The Economic Journal, 117 (January), 146–173. Ó The Author(s). Journal compilation Ó Royal Economic Society 2007. Published
by Blackwell Publishing, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.
THE ROAD NOT TAKEN: HOW PSYCHOLOGY WAS
REMOVED FROM ECONOMICS, AND HOW IT MIGHT BE
BROUGHT BACK*
Luigino Bruni and Robert Sugden
This article explores parallels between the debate prompted by Pareto’s reformulation of choice
theory at the beginning of the twentieth century and current controversies about the status of
behavioural economics. Before Pareto’s reformulation, neoclassical economics was based on theoretical and experimental psychology, as behavioural economics now is. Current Ôdiscovered preferenceÕ defences of rational-choice theory echo arguments made by Pareto. Both treat economics as a
separate science of rational choice, independent of psychology. Both confront two fundamental
problems: to find a defensible definition of the domain of economics, and to justify the assumption
that preferences are consistent and stable.
One of the most significant developments in economics over the last two decades has
been the growth of behavioural economics, which draws on the theoretical and
methodological approaches of psychology in explaining economic phenomena.
Behavioural economists take pride in grounding their explanations on empirical
hypotheses about how human beings really think and act, rather than on deductions
from a priori assumptions about rational choice, and in subjecting those hypotheses to
experimental test. Viewed in historical perspective, behavioural economists are trying
to reverse a fundamental shift in economics which took place from the beginning of
the twentieth century: the ÔParetian turnÕ.
This shift, initiated by Vilfredo Pareto and completed in the 1930s and 1940s by John
Hicks, Roy Allen and Paul Samuelson, eliminated psychological concepts from
economics by basing economic theory on principles of rational choice. From then to
the 1980s, almost all the main lines of development in economic theory were aimed at
extending the power and scope of rationality-based models. For example, a major
concern from the early 1950s was to extend the theory of rational choice to deal with
risk and uncertainty. The ÔmicrofoundationsÕ and Ôrational expectationsÕ literatures
sought to replace the psychologically and empirically-based assumptions of Keynesian
macroeconomics with assumptions about the preferences and beliefs of rational agents.
The new sub-disciplines of public choice, law and economics, and institutional economics extended rational-choice modelling to areas of social life that had previously
been thought of as non-economic. The work of John Harsanyi and John Rawls initiated
a literature in which social philosophy was grounded in rational individual choice.
Game theory, interpreted as modelling the strategic interactions of ideally rational
agents, was seen as the logical completion of rational-choice theory; there was a
widespread hope that, by providing a universal theory of choice, game theory would
* An earlier version of this article was presented to a seminar at the Centre for the Philosophy of the
Natural and Social Sciences at the London School of Economics. We thank the participants in that seminar,
and Nicolò Bellanca, Ken Binmore, Robin Cubitt, Ivan Moscati, Chris Starmer and three anonymous referees
for comments and advice. Robert Sugden’s work was supported by the Leverhulme Trust.
[ 146 ]
[ J A N U A R Y 2007 ]
THE ROAD NOT TAKEN
147
unify the social sciences. Against this background, behavioural economics stands out as
a counter-revolution.
Our premise is that there are significant parallels between current debates about the
status of behavioural economics and the debate which surrounded Pareto’s reformulation of the theory of choice. By looking at those two debates together, we believe one
can arrive at a better understanding of the fundamental question that is at issue in both
of them: should economic theory be based on assumptions about human psychology or
on assumptions about rational choice? Although we cannot help revealing our
sympathies with the behavioural approach, our aim in this article is not to act as
advocates on its behalf. Rather, we aim to elucidate the arguments that can be made for
and against these competing methodological positions.
In the official history of economics, the Paretian turn allowed the latent possibilities
of neoclassical theory to be exploited, and paved the way for the achievements of
mathematical economics from the middle years of the twentieth century.1 From the
standpoint of behavioural economics, however, this change of direction was not so
obviously progressive. It may have diverted economics from a path of development
which would have been less sharply separated from psychology, in which the concerns
of what is now behavioural economics might have been more central. That possibility –
the Ôroad not takenÕ of our title – provides the theme for our article.
1. An Overview of the Argument
It is a central claim of our article that, before the Paretian turn, neoclassical economics
was based on what was then state-of-the-art research on the psychology of sensation.
Although this research programme became less fashionable among psychologists in the
decades after Pareto’s reformulation of choice theory, it is now recognised as an
integral part of psychology; its subsequent developments are drawn on by modern
behavioural economists and by psychologists who investigate judgement and decision
making. Thus, what we see as the Ôroad not takenÕ is a potential continuation of
nineteenth-century neoclassical economics, leading in the direction of behavioural
economics.
In this respect, our interpretation of the Paretian turn – or of what Pareto turned
away from – is unconventional. Commentators who see Pareto’s reformulation as
progressive usually interpret the psychological assumptions of earlier neoclassical
economics as unscientific and redundant.2 But even commentators who are unsympathetic with the economics initiated by Pareto often see little of value in the
assumptions he sought to replace. Shira Lewin’s (1996) historical discussion of the role
of psychology in economics is an example of the latter type of interpretation. Championing a tradition which combines institutionalist economics, economic sociology and
the eclectic psychology of Henry James, Lewin sees pre-Paretian neoclassical economics
1
This Ôofficial historyÕ was first developed in the 1930s, when it seems almost as if there was a concerted
effort by a new generation of mathematical economists to induct Pareto into the hall of fame of economics.
Significantly, the first issues of both Econometrica and Review of Economic Studies began with articles on Pareto.
2
This was the first interpretation of Pareto’s revolution in the theory of choice, that of Hicks and Allen
(1934), Samuelson (1938) and Savage (1954). These works have determined subsequent readings of Pareto’s
shift in relation to earlier psychological assumptions. For a review, see Bruni and Guala (2001).
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
148
THE ECONOMIC JOURNAL
[JANUARY
as the first step down the wrong road of theoretical abstraction. On this account, Pareto
merely substituted mathematical formalism for a mechanistic form of psychology which
was already falling into disrepute among genuine psychologists.
If, however, we are right that nineteenth-century neoclassical economics rested on
a viable programme of scientific psychology, the relationship between this form of
economics and Pareto’s is parallel to that between behavioural and rational-choice
economics. We might therefore expect similarities between the arguments used by
Pareto in support of his reformulation and the arguments now used to defend rationalchoice theory against the behavioural challenge. Equally, we might expect similarities
between the arguments made by Pareto’s opponents and those made in support of
behavioural economics. In the course of the article, we will identify many such similarities between the two debates.
In each debate, the advocates of psychology in economics point to a successful programme of psychologically-based economic research. In each case, their opponents offer
an alternative model of economics as a separate science, independent of psychology,
whose theories are founded on deductive reasoning about the formal properties of
rational choice. We will give particular attention to those current defences of rationalchoice modelling which use the concept of Ôdiscovered preferenceÕ, and to parallel
arguments used by Pareto. The essential idea behind the discovered preference hypothesis is that rational-choice theory is descriptive of the behaviour of economic agents
who, through experience and deliberation, have learned to act in accordance with their
underlying preferences; deviations from that theory are interpreted as short-lived errors.
We will show that Pareto’s conception of economics as a separate science of Ôlogical actionÕ
rests on a similar idea. There are corresponding similarities between the counter-arguments used by Pareto’s opponents and by current behavioural economists. These counter-arguments identify significant problems for the discovered preference approach.
In Section 2, we provide evidence for our claims about the scientific status of the
psychological assumptions of neoclassical economics, prior to Pareto. We show that
economists who followed this approach made serious attempts to use psychological
principles in their work, and were aware of some of the limitations of conventional
theory that are now part of the subject matter of behavioural economics. In the light of
this interpretation of early neoclassical economics, Section 3 reviews Pareto’s arguments for separating economics from psychology. We note how narrowly Pareto has to
define the domain of economics in order to justify his approach, and how conscious he
is of the need to show – and how difficult he finds it to show – that the conditions that
define logical action also ensure that each individual’s preferences take the form of an
ordering. In Section 4, we review the discovered preference defence of rational-choice
theory, and note parallels with Pareto’s arguments. Section 5 reviews the evidence for
and against the discovered preference hypothesis. In Section 6, we look at some of the
main criticisms that behavioural economists have made of the discovered preference
hypothesis, and show that these correspond with the difficulties faced by Pareto.
2. The Role of Psychology in Early Neoclassical Economics
The debate initiated by Pareto’s reformulation of choice theory took place against the
background of an ongoing programme of research at the boundaries of economics and
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
2007 ]
THE ROAD NOT TAKEN
149
psychology. This programme was not, as behavioural economics is today, a selfconsciously distinct branch of the discipline: it was a central component of neoclassical
economics. Neoclassical economics and experimental psychology were both relatively
young enterprises, and the boundary between them was not sharply defined. According
to what was then the dominant interpretation, neoclassical theory was based on
assumptions about the nature of pleasure and pain. Those assumptions were broadly
compatible with what were then recent findings in psychophysics. Neoclassical economists could and did claim that their theory was scientific by virtue of its being
grounded in empirically-verified psychological laws.
In documenting these claims, we focus on three early neoclassical economists. The
first is William Stanley Jevons, whose Theory of Political Economy (1871/1970) is one of
the canonical texts of the Ômarginal revolutionÕ. The second is Francis Ysidro Edgeworth, whose Mathematical Psychics (1881/1967) is another canonical text, and is particularly sophisticated in its use of current psychological research. Our third theorist is
Maffeo Pantaleoni. Pantaleoni’s Pure Economics (1889/1898) is perhaps less original
than the books of Jevons and Edgeworth; but Pantaleoni has particular relevance to our
article as a leading economist of his day who engaged in a prolonged debate with
Pareto.3
The usual methodology in economics at this time was John Stuart Mill’s concrete
deductive method, by which theories about economic phenomena are arrived at by
deduction from a set of relatively simple empirical regularities or ÔlawsÕ in which (it
is claimed) the theorist can have great confidence.4 Some of these laws (for
example, the law of diminishing returns) were alleged properties of what would now
be called production functions. Others – more important in the context of this
article – were alleged properties of human psychology. These laws were interpreted
as tendencies in the human mind and in human behaviour – tendencies that are very
general and robust, but which interact with many other causal factors in determining behaviour in any particular economic environment. Thus the theories
deduced from those laws, when applied in any concrete setting, could be expected
to generate only inexact predictions; but if the theories were used with suitable
awareness of the factors that were not taken into account, their predictions would
be accurate to a reasonable degree of approximation in many applications. Mill
(1843/1967, Book 6, Ch. 3, pp. 586–589) himself, in a chapter cautiously entitled
ÔThat there is, or may be, a science of human natureÕ, used the theory of ocean
tides as an analogy for explanation in social science. That theory explains the tidal
effects of the sun and moon on the basis of deductions from the laws of gravitation,
while leaving out the more complex effects induced by the interaction of those laws
with particular oceanographical features. The early neoclassical economists, with
more grandiose scientific pretensions, drew analogies between theories of human
psychology and pure mechanics.5
3
Throughout this Section, references to Jevons, Edgeworth and Pantaleoni are to these three books.
For example, this method is endorsed explicitly by Jevons (pp. 87–91). Similarly, Pantaleoni defines
economics as Ôthe laws of wealth systematically deduced from the hypothesis that men are actuated exclusively
by the desire to realise the fullest satisfaction of their wants, with the least possible individual sacrificeÕ (p. 3).
5
The significance of analogies with physics in early neoclassical economics is discussed by Mirowski (1989).
4
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
150
THE ECONOMIC JOURNAL
[JANUARY
For the early neoclassical economists, economics rested on the fundamental
assumption that individuals act on self-interest. Thus, Jevons describes his theory as Ôthe
mechanics of utility and self-interestÕ (p. 90); Edgeworth’s Ôeconomical calculusÕ begins
with the famous declaration that ÔThe first principle of economics is that every agent is
actuated only by self-interestÕ (p. 16). This assumption was not to be treated literally, as
a claim about the actual motivation of all human beings in all situations. Rather, it was a
defining characteristic of economics as a pure deductive science. It was intended to
represent a core tendency in human motivation, which (it was claimed) had a dominant influence in those areas of human life studied by economists. As Jevons put it,
trying to justify his theory’s neglect of ethics, economics is concerned only with Ôthe
lowest rank of feelingsÕ (p. 93). Pantaleoni offered an evolutionary justification for
assuming self-interest, arguing that a powerful tendency to pursue pleasure and to
avoid pain is implanted in our species by natural selection (pp. 9–16).
Given the assumption of self-interest, economic theories of behaviour were to be
deduced from psychological laws about human wants, which in turn were understood
in terms of the pursuit of pleasure and the avoidance of pain. Pleasure and pain were
treated as sensations, of which the person who experiences them has direct knowledge.
By introspection, and by the study of other people’s reports of their introspections, an
investigator could arrive at knowledge of the laws governing pleasure and pain. For the
neoclassical economists, the most significant of these laws concerned the relationship
between stimuli and sensations. In slightly different ways, these economists advanced
the hypothesis that, as the amount of any stimulus increases, the increment of sensation
produced by a given increment of stimulus falls. The law of diminishing marginal utility
– that as consumption of any commodity increases, the increment of pleasure produced by a given increment of consumption falls – was seen as a special case of this
more general law of psychology.
Thus, one of the most fundamental principles of neoclassical economics was
grounded in a hypothesis about phenomena that are now seen as belonging to the
domain of psychology rather than economics. Consider, for example, the writer cited
by Jevons as having Ômost clearly appreciated the nature and importance of the law of
utilityÕ, Richard Jennings.6 Jevons quotes from Jennings’s book, Natural Elements of
Political Economy, published in 1855:
To turn from the relative effect of commodities, in producing sensations, to
those which are absolute, or dependent only on the quantity of each commodity, it is but too well known to every condition of men, that the degree of
sensation which is produced, is by no means commensurate with the quantity
of the commodity applied to the senses . . .
We may gaze upon an object until we can no longer discern it, listen until we
can no longer hear, smell until the sense of odour is exhausted . . . (quoted by
Jevons, p. 113)
6
In the preface to the 1879 edition of The Theory of Political Economy, Jevons acknowledges his recent
discovery that Hermann Heinrich Gossen, in a book published in 1854, Ôhas completely anticipated me as
regards the general principles and method of the theory of economicsÕ. Jevons characterises Gossen’s theory
as resting on a Ônatural law of pleasureÕ (p. 60–4).
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
2007 ]
THE ROAD NOT TAKEN
151
As the second sentence in this quotation makes clear, this is a general hypothesis
about diminishing sensitivity to sensory stimuli, and not simply about tastes for
economic commodities.
Contemporaneously with the development of neoclassical economics, the hypothesis
of diminishing sensitivity was being formulated and tested by people who are now seen
as founders of scientific psychology. Gustav Theodor Fechner proposed the revolutionary thesis that mental phenomena could be measured by finding quantitative
relationships between material stimuli and mental sensations. This idea was the foundation of the research programme of psychophysics. The publication in 1860 of Fechner’s book Elemente der Psychophysik is often considered as the birth of modern
psychology. Fechner’s work was based on a regularity found by the physiologist Ernst
Heinrich Weber: the change in magnitude of a given stimulus that produces a just
noticeable change in sensation is proportional to the total stimulus. Treating a just
noticeable change as the unit of sensation, Fechner used this regularity as a method of
measuring the magnitude of sensations: according to the Fechner-Weber law, if stimulus
increases geometrically, sensation increases arithmetically. This research programme
was continued by Wilhelm Wundt, who is generally regarded as the first experimental
psychologist. Wundt’s most influential book, Grundzuge der physiologischen Psychologie
(1874) – published at the moment of the marginal revolution in economics – proposed
the scientific investigation of people’s introspections about their experiences of
consciousness.
In understanding the relationship between psychology and economics at this time, it
is important to recognise that, in both disciplines, introspection was treated as a
legitimate source of data. The behaviourist movement in psychology, which denied the
scientific status of introspection, dates only from the second decade of the twentieth
century. (The founding text of behaviourism, John Broadus Watson’s paper ÔPsychology as the behaviourist views itÕ, was published in 1913.) And in assessing, with the
benefit of hindsight, the potentialities of the nineteenth-century form of psychologically-based economics, we must remember that the methodology of psychology is no
longer narrowly behaviourist.7 Modern psychology standardly uses ÔconstructsÕ which
have no directly observable correlates but which are treated as valid to the extent that
they play a useful role in explanatory theories; constructs (such as happiness) which
refer to individualsÕ affective states are in general use. The research programme of
psychophysics initiated by Fechner is an integral part of modern psychology.
As many neoclassical economists noticed, the hypothesis of diminishing marginal
utility of consumption can be interpreted as an implication of the Fechner-Weber law.8
Edgeworth was particularly aware of current work in psychophysics. Indeed, the title he
chose for his major work, Mathematical Psychics, suggests that he saw it as a contribution
to psychology as well as to economics. Edgeworth states as an axiom that ÔPleasure is
measurable, and all pleasures are commensurableÕ, by which he means not only that
7
It is perhaps unfortunate that Ôbehavioural economicsÕ has been adopted as the label for work at the
interface of psychology and economics. Behavioural economics is distinguished by an interest in actual
behaviour (as opposed to a normative ideal of rationality), but it is not behaviourist in the methodological
sense.
8
More specifically, the Fechner-Weber law could be interpreted as supporting the hypothesis advanced by
Daniel Bernoulli in the eighteenth century, that utility is a logarithmic function of wealth.
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
152
THE ECONOMIC JOURNAL
[JANUARY
different sorts of pleasures felt by one person can be measured on a single scale, but
also that the pleasures of different persons are commensurable. The method he proposes to use to establish commensurability is Ôthe Fechnerian method applied to
pleasures in generalÕ: the interpersonal unit of pleasure is the just-perceivable increment. Edgeworth appeals to the work of Fechner and Wundt to support his claim that
this method is both theoretically valid and practically feasible. He then proposes the
hypothesis that ÔThe rate of increase of pleasure decreases as its means increaseÕ; the
work of Fechner, Wundt and other psychological researchers is cited as supporting
evidence (pp. 59–62).
For a present-day reader, Edgeworth’s discussion of the determinants of pleasure has
a particularly interesting feature. Edgeworth proposes two distinct mechanisms of
diminishing sensitivity to stimuli. The first is the hypothesis we have just described,
which corresponds with the standard neoclassical concept of diminishing marginal
utility of consumption. But in addition:
But not only is the function connecting means and pleasure such that the
increase of means does not produce a proportionate increase in pleasure; but
this effect is heightened by the function itself so varying (on repetition of the
conditions of pleasure) that the same means produce less pleasure. (p. 62)
In other words, the function that specifies the amount of pleasure produced by
different quantities of consumption in any given period – the utility function for
consumption, as conventionally understood – shifts according to the individual’s consumption experiences in previous periods. The more a given pleasurable experience is
repeated, the less pleasure it gives.9 Edgeworth calls this the law of accommodation, which
he attributes to the Scottish philosopher and psychologist Alexander Bain.10 Once
again, Edgeworth cites supporting evidence from psychophysics, in this case concerning visual sensations (p. 62).
This second mechanism of diminishing sensitivity corresponds closely with one of
the most important ideas in modern behavioural economics, that preferences are reference-dependent (Tversky and Kahneman, 1991). If we take a person’s reference point to
be some weighted average of her previous consumption, Edgeworth’s hypothesis
implies that a person’s utility in any given period depends not only on the absolute
quantity she consumes of each good in that period, but also on differences or ratios
between those quantities and the corresponding quantities at the reference point. The
psychological concept of accommodation used by Edgeworth is essentially the same as
adaptation, which present-day psychologists have used to explain reference-dependence
(Kahneman and Varey, 1991, pp. 131–8). Adaptation is also one of the central ideas in
the now rapidly growing literature on the economic determinants of happiness. It
offers an explanation of the otherwise puzzling fact that, while the self-reported happiness of individuals is positively correlated with income in within-country cross-section
studies, the long-run upward trend in per capita income in western countries over the
9
Pantaleoni makes the same distinction between the two forms of diminishing sensitivity, which he
attributes to Gossen. However, he merges them into a single, conventional model of diminishing marginal
utility (pp. 28–38).
10
Bain’s analysis of pleasure and pain is also much cited by Jevons (pp. 84, 93, 97–8).
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
2007 ]
THE ROAD NOT TAKEN
153
last fifty years has no matching trend in average happiness (Easterlin, 1974; Scitovsky,
1976; Frederick and Loewenstein, 1999; Frey and Stutzer, 2002).
In fact, Edgeworth does not need to use the law of accommodation in his analysis of
markets, which focuses on exchange in a single time period. Even so, it is surely
significant that he was aware of the reference-dependence of preferences, and that he
could explain, if only in general terms, how that effect might be incorporated into his
theory. Reference-dependence had been put on the agenda of neoclassical economics,
as a well-grounded psychological hypothesis whose economic implications were open to
investigation. That hypothesis may have been one of the casualties of the Paretian turn.
The early neoclassical economists hoped to develop an analysis of pleasure which
would explain the main features of human wants. As Jevons put it, reacting against John
Stuart Mill’s assertion that there is no Ôdistinct scienceÕ of human enjoyment:
But it is surely obvious that economics does rest upon the laws of human
enjoyment; and that, if those laws are developed by no other science, they must
be developed by economists. We labour to produce with the sole object of
consuming, and the kinds and amounts of goods produced must be determined with regard to what we want to consume. Every manufacturer knows and
feels how closely he must anticipate the tastes and needs of his customers: his
whole success depends on it; and, in like manner, the theory of economics
must begin with a correct theory of consumption. (pp. 102–3)
A common concern among nineteenth-century economists was to explain how wants
vary with income. One obvious regularity – the law of variety proposed by Nassau Senior – is
that the variety of goods that a person buys tends to increase with her income. It seemed to
many economists that there was a hierarchy of wants, with ÔhigherÕ or Ômore complexÕ
wants assuming greater significance at higher levels of income. From the perspective of
early neoclassical economics, these regularities called for psychological explanations.
Jevons claims to provide just such an explanation. He explains the law of variety in
terms of diminishing marginal utility: the goods that satisfy the more basic wants have
high marginal utility at low levels of consumption, but their marginal utility declines
rapidly as consumption increases, while the goods that satisfy the higher wants have the
opposite characteristics (pp. 111–4). But, as Pantaleoni notices, there is a circularity in
this explanation. It makes a distinction between two classes of good (ÔnecessitiesÕ and
ÔluxuriesÕ), with allegedly different hedonic characteristics, and then uses this hypothesis to explain observed differences in income elasticity; but it offers no independent
criterion for identifying whether any particular good is a necessity or a luxury.
Pantaleoni responds to this challenge by proposing a theory of the Ôgenetic succession of wantsÕ, based on Ôpsychological analysis and the data of physiologyÕ. Developing
another idea from Jennings’s Natural Elements of Political Economy, Pantaleoni distinguishes between those sensations (such as temperature, hunger and thirst) that are
mediated by the ÔcommonÕ senses of the body as a whole, and those that are mediated
by the five ÔspecialÕ senses of sight, hearing, touch, taste and smell. Wants induced by
common sensations are ÔprimaryÕ; those induced by special sensations are ÔsecondaryÕ.
According to Pantaleoni’s theory, goods that satisfy primary wants have the hedonic
characteristics of Jevons’s necessities, while goods that satisfy secondary wants have the
hedonic characteristics of luxuries. More specifically, the satisfaction of one primary
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
154
THE ECONOMIC JOURNAL
[JANUARY
want is not substitutable with that of another, and the satisfaction of secondary wants is
ranked lexicographically below the satisfaction of primary wants (pp. 48–55). As far as
we know, the idea of a significant distinction between common and special sensations
has never been developed further. It may well be mistaken. Even so, it illustrates how
nineteenth-century neoclassical economists looked to psychology for explanations of
regularities in human tastes.
In the course of his discussion of the succession of wants, Pantaleoni points to what
he takes to be an established fact: that the Ôpositive expansionÕ of wants differs from the
Ônegative expansionÕ. Specifically, the tendency for increases in income to activate new
wants is stronger than the tendency for decreases in income to de-activate existing
wants. Pantaleoni recognises that this path-dependence is inconsistent with standard
economic theory. He suggests that, contrary to the assumptions of that theory, there
may in fact be a tendency for a person’s Ôhedonic scaleÕ at any given time to depend on
past consumption (pp. 53–4). What Pantaleoni is envisaging here seems similar to
Edgeworth’s concept of accommodation. Like Edgeworth, Pantaleoni does not pursue
this line of enquiry; but his psychologically-based methodology has allowed him to
entertain the hypothesis that preferences are reference-dependent and that hedonic
experience is subject to adaptation.
To sum up: for economists of the generation of Jevons, Edgeworth and Pantaleoni,
the psychology of sensation was an essential part of economics. The ideas they took
from psychology were, from the viewpoint of the time, products of a well-established
programme of psychophysical research. Although not all of those ideas have stood the
test of time, the research programme that generated them was fundamentally sound. It
would be quite wrong to think that, when economics turned away from psychology at
the beginning of the twentieth century, it was merely dumping obsolete or unscientific
intellectual baggage.
3. Pareto’s Science of ÔLogical ActionÕ
In proposing that economics should cut itself off from psychology, Pareto was proposing a major deviation from current understandings of the nature of economic
explanation. In this Section, we look at the arguments that he used to justify this move,
and at how these were received by contemporary economists. In subsequent Sections,
we will compare these arguments and counter-arguments with their present-day
analogues in the literature of discovered preference.
In proposing the separation of economics from psychology, Pareto was self-conscious
and explicit. In the introduction to the paper in which he first outlines his new
approach to the theory of choice, he claims as one of its main achievements that Ôevery
psychological analysis is eliminatedÕ (1900/1982, p. 214). In a letter dated 1897 to the
philosopher Adrien Naville, he writes:
It is an empirical fact that the natural sciences have progressed only when they
have taken secondary principles as their point of departure, instead of trying to
discover the essence of things. . . . Pure political economy has therefore a great
interest in relying as little as possible on the domain of psychology. (quoted in
Busino, 1964, p. xxiv).
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
2007 ]
THE ROAD NOT TAKEN
155
Pareto, like other economists of his day, sees himself as using Mill’s concrete
deductive method (Pareto, 1894/1966, pp. 156–7). However, he proposes that economic theories should be deduced from firmly-established empirical propositions
about choice rather than about sensation. He claims that everything that is significant in
the existing theory of economic equilibrium can be derived from the fact of indifference, without any recourse to psychological concepts such as utility or pleasure. If we
take his approach:
[T]his entire theory . . . rests on no more than a fact of experience, that is, on
the determination of the quantities of goods which constitute combinations
between which the individual is indifferent. The theory of economic science
thus acquires the rigor of rational mechanics; it deduces its results from
experience, without bringing in any metaphysical entity. (1909/1971, Ch. 3,
§36b)11
Or, as Pareto describes his position in a letter to the mathematician Herman Laurent: ÔI am not interested in the reason why man is indifferent between [one thing and
another]: I notice the pure and naked factÕ (1899/1989, p. 288).
Despite the way he uses the term ÔmetaphysicalÕ, Pareto does not deny the scientific
status of psychology, as practised at his time – that is, as an investigation of conscious
experience, as mediated by introspection. He also accepts that psychology is more
fundamental than economics, in the same sense that physics is more fundamental than
chemistry or biology: in principle, economic phenomena have psychological explanations, just as chemical phenomena have physical explanations. But, he believes, each
science makes progress by finding its own fundamental laws, and making deductions
from those laws (Ch 2, §1). His proposal is that economics should constitute itself as a
separate science – separate, in particular, from psychology and from sociology.
So what is economics to be the science of ? Economics – or at least pure economics – is
to be the science of logical action.
The distinction between ÔlogicalÕ and Ônon-logicalÕ action is fundamental to Pareto’s
methodology. Following classical physics, Pareto describes his methodology as one of
analysis and synthesis. To investigate complex social phenomena, we must break them
down into simpler components, and analyse each component separately. Analysis is the
role of science; synthesis – re-assembling the components – is essential for practice (Ch. 1,
§§21–26). When Pareto distinguishes between ÔpureÕ and ÔappliedÕ economics, pure
economics is understood as science while applied economics is understood as practice.
The implication is that applied economics might need to draw on the findings of
sociology or psychology. Nevertheless, pure economics is concerned only with ÔlogicalÕ
action.
For Pareto, an action is logical if and only if it is the result of valid instrumental
reasoning from objectively true premises. Thus, Pareto begins his exposition of Ôeconomic equilibriumÕ with the declaration that:
11
Throughout this Section, unless the contrary is stated, references to Pareto are to the Manual of Political
Economy (1909/1971). Page numbers in other quotations from Pareto refer to the Oeuvres Comple`tes (collected
works).
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
156
THE ECONOMIC JOURNAL
[JANUARY
We will study the many logical, repeated actions which men perform to procure the things which satisfy their tastes. . . . [W]e are concerned only with
certain relations between objective facts and subjective facts, principally the
tastes of men. Moreover, we will simplify the problem still more by assuming
that the subjective fact conforms perfectly to the objective fact. This can be
done because we will consider only repeated actions to be a basis for claiming
that there is a logical connection uniting such actions. (Ch 3, §1)
Pareto recognises that, as an implication of this definition of ÔlogicalÕ action, his theory
applies only to a restricted range of human behaviour: Ôby considering only one part of
man’s actions and, in addition, by assigning certain characteristics to them, we have
simplified the problem enormously. The study of these actions makes up the object of
political economyÕ (Ch. 3, §2).
So Pareto is proposing a particular definition of the domain of economic theory. Any
such proposal confronts two obvious questions. First, one can ask whether there are
operational criteria for determining whether any given class of behaviour falls in the
domain of the theory, prior to testing the theory’s predictions about that behaviour. (If
the domain of a theory were defined as the class of phenomena about which it makes
successful predictions, the theory could never be contradicted.) Second, one can ask
how far the theory’s domain, as defined by the proposal, corresponds with existing
understandings of the types of problem to which the theory can be applied. (If a
proposed definition of the domain of economics excludes significant areas of current
economic research, one might reasonably expect a convincing justification for that
exclusion.) We now consider how Pareto responds to these questions.
Notice that Pareto uses two criteria to identify logical actions in economics. The first
is that these are actions that are repeated many times. In such situations, he claims, it is
reasonable to assume that people’s beliefs about the world (Ôsubjective factsÕ) coincide
with how the world really is (Ôobjective factsÕ).12 In this respect, as we shall show in
Section 4, Pareto anticipates the discovered preference hypothesis: he restricts the
domain of economic theory to situations in which individuals have had adequate
opportunities for learning the consequences of alternative actions. The second criterion is that, for an economic action to be logical, it must be directed towards the
satisfaction of tastes through the acquisition of goods. Thus, the rationality of logical actions is
instrumental: economic actions (the buying and selling of goods and services) are the
means, while the satisfaction of tastes is the end. Taken together, Pareto’s two criteria
are intended to identify a category of actions that are instrumentally rational with
respect to objective facts. As Pareto puts it in his Treatise on General Sociology, actions are
logical to the extent that they Ôlogically conjoin means to ends not only from the
standpoint of the subject performing them, but from the standpoint of other persons
who have a more extensive knowledgeÕ (1916/1963, §150).
Pareto’s first criterion, that actions must be repeated many times in order to be
identified as logical, imposes a major restriction on the class of phenomena that economics can explain. This criterion may not have excluded much of what nineteenthcentury neoclassical economists sought to explain, but (as we shall detail in Section 4)
12
Pareto (1916/1963, §149) recognises that all human knowledge is ultimately subjective: what he calls
Ôobjective factsÕ are propositions that are generally taken to be true by well-informed people.
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
2007 ]
THE ROAD NOT TAKEN
157
it excludes a huge amount of what later generations of economists have treated as their
subject matter. Most of the economic theorists who have seen themselves as Pareto’s
followers have not applied this criterion. For many commentators, one of the great
advantages of grounding economic theory on the logic of rational choice rather than
on hedonism is that it widens the domain of the theory. Thus, reviewing the Manual of
Political Economy on its publication, Philip Wicksteed (1906, p. 817) criticises Pareto for
having restricted the domain of economics unnecessarily, by not realising how wide a
range of human behaviour can be explained in terms of indifference curves. Similarly,
Hicks and Allen, in the paper which relaunched Pareto’s ideas in the 1930s, write:
The methodological implications of [the new] conception of utility . . . are farreaching indeed. By transforming the subjective theory of value into a general
logic of choice, they extend its applicability over wide fields of human conduct.
(1934, p. 45)
As we shall argue in Section 4, the discovered preference hypothesis can be seen as a
movement back towards Pareto’s less ambitious conception of the applicability of
rational choice theory.
Pareto’s second criterion, that actions must be instrumental in order to be logical, is
essential for his larger conception of social science. Pareto is a sociologist as well as an
economist; he is trying to define distinct domains for these two sciences. At least as
much as economics, sociology deals with actions that are constantly repeated. Pareto
gives as examples a man who removes his hat whenever he enters a drawing-room, and
a Catholic who regularly attends mass. Notice that these actions satisfy Pareto’s first
criterion; just as in the case of repeated market transactions, we have reason to expect
that actorsÕ beliefs about the consequences of their actions will converge to the
objective facts. However, Pareto claims that these actions fail his second criterion.
Rather than being instrumentally rational, they are governed by norms; people perform them because, on their understanding of the world, Ôone ought to do soÕ (Ch. 2,
§2). On Pareto’s account, these actions belong to the domain of sociology, not
economics.
According to the instrumentality criterion, economic theory applies only to
choices that are directed towards the satisfaction of tastes. This immediately raises a
question about the operationality of Pareto’s definition of the domain of economics.
Considering only the naked facts of choice, how do we know whether or not a
person is motivated by her tastes? We cannot observe motivation itself, and (on
Pareto’s account) we cannot observe tastes independently of choices. Thus, we
cannot identify the situations to which the theory applies, prior to observing the
behaviour that it is intended to predict. Pantaleoni (1913/1925) drew attention to
this problem in defending the use of psychology in economics. For a hedonist
economist such as Pantaleoni, the domain of economics consists of those situations
in which individuals are motivated to seek pleasure. If – as the hedonists assert –
introspection is a valid mode of enquiry and if pleasure is measurable, we have an
operational criterion for distinguishing economic acts from non-economic ones. It is
not clear that Pareto does.
Leaving aside the problem noted by Pantaleoni, Pareto’s instrumentality criterion
presupposes some theoretical model of what it is to act to satisfy one’s tastes. If the
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
158
THE ECONOMIC JOURNAL
[JANUARY
criterion is to have content, we must be able to define a set of possible end states that
can be brought about by economic actions, and over which a person has tastes. And if
rational choice is to be defined in terms of the satisfaction of tastes, the problem of
choosing among actions so as to satisfy tastes must be well-defined. That requires that
tastes have some formal structure. Exactly what that structure is depends on how
rational choice is formalised. For Pareto, who interprets rational choice in terms of
constrained maximisation, the requirement is that tastes take the form of an ordering
over the relevant set of end states.
The need for this requirement presents Pareto with a fundamental problem: how,
consistently with his methodological position, can he explain this alleged property of
tastes? To understand his attempts to solve this problem, we must first ask what he
mean by ÔtastesÕ. As the passage about Ôobjective factsÕ and Ôsubjective factsÕ shows,
Pareto assumes that, after sufficient repetition of the relevant choice tasks, tastes
Ôconform perfectlyÕ to relevant objective facts. What kinds of facts are relevant, and what
it means for tastes to conform to those facts, are not clear. What is clear is the
assumption that, after sufficient repetition, each person has a stable system of tastes,
structured in such a way that problems of instrumentally rational choice are
well-defined.
Sometimes Pareto seems to be suggesting that a person can discover his true tastes, as
if tastes were objective facts in themselves. Thus:
A man who buys a certain food for the first time may buy more of it than is
necessary to satisfy his tastes, price taken into account. But in a second purchase he will correct his error, in part at least, and thus, little by little, will end
up by procuring exactly what he needs. We will examine this action at the time
when he has reached this state. Similarly, if at first he makes a mistake in his
reasoning about what he desires, he will rectify it in repeating the reasoning
and will end up by making it completely logical. (Ch. 3, §1)
It is tempting to read such passages as relying on an implicit assumption of
hedonism: one might think that the ÔtastesÕ or ÔdesiresÕ to which behaviour adapts
correspond with pleasure, as analysed by other economists of the time. On that
interpretation, the transitivity of the taste relation would be implied by the hedonistic
assumption that different pleasures for a given individual are commensurable on an
ordinal scale.
However, although Pareto permits a hedonistic interpretation of tastes, he does not
positively endorse it.13 To the contrary, he is explicit that he is not assuming hedonism.
Drawing a distinction between intrapersonal and interpersonal comparisons of sensations, he says that economics is principally concerned with that class of theories which
has as its object
13
How far Pareto rejected hedonism is a matter of dispute among historians of thought. Some commentators, such as Stigler (1950), argue that Pareto was inconsistent and failed to follow through the logic of
his rejection of ÔmetaphysicalÕ concepts. Bruni and Guala (2001) argue that, from 1900 on, Pareto was
consistent in eliminating psychology from the pure theory of economics. Since he did not deny the validity of an
introspectively-based analysis of sensations as psychology, his methodological position allowed him to use
hedonistic language when interpreting his theory and discussing its practical application.
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
2007 ]
THE ROAD NOT TAKEN
159
. . . to compare the sensations of a man in different situations, and to determine which of these he would chose. . . . [S]ince it is customary to assume that
man will be guided in his choice exclusively by consideration of his own
advantage, of his self-interest, we say that this class is made up of theories of
egotism. But it could be made up of theories of altruism (if the meaning of that
term could be defined rigorously), or, in general, of theories which rest on any
rule which man follows in comparing his sensations. It is not an essential
characteristic of this class of theories that a man choosing between two sensations choose the most agreeable; he could choose a different one, following a
rule which could be fixed arbitrarily. (Ch.3, §11)
Notice that Pareto is interpreting tastes as subjective comparisons of sensations. So
the set of end states over which tastes are defined is the set of possible sensations.
However, the ranking of different sensations for a given person need not be in terms of
more and less pleasure for that person. The theory assumes only that each person
compares sensations according to some fixed rule.
Given this understanding of tastes, Pareto’s problem is to justify the assumption that
tastes take the form of an ordering of sensations. What grounds do we have for this
assumption? Repetition alone does not provide an adequate explanation. The significance of repetition is that it allows the learning of what Pareto calls objective facts. The
relationship between actions and sensations is such an objective fact, and we might
expect that with repetition, individuals would become able to predict the sensations
that would result from alternative actions. But what is at issue is not individualsÕ
knowledge of the processes that induce their sensations, but the rules they use
to compare the sensations themselves. Unless we can assume that people compare
sensations as different quantities of some common objective attribute, we seem to have
no grounds for assuming that these comparisons are transitive.
Pareto was very conscious of this difficulty for his theory. In mathematical terms, this
is the integrability problem. Pareto treats a consumer’s marginal rates of substitution at
any given point in commodity space as Ôfacts of experienceÕ that in principle can be
discovered by observing the consumer’s behaviour. If there are only two goods, this
means that we can infer the slope of an indifference curve at each point; by repeated
use of this method we can plot a family of indifference curves. But if there are more
than two goods, how can we be sure that the marginal rates of substitution we observe at
the different points can be integrated into indifference surfaces? In other words: how
can we be sure that the consumer’s fixed rule – the rule that lies behind the regularities
we observe in his behaviour – takes the form of an ordering of end states? Pareto
struggled with the integrability problem for many years. He first mentions it in a letter
to Pantaleoni in 1891,14 and dedicated a good part of his energies to it in his last works
on pure economics (Pareto, 1909/1971, Appendix; Pareto, 1911/1982, pp. 597, 614).
14
In a letter to Pantaleoni dated 14 December 1891, Pareto (1984, I, p. 121) referred to the mathematician
Giovan Battista Antonelli (1886), who was the first to deal with the integrability problem. Pareto mentioned
the integrability problem again in a series of articles on demand theory, in which he specified that Ôwhen
there are only two goods, the equation [of the indifference curves] can always be integratedÕ (1892–93/1982,
p. 299, footnote).
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
160
THE ECONOMIC JOURNAL
[JANUARY
The truth is that the problem, as Pareto conceptualised it (by analogy with the problem
of ÔopenÕ and ÔclosedÕ cycles in thermodynamics), cannot be solved.15
The significance of the integrability problem for Pareto’s whole project seems to
have been missed by the theorists who took up his approach in the 1930s. Hicks (1939,
p. 19) criticises Pareto for spending so much energy on the Ômysterious problem of
open and closed cyclesÕ. According to Hicks, this problem Ôfascinates mathematicians,
but it does not seem to have any economic importance at allÕ. Of course, the problem is
swept under the carpet if one simply assumes that preferences are transitive, as became
standard practice in choice theory in the middle years of the twentieth century. But
Pareto’s methodological strategy is to create economic theories by deduction from
firmly-established empirical laws. If economics is to be a separate science, based on laws
whose truth is to be treated as axiomatic, we have to be very confident in those laws.
Otherwise, we are in danger of creating an complex structure of internally consistent
theory which has no correspondence with reality. For Pareto, it seems, local indifference
– the existence of stable marginal rates of substitution at each given point in commodity space – was a sufficiently solid fact on which to build a theory. But the idea that
indifference is transitive right across commodity space was only a speculative hypothesis, for which, at the time he was writing, no solid evidence existed.
Notice that the problem of justifying the transitivity assumption does not arise so
obviously in the hedonistic economics that Pareto is challenging. If the sensations
resulting from consumption can be measured along a single psychophysical dimension
of pleasure, it is immediately obvious that comparisons of sensations must be transitive.
Of course, the hypothesis that pleasure is one-dimensional is vulnerable to contradiction by psychophysical research, but the advocates of hedonistic economics were
entitled to treat it as an adequate working hypothesis. But, in the absence of direct
evidence of the transitivity of indifference, Pareto needs to find a theoretical derivation
of transitivity from non-psychological premises.
A similar issue arises in relation to the shape of indifference curves. Pareto recognises
that, in the overwhelming majority of cases confronted by economists, indifference
curves are convex to the origin. However, he also knows that there are occasional
anomalous cases in which indifference curves are concave (Ch. 4, §§34, 45). Presumably for this reason, he does not feel entitled to treat the convexity of indifference
curves as a firmly established empirical law. Thus, for Pareto, the fact that convexity
almost always holds can be registered only as a highly reliable but unexplained regularity in tastes. Noting the psychophysical findings of Fechner and Wundt, he explicitly
rejects their relevance in explaining convexity of preferences:
. . . in the great variety of economic uses [i.e. cases in which goods are used
in some way, but not necessarily consumed], there are many that are too far
removed from the phenomena to which Fechner’s law applies. It is better
to resort directly to experience, and the latter shows us that for a great many
uses and consumptions the elementary ophelimity [i.e. marginal utility] does
indeed diminish with an increase in the quantities consumed. (Ch. 4, §33)
15
On the integrability problem and its significance for Pareto, see Chipman et al. (1971), Mirowski (1989)
and Bruni (2002).
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
2007 ]
THE ROAD NOT TAKEN
161
Notice that Pareto does not challenge the scientific status of psychophysics but only
its external validity with respect to economic phenomena. There is an echo here of the
scepticism with which many present-day economists view the suggestion that results
from experimental psychology might be relevant for economics.
Pareto’s hedonist opponents took a contrary position. They could reply that, while
convexity is not a law of economics, the Fechner–Weber Law is a law of psychophysics,
and that that psychophysical law is useful in explaining the general tendency towards
convexity of preferences. As we showed in Section 2, the hedonists believed that other
reliable generalisations about preferences, beyond transitivity and convexity, could be
derived from psychological hypotheses. They no doubt expected that as psychology and
economics progressed, still more such generalisations would be found. Pareto’s proposal to constitute economics as a separate science seemed to them to be a perverse
refusal to use relevant data. In Pantaleoni’s words:
I claim that we cannot take away from economics the data coming from psychology. I cannot see what, by virtue of this renunciation, we gain, but I see
what we lose. . . . I cannot see the convenience of not utilising laws regarding
tastes and pains that we know to be true, and that are the reasons of economic
actions. (1913/1925, pp. 8–9)
Nevertheless, over the first half of the twentieth century, Pareto’s rationality-based
approach to the theory of choice gradually displaced the psychological approach
advocated by Pantaleoni. It seems that Pareto’s reservations about the general applicability of the concept of logical action and his concerns about the justification of
transitivity were quietly forgotten.
4. Behavioural Economics and the ÔDiscovered PreferenceÕ Defence of Rational
Choice Theory
We now move forward in time to current debates about the status of behavioural
economics.
In retrospect, the publication of Daniel Kahneman and Amos Tversky’s paper
ÔProspect TheoryÕ in Econometrica in 1979 can be seen as a defining moment for
behavioural economics. By presenting a body of experimental data which appeared to
contradict conventional economic theories of decision making, and by proposing an
alternative theory of non-rational behaviour based on psychological hypotheses,
Kahneman and Tversky challenged the prevailing methodology of economics.
Behavioural economics is one response to this challenge. It is an attempt to introduce into economics some of the theoretical and methodological approaches of psychology. Theoretically, the aim is to model economic agents in ways that take account
of the affective responses that decision problems evoke in human beings and of the
cognitive processes that are used in human decision making. Thus, for example,
behavioural economists have considered how decisions are influenced by loss aversion,
myopia and anticipated regret, and by perceptions of fairness and reciprocity; and they
have tried to model the heuristics that people use in processing probability information
and in assigning valuations to goods. Methodologically, there is an emphasis on
empirical investigation of individual behaviour, using experimental techniques
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
162
THE ECONOMIC JOURNAL
[JANUARY
modelled on those used in experimental psychology as well as more conventional
methods of empirical economics.
Other economists, however, have responded to the same challenge by defending
existing economic theory and methodology. Initially, many economists reacted to
Kahneman and Tversky’s work, and to behavioural economics more generally, by
denying that there was any case to answer. They objected to various features of the
experimental techniques customarily used by psychologists – particularly the common
practice of not providing incentives in decision tasks – as inappropriate for testing
economic hypotheses. Ex post rationalisations, consistent with conventional rationalchoice theory, could often be found for apparent anomalies observed in specific
experiments. Thus, anomalies were interpreted as artefacts of particular experimental
designs, rather than as evidence of fundamental properties of decision-making behaviour. But as experimental economics developed over the course of the 1980s and
1990s, the terms of debate gradually changed. Systematic violations of standard theory,
such as the common consequence effect, the common ratio effect, the Ellsberg paradox, preference reversal, the endowment effect, the rejection of positive offers in
ultimatum games and the choice of dominated strategies in public good games, were
replicated in experiments which controlled for the factors that previously had been
invoked in explaining anomalies as artefacts. Increasingly, economists have come to
accept that decision-making behaviour, as observed in laboratory environments,
diverges systematically from the predictions of standard theory, and that those divergences are in accord with the predictions of psychologically-based theories.16 Any
credible defence of the received theory of rational choice must take account of these
facts.
We are particularly concerned with one such defence, based on the idea of discovered
preference. The discovered preference strategy has been used by a number of leading
experimental economists, including Vernon Smith (1989, 1994), Charles Plott (1996)
and Ken Binmore (1999). The positions held by these writers are not identical but
share many common features. We shall distil these common features into a composite
argument, which we take to be broadly consistent with each writer’s own position. We
shall show that these writersÕ understandings of the relationship between psychology
and economics are remarkably similar to Pareto’s.
The general strategy of this defence is to claim that conventional theory captures
core elements of economic behaviour by isolating certain causal factors and abstracting
from the rest. The existence of systematic deviations between theoretical predictions
and real-world behaviour can then be accepted without calling into question the
validity of the ÔpureÕ theory – provided those deviations can be interpreted as resulting
from specific causal factors from which the theory has abstracted.17
16
For reviews of this evidence, see Camerer (1995; 2003, Ch. 2), Ledyard (1995) and Starmer (2000).
The discovered preference strategy must be distinguished from more pragmatic defences of rational
choice theory which appeal to parsimony. For example, John Hey and Chris Orme (1994) analyse the extent
to which alternative theoretical models of choice under risk fit a body of experimental data. They find that,
for a majority of their subjects, behaviour deviates significantly from the predictions of expected utility theory,
in the directions predicted by psychologically-based hypotheses; but, noting that the observed deviations are
relatively small, they argue that expected utility theory predicts behaviour to a Ôreasonable approximationÕ
(p. 1322). This latter kind of defence does not claim that the standard theory is valid in any specific limiting
case, but only that its predictions are accurate enough for economistsÕ purposes.
17
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
2007 ]
THE ROAD NOT TAKEN
163
This strategy has obvious precedents in natural science. To use an example which was
taken as a methodological model by many early neoclassical economists, consider how
theoretical mechanics may, on occasion, use such modelling assumptions as that a body
has mass located at a point, that a projectile meets with no air resistance and so on. For
practical purposes, these assumptions are justified on the grounds that they simplify the
analysis while permitting predictions that are correct to a reasonable approximation.
Notice, however, that this methodology rests on more than an appeal to parsimony.
The factors that are being Ôassumed awayÕ in the model are understood well enough for
us to be able to recognise circumstances in which they are (almost) absent. For
example, although no real bodies have mass but not extension, we can find ones for
which the distance between extremities is very small relative to mass. Thus, the predictions of pure mechanics can be tested by investigating what happens as we approach
the limiting cases referred to by the theory. And we can define the domain of the
theory in a non-arbitrary way, as consisting of those situations in which the idealising
assumptions are approximately true.
The discovered preference argument follows a similar methodological strategy. Its
central claim is that conventional theory describes the behaviour of individuals who
know which actions best satisfy their preferences. The theory abstracts from the processes by
which individuals discover how to satisfy their preferences: it simply assumes that those
processes, whatever they may be, have been completed. The domain of applicability of
the theory is restricted to situations in which, to an acceptable approximation, that
assumption holds. Just as pure mechanics has empirical content by virtue of operational criteria for identifying circumstances in which the impact of factors that are not
modelled (friction, air resistance and so on) is minimal, so rational-choice theory has
empirical content by virtue of operational criteria for identifying circumstances in
which the learning process can be expected to be almost at an end. Smith, Plott and
Binmore propose such criteria, intended to guide the design of experiments and the
interpretation of experimental findings.
The criteria proposed by these three writers are broadly similar. The essential
requirements are that, in relation to a given decision problem, the individuals whose
decisions are being studied have had adequate opportunities and incentives to gather
relevant information and to engage in relevant deliberation, and have faced that
problem sufficiently many times to have been able to learn by trial and error the
consequences of alternative actions. Smith, Plott and (with qualifications) Binmore
offer the hypothesis that rational-choice theory performs reasonably well in predicting
behaviour when these criteria are satisfied.18 Following Plott, we will call this the discovered
preference hypothesis.
The discovered preference hypothesis allows that, when the criteria for adequate
learning have not been satisfied, systematic violations of conventional theory may occur.
According to the hypothesis, those violations result from errors in decision making
which, given appropriate opportunities, incentives and feedback, decision makers can
18
Binmore treats his proposed criteria as preconditions for valid tests of rational-choice theory. He
maintains that some predictions of classical game theory – specifically, the prediction that Nash equilibria that
are not subgame perfect are not selected – are contradicted in valid tests. However, Binmore’s rhetoric is
mainly directed against the claim, which he attributes to Ôthe school of Kahneman and TverskyÕ, that economic theory fails in the laboratory (p. F16). See also footnote 18 below.
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
164
THE ECONOMIC JOURNAL
[JANUARY
learn to correct. It is accepted that psychologically-based theories may be able to
explain regularities in those errors. But that (it is said) does not challenge the validity
of rational-choice theory, because rational choice theory abstracts from the causal
factors that the psychologically-based theories are explaining.
Further, Smith and Plott argue that many real-world market institutions satisfy the
learning criteria that they propose. Indeed, it sometimes seems as if, for Smith and
Plott, the essential hypothesis of discovered preference is that the behaviour of agents
in repeated markets is as predicted by rational-choice theory, or (a weaker hypothesis) that
prices and traded quantities in repeated markets are as predicted by that theory.19
However, there would be an unsatisfactory element of mystery about a bald claim that
rational-choice theory explains behaviour inside but not outside repeated markets.
Such a claim would immediately prompt the question: What is special about repeated
markets? Since rational-choice theory is based on hypotheses about the preferences
and beliefs of individual agents, it seems that the most credible explanation of the
(supposed) observation that the theory makes reliable predictions about repeated
markets is that such markets induce agents individually to act according to the theory.
Then, the question becomes: What mechanisms within repeated markets induce agents
to act according to rational-choice theory? We take it that Smith and Plott are offering
the discovered preference hypothesis as an answer to that question.
It is important to realise that the discovered preference hypothesis implies much
more than that inconsistencies within the behaviour of a representative individual Ôgo
awayÕ as she accumulates experience. It is an essential part of the hypothesis that
preferences are discovered in whatever processes of learning occur, and not merely that
they are constructed. If a person’s preferences are constructed in response to the particular decision tasks that she faces, we have no reason to expect that the learning
process is path-independent, and hence no concept of underlying preferences that can
be ÔdiscoveredÕ (Plott, 1996, p. 227–8).20
This is not just a matter of semantics. Economic theory relies heavily on the method
of comparative statics, which compares alternative equilibrium states while holding
preferences constant. Thus, the consistency properties that are attributed to preferences
must hold across equilibria. If we are to treat the discovery of preferences as a
phenomenon of disequilibrium (as the discovered preference hypothesis does), we
must be able to assume that the end state of this discovery process is path and contextindependent. For example, consider the prediction of price theory that the imposition
of an excise tax on a good increases its price and reduces consumption. This depends
on the assumption that preferences are unaffected by the imposition of the tax. If the
discovered preference hypothesis is correct, individualsÕ first responses to the tax might
be quite unlike the theoretical prediction; responses will settle down only as preferences are discovered. If the comparative statics are to work, we must be able to model
this discovery process as a process of convergence to some underlying preference
ordering that is the same in both the without-tax and with-tax environments.
19
The distinction between these two hypotheses is discussed by Loomes et al. (2003).
In this respect, Binmore (personal communication) does not endorse the discovered preference
hypothesis as we have formulated it. He proposes that experience has a general tendency to eliminate
inconsistencies, but allows that learning may be path-dependent.
20
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
2007 ]
THE ROAD NOT TAKEN
165
However, the concept of discovered preference should not be interpreted so literally as to claim that preferences are never learned. It is obvious that individualsÕ
preferences can change over time, and that these changes sometimes occur in
response to experiences induced by participation in markets. Economists in general
have no need to deny that, nor do Smith, Plott and Binmore. Conventional comparative-static analysis works by holding preferences constant. This is not a general
claim about the nature of preferences but a modelling strategy which is useful to
the extent that the preferences of the relevant agents are reasonably stable with
respect to whatever experiences are induced by the specific workings of the market
being analysed. The advocates of the discovered preference hypothesis are committed only to a corresponding claim about the degree of stability of underlying
preferences.
It should by now be clear that there is a close parallel between the concept of
discovered preference and Pareto’s concept of logical action. Both concepts relate to
actions that are repeated sufficiently many times for errors to be eliminated. In each
case, there is an assumption that, after errors have been eliminated, actions will reveal
underlying preferences which satisfy standard conditions of internal consistency. The
main difference between the two concepts is that Pareto’s definition of Ôlogical actionÕ
includes an criterion of instrumentality, for which there is no explicit analogue in the
literature of discovered preference. In Section 6 we will ask whether a coherent concept
of discovered preference can dispense with that criterion.
5. Discovered Preference: The Evidence
As good experimental economists, Smith, Plott and Binmore treat the discovered
preference hypothesis as a working hypothesis, to be accepted only if it is confirmed by
the evidence. All three writers illustrate their arguments by describing cases in which,
they claim, apparent violations of rational-choice theory are compatible with the discovered preference hypothesis. Experimentally observed anomalies are attributed to
experimental designs that give insufficient opportunities and incentives for learning. In
designs which provide such opportunities, it is claimed, the frequency of anomalies
tends to fall as subjects gain experience, decaying to levels at which anomalies have
little practical significance.
There is significant dissonance here between the literature of discovered preference
and Pareto’s arguments. In the modern debate about behavioural economics, one of
the central issues is whether (or under what conditions) individual behaviour is as
predicted by standard economic theory; psychological mechanisms are invoked as
explanations of deviations from that theory. At the time that Pareto wrote, the possibility of such deviations was not a live issue. The dominant view was that the neoclassical
model of utility-maximising behaviour was descriptively valid, to a reasonable degree of
approximation and when applied within what was then seen as the domain of economics. This supposed regularity in human behaviour was thought to be explained by a
theory that was grounded in the psychology of sensation. Pareto agreed that this
regularity existed but proposed an alternative explanatory scheme. Thus, in arguing for
his reformulation of economics, Pareto did not feel the need to make systematic appeals to evidence about how economic agents behave.
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
166
THE ECONOMIC JOURNAL
[JANUARY
Because the empirical evidence bearing on the discovered preference hypothesis
has no obvious analogue in the earlier debate, we review it only briefly. We begin by
noting that, as so far advanced, that hypothesis does not specify how much opportunity and incentive for information-gathering and deliberation, or how much trialand-error learning, is sufficient for rational-choice theory to work. This might seem
to allow too much leeway for ex post rationalisations of observed behaviour: in
principle, failures of rational-choice theory can always be explained by claiming that
opportunities for deliberation and learning have been insufficient. Nevertheless, the
discovered preference hypothesis provides a useful organising framework for
empirical investigation of the effects of deliberation and experience on the predictive success of rational-choice theory. Ultimately, what matters are the quantitative impacts of specific kinds of deliberation and experience on the frequency of
specific anomalies, and one can investigate such impacts without taking any position
on how large they must be in order to confirm the discovered preference hypothesis.
A large part of the evidence cited by Smith, Plott and Binmore in support of the
discovered preference hypothesis comes from experimental designs in which preferences are induced. An induced-preference experiment is designed to investigate how
individuals behave, given that their preferences satisfy particular assumptions. For each
subject, hypothetical preferences, corresponding with the assumptions that are to be
investigated, are created by the experimenters. An incentive scheme is then put in place
which ensures that each subject’s monetary payoff from the experiment is monotonically increasing in the hypothetical preferences that have been created for her. The
experimental task is described to subjects in a way that makes the monetary payoffs as
salient as possible and discourages other motivations (such as a competitive desire to
win more than other subjects, or altruistic concerns about other subjectsÕ payoffs). Such
designs can be used to test whether laboratory subjects acting on induced preferences
behave in accordance with the predictions that economic theory makes for people with
the corresponding actual preferences.
The balance of evidence supports the following general conclusion. In the early
stages of induced-preference experiments, subjects often act contrary to theoretical
predictions; but over time, given adequate incentives, sufficiently simple decision
problems and sufficient repetition, they gravitate towards those actions that best satisfy
their induced preferences. In other words: if different actions in an experimental
environment consistently lead to different monetary payoffs, laboratory subjects who
are motivated to maximise their own payoffs can learn to choose actions which are in
fact payoff-maximising. A typical example of this general tendency is an experiment
cited by Binmore (1999, pp. F18–9), in which subjects played the same two-person zerosum game in repeated trials against changing opponents. Over successive plays, subjects
gravitated towards minimax strategies. In such a game, each player’s minimax strategy
is a payoff-maximising reply to the minimax strategy played by his opponent; so the
evidence is of gravitation towards behaviour that is payoff-maximising for each individual, given the behaviour of the others.
In addition to the evidence from induced-preference experiments, the proponents
of the discovered preference hypothesis cite evidence from two other broad classes of
experimental design.
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
2007 ]
THE ROAD NOT TAKEN
167
The first of these consists of experiments in which subjects interact with one another
in environments which provide cues for motivations other than individual payoffmaximisation – for example, other-oriented motivations such as altruism, reciprocity or
fairness. These experiments are designed to test whether subjects act on self-interest in
the presence of such cues. Examples of such experiments include ultimatum games,
prisoner’s dilemma games, and games in which individuals can make voluntary contributions to public goods. Plott (1996, pp. 233–5) cites the well-established result that,
in public goods games, voluntary contributions tend to fall as games are repeated.
Binmore (1999, pp. F19–20) cites an ultimatum game experiment in which, after
repetition, subjectsÕ behaviour became more self-interested. Plott and Binmore use
these examples to suggest that experimentally-observed behaviour that appears to
reveal other-oriented motivations may in fact result from errors which subjects learn to
correct. However, it is not at all clear that the balance of evidence supports this conclusion. For example, the decay of voluntary contributions in repeated public goods
games can be explained by the hypothesis that some subjects are self-interested while
others are motivated by reciprocity throughout the repetitions; contributions decline as
the reciprocators learn about the behaviour of their self-interested co-players (Bardsley
and Moffatt, 2005). The Ôrestart effectÕ found by James Andreoni (1988) – the tendency
for a break between rounds of play to induce an increase in contributions – is also
suggestive of a continuing motivation for reciprocity.
The second class of experiments investigates whether individualsÕ actual preferences
satisfy the various consistency properties assumed by conventional economic theory.
Such experiments typically compare subjectsÕ responses to two or more different tasks,
selected to test some consistency condition. In each task, considered in isolation,
conventional theory makes no specific predictions about how an individual will act; but
there is a prediction about how behaviour in one task relates to behaviour in another.
For example, preference reversal is an inconsistency between the ranking of two lotteries revealed in a binary choice task and the corresponding ranking revealed in two
valuation tasks. Smith (1994, pp. 117–8) and Plott (1996, pp. 229–31) both cite a
preference reversal experiment, carried out by James Cox and David Grether (1993), in
which subjectsÕ valuations were repeatedly elicited in second-price auctions. The usual
pattern of preference reversal was observed in relation to the valuations elicited in the
first auction that subjects faced; but after the auction had been run five times, the
number of reversals had fallen by about 40%, and the classic asymmetric pattern (in
which low-probability high-prize Ô$ betsÕ are given higher valuations than high-probability low-prize ÔP betsÕ, but P bets are preferred in the binary choice task) had almost
disappeared.21 Plott’s interpretation of this result is: ÔThe classical preference reversal
can be seen as a product of inexperience and lack of motivation, and it goes away with
experience in a market settingÕ (p. 231). This, he suggests, is an instance of a more
general phenomenon: apparent violations of the rationality assumptions of economic
theory, as observed in experiments, are Ôfostered by limited information, conditioned
21
Using the valuations elicited from subjects in the first of five repeated auctions, there were 24 ÔpredictedÕ
reversals and 2 ÔunpredictedÕ reversals among 60 subjects. The corresponding figures for the fifth auction
were 10 and 8 (Cox and Grether, 1996, p. 390).
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
168
THE ECONOMIC JOURNAL
[JANUARY
by the nature of perceptions and attention which, with experience under suitable
incentives, does not persistÕ (p. 229).
Again, however, other interpretations are possible. The second-price auction mechanism used by Cox and Grether to elicit valuations has the property that, after each
auction has been completed, subjects are told the market price (since subjects are
bidding to sell, this is the second lowest bid). Cox and Grether report that subjectsÕ bids
in later auctions were positively correlated with market prices in earlier auctions, and
interpret this as an indication that market mechanisms provide information which
helps individuals to eliminate inconsistencies in their preferences (p. 400). But if
subjects adjust their reported valuations in the direction of previously-observed market
prices – which are summary statistics of the reported valuations of all participants in the
auction – the adjustment process is path-dependent: preferences are being shaped by
the institution in which they are expressed, and not merely being discovered. There is
evidence from other experiments that this kind of shaping effect occurs and can work
to reduce the frequency of preference reversals as second-price auctions are repeated
(Knetsch et al., 2002; Loomes et al., 2003).
Since the publication of the papers by Smith, Plott and Binmore, the question of how
far the frequency of anomalies decays with experience has become an important focus
of research in behavioural and experimental economics. We do not have space to
review this work here, but we believe it is a fair summary to say that the evidence is
mixed, and that no general answer to the question has yet emerged. The indications so
far are that some anomalies do tend to decay with experience, while others do not. For
example, there is growing evidence that, as experience increases, disparities between
willingness-to-pay and willingness-to-accept valuations are reduced. This seems to be
due partly to people learning not to use the bargaining ploy of over-stating valuations
when selling and under-stating them when buying, and partly due to their becoming
less loss-averse as they gain experience of selling (List, 2003; Loomes et al., 2003; Plott
and Zeiler, 2005). In contrast, the tendency for stated valuations to be influenced by
salient but objectively irrelevant ÔcuesÕ has been found to persist despite repeated
experience of trade and consumption (Ariely et al., 2003).
6. Discovered Preference: Methodological Controversies
The discovered preference hypothesis has generally met with scepticism from behavioural economics. Among such responses are papers by Colin Camerer (1996),
Kahneman (1996), George Loewenstein (1999), Graham Loomes (1999), Chris Starmer (1999) and Robin Cubitt, Starmer and Robert Sugden (2001). These commentators have pointed to the possibility of alternative interpretations of the evidence of
the effects of experience, along the lines we sketched in Section 5. More significantly
for our comparative purposes, they have made theoretical and methodological criticisms of the discovered preference hypothesis which parallel problems faced by Pareto.
We focus on two particularly important criticisms.
The first of these concerns the limits that the discovered preference hypothesis places
on the domain of rational-choice theory. If the theory applies only to decision problems
that have been repeated many times, many economically significant decisions – for
example, choices about education, between alternative careers, about buying and selling
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
2007 ]
THE ROAD NOT TAKEN
169
homes – lie outside the domain of the theory. While some decisions that involve known
risks (for example, modest gambling on games of chance) fall within the domain, all
problems of choice under uncertainty (interpreted as referring to events for which relative frequencies are unknown) must also be excluded. Public goods are outside the
domain of the theory too. The standard economic theory of public goods assumes that
individuals have consistent preferences with respect to public goods; but since those
preferences are not directly revealed in any decision problems faced by those individuals,
there is no mechanism for the correction of error.
It should be said that the proponents of the discovered preference hypothesis
acknowledge that it implies the need for some serious rethinking about the scope of
conventional economic theory. For example, Plott (1996, p. 226) accepts that Ônew tasksÕ –
decision problems that are faced without experience – Ôabound in economicsÕ. Binmore
(1999, p. F17) notes that he is proposing significant restrictions on the domain of economic theory, and asks rhetorically: ÔBut have we [economists] not got ourselves into
enough trouble already by claiming vastly more than we can deliver?Õ He goes on to deny
believing that consumer theory is relevant to the behaviour of customers buying low-cost
items in supermarkets. One might have thought that such behaviour would be a paradigm
case of consumer choice but Binmore claims that the theory applies only if customers can
Ôfind the time to research the value of the products on saleÕ. Still, the radical implications
of these domain restrictions give pause for thought. The implication is that, if economics
is to define itself as the application of rational-choice theory to situations of repeated
choice, it must retreat from much of its established territory.
The extent of retreat that is now required is much greater than it was when Pareto
proposed defining economics in terms of logical action. One might say that this is the
consequence of the fact that twentieth-century economists ignored Pareto’s reservations about the applicability of rational-choice theory: if Pareto was right, large parts of
the work of his successors were misguided. But one might reach a different conclusion:
that economics has found it needs theories of behaviour and preference which apply to
situations other than those of repeated choice, and so cannot restrict itself to the
analysis of discovered preferences.
The second criticism of the discovered preference approach is that it lacks an adequate theoretical explanation of the consistency properties that it attributes to the
underlying preferences that are ÔdiscoveredÕ. Of course, whether the discovered preference hypothesis is true or false is ultimately an empirical matter; if it turns out to be
confirmed by the evidence, behavioural economists will have to accept it. But if it is
true, it identifies a major regularity in human behaviour which calls out for an
explanation. Conversely, if there were a credible theory which predicted such a regularity, there would be more grounds for confidence in the discovered preference
interpretation of the existing evidence.
The discovered preference approach treats rational-choice theory as an abstraction,
arrived at by assuming that the processes by which individuals learn how to satisfy their
preferences have been completed. But the assumption that individuals are capable of
learning what their preferences are does not license the hypothesis that those preferences are context-independent or that they satisfy axioms such as transitivity or the
sure-thing principle. This is essentially the same problem that Pareto faces when he
tries to justify the transitivity of indifference.
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
170
THE ECONOMIC JOURNAL
[JANUARY
In modern versions of the theory of choice, axioms of consistency of preferences are
usually justified as formal principles of rationality. That is, no claim is made about
which particular set of preferences is rational for a given individual. Instead, the claim
is that any preferences that fail to satisfy the axioms are internally inconsistent, and
thereby irrational. A variant of this argument asserts that preferences that are inconsistent in this sense are vulnerable to ÔDutch booksÕ or Ômoney pumpsÕ – that is, vulnerable
to exploitation by arbitrageurs. At best, such arguments might explain why, after
learning processes have been completed, those preferences that are revealed in
repeated choices satisfy the axioms.22 But they would not explain the (assumed) context-independence of the discovery process. For example, suppose that an individual’s
preferences do not automatically satisfy the standard consistency conditions, but are
revised in the direction of consistency whenever inconsistencies come to light – either
because the individual has a desire to be consistent, or because he learns that inconsistent preferences are exploited to his disadvantage by arbitrageurs. That process
might ultimately generate preferences which – in so far as they are revealed in whatever
decision problems are faced repeatedly – satisfy the consistency axioms. But there
seems no reason to assume that such preferences are independent of the decision
problems through which they are generated. So, if preference consistency is interpreted merely as a matter of formal rationality, it is hard to explain the contextindependence attributed to discovered preferences.
This problem might be overcome by invoking a substantive, rather than formal,
concept of rationality. That is, we might stipulate that a person’s underlying preferences rank outcomes in terms of some objective measure which serves as the standard
of rationality. For example, if an objective measure of pleasure could be defined for
each outcome, we might stipulate that a rational person always prefers more pleasure to
less. Then, underlying preferences inherit the consistency properties of the relevant
measure, and it would be coherent to postulate that individuals learn their underlying
preferences through experience (for example, through experience of the amounts of
pleasure generated by different actions). In Plott’s presentation of the discovered
preference hypothesis, there are hints that he may have in mind some instrumental
concept of rationality, in which an individual’s choices are rational to the extent that
they deliver what that individual really ÔwantsÕ:
People acquire an understanding of what they want through a process of
reflection and practice. In a sense, they do not know what they want and it may
be costly, or even unpleasant, to go through the process of discovery. Attitude
discovery is a process of evolution which has a direction, and in the final stage
results in the ÔdiscoveryÕ of a consistent and stable preference. (1996, p. 227)
But if this is what is intended by the proponents of the discovered preference
hypothesis, a critic is entitled to ask what objective measure is being used as the
standard of rationality, and what evidence there is that, after learning processes
are complete, people choose the actions that maximise the value of that particular
22
Cubitt and Sugden (2001) show that an individual’s responses to decision problems can be invulnerable
to money pumps without those responses being rationalisable in terms of context-independent and consistent
preferences.
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
2007 ]
THE ROAD NOT TAKEN
171
measure. If (as one might naturally expect) the measure is of some mental experience,
such as pleasure, answering such questions will require the concepts and methods of
psychology: economics will not be a separate science after all. That is exactly the
problem that led Pareto to start on his doomed quest for a solution to the integrability
problem.
More concretely, suppose the discovered preference hypothesis were to be based on
the more fundamental hypothesis that individuals seek to maximise pleasure. Then a
whole range of additional tests would be possible and new theoretical questions would
be opened up. It might be possible to investigate directly whether people maximise net
pleasure. Indeed, this question is already being investigated by behavioural economists.
Some of the results of these investigations suggest that human beingsÕ mental capacities
for recording affective experiences in memory and retrieving them later are subject to
systematic limitations and errors (Kahneman et al., 1997). If individualsÕ memories of
affective states are systematically biased, it is hard to see how any experiential learning
process could discover the true relationship between actions and affective states.
7. Conclusion
Pareto and the modern exponents of the discovered preference approach can be seen
as pursuing a common project: to show that economics can be a separate science of
rational choice, independent of psychology. In trying to achieve this objective, they
confront a common set of methodological problems, of which two are particularly
fundamental and particularly difficult to solve. The first problem is to find and to
justify a definition of the domain of economics within which rationality-based theories
predict successfully, which is not vacuous (as it would be if the domain were defined to
consist of exactly those choice problems for which the predictions of rational choice
theory succeed) and which is wide enough for economics to have something useful to
say about the real world. The second problem is to find consistency conditions for
ÔrationalÕ preferences, secure enough to serve as the basis for reliable deductive
inferences and with enough substance to allow a science of rational choice to have
predictive power, without appealing to contestable hypotheses that derive from psychological theory or experiment. Pareto and the discovered preference theorists
grapple with these problems – with what success, we leave the reader to judge.
For most of the twentieth century, however, mainstream economics represented itself
as a separate science without bothering much about these problems. Limits to the
domain of rational choice theory were not discussed; there was an implicit presumption
that the theory was universal in its application – applying, for example, under uncertainty as well as under certainty and objective risk, to public as well as to private goods,
to altruistic as well as to self-interested behaviour, and to politics as well as to economics. The preferences of economic agents were assumed to satisfy strong axioms of
consistency; these axioms were motivated on a priori grounds but not tested against the
evidence of real decision-making behaviour. But now behavioural economics has called
the bluff. It can no longer be taken for granted that the Paretian turn – the project of
separating economics from psychology, of grounding economics on principles of
abstract rational choice – was the path of progress. The road not taken was perhaps not
a dead end after all.
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
172
THE ECONOMIC JOURNAL
[JANUARY
University of Milan-Bicocca
University of East Anglia
Submitted: 3 February 2004
Accepted: 13 September 2005
References
Andreoni, J. (1988). ÔWhy free ride? Strategies and learning in public goods experimentsÕ, Journal of Public
Economics, vol. 37, pp. 291–304.
Antonelli, G. B. (1886). Sulla teoria matematica dell’economia politica, Pisa: Edizioni Fochetto.
Ariely, D., Loewenstein, G. and Prelec, D. (2003). Ô‘‘Coherent arbitrariness’’: stable demand curves without
stable preferenceÕ, Quarterly Journal of Economics, vol. 118, pp. 73–105.
Bardsley, N. and Moffatt, P. (2005). ÔThe experimetrics of public goods: inferring motivations from contributionsÕ, CeDEx Working Paper 2005-09, University of Nottingham.
Binmore, K. (1999). ÔWhy experiment in economics?Õ, Economic Journal, vol. 109, pp. F16–24.
Bruni, L. (2002). Vilfredo Pareto and the Birth of Modern Microeconomics, Cheltenham: Elgar.
Bruni, L. and Guala, F. (2001). ÔPareto and the epistemological foundations of rational choiceÕ, History of
Political Economy, vol. 33, pp. 21–49.
Busino, G. (1964). ÔNote bibliographique sur le CoursÕ, in Vilfredo Pareto. Epistolario, pp. 1165–72, Rome:
Accademia Nazionale dei Lincei.
Camerer, C. (1995). ÔIndividual decision makingÕ, in (J. H. Kagel and A. E. Roth, eds.), Handbook of Experimental Economics, pp. 587–703, Princeton, N.J.: Princeton University Press.
Camerer, C. (1996). ÔRules for experimenting in psychology and economics, and why they differÕ, in (W.
Albers et al, eds.), Experimental Studies of Strategic Interaction: Essays in Honor of Reinhard Selten, pp. 313–27,
Berlin: Springer-Verlag.
Camerer, C. (2003). Behavioral Game Theory, Princeton, N.J.: Russell Sage Foundation and Princeton University
Press.
Chipman, J. S., Hurwicz, L., Richter, M. K. and Sonnennschein, H. F. (1971), Preferences, Utility, and Demand:
A Minnesota Symposium, New York: Harcourt Brace Jovanovich.
Cox, J. C. and Grether, D. M. (1993). ÔThe preference reversal phenomenon: response mode, markets and
incentivesÕ, Economic Theory, vol. 7, pp. 381–405.
Cubitt, R., Starmer, C. and Sugden, R. (2001). ÔDiscovered preferences and the experimental evidence of
violations of expected utility theoryÕ, Journal of Economic Methodology, vol. 8, pp. 385–414.
Cubitt, R. and Sugden, R. (2001). ÔOn money pumpsÕ, Games and Economic Behavior, vol. 37, pp. 121–60.
Easterlin, R. (1974). ÔDoes economic growth improve the human lot? Some empirical evidenceÕ, in (P. David
and M. Reder, eds.), Nations and Households in Economic Growth: Essays in Honor of Moses Abramowitz,
pp. 89–125, New York: Academic Press.
Edgeworth, F. (1881/1967). Mathematical Psychics, New York: Kelley.
Fechner, G. T. (1860/1964). Elemente der Psychophysik, Amsterdam: Bonset.
Frederick, S. and Loewenstein, G. (1999). ÔHedonic adaptationÕ, in (D. Kahneman et al, eds.), Well-Being: The
Foundations of Hedonic Psychology, pp. 302–29, New York: Russell Sage Foundation.
Frey, B. and Stutzer, A. (2002). ÔWhat can economists learn from happiness research?Õ, Journal of Economic
Literature, vol. 40, pp. 402–35.
Hey, J. and Orme, C. (1994). ÔInvestigating generalizations of expected utility theory using experimental dataÕ,
Econometrica, vol. 62, pp. 1291–326.
Hicks, J. (1939). Value and Capital, Oxford: Oxford University Press.
Hicks, J. and Allen, R. (1934). ÔA reconsideration of the theory of valueÕ, Economica, vol. 1, pp. 52–76; 196–219.
Jevons, W. S. (1871/1970). The Theory of Political Economy, London: Penguin.
Kahneman, D. (1996). ÔCommentÕ [on Plott (1996)], in (K. J. Arrow et al., eds.), The Rational Foundations of
Economic Behaviour, pp. 251–4, Basingstoke: International Economic Association and Macmillan.
Kahneman, D. and Tversky, A. (1979). ÔProspect theory: an analysis of decision under riskÕ, Econometrica,
vol. 47, pp. 263–91.
Kahneman, D. and Varey, C. (1991). ÔNotes on the psychology of utilityÕ, in (J. Elster and J. Roemer, eds.),
Interpersonal Comparisons of Well-Being, pp. 127–63, Cambridge: Cambridge University Press.
Kahneman, D., Wakker, P. and Sarin, R. (1997). ÔBack to Bentham? Explorations of experienced utilityÕ,
Quarterly Journal of Economics, vol. 112, pp. 375–405.
Knetsch, J., Tang, F.-F. and Thaler, R. (2002). ÔThe endowment effect and repeated market trials: is the
Vickrey auction demand revealing?Õ, Experimental Economics, vol. 4, pp. 257–69.
Ledyard, J. (1995). ÔPublic goods: a survey of experimental researchÕ, in (J. H. Kagel and A. E. Roth, eds.),
Handbook of Experimental Economics, pp. 111–94, Princeton, N.J.: Princeton University Press.
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
2007 ]
THE ROAD NOT TAKEN
173
Lewin, S. (1996). ÔEconomics and psychology: lessons for our own day from the early twentieth centuryÕ,
Journal of Economic Literature, vol. 34, pp. 1293–323.
List, J. (2003). ÔDoes market experience eliminate market anomalies?Õ, Quarterly Journal of Economics, vol. 118,
pp. 41–71.
Loewenstein, G. (1999). ÔExperimental economics from the viewpoint of behavioural economicsÕ, Economic
Journal, vol. 109, pp. F25–34.
Loomes, G. (1999). ÔSome lessons from past experiments and some challenges for the futureÕ, Economic
Journal, vol. 109, pp. F35–45.
Loomes, G., Starmer, C. and Sugden, R. (2003). ÔDo anomalies disappear in repeated markets?Õ, Economic
Journal, vol. 113, pp. C153–66.
Mill, J. S. (1843/1967). A System of Logic, London: Longmans.
Mirowski, P. (1989). More Heat than Light: Economics as Social Physics, Physics as Nature’s Economics, Cambridge:
Cambridge University Press.
Pantaleoni, M. (1889/1898). Pure Economics, London: Macmillan. English translation of the Italian 1889
edition (Florence: Barbera).
Pantaleoni, M. (1913/1925). ÔDefinizione dell’economia. Una prolusioneÕ, Erotemi di economia, vol. I, pp. 1–66,
Bari: Laterza.
Pareto, V. (1892–93/1982). ÔConsiderazioni sui principi fondamentali dell’economia puraÕ, Giornale degli
Economisti May, pp. 390–420; June, pp. 485–512; August, pp. 119–57. Reprinted in Oeuvres Comple`tes of
V. Pareto, vol. 26 (1982), Geneva: Droz.
Pareto, V. (1894/1966). ÔLeçon d’ouverture dÕéconomie politique de MÕ. ÔV PÕ, professeur extraordinaire,
Lausanne: Editions de l’Université. Reprinted in Oeuvres Comple`tes of V. Pareto, vol. 9 (1966), Geneva:
Droz.
Pareto, V. (1899/1989). Letter to Laurent, 19 January 1899. In Lettres et Correspondances, vol. 30 of Oeuvres
Comple`tes of V. Pareto, Geneva: Droz.
Pareto, V. (1900/1982). Sunto di alcuni capitoli di un nuovo trattato di economia politica del prof. Pareto.
Giornale degli Economisti, X: 216–35; 511–49. Reprinted in Oeuvres Comple`tes of V. Pareto, vol. 26 (1982),
Geneva: Droz.
Pareto, V. (1909/1971). Manual of Political Economy, New York: Kelley. English translation of the 1909 French
edition of the 1906 Italian Manuale d’economia politica con una introduzione alla scienza sociale, Milan: Societa
Editrice Libraria.
Pareto, V. (1911/1982). LÕéconomie mathématique. In Enciclope´die des sciences mathe´matiques pures et applique´es,
publie´e sous les auspices des Acade´mies des Sciences de Gottingue, de Leipzig, de Munich et de Vienne avec la
collaboration de nombreux savants, Paris: Gauthier-Villars, tome I, vol. IV. Reprinted in Oeuvres Comple`tes of
V. Pareto, vol. 26 (1982), Geneva: Droz.
Pareto, V. (1916/1963). Trattato di sociologia generale, Florence: Barbera, Translated into English as The Mind
and Society: A Treatise on General Sociology, New York: Dover, 1963.
Pareto, V. (1984). Lettere a Maffeo Pantaleoni. Edited by G. De Rosa, vol. 3, Banca Nazionale del Lavoro, Rome.
Reprinted in Oeuvres Comple`tes of V. Pareto, vol. 28, Geneva: Droz.
Plott, C. (1996). ÔRational individual behaviour in markets and social choice processes: the discovered preference hypothesisÕ, in (K. J. Arrow et al, eds.), The Rational Foundations of Economic Behaviour, pp. 225–50,
Basingstoke: International Economic Association and Macmillan.
Plott, C. and Zeiler, K. (2005). ÔThe willingness to pay – willingness to accept gap, the ÔÔendowment effectÕÕ.
Subject misperceptions, and experimental procedures for eliciting valuationsÕ, American Economic Review,
vol. 95, pp. 530–45.
Samuelson, P. (1938). ÔA note on the pure theory of consumer’s behaviourÕ, Economica, vol. 5, pp. 61–7.
Savage, L. (1954). The Foundations of Statistics, New York: Wiley.
Scitovsky, T. (1976). The Joyless Economy, Oxford: Oxford University Press.
Smith, V. (1989). ÔTheory, experiment and economicsÕ, Journal of Economic Perspectives, vol. 3, pp. 151–69.
Smith, V. (1994). ÔEconomics in the laboratoryÕ, Journal of Economic Perspectives, vol. 8, pp. 113–31.
Starmer, C. (1999). ÔExperimental economics: hard science or wasteful tinkering?Õ, Economic Journal,
vol. 109, pp. F5–15.
Starmer, C. (2000). ÔDevelopments in non-expected utility theory: the hunt for a descriptive theory of choice
under riskÕ, Journal of Economic Literature, vol. 38, pp. 332–82.
Stigler, G. (1950). ÔThe development of utility theory: I, IIÕ, Journal of Political Economy, vol. 58, pp. 307–27 and
373–96.
Tversky, A. and Kahneman, D. (1991). ÔLoss aversion in riskless choice: a reference-dependent modelÕ,
Quarterly Journal of Economics, vol. 106, pp. 1039–61.
Watson, J. B. (1913). ÔPsychology as the behaviourist views itÕ, Psychological Review, vol. 20, pp. 158–177.
Wicksteed, P. (1906). Review of Pareto’s Manuale di economia politica, Economic Journal, vol. 16, pp. 553–57.
Wundt, W. (1874). Grundzuge der physiologischen Psychologie, Leipzig: W. Engelmann.
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007