Framing effects –An outline of their impact on Decision Theory and Economics

Gustavo Marqués (CIECE, FCE, UBA-UNLZ)

Diego Weisman (CIECE, FCE, UBA- CONICET)

Abstract
The paper examines the impact that framing effects have on decision theory and economics. In Section I the
specific nature of framing effects and their point of impact on EUT –the assumption of extensionality- are
clarified and their implications for rationality explored. Section II shows that extensionality may be
understood as a bare linguistic (logical) requirement or as a condition for preference theory. Section III
points out the existence of a trade-off between the assumption of rationality and the descriptive success of
conventional decision theory. It is argued that because human beings are not extensional decision-makers,
the capacity of the conventional approach for measuring utility in the domain of human choices as well as
achieving predictive success fails. Section IV describes how Prospect Theory manages to avoid the
conventional shortcomings. Introducing the alternative descriptions that give raise to framing effects into its
equations, Prospect Theory is able to define prospective utility functions for the individuals, which can be
maximized. Besides, a particular kind of predictions (prospective predictions) may be obtained. The paper
comes to an end suggesting some philosophical and epistemological problems posed by the nature of
Prospect Theory and its relation with conventional decision theory. Particularly, it focuses on the changes
that Prospect Theory introduces in the notion of utility, its incommensurability with traditional expected
utility, the particular type of predictions that can be achieved assuming intensional decision-makers, and the
impact the rejection of extensionality has on the normative approach to conventional theory and economic
theory. Some remarks about the logical relationship between conventional decision theory and prospect
theory are also advanced.

Introduction

In 1944 John von Neumann and Oskar Morgenstern proposed an axiomatic theory of choice under
risk, which was called Expected Utility Theory (EUT). It is difficult to exaggerate the impact EUT
had on economics. In the past utility theory had been successfully applied under conditions of
certainty (consumption theory), but it was considered useless for approaching risky economic
decisions. The Neumann–Morgenstern theory of expected utility, as well as later versions inspired
on it, like Marschack‟s theory (1950), seemed to open the possibility of a unified explanation of
economic phenomena. Hopefully, adequately developed, it could explain some type of decisions
involving risk which where previously considered beyond economics and attributable to the
preeminence of psychological factors or plain arbitrary behavior.

It was taken as granted that any scientific empirical theory of decision making should be built
assuming (ideally) rational agents. This claim seemed to be undisputable at that time (and it
remains still beyond question for most economists and decision theorists). Rationality is a complex
notion and for the purposes of our paper it is convenient to approach it at two different levels.
EUT‟s rational agents are those who have consistent probability beliefs and well behaved
preferences1. At a deeper level, for having such a kind of preferences, rational agents must be
extensional decision-makers (in a sense that will be clarified later). Extensionality was considered
an essential (unavoidable) feature of any rational theory of decision making. To take account of this
distinction we will refer to this narrower notion as “rationality-k”, alluding to the fact that it belongs
to the kernel of any kind of rational decision theory.

Endowed with this kind of agents EUT exhibited two outstanding virtues. On the one hand, it was
successful in developing an acceptable method for measuring utility in risky contexts –a procedure
that freed the utility concept from any psychological connotation-2. On the other, it was believed

1
that assuming individuals who behave rationally in the sense defined by the theory, it was possible
to predict the choices that agents would make among alternative prospects, provided some initial
choices were known. Under the influence of the dominant empiricist view, these two features of
EUT were considered a remarkable achievement of the theory that suited the scientific standards
accepted at the time.

Modeling agents as rational was believed to have another advantage. The empirical success of the
theory was thought as dependent on the normative properties of rationality. Even if at first
individuals behaved in an irrational way, a good prescriptive theory of rational behavior should be
able to self validate its consequences influencing the individual decisions in the right direction and
closing step by step the gap between its recommendations and the deviated behavior. The dream of
an adequate rational theory of decision making was more than a dream: in principle, it seemed to be
a clever strategy for reaching an agreement between advised and observed behavior.

However, at the beginning of 1950 some counterexamples were found and when this happened and
they proved to be persistent, EUT was amended in different ways for allowing their accommodation
into the theory. This was done relaxing the commitment to well behaved preferences. In all these
attempts however the key condition of extensionality was preserved. In this paper for conventional
decision theories we refer to all the latter attempts to modify EUT in order to obtain a match
between its consequences and the facts without resigning this narrower notion of rationality
(rationality-k).

The discovery of a certain type of counterexamples – the so called framing effects- which imply that
human beings violate extensionality, showed the limits of that strategy of development and seemed
to jeopardize conventional decision theory in a far more decisive way than usual anomalies do.
First, as long as the measurability of the utility concept of conventional theory depends on the
assumption of extensional decision-makers, framing effects entail that utility –i.e., the “object” (or
property) measured by the usual procedure – cannot be applied to human decisions. This means that
human choices exhibit no well defined order among prospects. Besides, to the extent that the
predictive power of conventional theory rested on the assumption of rational agents (that includes
extensionality at its very basis) framing effects imply that the development of any adequate
conventional descriptive theory of decision making under risk is condemned to failure.

Fortunately, the link between rationality –and particularly rationality-k-, on the one hand, and
descriptive and predictive power, on the other, can be safely undone. Prospect Theory was
successful in showing that it is perfectly possible to reach descriptive adequacy without assuming
extensional decision makers. To the extent that an initially unsuspected trade–off between
rationality and descriptive adequacy seems to exist, PT could pave the way for the development of
what hopefully might be a successful purely descriptive theory in the economic realm: Behavioral
Economics.

The fact that human beings are not extensional-decision makers, however, poses some questions
regarding the epistemic status of Prospect Theory and the future of an economics built on such a
basis. What kind of decision theory may be constructed assuming intensional decision makers?
What are their implications for the measurability of utility? Particularly, what about the traditional
goal of counting with a predictive decision and economic theory? Is there a sub-domain of
economic research in which extensional decision makers could be properly assumed? In the present
article the challenges that framing effects pose for economics and decision theory will be examined,
and some remarks on the problems that arise if economic theory has to be built assuming
intensional economic agents will be advanced.

2
I. Framing Decisions and Framing Effects
Empirical (psychological) investigations have shown that having to choose among alternatives
individuals always start from a reference point. Consider the following cases: at the end of a day an
individual who has lost $140 has to decide whether to pay $10 for a prospect that may pay $150
with probability 1/15 or nothing with probability 14/15 (Tversky and Kahneman, 1981, p. 456).
Clearly his options are to bet or not to, but these two alternatives may be described in many
different ways. For instance, he may take into account his accumulated loss of $140 or he may leave
the past performance out of consideration and start anew. The prospects he is going to consider will
depend on where his reference point is located.

Case 1. The individual does not take into account past losses. His reference point is in this case $0, and his options are the following:

A: Not to Bet (A) = 0

B: To Bet (B) = (-$10, 14/15; +$140, 1/15)

Case 2. The individual considers past losses. His reference point is now -$140, and his options change accordingly:

C: Not to Bet C: (-$140)

D: To Bet D: (-$150, 14/15; $0, 1/15)

Interestingly, no privileged frame which would deserve to be called “correct” exists, and nothing in
this picture allows to say that when choosing one particular frame the subject is making a mistake.

Compare now the situation described above with the one involved in the so-called Asian disease
(Kahneman, 2003ª, pp. 1458).

Imagine that the United States is preparing for the outbreak of an unusual Asian disease, which is expected
to kill 600 people. Two alternative programs to combat the disease have been proposed:

If Program A is adopted, 200 people will be saved.

If Program B is adopted, there is a one-third probability that 600 people will be saved and a two-thirds
probability that no people will be saved.

Framed this way most people choose Program A over Program B, showing risk aversion. Consider
now this new pair of options:

If Program A’ is adopted, 400 people will die.

If Program B’ is adopted, there is a one third probability that nobody will die and a two-thirds probability
that 600 people will die.

When they have to choose between them most people prefer Program B‟ over A‟, even when it is
easy to note that A ≡ A‟ and B ≡ B‟. In both cases exactly the same options are described. The only
difference is that in the first case the options are described in terms of saved (gained) lives while in
the second one in terms of lost lives. This change in terminology should be unsubstantial for truly
rational men but apparently these subjects think differently because their preferences within one
frame are reverted when the frame is shifted. The way in which options are described seems to be
more important for them than the options themselves. In this paper we use the expression framing
effects only in reference to this kind of reversion of preferences3.

From the standpoint of EUT, framing effects are unexpected and unwelcomed, and their existence
demands that some of its premises be revised. The problem is which of them must be blamed for the
observed deviation of rational behavior. Broadly speaking a theory has two kinds of premises:
axioms, that concentrate all its particular informative content, and presuppositions, that incorporate
pre-existent knowledge from other theories or disciplines whose validity is taken for granted. A

3
theory will be unable to deliver its own message without profiting from an available background of
theoretical resources. EUT, for instance, assumes at least basic algebra and standard logic.

Here is a simplified enumeration of the components of both categories (for a full account of them
see Starmer, 2000)

Axioms of EUT Presuppositions of EUT

Rational preferences on prospects have the -Humans are extensional decision-makers regarding
followings properties: his preferences.
-Continuity
-Transitivity
-Completeness
-Independence

Traditional counterexamples of standard decision theory -anomalies and paradoxes- were supposed
to affect its axioms and for this reason, successive developments of conventional theory tried to
modify some of them aiming at a reconciliation between the theory and the observed behavior.
Machina (1982), for instance, debilitated the independence axiom and Regret Theory (Loomes and
Sugden, 1982) dropped transitivity. Other proposals have relaxed completeness (Levi, 1980;
McCleennen, 1990) and continuity (Simon, 1955). In this sense framing effects do not belong to the
class of traditional counterexamples: no axiom is taken as responsible for them. Rather, they are
traced back to one of the basic presuppositions of EUT – the presupposition of extensional
decision-makers- which is at the very basis of conventional decision theory.

As the borderline between framings and framing effects is somewhat fuzzy some remarks about this
issue may be of help. Framing effects are a special case of framings. All decisions are framed, but
not all of them generate framing effects. The specific feature of scenarios like those described in the
case of Asian Disease is that humans violate the extensionality principle, something that does not
necessarily happen every time decisions are framed. Let‟s return for a moment to the case in which
after several bets a man has to decide whether betting again or not. The way in which he frames his
options crucially determined the prospects he faces. More importantly, no violation of the
extensionality principle occurs because in none of the alluded cases is the same situation described
in different ways. A and B are not the same options as C and D. The two cases involve different
prospects. Shifts in the point of reference change the prospects, not just their descriptions.

By itself, framings do not challenge the basic idea of conventional decision theory that people have
a unique well defined order of preferences which is revealed through their choices. Framing effects,
instead, show that in those cases in which the same options are described differently two different
(opposite) orders of preference may emerge. The conventional presumption of rationality-k is
refuted in the domain of humans.

II. Are humans intensional or extensional decision makers?

According to conventional methodology scientific theories have two broad kinds of expressions:
logical (like usual truth-functions of standard logic, namely “and” and “or”) and descriptive (for
instance, sentences and terms). In substantially richer languages the identity of expressions has to be
defined. This is done asserting that two (or more) syntactically different (logical or descriptive)
expressions of a language are equivalent when their referent is the same (i.e., they are co-extensive).
The Extensionality Principle (EP), asserts that two different but coextensive descriptions D1 and D2
belonging to a language L, may be exchanged one for the other within L salva veritate. Outlining a

4
very simplistic semantic theory, we may call descriptions the material aspect of expressions. Thus
the fact that it is raining may be described in many different ways, like “it rains”, “it is raining”, “it
is raining now”, etc. Analogously, individuals (or classes of individuals) and predicates may be
described by different expressions. “The author of „El Quijote‟” and “The one-armed man of
Lepanto” are equivalent descriptions for the same individual: Cervantes. According to EP
“Cervantes” may be substituted for “The one-armed man of Lepanto” within the sentence “The one-
armed man of Lepanto wrote „El Quijote‟” with no alteration in the truth-value of the original
sentence.

Let‟s move now to the case in which different descriptions of a same choice-situation may be
formulated. What we called extensional decision-makers are those agents who having to choose
among prospects do not pay attention to the particular way in which the options are described and
focus only in the “real” options alluded to by these descriptions. In contrast, intensional decision-
makers are those people whose choices may depend on the particular descriptions of the options
they face. To illustrate what undesired consequences may emerge from the violation of
extensionality we start with a very simple case. Suppose an individual has to choose among two
pairs of options:

I A or B
II B or C

And suppose that he chooses A over B and B over C. As long as he only pays attention to the
options involved (i.e., he is an extensional decision-maker), he must also choose A over C. For
instance suppose that A stands for Cervantes, B for Shakespeare, and C for Goethe. Our extensional
man will prefer Cervantes to Goethe. But suppose now that an alternative name for Cervantes: “The
one-armed man of Lepanto” is given. How is our man going to choose this time among the options
so described? Would he behave as a truly extensional decision-maker? What kind of choosers are
real men is something that has to be established by empirical investigation. And a huge amount of
evidence shows that many people are intensional decision-makers. Now, it is possible that having to
choose between the work of “The one-armed man of Lepanto” or that of Goethe, an individual who
shows a negative attitude to disabilities, picked up the second option (in spite of his preference of
Cervantes‟ work over that of Goethe‟s)4.

Let‟s see now the particular form in which the extensionality principle is violated in the Asian
Disease case. It has a prominent feature: it sets an upper boundary on the universe of individuals
upon which the consequences are later built. In the example, 600 individuals are neutrally described
as being in danger of death. The two alternative therapies are then introduced in terms of lives saved
and lost. Closing the universe is a crucial step. Given this, it is possible then to establish the
following equivalences:

Y: 200 lives will be saved = Y‟: 400 lives will be lost

Z: 600 lives will be saved = Z‟: Nobody will die.
J: Nobody will be saved = J‟: Everybody will die.

These equivalences are valid only in the context of this illustration in which the total number of
endangered population is fixed at 600. The following step is to represent the matrix of the first two
options (the ones described in terms of saved lives):

Program A: (Y)
Program B: (Z, 1/3; J, 2/3) (I)

5
Given extensionality, we can now replace Y‟ by Y in Program A, and Z‟ by Z and J‟ by J in
Program B:

Program A: (Y‟)
Program B: (Z‟, 1/3; J‟, 2/3) (II)

It is obvious that (I) and (II) are equivalent formulations of the same choice problem and deserve
the same response on the part of rational agents.5

The role of the extensionality principle in logic and decision theory and its consequences for
consistency may be now briefly compared. Within an assertive context extensionality is needed for
substituting strictly different (but co-extensive) expressions for one another, salva veritate. Within a
choice-situation extensionality guarantees that any description D1 of the options involved may be
substituted for another equivalent description D2, salva choices. The significance of framing effects
is that they show that humans violate the extensionality principle in the realm of choices.
In the following sections we will consider two opposite disciplinary reactions to the fact that
humans are not truly extensional decision-makers: the conventional strategy, which refuses to
accept it and the more heterodox (revolutionary?) approach that accept the fact and try to live with
its consequences.

III. How damaging are framing effects for the conventional approach to decision theory?

An adequate descriptive theory of decision making under risk must allow the observed patterns to
be derived from its set of premises. Building such a good theory is what many conventional
theorists want to achieve6. They pursue this goal preserving their commitment to rationality-k. This
strategy, however, is jeopardized and may be criticized on epistemological grounds. The fact that
human beings are intensional decision-makers affect conventional approach in its deepest form:
questioning its capacity for measuring utility in the domain of human choices as well as its
predictive performance.

Let‟s start with the measurability issue. As was said earlier, the very core of conventional theories –
namely, EUT-, incorporates a procedure for the measurement of utility. This property was (and still
is) considered an extraordinary achievement that invests economics with scientific rigor and helps
to distinguish it from the rest of the social sciences. The procedure gives information about the
utilities that an individual assigns to a set of particular results included in a set of prospects. The
information thus obtained may be used for ranking these prospects in an unambiguous way.

In 1971 Lichtenstein and Slovic discovered the phenomenon of preference reversals under
conditions of risky choices, which put in question the validity of the method in the human domain.
According to measurability theory, having more than one procedure for ranking prospects they
should be compatible with one another. But in the anomalous case referred by Lichtenstein and
Slovic what seems to be two irreproachable ways for eliciting preferences lead to opposite results7.
The experiment suggests that the notion of utility involved in the standard decision theory cannot be
measured after all in the domain of human beings. This result has direct relevance for economics:
individuals do not have a well behaved utility function defined over prospects8, and given that the
maximization of the expected utility depends on the possibility of counting with such utility
function, it cannot be maximized in the conventional sense. In this way what at first appeared to be
an apparently restricted (technical) problem spreads all over economic theory.

What is supposed to be violated in the experiment performed by Lichtenstein and Slovic is the
assumption of procedural invariance, and someone could object that framing effects are not directly

6
related to this assumption. However framing effects are a particular kind of preference reversals that
affects the measurement of conventional utility at an even deeper level than ordinary preference
reversals do. Even in the more favorable case in which the violation of procedural invariance could
be overcome and a single eliciting procedure could be identified, the options faced by decision-
makers have anyway to be described and the ensuing ranking of preferences should be independent
of the particular descriptions adopted at the time. This is precisely what the presupposition of
descriptive invariance (i.e., extensionality) demands. Extensionality then is a more basic
requirement than procedural invariance because it has yet to be fulfilled once this last assumption
has been satisfied9.

Turning now to predictive (descriptive) capacity, things are even worse for the conventional
approach to choices. Its compromise with the broader notion of rationality has proved to be
excessively demanding and, in trying to explain away the many anomalies that plagued standard
decision theory, it has been progressively debilitated10. Because its commitment with rationality-k
represents a borderline that cannot be crossed without giving up the realm of rationality altogether,
conventional decision theories have to remain faithful to extensionality and cannot accommodate
the decisions taken by intensional agents. Given that framing effects are ubiquitous in artificial
contexts as well as in real life, the conventional strategy –that once upon a time was thought to be
the key for developing any empirically successful theory- is ultimately self-defeating because it
precludes descriptive success in a wide range of situations, a devastating result for conventional
theories.

At this point a classical objection may be raised against our argument. Someone could be tempted
to think that the particular damaging effect that framing effects have on conventional theory is only
apparent and that they are ultimately irrelevant. As long as the descriptions involved in framing
effects are equivalent descriptions of the same choice-setting sooner or later agents would become
aware of this fact. People will learn to avoid the rough mistakes involved in framing effects and
they would tend to disappear. This claim seems plausible at first sight, but it is largely ungrounded:
different empirical results show that it cannot be taken for granted that agents can learn so easily11.
Apparently, the biases originated in framing effects are unavoidable given the cognitive and
computational limitations of the human mind. The usual mainstream argument according to which
agents can behave rationally given satisfactory incentives and time enough for collecting
information and learning is falsified. This virtuous mainstream process is a psychological
impossibility. Most people are unable to escape the emotional or ethic impact of the terms
employed to describe the options they face and consequently, it cannot be assumed that in the long
run, through learning, individuals are going to become extensional decision makers.

IV. The prospect of a purely descriptive non conventional decision theory

There exists another strategy for developing a successful descriptive theory of risky choices.
Modeling agents as intensional decision-makers, Prospect Theory is able to predict most of the
known anomalies of conventional decision theory, including framing effects. It is not a surprising
result since it was constructed ex-professo for this purpose. Kahneman and Tversky knew in
advance what kind of facts have to be “predicted” by the theory. The merit of PT lies rather in
finding the way to achieve the intended results. In sharp contrast with many of the “realist”
proposals of the past, Prospect Theory manages to introduce the alternative descriptions that give
rise to framing effects into its equations inspiring the construction of what may be called
prospective preference functions that progressively have been included in several economic
models12. This device is welcome in economics: it meets the desiderata of formal accuracy and
predictive power. Unlike the original conception of bounded rationality as well as some of its

7
current versions that dispute the very idea of maximization13, prospective preference functions can
be maximized.

Going back to predictive power, prospective predictions have a distinctive feature. In one sense all
scientific predictions are conditional by nature. This feature gives rise to the usual distinction
between prediction and prophesy14. However, the way in which Prospect Theory and conventional
theory conditionalize their predictions is different. While traditional vision takes predictions to be
dependent on the assumption of rationality (i.e., the axioms of conventional theory), Prospect
Theory makes them dependent on the possibility of knowing the way in which the choice setting is
framed. It provides conditional predictions of the form: given the alternative prospects A and B, if
the choice setting is framed by the agent in the particular way F, the pattern of choices will be P.

The fact that for saying something definitive about individual choices Prospect Theory needs to
know how the agents are framing their options seems to point out to what can be considered an
Achilles‟ heel for the theory: as long as we don‟t know the agents‟ point of references we are unable
to anticipate their choices. This limitation seems to involve the existence of a potential gap between
lab and field conditions. Experimentalists are better prepared to predict choices under conditions of
experimental control, in which the way people frame their choices is better known, than in real
situations. Besides, within an experimental-setting it is possible to impose the frame to the decision-
makers, affording the relevant descriptions of the results incorporated into the prospects or the
probabilities associated with them. This imposition is facilitated because individuals tend to react
passively in face of the exogenously introduced descriptions.

However, even if some difference in manipulability exists between artificial and real situations, it is
only a matter of degree. True, in real life it is often more difficult to assert in advance what is the
point of reference agents are going to consider when making choices, but in many cases it is fairly
possible to influence the way individuals will frame their decisions15. While conventional
predictions are the result of a purely intellectual task, which demands calculation on the part of the
“observer”, Prospect Theory‟s predictions allows for an intrusion of the observers on real life
agents‟ choices. Prospective predictions have a technological dimension that is absent in traditional
decision theory. The fact that agents‟ decisions can be predicted through manipulations is a
remarkable (and also somewhat disquieting) result. Leaving aside considerations of social power or
individual freedom, the capacity for framing decisions outside the Lab makes it possible to apply
the theory on real markets. Marketing, as a discipline, is based on this possibility.

We conclude with some final remarks about the capacity of Prospect Theory for providing a
systematic account of risky choices. It succeeded where conventional theory failed because it gave
up the assumption that agents are extensional decision-makers and someway a sort of homo
intensionalis was incorporated into the theory. If it had happened that an indefinite (maybe
numerous) number of different descriptions for every choice situation had existed, a conceivable
successful “theory” would look like a collection of separate pieces each one designed for one
particular case, instead of a unified single theory. As long as the number of potential descriptions
increases, the problem of achieving a systematic account of all of them aggravates. Fortunately, in
the special case of risky economic choices a few kinds of framing effects may be identified and the
issuing patters recognized, reducing the problem to a proportion that can be easily handled. For
instance, in cases in which what are framed are the outcomes of the lotteries the relevant number of
frames is reduced to two: losses and gains. When “contingencies” are framed, a small number of
relevant cases and their associated responses may be identified. These circumstances allows that
some definite pattern of choices may be associated to each description (i.e., frame). This is a
remarkable achievement of Prospect Theory16.

8
Discussion

Conventional decision theory incorporates rational decision makers as an essential step for reaching
an adequate descriptive theory of decision making under risk. It was mistaken: given the extent of
framing effects the assumption of rationality is rather an obstacle for the acquisition of empirical
success. At least it is clear that an originally unsuspected trade-off between rationality and
descriptive accuracy exists. After some immediate benefits of assuming rationality have been
reached, meaningful progress has been obtained by weakening this assumption (or setting it aside
altogether).

But the abandonment of the rationality assumption –especially in its weaker form: the rejection or
extensionality- has major consequences for economic theory as well as decision theory. In what
follows we examine some theoretical and epistemological features that distinguish Prospect Theory
from conventional theory and may be traced back to their opposite stance towards rationality.

Particularly, we focus on the changes produced in the notion of utility and their philosophical
implications, the type of predictions that can be achieved assuming intensional decision-makers, the
impact of the rejection of extensionality on the normative approach to conventional decision theory
and (conventional) economic theory. Finally, an application of incommensurability to the utility
concept and some remarks on the importance of these issues for behavioral economics are also
advanced. These are all highly sensitive issues but have been rarely discussed explicitly. Our
purpose in this paper is to show the way in which philosophy can shed light on economic problems
and to stimulate a long needed debate.

Utility and incommensurability

In EUT as well as in any of its conventional versions, expected “utility” is a property of prospects:
given two prospects X and Y that the individuals may rank in a well defined manner, the numbers
(or symbols) that allow ranking the options are the utility they get from them. This feature of
conventional theory is crucially dependent on extensionality. As long as individuals are intensional
decision-makers their choices do not reveal a well defined order of preferences over prospects: the
same choice-setting triggers two opposite rankings when the options are described alternatively as
gains or losses. In this situation “utility” in the conventional sense is no longer valid in the human
domain (and, then, no maximization of such property becomes possible).

In a domain where extensionality has lost validity, the conventional utility notion must be given up.
Prospect Theory introduces a major break with this position allowing individuals to have
preferences (and consequently, utility functions) defined as descriptions of prospects rather than
prospects themselves. And this idea has been built into functions and equations able to represent a
unique order of preferences even when framing effects, which can be maximized, prevail.
Following the tradition it would be perfectly possible to call this new numbers “utility”, but it is
important to keep in mind that this sort of utility is a very different notion from the one involved in
conventional decision theory: it refers to descriptions of changes in endowments (weighed by their
probabilities) relative to a certain point of reference. Following Barberis and Thaler (2003) we have
called it prospective utility.

A conceptual change like the one described above which affects the utility concept invites
methodological elaboration. It brings to mind the Kuhnean description of the transformation of the
concept “mass” when exported from Newton‟s (or Galileo‟s) theoretical framework to Einstein‟s.
The difficulties that mainstream and heterodox economists find in exchanging views about the
utility concept –or some other critical issue involving this notion- may be due to the fact that

9
“utility” is used in different senses for each community and a sort of incommensurability arises
around this particular point. If this diagnosis is correct, a lot of philosophical work may be required
for improving the current conversation about the issue and reaching an integrated view of the many
utility notions referred to in the current literature (mainly, among conventional utility, prospective
utility and, ultimately, experienced utility).

The unique character of prospective predictions

Two essential features of Prospect Theory‟s capacity for predicting choices deserve to be noted. In
the first place, the domain of causal variables that have a crucial role in decisions is expanded. In
conventional theory the elements of the prospects (results and probabilities) are the only causal
variables responsible for agents‟ decisions. Given their values the experimentalist can predict what
choices agents will make computing the expected utility of every prospect they face. In Prospect
Theory the point of reference is introduced as an additional causal variable that proves to be crucial
for ranking the alternatives. A second point of departure of prospective predictions from
conventional ones is that the point of reference can be somewhat manipulated and, through it, the
agent‟s consideration of the traditional variables (results and probabilities) may also be influenced.
All conventional theories preserve the assumption that the observer does not influence the elements
of the choice situation. The experimenter is a true observer and interacts neither with the elements
of the choice situation nor the individual‟s view of the choice situation. In Prospect Theory, instead,
the way in which individuals frame the options they face may be crucially influenced by the point
of reference afforded by the experimenters. So, prospective predictions are possible due to an
intrusion of the observer who helps to shape the choice-setting in which agents‟ decisions are taken.
Combined, both features give a unique character to what has been called prospective predictions,
and explain its capacity for dealing with observed patterns of behavior that were considered
anomalous in conventional theory.

The impact of framing effects on EUT and Economics

Some comments about the consequences of adopting a theory able to deal with the impact of
framing effects on conventional decision theory and economics will be helpful. There are no
conceptual (logical) problems in considering ideal decision makers who meet all the relevant
requirement of an adequate rational theory of choices under risk. Framing effects are not an obstacle
for a purely normative research in choice theory. They rather point out to the existence of an
(apparently insuperable) empirical fact: the human vulnerability to the impact of ethical and
emotional (psychological) factors when adopting decisions. As long as framing effects are
ubiquitous no successful descriptive theory may be constructed assuming extensional decision-
makers.

Locating in an accurate way the point of impact of framing effects on conventional decision theory
is also relevant for appraising their consequences for economics. Particularly, it helps to rescue
most of the so-called mathematical economics from the charge of vacuity or invalidity. Though
applied economics is a growing field that has expanded at a fast rate, a good deal of theoretical
work in economics is normative in nature. The legitimacy of a normative interpretation of EUT is
attractive because it helps to validate some crucial goals of conventional economic theory that are
difficult to accommodate into its descriptive-oriented developments. Idealizations without intended
empirical counterpart lies at the proper base of many economic models. Ideal agents, assumed as
perfectly rational, are frequently incorporated into theories and models with the intention of
studying their behavior under different (also ideal) choice settings. Even in those cases in which
nobody expects that these ideal theories may offer observable consequences, they can offer the base
for the construction of a theoretical project designed for exploratory purposes and aiming at the
acquisition of well defined results. Besides, such theories could be interesting in another sense: they

10
might be adequate for securing micro foundations to General Equilibrium Theory or
Macroeconomics. Usual textbooks illustrate this particular use of standard theory, and nothing said
in this paper prevents this practice.

Up to this point, our analysis was limited to theories of decision making (conventional theories and
Prospect Theory), and within this specific area it has been argued that as long as framing effects
prevail, descriptive conventional theories can not be completely successful in the domain of human
risky decisions. Can the results just obtained be extended to other economic theories that
“unrealistically” incorporate rational agents?

To the extent that at the basis of any descriptive economic theory lies a successful descriptive
decision theory and any rational decision theory cannot be fully adequate within the human realm,
one could think that, to move going forward, descriptive economic theory has to incorporate
Prospect Theory at its very basis. This is the path followed for Behavioral Economics. However
things are not so simple. It cannot be dismissed a-priori that a handful of economic phenomena
could be fairly approached with the prevalent rational vision. Indeed, beyond some explicit
commitment to psychological realism Behavioral Economics has restricted itself to modeling some
especially anomalous cases and made no attempt to give an alternative account in most of the
domains in which standard theory has not been disproved. Besides, even in these domains in which
Behavioral Economics claims pertinence, the rationalist approach may be of some utility after all.
As Barberis and Thaler (2002, p. 1076) recognize, “both the rational approach and behavioral
approaches to finance have made progress in understanding the three puzzles singled out at the start
of this section” (namely, the equity premium, the volatility and the predictability puzzles). So, a
prudent stance toward this issue is to acknowledge that what kind of problems is sound to approach
with the rationalist view and where it is necessary to resort to a behavioral view is something that
has to be decided on empirical and theoretical grounds.

Of course, given that the basic theoretical terms used in conventional and prospective
approaches are irreducible into one another, and as long as the coexistence of these views
endure, the near future of economic theory could be fragmentation rather than unity. Looking
forward one could expect that rationalistic models and behavioral models will have their own and
separate set of explained phenomena, each vision using different notions of “utility” and assuming
different kinds of agents: extensional or intensional in each case.

The way in which some behavioral economists see the relation between traditional economics and
Behavioral Economics seems to give some support to the point we are developing here. They
advance the thesis that Behavioral Economics is intended as a generalization of standard economics.
A theory T‟ is thought to be a generalization of another theory T when T‟ preserves the parameters
present in T but incorporates additional ones. When the new factors become zero the wider theory
reduces to the older one17. Is this possible?

Generalization and Incommensurability

At the beginning of this last section we stated that the concepts of expected utility and prospective
utility are incommensurable. However, it is difficult to reconcile this result with the vision that
behavioral economists have of the relation between their own approach and the traditional one. The
idea that unconventional theory may be reduced to traditional theory when the additional
parameters become null involves a sort of cumulative (empiricist) view of scientific change. More
precisely, it takes for granted that the reference is preserved when expected utility is substituted by
prospective utility (or the other way around). If, instead, a sort of incommensurability is involved in
the relation between these concepts it cannot be said that when the additional parameters become

11
zero the new utility concept collapses to the old one. A comparison with the way in which PT
generalizes EUT will help. Consider the expected utility hypothesis:

∑ pi u(xi) Expectation Principle

Equation (1) of PT “generalizes” the Expectation Principle.

(1) ∑ ∏ (pi) v (xi)

This “generalization” does not incorporate new parameters into the old equation, but admits a wider
range of values for the old ones. More importantly, equation (1) changes in an irreducible way the
notion of utility found in Expectation: v (x) is not u (x). If equation (1) could be reduced to
Expectation, prospective utility (v (x)) would be reduced to expected utility (u (x)). Is this possible?
As long as the incommensurability thesis prevails, the answer seems to be negative.

Consider now the relation between equations (1) and (2) of PT. In this case too Kahneman and
Tversky said that a generalization exists (and a reduction is possible). In fact, equation (2) may be
reduced to (1) given some restrictive conditions. But here the utility concept assumed in both
equations is the same. This reduction –in contrast with the one mentioned first- involves no
incommensurability issues. The first kind of reduction (generalization) is philosophically
problematic and many economists don‟t pay full attention to the many problems involved in this
theoretical transformation.

Bibliography

Arrow, K., Colombatto, E., Perlman, M., and Schmidt, M., (eds), (1999), The rational Foundations
of Economic Behaviour –Proceedings of the IEA Conference held in Turin, Italy, Great Britain,
MACMILLAN PRESS LTD.

Barberis, N. and Thaler, R., 2003, “A Survey of Behavioral Finance”, in G.M Constantinides, M.
Harris and R. Stuls, Ed., Handbook of the Economics of Finance, Elsevier Science B.V.

Camerer, C., and Loewenstein, G., 2002, “Behavioral Economics: Past, Present and Future
(http://www.hss.caltech.edu/~camerer/ribe239.pdf)

Friedman, Milton and Savage, L. J., (1948), The Utility Analysis of Choices Involving
Risk”, Journal of Political Economy, LVI, pp. 279 – 304.

Friedman, M, and Savage, L., 1952, “The Expected utility Hypothesis and the
Measurability of utility” (The Journal of Political Economy, Vol. LX, Nº 6).

Friedman, M, (1953), “The Methodology of Positive Economics”, in Essays in Positive

Economics, Chicago: University of Chicago Press.

Friedman, M., 1955, “What All is Utility”, The Economic Journal, Vol. 65, Nº 259, pp. 405 – 409.

Gigerenzer, G., and Selten, R., eds., (2001), Bounded Rationality –The Adaptive Toolbox,
Massachusetts, The MIT Press.

Grether, D., and Plott, C., (1979), “Economic Theory and the Preference Reversal Phenomenon”,
American Economic Review 69: 623 – 638.

12
Hausman, D., (1992), The Inexact and Separate Science of Economics, Cambridge
University press, Cambridge.

Ho, T.H., Lim, N. and Camerer, C., 2005, “Modeling the Psychology of Consumer and Firm
Behavior with Behavioral Economics” (http://www.hss.caltech.edu/~camerer/JMRFinal.pdf)

Holt, Ch., (1986), Preference Reversals and the Independence Axiom, The American Economic
Review, vol. 7, Nº 3.

Kahneman, D, (2003a) “Maps of Bounded Rationality: Psychology for Behavioral Economics”,

AER, Vol. 93, Nº 5, pp. 1449 – 1475.

Kahneman, D., (2003b), “A psychological Perspective on Economics”, AER, Vol. 93, Nº 2, pp. 162
- 168.

Kahneman, D. and Tversky, A, (1979), “Prospect Theory: An Analysis of Decision

Under Risk”, Econometrica 47, pp. 263-291.

Levi, I, (1980), The Enterprise of Knowledge, Cambridge, MA: MIT Press.

Lichtenstein, S., and Slovic, P., (1971), “Reversal Preferences between bids and choicen in
gambling decision”, Journal of Experimental Psychology 89, pp. 46 – 55.

Loomes, G., and Sugden, R., (1982), Regret Theory: An Alternative Theory of Rational Choice
Under Uncertainty, The Economic Journal, Vol. 92, No. 368 (Dec., 1982), pp. 805-824.

Machina, M., (1982), “‟Expected Utility‟ Analysis Without the Independence Axiom”,
Econometrica, Volume 50, Nº 2.

Marschack, J., (1950), “Rational Behavior, Uncertain Prospects and Measurable Utility”,
Econometrica, Vol. 18, Nº 2, pp. 111 – 141.

Mongin, Ph., (1998), “Expected Utility Theory”, in Davis, J., B., Hands, D., W., and Mâki, U.,
(eds), The Handbook of Economic Methodology, Cheltenham, Edward Elgar, pp. 171 - 178.

Nurmi, H., 1998, Rational Behaviour and the Design of Institutions –Concepts, Theories and
Models, Edward Elgar, Cheltenham, UK.

Popper, K., (1965), Conjectures and Refutations –The Growth of Scientific Knowledge, London,
Routledge and Kegan Paul.

Schoemaker, P.J., The Expected Utility Model: Its Variants, Purposes, Evidence and Limitations,
JEL, June 1982.

Slovic, P., and Lichtenstein, S., (1983), Preference Reversals: a Broader Perspective”, The
American Economic Review, Vol. 73, Nº 4, pp. 596 – 605.

Starmer, Chris, (2000), “Developments in Non-Expected Utility Theory: The Hunt for a Descriptive
Theory of Choice under Risk”, Journal of Economic Literature, Vol., XXXVIII, pp. 332-382.

13
Thaler, Richard, 1986, “The Psychology and Economics Conference Handbook: Comments on
Simon, on Einhorn and Hogarth, and on Tversky and Kahneman”, The Journal of Business, Vol 59,
Nº 4, Part 2, pp. S279 – S284.

Thaler, Richard, 2000, “From Homo Economicus to Homo Sapiens”, Journal of Economic
Perspectives, Volume 14, Number 1, Winter 2000, pp. 133-141).

Tversky, A., (1999), “Rational Theory and Constructive Choice”, in Arrow, K., Colombatto, E.,
Perlman, M., and Schmidt, M., (eds), The rational Foundations of Economic Behaviour”, Great
Britain, MACMILLAN PRESS LTD.

Tversky, A. and Kahneman, D, (1981), “The Framing of Decisions and the Psychology of
Choice”, Science, New Series, Vol. 211, Nº 4481, pp. 453 – 458.

Tversky, A. and Kahneman, D. (1986), “Rational Choice and the Framing of Decisions”, The
Journal of Business, Vol. 59, Nº 4, Part 2, pp. S251 – S278.

Von Neumann, J. and Morgenstern, O., 1944, Theory of Games and Economic Behavior; Princeton
University Press.

1
Beliefs have to meet the axioms of probability theory and preferences should be complete, transitive and continuous.
2
Suppose you have three alternative options, namely A, B and C. The procedure for measuring utility can be posed in
three steps:
1) Choose the most and the less preferred options. Assign the numeral 1 to the former and 0 to the later. In our
example, suppose A=1 and C=0
2) To assign a numeral for B, we need to answer the following question: what is the probability p for which you
would be indifferent between B with certainty and the lottery that brings you A with probability p and C with
probability (1-p) ?
Explicitly written, you must choose p so that: B ˜ A(p) + C (1-p)
3) Call the numbers so assigned the utility of A, B and C.

See Alchian, 1953, for a detailed account of this method. Some comments would be useful here. This procedure
clarifies why Friedman reminds us several times that “utility is as measurable as temperature and length”. Once you
choose the extreme values of the scale (in our example, A=1 and C=0), the intermediate points in the scale are
determined in a subjective but non-arbitrary way. It is worth noticing also that this procedure presupposes the validity of
the extensionality thesis. To illustrate, suppose that A is for “Shakespeare”, B for “Cervantes” and C for “Goethe”.
Following the steps exposed above, the order remains A, B , C. Now, replace “Cervantes” with “ The one-armed man
of Lepanto”. It is possible now that the order be shifted, and that the individual chose A, C, B this time. Then under two
extensionally equivalent descriptions the same object “B” has two different utilities! But, so to speak, if the length of a
table changes according to the procedure used to measure it, would you still say that there exists something like “table
length”?
3
Preference reversals may be obtained in many other ways. Lichtenstein and Slovic (1971) discovered a form of
preference reversals attributed to the violations of the procedural assumption (that asserts that the way in which
preferences are elicited is independent of the particular method employed for that purpose). Another form of preference
reversals arise when the preference relation on two options is affected by the presence of a third alternative. This is the
so called “compromise effect” which results from a violation of the context-free requirement (Nurmi, 1998). This kind
of preference reversals have important social and political consequences.
4
As long as choices “reveal” attitudes, and these are emotional reactions that may be influenced by the particular
descriptions of the options involved, choices may not be represented in a consistent way. Emotions are beyond
consistency constraints (see Kahneman and Sugden, 2005, p.166-167)
5
It should be noted that this underlying common matrix cannot always be constructed (or discovered) by the mere fact
that choices are framed. The possibility of doing this is the distinguishing feature of framing effects.

14
6
This particular vision of the theory has been developed as an independent field of research, completely purged from
normative concerns and relatively insulated from the rest of the economic building, and many theorists advocate this
project without hesitation (Starmer, 2000).
7
“Preference reversals occur when individuals are presented with two gambles, one featuring a high probability of
winning a modest sum of money (the P bet), the other featuring a low probability of winning a large amount of money
(the $ bet). The typical finding is that people often choose the P bet but assign a larger monetary value to the $ bet”
(Slovic and Lichtenstein, 1983, p. 596)
8
“Procedure invariance plays an essential role in measurement theories. For example, the ordering of objects with
respect to mass can be established either by placing each object separately on a scale, or by placing both objects on the
two sides of a pan balance; the two procedures yield the same ordering, within the limit of measurement errors.
Analogously, the classical theory of preference assumes that each individual has a well defined preference order (or a
utility function) that can be elicited either by offering a choice between options, or by observing their reservation price.
Procedure invariance provides a test for the existence of a measurable attribute. It would have been difficult to attribute
mass to objects if the orderings of these objects with respect to mass were dependent on the measuring device.
Similarly, it is difficult to defend the proposition that a person has a well-defined preference order (or equivalently a
utility function) if different methods of elicitation give rise to different choices” (Tversky, 1999, p. 189).
9
The crucial role of extensionality in the measurement procedure is shown in note 2.
10
See Section I.
11
It seems that learning isn‟t an easy task at all, and only takes place when relatively exceptional conditions are
fulfilled. “The problem with many economic models of learning is that they seem to apply to a very static environment.
In fact, such models seem to be directly applicable only to the situation in which Bill Murray finds himself in the movie
Groundhog Hog Day. In the movie, Bill Murray is a TV weatherman sent to report on whether the groundhog sees his
shadow on Feb. 2. Murray‟s character ends up reliving the same day over and over again. Although he is a slow learner,
the opportunity to rerun the same day repeatedly, and to learn from the consequences of his actions each time, creates a
controlled experiment in which he is able to learn many things eventually, from how to prevent accidents to how to play
the piano. Alas, life is not like Groundhog Day. In life, each day is different, and the most important of life‟s decisions,
such as choosing a career or spouse, offer only a few chances for learning!” (Thaler, 1986, p.p. 135-136) . See also
Tversky and Kahneman (1986, pp. S274.S275).
12
See Barberis and Thaler (2003),, and Ho, Lim and Camerer (2005)
13
See Simon (1955) and Gigerenzer and Selten (2001)
14
See Popper (1965), chap. 16.
15
Due to the fact that the adopted point of reference determines the set of prospects to be considered, the predictive task
in spontaneously formed choice-settings is increasingly difficult because normally there is an indefinite number of
potential reference points, each one as acceptable as the others. Turning once more to the case of the betting agent, he
could have considered a wider scope of options than the ones mentioned earlier. He could have taken into account not
just the present day, but his past performance along the course of the week. Today he has lost $140 during the whole
journey, but yesterday he won $280. This means that into the last 48 hours he has a net gain of $140. If the individual
frames his situation in this form we have an edition of his options that is quite different from the two ones first
described.

E: Not to bet E = $140

F: Bet F = ($280, 1/15; $130, 14/15)

It is easily seen that an arbitrarily number of points of departure may be chosen by the agent as reference point.
However, though the pair of resultant prospects may be different in each of the many cases, they all may be grouped
into just two categories: gains and losses. And PT predicts a well defined pattern in each case: the agent will show risk
aversion regarding gains and risk seeking regarding losses. In the particular case described above, he will choose E.
16
How far this strategy can be successfully pushed beyond the usual economic realm is hard to say. As long as the
complex emotional (and ethical) feelings that have an impact on real life decisions cannot be introduced into the simple
taxonomy of Prospect Theory (not to mention the ethical reactions that do not seem to be captured in terms of
probabilities and gains and losses), the project of the so called “economic imperialism” in the domain of decisions is
doomed to fail.

15
17
“Theories in behavioral economics also strive for generality –e.g., by adding one or two parameters to standard
models. Particular parameter values then often reduce the behavioral model to the standard one…” (Camerer and
Loewenstein, 2002, p. 2).

16