp. 493-
121. LEWIN, K. Behavior and development 506.
as a function of the total situation. 137. MARSHALL, A. Principles of economics.
In L. Carmichael (Ed.), Manual of (8th Ed.) New York: Macmillan, 1948.
child psychology. New York: Wiley, 138. MAY, K. O. Transitivity, utility, and
1946. Pp. 791-844. aggregation in preference patterns.
122. LEWIN, K., DEMBO, TAMARA, FESTINGER, Econometrica, 1954, 22, 1-13.
L., & SEARS, PAULINE S. Level of 139. MELVILLE, L. G. Economic welfare.
aspiration. In J. McV. Hunt (Ed.), Econ. J., 1939, 49, 552-553.
Personality and the behavior disorders. 140. MISHAN, E. J. The principle of com-
Vol. I. New York: Ronald, 1944. Pp. pensation reconsidered. J. polit. Econ.,
333-378. 1952,60,312-322.
123. LEWISOHN, S. A. Psychology in eco- 141. MORGAN, J. N. Can we measure the
nomics. Polit. Sci. Quart., 1938, 53, marginal utility of money? Econo-
233-238. metrica, 1945, 13, 129-152.
124. LITTLE, I. M. D. The foundations of 142. MOSTELLER, F., & NoGEE, P. An experi-
welfare economics. Oxf. econ. Pap., mental measurement of utility. /.
1949, 1, 227-246. polit. Econ., 1951, 59, 371-404.
125. LITTLE, I. M. D. A reformulation of the 143. NEISSER, H. The strategy of expecting
theory of consumer's behavior. Oxf. the worst. Soc. Res., 1952, 19, 346-
econ. Pap., 1949, 1, 90-99. 363.
126. LITTLE, I. M. D. The theory of con- 144. NEYMAN, J., & PEARSON, E. S. The test-
sumer's behavior—a comment. Oxf. ing of statistical hypotheses in rela-
econ. Pap., 1950, 2, 132-135. tion to probability a priori. Proc.
127. LITTLE, I. M. D. Social choice and indi- Cambr. phil. Soc., 1933, 29, 492-510.
vidual values. /. polit. Econ., 1952, 145. PAPANDREOU, A. G. An experimental
60, 422H132. test of an axiom in the theory of
128. MACFIE, A. L. Choice in psychology and choice. Econometrica, 1953, 21, 477.
as economic assumption. Econ. J., (Abstract)
1953, 63, 352-367. 146. PARETO, V. Manuale di economic, po-
129. McKiNSEY, J. C. C. Introduction to the litica, con una introduzione utta scienta
416 WARD EDWARDS
sociale. Milan, Italy: Societa Editrice 165. SAMUELSON, P. A. Consumption theory
Libraria, 1906. in terms of revealed preference. Eco-
147. PHELPS-BROWN, E. H. Note on the de- nomica, 1948, 15, 243-253.
terminateness of the utility function. 166. SAMUELSON, P. A. Evaluation of real
Rev. econ. Stud., 1934, 2, 66-69. national income. Oxf. econ. Pap.,
148. PIGOU, A. C. Some aspects of welfare 1950, 2, 1-29.
economics. Amer. econ. Rev., 1951, 41, 167. SAMUELSON, P. A. The problem of in-
287-302. tegrability in utility theory. Eco-
149. PRESTON, M. G., & BARATTA, P. An nomica, 1950, 17, 355-385.
experimental study of the auction- 168. SAMUELSON, P. A. Probability, utility,
value of an uncertain outcome. Amer. and the independence axiom. Econo-
J. Psychol., 1948, 61, 183-193. metrica, 1952, 20, 670-678.
150. RAMSEY, F. P. Truth and probability. 169. SAMUELSON, P. A. Consumption theo-
In F. P. Ramsey, The foundations of rems in terms of overcompensation
mathematics and other logical essays. rather than indifference comparisons.
New York: Harcourt Brace, 1931. Economica, 1953, 20, 1-9.
151. RICCI, U. Pareto and pure economics. 170. SAVAGE, L. J. The theory of statistical
Rev. econ. Stud., 1933, 1, 3-21. decision. J. Amer. statist. Ass., 1951,
152. ROBBINS, L. Interpersonal comparisons 46, 55-67.
of utility: a comment. Econ. J., 1938, 171. SAVAGE, L. J. An axiomatic theory of
48, 635-641. reasonable behavior in the face of un-
153. ROBBINS, L. Robertson on utility and certainty. Unpublished manuscript,
scope. Economica, 1953, 20, 99-111. Statistical Research Center, Univer.
154. ROBERTSON, D. H. Utility and all that of Chicago, No. SRC-21222S14.
and other essays. London: George 172. SCHULTZ, H. The theory and measure-
Allen & Unwin, 1952. ment of demand. Chicago: Univer. of
155. ROTHENBERG, J. Conditions for a social Chicago Press, 1938.
welfare function. /. polit. Econ., 1953, 173. SCITOVSKV, T. A note on welfare propo-
61, 389-405. sitions in economics. Rev. econ. Stud.,
156. ROTHSCHILD, K. W. The meaning of 1941, 9, 77-88.
rationality: a note on Professor Lange's 174. SCITOVSKY, T. The state of welfare eco-
article. Rev. econ. Stud., 1946, 14, 50- nomics. Amer. econ. Rev., 1951, 41,
52. 303-315.
157. ROUSSEAS, S. W., & HART, A. G. Ex- 175. SHACKLE, G. L. S. Expectations in eco-
perimental verification of a composite nomics. Cambridge, Eng.: Cambridge
indifference map. J. polit. Econ., 1951, Univer. Press, 1949.
59,288-318. 176. SHAPLEY, L. S., & SHUBIK, M. Solutions
158. SAMUELSON, P. A. A note on measure- of n-person games with ordinal utili-
ment of utility. Rev. econ. Stud., 1937, ties. Econometrica, 1953, 21, 348-349.
4,155-161. (Abstract)
159. SAMUELSON, P. A. Empirical implica- 177. SLUTSKY, E. E. Sulla teoria del bilancio
tions of utility analysis. Econometrica, del consumatore, Giornale degli econo-
1938, 6, 344-356. misti, 1915, SI, 1-26. (Trans, by O.
160. SAMUELSON, P. A. A note on the pure Ragusa and reprinted in G. J. Stigler
theory of consumer's behavior. Eco- & K. E. Boulding [Eds.], Readings in
nomica, 1938, S, 61-71. price theory. Chicago: Richard D.
161. SAMUELSON, P. A. A note on the pure Irwin, 1952. Pp. 27-56.)
theory of consumer's behavior. An 178. SPROWLS, R. C. Psychological-mathe-
addendum. Economica, 1938, S, 353- matical probability in relationships of
354. lottery gambles. Amer. J. Psychol.,
162. SAMUELSON, P. A. The numerical repre- 1953, 66, 126-130.
sentations of ordered classifications 179, STIGLER, G. J. The limitations of sta-
and the concept of utility. Rev. econ. tistical demand curves. J. Amer.
Stud., 1938, 6, 65-70. statist. Ass., 1939, 34, 469-481.
163. SAMUELSON, P. A. The end of marginal 180, STIGLER, G. J. The development of
utility: a note on Dr. Bernardelli's utility theory. /. polit. Econ., 1950,
article. Economica, 1939, 6, 86-87. 58, 307-327, 373-396.
164. SAMUELSON, P. A. Foundations of eco- 181, STONE, J. R. N. The theory of games.
nomic analysis. Cambridge, Mass.: Econ. J., 1948, 58, 185-201.
Harvard Univer. Press, 1947. 182. STONE, R. (J. R. N.) The role of measure-
THEORY OF DECISION MAKING 417
ment in economics. Cambridge, Eng.: Theory of games and economic behavior.
Cambridge Univer. Press, 1951. (1st Ed.) Princeton: Princeton Univer.
183. STROTZ, R. H. Cardinal utility. Amer. Press, 1944.
econ. Rev. SuppL, 1953, 43, 384-405. 197. VON NEUMANN, J., & MORGENSTERN, O.
184. SWEEZY, A. R. The interpretation of Theory of games and economic be-
subjective value theory in the writings havior. (2nd Ed.) Princeton: Prince-
of the Austrian economists. Rev. econ. ton Univer. Press, 1947.
Stud., 1933, 1, 176-185. 198. WALD, A. Statistical decision functions.
185. THURSTONE, L. L. The indifference New York: Wiley, 1950.
function. /. soc. Psychol., 1931, 2, 199. WALKER, K. F. The psychological as-
139-167. sumptions of economics. Econ. Rec.,
186. THURSTONE, L. L. The measurement of 1946, 22, 66-82.
values. Psychol. Rev., 1954, 61, 47-58. 200. WALLIS, W. A., & FRIEDMAN, M. The
187. TINTNER, G. The theory of choice under empirical derivation of indifference
subjective risk and uncertainty. Econ- functions. In O. Lange, F. Mclntyre,
ometrica, 1941, 9, 298-304. & T. O. Yntema, (Eds.), Studies in
188. TINTNER, G. A contribution to the non- mathematical economics and economet-
static theory of choice. Quart. J. rics. Chicago: Univer. of Chicago
Econ., 1942, 56, 274-306. Press, 1942.
189. T6RNQVIST, L. On the economic theory 201. WECKSTEIN, R. S. On the use of the
of lottery-gambles. Skand. Aktuar- theory of probability in economics.
Tidskr., 1945, 28, 228-246. Rev. econ. Stud., 1953, 20, 191-198.
190. VAIL, S. V. Expectations, degrees of be- 202. WEISSKOPF, W. A. Psychological as-
lief, psychological probabilities. Un- pects of economic thought. /. polit.
published manuscript Univer. of Mich- Econ., 1949, 57, 304-314.
igan, Seminar on the Application of 203. WELDON, J. C. A note on measures of
Mathematics to the Social Sciences, utility. Canad. J. Econ. polit. Sci.,
October 23, 1952. 1950, 16, 227-233.
191. VAIL, S. V. A stochastic model of utili- 204. WOLD, H. A synthesis of pure demand
ties. Unpublished manuscript, No. 24, analysis. Part I. Skand. Aktuar~
Univer. of Michigan, Seminar on the Tidskr., 1943, 26, 85-118.
Applications of Mathematics to the 205. WOLD, H. A synthesis of pure demand
Social Sciences, April 23, 1953. analysis. Part II. Skand. Aktuar-
192. VAIL, S. V. Alternative calculi of sub- Tidskr., 1943, 26, 220-263.
jective probabilities. In R. M. Thrall, 206. WOLD, H. A synthesis of pure demand
C. H. Coombs, & R. L. Davis (Eds.), analysis. Part III. Skand. Aktuar-
Decision processes. New York: Wiley, Tidskr., 1944, 27, 69-120.
in press. 207. WOLD, H. Ordinal preferences or cardi-
193. VICKREY, W. S. Measuring marginal nal utility? Econometrica, 1952, 20,
utility by reactions to risk. Econo- 661-664.
metrica, 1945, 13, 319-333. 208. WOLD, H., & JUR^EN, L. Demand analy-
194. VINER, J. The utility concept in value sis. New York: Wiley, 1953.
theory and its critics. /. polit. Econ., 209. ZEUTHEN, F. On the determinateness of
1925, 33, 369-387, 638-659. the utility function. Rev. econ. Stud.,
195. VON NEUMANN, J. Zur Theorie der 1937, 4, 236-239.
Gesellschaftsspiele. Math. Ann., 1928,
100, 295-320.
196. VON NEUMANN, J., & MORGENSTERN, O. Received for early publication A pril 8, 1954.