Psychology and Economics

by Matthew Rabin
Citation
Title:
Psychology and Economics
Author:
Matthew Rabin
Year: 
1998
Publication: 
Journal of Economic Literature
Volume: 
36
Issue: 
1
Start Page: 
11
End Page: 
46
Publisher: 
Language: 
English
URL: 
Select license: 
Select License
DOI: 
PMID: 
ISSN: 
Abstract:

Psychology and Economics

MATTHEW RABIN

University of Calijornia at Bel-keley

For co~)lr~icrlti this es.iny or rlisctr.siior, 011

o!lr/ Ricllnrrl Thnlcr. Co-oiitl~orr orr rereor-cli rclnler/ to the to}~ic.i es.soy i~lc~'rtrle Dntiirl Boccr~in~,, Debornll Jfi~ic/zo~t,

Ted O'Do~toghire, n~trl Joel Schrng Helpful resenrcll n.ssistn~,ce tcos pr-ocirlcrl by Gndi Bnrlccy, Sikki Blnsbcrg, Gnil Brcri~rn~i,

Pnlrl Ellick- soil, April Frrl~lco, hfnrcns Hcrlg, Brtice H.911, Ji~i I'I'oo Jzirig, orlrl cspcciolly Stccerl Blntt, Jirrlr,iy Chnn. Erik Eyster, orirl Clnr-o IVo~tg. I orlt extre~)lcly gr-otcjirl to tllc Votio~tol Scicrrcr, R11s.icl1 Soge, Alfred P. Slon~z, nrld JfncA~iltur Foirndntior~s forj~io~,cinlstrppo~t or1 ~wsenrch relofed to thiv cssny. A rnore cerbo.se nllrl citntio~l-lieouy cersioll oftllis e.ssny, eorerirlg o fc~c orlditiorlol topics, is nuoiloblc or Crliccr.sify of Cnllforrlio nt Berkeley Deportrllort ofEcorio~)l- ics 1Vorlti1ig Poper 1'0. 97-2.51 ,

1. Introduction person's preferences are often determined by changes in outcomes relative ECAUSE PSYCHOLOGY systemati-to her reference level, and not merely cally explores human judgment, be- by absolute levels of outcomes. In par- havior, and well-being, it can teach us ticular, relative to their status quo (or important facts about how humans differ other reference points), people dislike from the way they are traditionally de- losses significantly more than they like scribed by economists. In this essay I gains. I then discuss how people depart discuss a selection of psychological find- from pure self-interest (as narrowly de- ings relevant to economics. fined) to pursue "other-regarding" goals

Economics has conventionally as-such as fairness, reciprocal altruism, su~ned that each individual has stable and revenge. and coherent preferences, and that she By focusing on evidence consistent rationally maximizes those preferences. with rational choice, Section 2 reviews Given a set of options and probabilistic evidence that requires relatively small beliefs, a person is assumed to maxi-modifications of the familiar econoinic mize the expected value of a utility framework. Section 3 reviews research function, U(x). Psychological research on biases in judgment under uncersuggests various modifications to this tainty; because these biases lead people conception of human choice. Section 2 to make errors when attempting to provides examples of what psychological maximize U(r), this research poses a research can teach us about making more radical challenge to the economU(r) more realistic than under standard ics model. I discuss two biases in de- economic assumptions. I begin by dis- tail-how we infer too much from too cussing research which suggests that a little evidence and how we misread evi-

11

dence as confirming previously held hy- potheses-and briefly explain a few more. I also discuss some of the psycho- logical evidence on whether and when experience and learning lead people to overcome these biases.

The array of psychological findings reviewed in Section 4 points to an even more radical critique of the economics model: Even if we are willing to modify our standard assurnptions about U(x), or allow that people make systematic errors in judgment when attempting to maximize U(x), it is sometimes mislead- ing to conceptualize people as attempt- ing to maximize a coherent, stable, and accurately perceived U(x). I begin by reviewing evidence that people have difficulties evaluating their own prefer- ences-we don't always accurately pre- dict our own future preferences, nor even accurately assess our experienced well-being from past choices. I then dis- cuss research on framing effects, pref- erence reversals, and related phenorn- ena in which people prefer some option x to y when the choice is elicited one way, but prefer y to r when the choice is elicited another way. I conclude Sec- tion 4 with a discussion of self-control problems and other phenomena that arise because people have a short-run propensity to pursue immediate gratifi- cation that is inconsistent wit11 their long-run preferences.

The customary disclaimer of review essays-that they do not pretend to be exhaustive-applies even more here than usual: The topics covered are only a sinall fraction of economically relevant psychology. Some omitted topics of obvious economic relevance are non-expected utility; status, envy, and social comparisons; conforlnity and herd be- havior; self-serving biases and motivated cognition; the tendency of "extrinsic motivation" (e.g., organizational incentive schemes) to drive out "intrin- sic motivation" (e.g., the internal drive to excel at your vocation); and a mass of research on learning from cognitive and developmental psychology. 1

Two other limitations of scope are conspicuous. First, I emphasize what psychologists and experimental economists have learned about people, rather than IZOW they have learned it. Conse- quently, the focus of this essay is not at all on experimental methods per se.2 Second, my primary emphasis is on re- viewing what psychology tells us about modifying our general assumptions about individual behavior, rather than on any of its specific economic applica- tions. For instance, while fairness and reference-level effects (reviewed in Section 2) and framing effects (reviewed in Section 4) are likely to con- tribute to downward stickiness in wages, I leave the exploration of these implications to other forums.

Mainstream economics employs a powerful combination of methods: lnethodological individualism, mathematical for~nalization of assumptions, logical analysis of the consequences of those assumptions, and sophisticated empirical field testing. I believe these methods are tremendously useful, and

' Another to ic I have omitted is "non-psycho- logical" mode?s of bounded rationality Researchers have formulated models of bounded ra- tionality (based on intuition, computer science, or artificial intelligence) meant to ca ture cognitive limits of ecollornic actors, but whic R do not invoke research on the specific patterns of errors that hu- man beings make. For an excellent review of the role of bounded rationality in economics, see John Conlisk (1996).

2 Descriptions of experimental methods employed by psychologists can be found in virtually any psychology text. For an excellent set of reviews of the results of experimental economics, see Tohn Kagel and Alvin Roth (1995). While most of the findings 1 resent in this essay come from psychologists, L oken draw on evidence from ex- perimental economics as well. My focus, however, is on experiments that explore the sychological nature of individuals, not 011 resear2 seeking to replicate economic institutions in the laboratory.

an underlying premise of this essay is that we should strive to understand psy- chological findings in light of these methods.3 These neth hods raise problems for doing full justice to behavioral reality: Because of the high premium economics places on the logic and pre- cision of arguments and the quantifica- tion of evidence, attending to all facets of human nature is neither feasible nor desirable. The realization that many de- tails of human behavior must be ignored, however, should not license in- stitutionalized cornplacency about the behavioral validity of our assumptions; "tractability" and "parsimony" should be guiding principles in our efforts to make our research more realistic, not pretexts for avoiding this task. As it now stands, some important findings seem tractable and parsimoni- ous enough that we should begin the process of integrating them into economics. Incorporating other findings will take longer. But even in those cases, economists ought to become

aware of the shortco~nings of our mod- els, regret these shortcomings, and keep our eyes open for ways to remedy them. While I conclude in Section 5 by briefly discussing the general endeavor of incorporating psychological findings into econornics, I have omitted from this essay many of the classical meta- arguments about whether it is "possi- ble" that behavioral research identifies important departures from economists' habitual assu~nptions about human be- havior. Of course it is possible, and in fact it is true. In this essay, I review

7 Indeed, I have organized the essay to reflect this premise. At times, difficulties in orgallizatioll belie deeper difficulties of how to fit this research into the econornics framework. For instance, while Section 2 discusses ap roaches to rnodeling loss aversion fully withill 1; frame-

t le rational-choice work, material in Section 4 suggests loss aversion isn't always best conceived of as coinpatible with utility maximization.

what research shows some of those de- partures are.

2. Preferences

This section discusses how psychological findings suggest we modify the utility functions economists employ.

A. Reference Lecels, Adaptation, ccnd Losses

Overwhelming evidence shows that humans are often rnore sensitive to how their current situation differs froin some reference level than to the abso- lute characteristics of the situation (Harry Helson 1964). For instance, the same temperature that feels cold when we are adapted to hot temperatures may appear hot when we are adapted to cold temperatures. Understanding that people are often more sensitive to changes than to absolute levels suggests that we ought incorporate into utility analysis such factors as habitual levels of consumption. Instead of utility at time t depending solely on present con- sumption, ci, it may also depend on a "reference level," rt, determined by fac- tors like past consumption or expecta- tions of future consumption. Hence, in- stead of a utility function of the form ut(rt;ct), utility should be written in a more general form, ur(rt;ct). IVhile solne economists have over the years incorpo- rated reference dependence into their economic analysis, it is fair to say that the ways and degrees to 11~11ich refer- ence points influence beha\,ior have not fully been appreciated by econo~nists.~ Researchers have identified a per\,asive feature of reference dependence: In a wide variety of domains, people are sig- nificantly more averse to losses than

4 There have been sonle earlier esarnples of economists considering reference depeildellce (Jaines Duesenben . 1949; Richard. Easterlin 1974; and Harl Ryder mdceoffrey Heal 1973).

they are attracted to same-sized gains (Kahneman, Jack Knetsch, and Thaler 1990). One realm where such loss aver- sion plays out is in preferences over wealth levels. Tversky and Kahnernan (1991) suggest that in the domain of money (and in others where the sizes of losses and gains can be measured), people value modest losses roughly twice as much as equal-sized gains. That the displeasure from a monetary loss is greater than the pleasure from a same-sized gain is also implied by a concave utility function, which economists typically use as the explanation for risk aversion. But loss aversion says that the value function abruptly changes slope at the reference level, so that peo- ple dislike even small-scale risk. For in- stance, most people prefer their status quo to a 50/50 bet of losing $10 or gain- ing $11. The standard concave-utility- function explanation for risk aversion is simply not a plausible explanation of such risk attitudes. Rajnish Mehra and Edward Prescott (1985) and Larry Ep- stein and Stanley Zin (1990) have, for instance, observed that the expected-utility framework cannot simultaneously explain both the small-scale and large- scale risk attitudes implied by macro data, and Rabin (1997) provides a "cali- bration theorem" that indeed shows that no concave utility function can si- multaneously explain plausible small-scale and large-scale risk attitudes. A reference-based kink in the utility func- tion is required to explain such risk atti- tudes within the expected-utility frame- work.5

Loss aversion is related to the strik- ing endotuinent effect identified by Thaler (1980): Once a person comes to possess a good, she immediately values

Uzi Segal and Avia Spivak (1990),aid others, however, develop a model of such first-order risk aversion outside the expected-utility framework.

it more than before she possessed it. Kahneman, Knetsch, and Thaler (1990) nicely illustrate this phenomenon. They randomly gave mugs worth about $5 each to one group of students. Minimal selling prices were elicited from those given the mugs (with an incentive-com- patible procedure that ensured honest reports). Minimal "pricesn-sums of money such that they would choose that sum rather than the mug-were elicited from another group of subjects not given mugs. These two groups, "sellers" and "choosers," faced precisely the same choice between rnoney and mugs, but their reference points differed: Those who were randomly given mugs treated the mugs as part of their refer- ence levels or endowments, and consid- ered leaving without a mug to be a loss, whereas individuals not given mugs con- sidered leaving without a mug as remaining at their reference point. In one experiment, the median value placed on the mug was $3.50 by choosers but $7.00 by sellers. Such results have been replicated repeatedly by many researchers in many contexts.

As established by Knetsch and John Sinden (1984), William Samuelson and Richard Zeckhauser (1988), and Knetsch (1989), a comparable phenomenon-the status quo bias-holds in multiple-good choice problems. Here, loss aversion implies that individuals tend to prefer the status quo to changes that involve losses of some goods, even when these losses are offset by gains of other goods. Knetsch and Sinden (1984) and Knetsch (1989), for instance, randomly gave stu- dents either candy bars or decorated mugs. Later, each student was offered the opportunity to exchange her gift for the other one-a mug for a candy bar or vice versa. Ninety percent of both mug- owners and candy-owners chose not to trade. Because the goods were allocated randomlv and transaction costs were minimal, the different behavior for the two groups of subjects presumably reflected preferences that were induced by the initial allocation. Knetsch (1989) and Tversky and Kahneman (1991) show that such preferences can be use- fully captured by utility functions de- fined over reference levels as well as consumption levels.

In addition to loss aversion, another important reference-level effect is di- minishing sensitivity: The marginal ef- fects in perceived well-being are greater for changes close to one's refer- ence level than for changes further away. As Kahnernan and Tversky (1979) note, diminishing sensitivity is a perva- sive pattern of human perception, where our perceptions are a concave function of the magnitudes of change. For instance, we are more likely to dis- crilninate between 3" and 6" changes in room temperature than between 23" and 26" changes. This applies to both increases and decreases in temperature. In the context of preferences over uncertain monetary outcomes, diminishing sensitivity implies that the slope of a person's utility function over wealth becomes flatter as her wealth gets further away from her reference level. Because for losses relative to the reference level "further away" implies lower wealth levels, di- minishing sensitivity has a provocative implication: 1,T7hile people are likely to be risk averse over gains, they are often risk-loving over losses. Kahneman and Tversky (1979) found that 70 percent of subjects report that they would pre- fer a 3/4 chance of losing nothing and 1/4 chance of losing $6,000 to a 2/4 chance of losing nothing and 1/4 chance each of losing $4,000 or $2,000.. Be- cause the preferred lottery here is a mean-preserving spread of the less-pre- ferred lottery, the responses of 70 per- cent of the subjects are inconsistent with the standard concavity assumption."

In order to study the effects of refer- ence points in a dynamic utility-maximi- zation framework. we need to take into account how people feel about the ef- fects their current choices have on their future reference points. To maximize their long-run utilities when reference points matter, people must determine two things beyond how they feel about departures from reference points: How current behavior affects future reference points and how they feel about changes in their reference points. Economists Rvder and Heal 11973)

,

model the process by which reference points change with the formula rt = act-I+ (1-a)r,-l, where a E (0,l) is a parameter measuring how quickly peo- ple adjust their reference points. In a rational-expectations model, people will take this formula into account when maximizing their long run well being. Such an account of how reference levels are determined seems intuitive, though there seems to be little evidence on this topic.7 Evidence is similarly sparse about how people's preferences depend on changes in reference points? \T7ith-

6 Because this "risk-loving" tendency is in con- flict with the diminishing marginal utility of in- come, however, it has often been conjectured that risk aversion may reappear for lar e losses that might push a person to extremely Bow consuInp- tion levels.

7 Bowman, Minehart, and Rabin (1997) combine the Ryder and Heal approach of rational-eqectations, ref- erence-dependent utilities with a utility function that incorporates loss aversion and di-minishing sensitivity. Duesenberry (1949) implicitly posited a reference function closer to r-, = Mau {c, : T < ti-that is, a per- son's reference level was her highest past consulnption level.

8 For exceptions, see Loewenstein and Nachuln Sicherlnan (1991), Robert Frank (1985 ch. 15, 1989), and Frank and Robert Hutchens (1993), who have identified a tendency for people to pre- fer income profiles that are steady or increasing over time to same-sized decreasing profiles, strongly indicating that people prefer not to be- come accustolned to levels of consuin tion they knoiv they cannot inaintain For consiAration of

out assumptions about these relationships, there will be a relatively small set of circumstances where loss aversion and diminishing sensitivity can be inte- grated into models of dynamic utility maximization.

There have been some initial attempts to study loss aversion, the endowment effect, and the status quo bias in economic contexts. Raymond Hartman, Michael Doane, and Chi-Keung Woo (1991) find empirical evidence for the existence of a status quo bias in consumer demand for electricity; Bow- man, Minehart, and Rabin (1997) repli- cate evidence by John Shea (1995a, 199513) that consumers are more averse to lowering consumption in response to bad news about income than they are to increasing consumption in response to good news, and argue that this behavior is a natural implication of loss aversion.

B. Social Preferences and Fair Allocations

It is common for undergraduates to encounter the following quote from The Wealth of Nations (Adam Smith 1776, pp. 26-27) at the beginning of Econom- ics 1:

It is not from the bellevolellce of the butcher, the brewer, or the baker that we expect our dinner, but from their regard for their own interest. We address ourselves not to their humanity, but to their self-love, and never talk to them of our necessities, but of their advantage.

There is not much to disagree with in Smith's poetic analysis of the motivations driving most market behavior, and probably no other two-word description of human motives comes close to "self- interest" in explaining economic behav- ior. Yet pure self-interest is far from a

some other factors involved in the preference for increasing wage profiles, see Loewenstein and Sicherinan (1991, pp. 76-82).

complete description of human motiva- tion, and realism suggests that economists should move away from the presumption that people are solely self-interested. Robyn Dawes and Thaler (1988, p. 195) eloquently set parame- ters for this endeavor:

In the rural areas around Ithaca it is coininon for farmers to put some fresh produce on the table by the road. There is a cash box on the table, and customers are expected to put money in the box in return for the vegetables they take. The box has just a small slit, so money can only be put in, not taken out. Also, the box is attached to the table, so no one can (easily) make off with the money. We think that the farmers have just about the right model of human nature. They feel that enough people will volunteer to pay for the fresh corn to make it worthwhile to put it out there. The farmers also know that if it were easy enough to take the money, someone would do so.

Examples of economic behavior induced by social goals are donations to public television stations, voluntary re- ductions of water-use during droughts, and conservation of energy to help solve the energy crisis (Kenneth Train, Daniel McFadden, and Andrew Goett 1987). One context in which fairness has been studied is inonopoly pricing (Thaler 1985; and Kahneman, Knetsch, and Thaler 1986a, 1986b). Might consumers see the conventional monopoly price as unfair, and refuse to buy at that price even when worth it in material terms? If this is the case, then even a profit-maximizing monopolist would price below the level predicted by stan- dard economic theory. Finally, hundreds of researchers in psychology, in- dustrial relations, and economics have investigated how equity, fairness, status-seeking, and other departures from self-interest are important in em- ployee behavior. Indeed, the massive psychological literature on equity the- ory was developed largely in the context

17

of industrial relations (Stacy Adams 1963; Akerlof 1982; and Akerlof and Janet Yellen 1990).

Experimental research makes clear that preferences depart from pure self- interest in non-trivial ways: Subjects contribute to public goods more than can be explained by pure self-interest; they often share money when they could readily grab it for themselves; and they often sacrifice inoney to retaliate against unfair treatment. Debates remain, as some researchers have empha- sized the possibility of producing laboratory environments (e.g., well-or- ganized, private-information double-auction spot markets) that induce behavior that is closer to purely self-interested than is behavior in other settings. But, disentangling debates over the nature of preferences from strong ancillary assumptions about which institutions and environments matter in the real world, there does not appear to be debate among behavioral researchers about whether underlying preferences depart non-trivially from pure self-interest.

One form of social motivations on which economists have focused is al-ttmUism,in the sense that put posi-

tive On the of others. Roughly, this approach says that person 1 acts as if she is maximizing prefer- ences of the form ul(x) (1-')'~I(T) +r.nz(x),where nl(x)is person 1's "ma- terial well-beingn from x, and n2(x) is person 2's material well-being. By letting r be small, we can capture the idea that are mostly self-in-

terested; assulning >O, we can investitlate when and how concern for others affects behavior and welfare. 'Itruism may P ~ capture important phenomena in many contexts. ~~t there is a Inass of experievidence that indicates it is often a very wrong description of social

preferences."~ get a sense for how so- cial preferences differ from simple al- truism, consider what can be called "be- havioral distributive justice": How do people choose to divide resources?lO There are two aspects to this question: First, what do people, when disinterested, feel are proper rules for alloca- tion? Second, to what degree do people sacrifice self-interest for the sake of these principles? Very roughly, imagine that person 1's utility function takes the form U1= (1-r).nl+ ~.117~(n~,n~),

where W1 is person 1's view of the proper allo- cation, and IIl is (as above) her self-in- terested payoff, and 0 $r 5 1 measures the weight this person puts on self-interest versus proper allocation. To understand the implications of fairness and justice for behavior, we need to know both the nature of the W1 func- tion, and the level of r.

To address the question of disinter- ested assessments, suppose two people together find $10 worth of money or other goods on the ground. How would the average person, acting as a third

9See Dennis Krebs (1970, 1982) for some ps>- chological evidence on altruism and need-based helpiGg motives. Even where simple altruism may adequately describe behavior (e.g., in donating to charity), psychological research may be of value for welfare economics. In particular, research has

explored whether people who help others do so for "truly" altruistic reasons, in the sense that the actions they take will lower their experienced well-being., or whether thev do so "onlv" to allevi- ate painfJ guilt, etc, DO ~Auhelp a stl-bnger when inconvenient because it is the right thing to do even tllough it makes you worse Or do you like bringing joy to others in the same way you like eating apples? Or do you know you will have un- pleasant, guilt-induced nightmares if you don't help? These distillctiolls will be important in wel- fare analvsis: Even if we think behavior is de- scribed 6y Max C~(.Y)=(1-r-).nl(r)+rnp(.x),we must still resolve whether Person 1's experienced

~~~~

well-being is described by Ul(i) or FIl(x),

101 use the term "behavioral distributive jus- tice" to emphasize that I ain reviewin evidence 011 how people actually feel about distr&utive jus- tice, not normative or philosophical questions of what is the proper notion of distributive justice.

party, decide to split the surplus be- tween the two?ll The simple-altruism perspective would suggest solutions such as giving the surplus to the person who is poorer-or for non-money goods-the person who values the goods more. But research shows that many people in many contexts do not find the "maximal-benefits" criterion at- tractive. One simple alternative norm is prevalent: Resources should be split 50-50. Except in extreme cases, often we ignore issues of relative usefulness and feel that goods should be divided equally.

But the maximal-benefits criterion fares even worse than this. Many people feel goods should be allocated accord- ing to a "maximin" criterion which equal- izes welfare improvements between the two people. That is, disinterested peo- ple often seem to maximize preferences of the form WO= Min[TIl,n2], where IIl and nzare the gains in utility from di- viding resources. Because it takes a greater allocation to increase the utility of a person who values a resource less, the maximin criterion typically implies that more than half the resources are al- located to the person who values those resources less. Consider the following hypothetical situation that Maya Yaari and Menahem Bar-Hillel (1984, p. 8) posed to 163 subjects:

Q1:A shipment containing 12 grape- fruits and 12 avocados is to be distrib- uted between Jones and Smith. The following information is given, and is knov7n also to the two recipients:

IJ I assume the surplus is found so as to con- sider the thought experiment that neither party deserves the money more than the other. Desert will obviously be relevant in many situations-and the massive psychological literature on "equity theory" shows that people feel that those who have put more effort into creating resources have more claim on those resources (Ellen Berscheid, David Boye, and Elaine Walster 1968).

-Doctors have determined that Jones's metabolism is such that his body derives 100 milligrammes of vita- min F from each grapefruit consumed, while it derives no vitamin F whatsoever from avocado.

-Doctors have also determined that Smith's metabolism is such that his body derives 50 milligrammes of vita- inin F from each grapefruit consumed and also from each avocado consumed.

-Both persons, Jones and Smith, are interested in the consumption of grapefruit or avocados only insofar as such consumption provides vitamin F-and the more the better. All the other traits of the two fruits (such as taste, calorie content, etc.) are of no consequence to them.

-No trades can be made after the di- vision takes place.

How should the fruits be divided be- tween Jones and Smith, if the division is to be just?

Grapefruits are worth more to Jones than to Smith, and avocados are worth- less to Jones. The "socially efficient," metabolism-maximizing allocation, therefore, is for Jones to get all the grapefruits, and Smith to get all the avocados. (Such an allocation would seem to accord somewhat to 50-50 norms.) The maximin allocation, how- ever, would be to give eight grapefruits to Jones, and give the remaining grape- fruits and all the avocados to Smith.

Subjects were given a menu of five different allocations, and asked to state which allocation they found the most "just." Yaari and Bar-Hillel (1984, p. 10) report the percentage of respondents who chose each of the five alloca- tions. The five allocations are denoted by the number of grapefruits and avoca- dos to be distributed to Jones and Smith (denoted by their initials):

Distribution % of respondents (J:fX, S:6-6) 8 (J:&O, S.6-12) 0 (J:%O, S:412) 82 (J:9-0, S:3-12) 8 (J:12-0, S:O-12) 2

The vast majority of respondents chose to equalize welfare gains (J:8-0, S:4-12), rather than maximize total wel- fare gains (J:12-0, S:0-12). Because the maximin solution is subtle, the results strongly suggest subjects thought about the problem, and did not merely choose some simple focal point.l3

Yaari and Bar-Hillel (1984, p. 11) then tested the robustness of the maxi- min criterion by posing a variant of the original question, where subjects were told that Smith derives only 20 milligrams of Vitamin F from both fruits, rather than 50 milligrams. Subjects sup- ported the maximin criterion just as strongly-now imposed at a greater cost to total social benefits. Indeed, Smith is now given more than half the grapefruits, though they are far less valuable to Smith than they are to Jones. Posit- ing that, "Sooner or later . . . [the maximin criterion] runs the risk of be- coming morally unsound," Yaari and Bar-Hillel push the limits of the maxi- min criterion by telling subjects that Smith derives only 9.1 milligrams of Vi- tamin F from both fruits, so that grape- fruits are 11 times more valuable to Jones than Smith. While a far greater number of respondents now seemed

'"he very subtlety of the maximin solution, however, points to a concern with these data that Yaari and Bar-Hillel themselves draw attention to: The questions here were posed in writing at the end of a college entrance exam, which may have induced subjects to treat the questions as lelns to solve, rather than as opinions they s rob

ould express. Informal surveys by Yaari and Bar-Hillel in very different contexts, and experimental results reported below su gest that the problem-solving context does not Bully explain the patterns they found in this study.

willing to tolerate unequal welfare gains, still only 12 percent of respon- dents felt that all 12 grapefruits should be given to Jones. Moreover, only 18 percent thought that Jones should get more than half the grapefruits while 38 percent thought Smith should get more than half.

While Yaari and Bar-Hillel study what disinterested people consider a proper allocation rule, to address the question of how people trade off self- interest against justice, we need a situ- ation where the allocator is not disinterested. If, for instance, an allocator must unilaterally choose how to divide money between herself and a second party, her choice will depend both on what she feels is a just allocation and on how much she values a just outcome relative to her self-interest. Andreoni and John Miller (1996) consider just such a situ- ation.13 They asked each subject to uni- laterally allocate money between herself and an anonymous second party. Sub- jects were given various "exchange rates" for allocating money; for instance, some subjects were told that for every $1 they sacrificed, the other party would get $3. Over half the subjects be- haved in a way significantly inconsistent with pure self-interest. Among those whose behavior was least self-interested, two thirds chose approximately the maximin allocatio~l rule, and one third chose approximately the dollar-maximizing allocation. These results

1:3Andreoni and Miller's and other monetary- stakes experiments are also useful for allayin concerns one might have re arding the Yaari a13 Bar- Hillel survey, which asffed subjects to say what they considered the jtrst allocation. If subjects considered "justice" to be only one cornpollent of a proper allocation rule, they may not have inter- preted the question as meaning "How would you choose to allocate between these two parties?" An- dreoni and Miller (1996) and bargaining ex eli ineirts typically do not prompt subjects to evakaik allocations according to any criterion-but merely to make a choice.

suggest that the maximin rule resonates with Inany people even when allocating money, and even when self-interest is at stake. Allowing proportions of a "pie" to have different monetary values to different parties has also been a theme in the experimental bargaining research (Roth and J. Keith Murnighan 1982; and Kagel, Chung Kiln, and Donald Moser 1996). In these settings, disentangling self-interested bargaining strategies from preferences for just allo- cations is quite difficult, but results seem comparable to Andreoni and Miller's.

In moving from abstract, context-free allocation problems to everyday economic fairness judgments, things become significantly more complicated. First, as elsewhere, reference levels are crucial. Thaler (1985) and Kahneman, Knetsch, and Thaler (1986a, 1986b) demonstrate that loss aversion plays a very strong role in people's notion of fairness; firms have more of an obliga- tion not to hurt workers or customers relative to reference transactions than they have to improve the terins of trade. Relatedly, people's general per- ceptions of fair behavior may adjust over time. Kahneman, Knetsch, and Thaler (1986a, p. 730) argue that, "Terms of exchange that are initially seen as unfair may in time acquire the status of a reference transaction. Thus, the gap between the behavior that people consider fair and the behav- ior that they expect in the marketplace tends to be rather small." Robert Franciosi et al. (1995) experimentally support this hypothesis by testing reactions to unfair price increases in a laboratory posted-offers market; they show that the role of fairness considerations in price-determination diminishes with repetition, suggesting that in competitive spot markets people inay eventually come to believe that the prevailing market price is fair. Other experiments find virtually no change in either behavior or perceptions of fairness over time (Fehr and Armin Falk 1996). In any event, be- cause adjustments of fairness judgments are not immediate, fairness considera- tions may help explain the sort of medium-run wage and price stickiness studied by macroeconomists, and evidence that market outco~nes are likely to be self-interested exists only for com- petitive spot markets. See, for example, Fehr, Erich Kirchler, and Andreas Weichbold (1994) for an experimental study of labor markets where behavior never converges to the self-interested outcoine.

Finally, in attempting to capture behavioral findings with models of social preferences, it is important to note that people seem to iinplicitly (but pervasively) consider equitable sharing over clznnges in total endowments, not total endowments theinselves. Preferences defined over final wealth states cannot plausibly explain rules such as 50/50 sharing or the maxi- min criterion. With plausible assumptions about initial endowments entering any moderate size division-of-the-pie situation, any social welfare function defined with respect to overall consumption levels will almost always yield all-or-nothing allocations. Apparently, people generally have a one-pie-at-atime conception of fair-division problems. This is not an insurmountable tendency: If people are presented with several allocation probleins together, they will likely attend to the overall implications of their several choices. Nevertheless, any attempt to capture behavioral norms of fairness and dis- tributional justice with formal models of social preference must confront the "piecemeal" nature of these norms.

Evidence in support of reciprocal al- truism comes from experimental studies of the voluntary provision of public goods. Dawes and Thaler (1988) con- clude that, for most experiments of one- shot public-good decisions in which the individually optimal self-interested con- tribution is close to zero percent, the contribution rate varies between 40 percent and 60 percent of the socially opti- mal level.14 Many of these experiments hint that contributions toward public goods are not the result of simple altru- ism, though the evidence for reciprocal altruism is varied, and often indirect. For instance, Rachel Croson (1995) finds a strong positive correlation be- tween subjects' contribution levels to a public goo"d and their beliefs about how much others were contributing. Further indirect evidence is that pre-decision communication greatly enhances coop- eration (David Sally 1995). One reason why coinmunication enhances contribu- tions may be that reciprocal altruism es- sentially turns public-goods situations into "coordination games," where high contributions are efficient equilibria, and low contributions are inefficient equilibria. As in general coordination games, therefore, pre-game communication can help players coordinate on the efficient equilibria.

Indeed, the common emphasis when describing the prisoner's dilemma on

1-1Many of these experiments are problematic in that their null hypothesis of completely self-inter- ested behavior corresponds to zero contributions, where zero colltributions is also the most extreme behavior subjects could exhibit. Therefore, all de- partures from full rationality are necessarily in the direction of "generous" behavior. Andreoni (1995) shows that, by a very conservative estimate, at least half the contributions to public goods are in- tentional rather than "errors." See also Claudia Keser (1996) for evidence that generous contribu- tions are not merely an artifact of experimental design.

the inability of the two captured prison- ers to communicate with each other in- dicates that we implicitly believe that the prisoner's dilemma also really amounts to a coordination game. If de- fecting were truly a dominant strategy, pre-game communication would not matter. More direct evidence of recip- rocal altruism, in the context of the prisoner's dilemma, comes from Shafir and Tversky (1992). When subjects were told that their anonymous partner in a prisoners' dilemma had cooperated, 16 percent also cooperated; when sub- jects were told that their partner did not cooperate, only 3 percent cooperated.

Reciprocity motives manifest themselves not only in people's refusal to cooperate with others who are being uncooperative, but also in their willing- ness to sacrifice to hurt others who are being unfair. A consumer may refuse to buy a product sold by a monopolist at an "unfair" price, even if she hurts her- self by foregoing the product. An employee who feels he has been mistreated by a firm may engage in costly acts of sabotage, perhaps to the point of violently retaliating against his employ- ers. Members of a striking labor union may strike longer than is in their mate- rial interests because they want to pun- ish a firm for being unfair.

A crucial feature of the psychology of reciprocity is that people determine their dispositions toward others accord- ing to motives attributed to these oth- ers, not solely according to actions taken. When motivated by reciprocal al- truism, for instance, people differenti- ate between those who take a generous action by choice and those who are forced to do so. Demonstrating both the basic principle of reciprocity and the role of volition, Richard Goranson and Leonard Berkowitz (1966, p. 229) con- ducted an experiment in which confed- erates posing as subjects were in a posi- tion to help real subjects fill out some worksheets. One third of the subjects were told that the confederate had vol- untarily offered to help; one third were told that the experimenter had instructed the confederate to help; and one third were told that the confederate might be willing to help, but the confederate was instructed to refuse to help. When the subjects were later given an opportunity to assist the confederates, they reciprocated earlier help, but did so significantly more when it was voluntary than when it was invol- untary.

Volition is also central to the propen- sity to retaliate against negative actions. Sally Blount (1995) asked subjects about their willingness to accept take- it-or-leave-it offers made by anonymous other parties on how to split $10.15 One group of subjects was told that the "ulti- matum" was coming from anonymous other students, and that their responses would determine the division between them and these anonymous other students. Another group was told that a third party (also an anonymous student) was to determine the offer made. In this variant, the person who would be hurt by a subject's decision to reject an offer did not participate in the offer, and the third party who made the offers would not be affected by the subject's decision. A final group of subjects were told that the offer would be generated randomly by a computer-simulated rou- lette wheel. In one study, the average minimal acceptable offers for those groups were $2.91, $2.08, and $1.20. That is, people did reject very low of-

1"he "ultimatum ame" of the sort studied b

K

Bloullt war first deve oped by Werner Guth, Roi Schmittberger, and Bernd Schwarze (1982). For reviews of the (massive) literature developed since, see Thaler (1988), Guth and Reinhard Tietz (1990), and Calnerer and Thaler (1995).

fers even if computer or third-party generated, but were less keen to reject offers which were not the result of voli- tion by the person who would be hurt by the rejections.

The importance of intentions goes even further, however, than considera- tion of whether a person's actions are voluntary. Suppose, for instance, you are eating lunch with parents and a child when all of a sudden your hands flail across the table and knock a pitcher of water all over the child. How do the parents react? If they thought your goal was to spill water on the child, they are probably angry. If they thought you were worried that pitcher was precariously perched next to the child, and that your flailing arms were an uncoordinated attempt to prevent a spill, they are probably less angry.1"

Such examples indicate that interpreting other peoples' motives depend on what we believe their beliefs about the consequences of their actions are. Another example of the importance of beliefs is, if you think somebody has been generous to you solely to get a bigger favor from you in the future, then you do not view his generosity to be as pure as if he had expected no reci- procity from you. For example, Arnold Kahn and Thomas Tice (1973) found that subjects' reactions to others' state- ments of intentions depended on whether they thought those making statements knew that their intentions would be made known to the subjects.17

161 have confirmed this hypothesis in a field study conducted in North London.

17Frank (1994, p. 21) tells the following story that colorfully summarizes the importance of in- tentions: There is an often told story of a boy who found two ripe apples as he was walking home from school with a friend. He kept the larger one for himself, and gave the smaller one to his friend. "It wasn't fair to keep the larger one for yourself', the friend replied. "What would you have done?" the first boy asked "I'd have ~iven you the larger one and kept the smaller one or myself," said the

The role of reciprocity and volition appears in some important economic contexts. Akerlof (1982) posits that firms and workers can be thought of as engaging in "gift exchange," a view of social exchange emphasized in sociology and especially anthropology. If a firm pays a higher wage to an employee, that employee is likely to reciprocate with higher quality work. Consequently, firms may give higher wages hoping workers will reciprocate with higher ef- fort. Similarly, Akerlof (1982, 1984) and Akerlof and Yellen (1990) propose that "efficiency wages," above the market-clearing wages, will be paid to workers to induce higher effort by those work- ers. Fehr, Georg Kirchsteiger, and Arno Riedl (1993) tested this hypothesis in laboratory models of labor markets. Subjects were assigned roles as "firms" or "workers." Firms offered a wage-in- volving a real monetary transfer from firm to worker-and workers responded by choosing an "effort" level, where this effort was monetarily costly to workers. The results were that most workers chose effort levels higher than their money-maximizing levels. Moreover, while low wages induced little or no ef- fort by workers, workers rewarded firms for setting high wages by providing high effort.

What is the source of high effort lev-

friend. To which the first boy responded, "Well, we each got what you wanted, so what are you complaininga~out?"

The punc ine of this story plays off the obvious silliness of presuming that the second boy's satisfaction with events on1 depends on the resulting allocation. Yet this comicJpresumption is the basis of most at- tempts by economists to model "social preferences" defined solely over outcomes. To formalize the role of intentions in fairness judgments, Rabin (1993) adopts the framework developed by John Geanakoplos, David Pearce, and Ennio Stacchetti (1989), who modify con- ventional game theory by allowing payoffs to depend on players' beliefs as well as their actions. By positing that my beliefs about your beliefs are arguments in my utility function, we can model my beliefs about your motives as directly influencing my fairness judgments.

els by workers in response to high wages by firms? While workers may simply be choosing to share some of their additional wealth from higher wages with the firm, they may also be reciprocating the volitional generosity of firms. Charness (1996) conducts ex- periments that helps us differentiate these hypotheses. In Fehr, Kirchler, and Weichbold (19941, it is clear to the worker-subjects that the firms choose wages of their own volition. Charness (1996) replicates this condition, but also conducts variants of the experiment where wages are either chosen randomly, or by a "third party" (the experi- menter). In these conditions, a high wage is not an act of kindness by a firm, and a low wage is not act of meanness; both are beyond a firm's control. Re- sults indicated that the high-wagesyields-high-effort reaction has both a "share-the-wealth" and an attribution element: Workers were substantially more likely to reward high wages with high effort and punish low wages with low effort when the wages reflected the volition of the firm.

3. Biases in Judgment

Economists traditionally have assumed that, when faced with uncertainty, people correctly form their

subjective probabilistic assessments cording to the laws of probability. But have documented many sys- in judgment under uncertainty. Tversky and ~~h~~~~~ (1974, p, 1124) help observed perfect rationality by noting that people rely on "heuristic principles wllicll reduce tasks of assessing probabilities and predicting values to simpler judgmental operations.m In gen- heuristics are quite but sometimes they lead to severe and

systematic errors. As the quote clearly suggests, the research described here does not at all suggest economists should abandon the assumption that people are intelligent and purposive in their decision making. Rather the research explores how people depart from perfect rationality, positing biases that represent specific and systematic ways that judgment departs from perfect ,ra- tionality. For the remainder of this sec- tion, I describe some of this research, presenting two biases at length and more quickly outlining several others. I conclude by discussing some evidence for when people do and don't learn to correct biases.

A. The Law of Small Numbers

According to a bias called "the law of small numbers" (Tversky and Kahneman 1971), people exaggerate how closely a small sample will resemble the parent population from which the sam- ple is drawn.18 We expect even small classes of students to contain very close to the typical distribution of smart ones and personable ones. Likewise, we un- derestimate how often a good financial analyst will be wrong a few times in a row, and how often a clueless analyst will be right a few times in a row. Be-

IS Kahneman and Tversky relate the law of small

numbers to people's tendency to under-use base rates, T\ersk\. and Kahnrlnan (1974) evidence for the representntizjeness heuristic. Bayes's Law tells us that our assessment of likelihoods should combine represe~ltativeness with base rates (the percentage of the population falling into vari- ous groups). Yet people under-use base-rate infor- mation in forming their judgments. If we see soinebody who looks like a criminal, our assessment of the probability that he is a criminal tends to under-use knowledge about the percentage of peo le who are criminals. Similarly, if a certain megcal test always comes out ositive anlong peo- ple with a rare disease, an1 only occasional1 among people without the disease, people rib tend to exaggerate the likelihood of having the dis- ease given a positive result. Given the rarity of the disease, the total number of false positives may be far greater than the number of true positives.

cause we expect close to the same prob- ability distribution of types in small groups as in large groups, for example, we tend to view it as comparably likely that at least 80 percent of 20 coin flips will come up heads than that at least 80 percent of 5 coin flips will come up heads; in fact, the probabilities are about 1percent and 19 percent, respec- tively. Kahneman and Tversky (1982a,

p. 44) asked undergraduates the follow- ing question:

A certain town is served by two hospitals. In

the larger hospital about 45 babies are born

each day, and in the smaller hospital about 15

babies are born each day. As you know, about

50 percent of all babies are boys. However,

the exact percentage varies from day to day.

Sometimes it may be higher than 50 percent,

sometimes lower.

For a period of 1 year, each hospital re

corded the days on which more than 60 per-

cent of the babies born were boys. Which

hospital do you think recorded more such

days?

Twenty-two percent of the subjects said that they thought that it was more likely that the larger hospital recorded more such days, and 56 percent said that they thought the number of days would be about the same. Only 22 per- cent of subjects correctly answered that the smaller hospital would report more such davs. This is the same fraction as guessed exactly wrong. Apparently, the subjects simply did not see the relevance of the number of child births per day.19

The law of small numbers implies

19While people believe in the law of small num- bers, they apparently don't believe in the law of large numbers: We zinderestinzate the resemblance that large samples will have to the overall popula- tion. Kahneman and Tversky (1982a),for instance, found that subjects on average thought that there was a more than 1/10 chance that more than 750 of 1000 babies born on a given day would be male. The actual likelihood is way less than 1 percent. To overstate it a bit, people seem to have a univer- sal probability distribution over sample means that is insensitive to the sample size.

that people exaggerate the likelihood that a short sequence of flips of a fair coin will yield roughly the same number of heads as tails. What is commonly known as "the gambler's fallacy" is a manifestation of this bias: If a fair coin has not (say) come up tails for a while, then on the next flip it is "due" for a tails, because a sequence of flips of a fair coin ought to include about as many tails as heads.

When the underlying probability dis- tribution generating observed sequences is uncertain, the gambler's fal- lacy leads people to over-infer the probability distribution from short sequences. Because we underestimate the frequency of a mediocre financial analyst making lucky guesses three times in a row, we exaggerate the likelihood that an analyst is good if she is right three times in a row. This tendency to over- infer from short sequences, in turn, leads to misperception of regression to the mean. Because we read too much into patterns that depart from the norm, we don't expect that further ob- servations will look more normal. As teachers, we exaggerate the extent to which one good or bad performance on a test is a sign of good or bad aptitude, so we don't expect exceptional perfor- mances to be followed by unexceptional performances as often as they are. Mis- understanding regression to the mean gives rise to spurious explanations for observed regression. When a student performs poorly on the midterm but well on the final, teachers infer that the student has worked harder; if the stu- dent performs well on a midterm but poorly on the final, teachers infer that the student has slacked off. Tversky and Kahneman (1974) give another example. Flight-training instructors observed that when they praised pilots for smooth landings, performance usually deteriorated on the next landing, but when they criticized pilots for poor landings, performance improved the next time. But random performance will lead to "deterioration" following a good landing and "improvement" following a poor landing. These flight instructors developed a wrong theory of incentives based on erroneous statistical reasoning.

Another implication of the law of small numbers is that people expect too few lengthy streaks in sequences of ran- doin events. As with regression to the mean, therefore, people tend to gener- ate spurious explanations for long streaks that are determined by chance. For instance, there is widespread belief in the "hot hand" in basketball-that particular basketball players are streak shooters who have "on" nights and "off' nights which cannot be explained by randomness. Thomas Gilovich, Robert Vallone, and Tversky (1985) and Tver- sky and Gilovich (1989a, 1989b) have argued that the almost universally accepted phenomenon of the hot hand is non-existent in basketball. The exagger- ated belief in hot hands seems partly ex- plained by the misperception that purely random streaks are too long to be purely random.

B. Belief Perseverance and Confirmatory Bias

A range of research suggests that once forming strong hypotheses, people are often too inattentive to new infor- mation contradicting their hypotheses. Once you become convinced that one investment strategy is more lucrative than another, you may not sufficiently attend to evidence suggesting the strat- egy is flawed. A particularly elegant demonstration of such "anchoring" is found in Jerome Bruner and Mary Pot- ter (1964). About 90 subjects were shown blurred pictures that were gradu- ally brought into sharper focus. Differ- ent subjects began viewing the pictures at different points in the focusing pro- cess, but the pace and final degree of focus were identical for all subjects. Of those subjects who began their viewing at a severe-blur stage, less than a quar- ter eventually identified the pictures correctly, whereas over half of those who began viewing at a light-blur stage were able to correctly identify the pic- tures. Bruner and Potter (1964, p. 424) conclude that "Interference may be ac- counted for partly by the difficulty of rejecting incorrect hypotheses based on substandard cues." That is, people who use weak evidence to form initial hy- potheses have difficulty correctly inter- preting subsequent, better information that contradicts those initial hypothe- ses. David Perkins (1981) argues that such experiments provide support for the perspective that "fresh" thinkers may be better at seeing solutions to problems than people who have medi- tated at length on the problems, because the fresh thinkers are not overwhelmed by the "interference" of old hypotheses.

This form of anchoring does not nec- essarily imply that people misinterpret additional evidence, only that they ig- nore additional evidence. Psychological evidence reveals a stronger and more provocative phenomenon: People tend to misread evidence as additional support for initial hypotheses.20 If a teacher initially believes that one student is smarter than another, she has the propensity to confirm that hypothe- sis when interpreting later performance.

Some evidence for confirmatory bias is a series of experiments demonstrating how providing the same ambiguous in- formation to people who differ in their

"For a formal model of confirmatory bias, see Rabin and Schrag (1997).

initial beliefs on some topic can move their beliefs further apart. To illustrate such polarization, Charles Lord, Lee Ross, and Mark Lepper (1979, pp. 2102) asked 151 undergraduates to complete a questionnaire that included three questions on capital punishment. Later, 48 of these students were recruited to participate in another experi- ment. Twenty-four of them were se

lected because their answers to the earlier questionnaire indicated that they were "'proponents' who favored capital punishment, believed it to have a deter- rent effect, and thought most of the relevant research supported their own beliefs. Twenty-four were opponents of capital punishment, doubted its deter- rent effect and thought that the rele- vant researcl~ supported their views." These subjects were then asked to judge the merits of randomly selected studies on the deterrent efficacy of the death penalty, and to state whether a given study (along with criticisms of that study) provided evidence for or against the deterrence hypothesis. Sub- jects were then asked to rate, on 16 point scales ranging from -8 to +8, how the studies they had read moved their attitudes toward the death penalty, and how they had changed their beliefs re- garding its deterrent efficacy. At confi- dence levels of p < .O1 or stronger, Lord, Ross, and Lepper found that pro- ponents of the death penalty became on average more in favor of the death pen- alty and believed more in its deterrent efficacy, while opponents became even less in favor of the death penalty and believed even less in its deterrent effi- cacy. Scott Plous (1991) replicates the Lord, Ross, Lepper results in the con- text of judgment about the safety of nu- clear technology.~l

"Lord, Ross, and Lepper posit that even pro- fessional scientists are susceptible to such "same- evidence polarization." Indeed, many econoinists

John Darley and Paget Gross (1983) demonstrate a related and similarly striking form of polarization. Seventy undergraduates were asked to assess a nine-year-old girl's academic skills in several different academic areas. Before completing this task, the students received information about the girl and her family and viewed a video tape of the girl playing in a playground. One group of subjects was given a fact sheet that described the girl's parents as col- lege graduates who held white-collar jobs; these students viewed a video of the girl playing in what appeared to be a well-to-do suburban neighborhood. The other group of subjects was given a fact sheet that described the girl's par- ents as high school graduates who held blue-collar jobs; these students viewed a video of the same girl playing in what appeared to be an impoverished inner city neighborhood. Without being sup- plied any more information, half of each

group of subjects was then asked to evaluate the girl's reading level, meas- ured in terms of equivalent grade level. There was a small difference in the two groups' estimates-those subjects who had viewed the "inner-city" video rated the girl's skill level at an average of 3.90 (i.e., 9/10 through 3rd grade) while those who had viewed the "suburban" video rated the girl's skill level at an av-

and other academics have robably observed how differing schools of thougl!t interpret ambiguous evidence. A colleague saw the same model-cali- brating the elasticity of demand facing a Cournot oligopolist as a function of the number of firms in an industry-described at the University of Chi- cago and at M.I.T. A Chicago economist derived the formula and said, "Look how few firms you need to get close to infinite elasticities and perfect competition." An M.I.T. economist derived the same formula and said, "Look at how large n has to be before you get anywhere close to an infinite elasticity and perfect competition." These differ- ent schools each interpreted the same mathenzatical formula as evidence reinforcing their respec- tive views.

erage of 4.29. The remaining subjects in each group were shown a second video of the girl answering (wit11 mixed success) a series of questions. Afterwards, they were asked to evaluate the girl's reading level. The inner-city video group rated the girl's skill level at an average of 3.71, significantly below the

  1. estimate of the inner-city subjects who did not view the question-answer video. Meanwhile, the suburban video group rated the girl's skill level at an average of 4.67, significantly above the
  2. estimate of the suburban subjects who did not view the second video. Even though the two groups viewed the identical question-and-answer video, the additional information further po- larized their assessments of the girl's skill level. Darley and Gross (1983) in- terpret this result as evidence of confir- matory bias-subjects were influenced by the girl's background in their initial judgments, but their beliefs were evidently influenced even more strongly by the effect their initial hypotheses had on their interpretation of further evi- dence.

Certain types of evidence flows seem to be most conducive to confirmatory bias. Ambiguity of evidence is widely recognized to be an important mediat- ing factor in both confirmatory bias and overconfidence (see, e.g., Gideon Keren 1987; and Dale Griffin and Tver- sky 1992). Keren (1988) notes the lack of confirmatory bias in visual perceptions, and concludes that confirmatory tendency depends on some degree of abstraction and the need for interpreta- tion not present in simple visual tasks. Lord, Ross, and Lepper (1979, p. 2099) posit that when faced with complex and ambiguous evidence, we emphasize the strength and reliability of confirming evidence but the weaknesses and unreli- ability of disconfirming evidence. They also report an impression that may re- sound with those observing economists' reactions to behavioral evidence that might be damaging to habitual econom- ics assumptions:

With confirming evidence, we suspect that both lay and professional scientists rapidly re- duce the complexity of the information and remember only a few well-chosen supportive impressions. With disconfirming evidence, they continue to reflect upon any information that suggests less damaging "alternative inter- pretations." Indeed, they may even come to regard the ambiguities and conceptual flaws in the data opposing their hypotheses as somehow suggestive of the fundamental correctness of those hypotheses.

The above passages hint at the role that selective scrutiny of evidence plays in confirmatory bias. Another form of "scrutiny-based" confirmatory bias is what I shall call hypothesis-based jilter- ing. While it is sensible to interpret am- biguous data according to current hypotheses, people tend to use the consequent "filtered" evidence inappro- priately as further evidence for these hypotheses. If a student gives an unclear answer to an exam question, it is perfectly reasonable for a teacher to be influenced in his evaluation of the an- swer by his prior perceptions of that student's mastery of the material. How- ever, after assigning differential grades to students according differential inter- pretation of comparable answers, it is a mistake to then use differential grades on the exam as further evidence of the differences in the students' abilities. Lord, Ross, and Lepper note a similar distinction in reflecting on the bias in their experiment: It is legitimate for people to differentially assess probativeness of different studies according to their current beliefs about the merits of the death penalty. The "sin" is in us- ing their hypothesis-based interpretations of the strength of different studies as further support for their beliefs.

Even when each individual datum is

unambiguous, confirmatory bias can be generated when people must statistically assess correlations extended over time. Richard Nisbett and Ross (1980) argue that the inability to accurately perceive correlation is one of the most robust sl~ortcomings in human reasoning, and people often imagine correla- tions between events when no such correlation exists." Dennis Jennings, Teresa Amabile, and Ross (1982) argue that illusory correlation can play an im- portant role in the confirmation of false hypotheses, finding that people under- estimate correlation when they have no theory of the correlation, but exagger- ate correlation and see it where it is not when they have a preconceived theory of it.

C. Other Biases

I briefly outline a few more biases that might interest economists."3 The first is anchoring and adjustment. Slovic and Sarah Lichtenstein (1971) demon- strate that, in forming numerical esti- mates of uncertain quantities, adjust- ments in assessments away from (possibly arbitrary) initial values are typically insufficient. Tversky and Kahneman (1974, pp. 1128) provide the following example: "Loren Cha man and Jean Chapman (1967, 1969, 1971) &monstrate that clinicians and laypeople often perceive entirely illusory correla- tion among (for instance) pictures and the person- ality traits of the people who drew the pictures. Charles Stangor (1958) and David Hamilton and Terrence Rose (1950) also discuss the role of illu- sory correlation in the context of confirmatory- like phenomena More generally, as Jenninf, Amabile, and Ross (1952, p. 212) put it, even t e staunchest defenders of the lay-person's capacities as an intuitive scientist . . . have had little that was flattering to say about the layperson's han- dling of bivariate observation." 23 For a more thorough introduction to this lit- erature, see Kahneman, Paul Slovic, and Tversky (1982), or, for an outstanding review of this mate- rial, and of individual decision making more gen- erally, see Camerer (1995). [Slubjects were asked to estimate various quantities, stated in percentages (for exam- ple, the percentage of African countries in the United Nations). For each quantity, a number between 0 and 100 was determined by spinning a wheel of fortune in the sub- jects' presence. The subjects were instructed to indicate first whether that number was hipher or lower than the value of the auan 0 tity, and then to estimate the value of the quantity by illoving upward or downward from the given nuinber. Different groups were given different numbers for each quan- tity, and these arbitrary ilumbels had a marked effect on estimates. For exam~le. the I median estimates of the percentage of Afri- can countries in the United Nations were 25 and 45 for groups that received 10 and 65, respectively, as starting points. Payoffs for accuracy did not reduce the anchoring ef- fect. While this example is somewhat arti- I

i predictions, with the first procedure yielding a probability distribution more concentrated around the median than the second. One of the most widely studied biases in the judgineilt literature is the hindsight bias." Baruch Fischhoff (1975, "For two excellent recent reviews of the hind- sight bias, see Scott Hawkins and Reid Hastie (1990) and Jay Christensen-Szalanski and Cynthia Willham (1991). Christensen-Szalanski and Will-ham co~lduct a meta-analysis of the literature-ag- gregating the findings of 122 different studies, gathered through an unbiased procedure, to test p. 288) first proposed this bias by observing that "(a) Reporting an outcome's occurrence increases its perceived probability of occurrence; and (b) people who have received out- come knowledge are largely unaware of its having changed their perceptions [along the lines of (a)]." Combin- ing these, the literature on the hind- sight bias shows that people exaggerate the degree to which their beliefs before an informative event would be similar to their current beliefs. We tend to think we "knew it would happen all along." After a politician wins election, people label it as inevitable-and believe that they always thought it was in- evitable. One example of Fischhoffs (1975) original demonstration of this effect was to give subjects a historical passage re- garding British intrusion into India and military interaction with the Gurkas of Nepal. Without being told the outcome of this interaction, some subjects were asked to predict the likelihood of each of four possible outcomes: 1) British victory, 2) Gurka victory, 3) military stalemate with a peace settlement, 4) military stalemate without a peace set- tlement. Four other sets of subjects were each told a different one of the four outcomes was the true one (the real true outcome is that the two sides fought to a stalemate without reaching a peace settlement). For each reported outcome, when compared to a control group not told any outcome, subjects' average ex post guesses of their hypo- thetical ex ante estimates were 15 per- cent higher than those of the control group. People don't sufficiently "sub- tract" information they currently have about an outcome in imagining what for the existence of the bias. They conclude that the bias is very real. (They also argue that the ef- fects are "small.") they would have thought without that informati~n.~s A pervasive fact about human judg- ment is that people disproportionately toeight salient, menaorable, or uioid evidence even when they have better sources of information." For instance, our assessment of a given city's crime rate is likely to be too influenced by whether we personally know somebody who has been assaulted, even if we are familiar with much more relevant gen- eral statistics. Likewise, dramatic stories by people we know about difficul- ties with a brand of car are likely to be overly influential even if we are famil- iar, via Consumer Reports, with general statistics of the reliability of different brands. In both these cases, and in many others, the more salient informa- tion should have virtually no influence on our beliefs in the face of much more pertinent statistical information. Tver- sky and Kahneman (1973) discuss, for example, how salience may distort clini- cians' assessments of the relationship between severe depression and suicide. Incidents in which patients commit sui- '5The definition of hindsight bias regards peo- ple's perceptions of how they thenzselves would have answered a particular question absent infor- mation they now have. As economists, we are likelv to care mostlv about a verson's beliefs about oth& people, not ;bout her'self. In general, it is hard to control experimentally for the fact that people have different information, and hard to iso- late the hindsight bias when asking subjects what others would have believed absent certain infor- mation. Subjects may believe that others have dif- ferent beliefs for a variety of reasons (e.g., you could believe that other people are not as smart as you). I feel, however, that the evidence suggests that we have a tendency to think that other people "should have known" as well (Hawkins and Hastie 1990, p. 319). "In Tversky and Kahneman's (1973) formula- tion: "[A] person is said to employ the availability heuristic whenever he estimates frequency or probability by the ease with which instances or as- sociations could be brou ht to mind." For more general reviews of the ro 'fe of salience and vivid- ness, see Susan Fiske and Shelley Taylor (1991, chs. 5,7). cide are much more likely to be remem- bered than are instances where patients do not cominit suicide. This is likely to lead to an exaggerated assessment of the probability that depressed patients will commit suicide. Finally, there is a mass of psychologi- cal research that finds people are prone toward overconfidence in their judgments. The vast majority of researchers argue that such overconfidence is per- vasive, and most of the research concerns possible explanations (of which confirmatory bias discussed above is one).zT D. Do Learning and Expertise Eliminate Biases? The conjecture that experience helps overcome biases often leads economists to doubt the relevance of laboratory evi- dence from inexperienced subjects. It is commonly argued that if important eco- nomic activity is performed by special- ists and experts, or consists of tasks done repeatedly by the saine individu- als, the assumption of full rationality fares much better than some of the psy- chological evidence indicates. Do experience, expertise, and learn- ing virtually eliminate biases? These are reasonable conjectures, and such factors probably do on average moderate biases. But the conjectures do not appear to be nearly as valid as econoinists imagine. Kahneman and Tversky (1982a) and Tversky and Kahnemail (1982), for instance, present experiments with subjects who vary in their level of statistical sophistication, to test whether general knowledge of statistics "See, e.g., Stuart Oskamp (1965), Jayashree nilahajan (1992), and Paul Paese and nilaryellen Kinnaly (1993). An early paper arguing this is Fischhoff, Slovic, and Lichtenstein (19771, who also tested the robustness of overconfidence with lilonetary stakes rather than reported judgments. No decrease in overconfidence was found relative to the no-money-stakes co~ldition. reduces or eliminates observed biases. The results are surprisingly negative. More generally, the research leads to mixed conclusions about when and how learning takes place, but very much does not support the strong versions of the experts-get-things-right and in-the- real-world-people-learn hypotheses. Research also suggests we should use extreme caution in defining the relevant notion of learning, because inany peo- ple who do learn general principles do not apply those principles in particular situations. In the context of overconfi- dence, for instance, Griffin and Tversky (1992) and Andrea Baumann, Raisa De- ber, and Gail Thompson (1991) con- clude that people who are aware of their own accuracy overall are overcon- fident on a case-by-case basis. When people understand the liinits in their abilities to predict events accurately, they tend not to apply this general knowledge in calibrating the appropri- ate confidence in individual cases: Kahnemail and Tversky (198213, p. 495) call such errors el-rors of applications, and note that "An error of application is most convincingly demonstrated when a person, spontaneously or with minimal prompting, clutches his head and exclaims: 'How could I have missed that?"' Even if people learn the rele- vant statistical truths of their en\' iironment, they may continue to make errors in their judgments and decision ~naking in every single case. One fears that economists may sloppily interpret such head-clutching as evidence for the rationality l~ypothesis, rather than against it. But evidence that people see their errors when confronted wit11 them does not boost the rationality assumption as economists use it. Our models use the rationality assumption as a realized feature of huinan behavior, not merely a human potential. As was demonstrated in the context of confirmation bias discussed above, "learning" can even sometimes tend to exacerbate errors. Relatedly, Griffin and Tversky (1992), address the relationship between expertise and overconfidence. When certain forms of pre- dictability are high and when feedback takes the form of unambiguous statisti- cal evidence, experts tend to have a pretty good sense of how accurate their predictions are. In such cases, experts not only know more, but are more real- istic than laypersons about how much they know. But when predictability is low, experts are often more susceptible to overconfidence than are laypersons. Griffin and Tversky (1992, p. 430) pro- vide illustrations: If the future state of a mental patient, the Russian economy, or the stock market cannot be predicted from present data, then experts who have rich models of the system in ques- tion are more likely to exhibit overconfidence than lay people who have a very limited un- derstanding of these systems. Studies of clini- cal psychologists (e.g., Oskamp 1965) and stock market analysts (e.g., Yates 1990) are consistent with this hypothesis. While a reasonable conjecture is that greater thoughtfulness and intelligently searching for patterns in the world would help douse biases, the quote specifically targets (economists take note . . .) as sus- ceptible to overconfidence those "experts who have rich models of the system in question." Indeed, many authors have hypothesized the role of the reasoning process itself in exacerbating the confir- matory bias discussed above, which in turn leads to overconfidence (see, e.g., Craig Anderson, Lepper, and Ross 1980; Ross et al. 1977; and Timothy Wilson and Jonathan Schooler 1991). Wilson and Suzanne LaFleur (1995, pp. 23-24) conduct an experiment on the role of "rea- soning" in strengthening confidence: Members of six sororities at the Univer- sity of Virginia were asked at the begin- ning of the semester to predict their own future behaviors toward fellow sorority members. Each subject was asked to pre- dict "yes" or "no" whether she would en- gage in each of six different behaviors- and to assess her confidence in her prediction. Randomly, some of the mem- bers were asked to list "why they might or might not perform each of the . . .behaviors . . . People were given a separate page on which to list their reasons for each behavior. They were told that the purpose of this task was to 'organize their thoughts,' that 'no one will actually read what you have written,' and that 'your reasons will be discarded.' They were urged to 'list as many reasons as you can think of, filling up this page if you can'." Other subjects were asked to make the same six predictions, but not asked to think of reasons for their behav- ior. At the end of the semester, subjects were asked if they had actually performed each of the six activities. The results from this experiment showed that the act of reasoning increased sub- jects' overconfidence regarding their predictions of their own behavior. While those who reasoned about their predictions were roughly as confident as those who did not reason (reasoners predicted their own Yes/No predictions would be accurate 80 percent of the time, while the control group predicted their accuracy at 82 percent), the reasoners were in fact significantly less accurate than the control group in their predictions; reasoners' predictions were accurate 71 percent of the time, compared to 79 percent for the control group. 4. Is "Maximizing Utility" the Right Model? The varied material of this section suggests that it may be wrong to con- ceptualize some types of economic be- havior in terms of an agent who maxi- mizes a stable, coherent utility function. Indeed, some of the material here also calls into question some of the interpre- tations presented earlier. For instance, loss-averse behavior may often reflect a flawed rule of thumb employed because people misperceive their own long-run well-being, rather than a modified util- ity function as suggested in Section 2. By calling into question the utility-maximization interpretation of behavior given in Section 2, such material sug- gests greater difficulty in improving the behavioral realism of formal economics. On the other hand, the material at the end of the section on self-control prob- lems provides a great opportunity to im- prove the realism of economics: Researchers have shown that a (relatively) simple multiple-self model of time-in- consistent discounting tractably modifies our familiar exponential model to yield a model that is manifestly more realistic behaviorally and surely has im- portant econoinic consequences. A. Do We Knou: bV1zat Makes Us Happy? The research on biases reviewed in Section 3 indicates that people misjudge the probabilistic consequences of their decisions. But other research sug- gests that, even when they correctly perceive the physical consequences of their decisions, people systematically misperceive the well-being they derive from such outcomes. We often system- atically mispredict our future experienced utility, even when those predic- tions rely only on accurate assessments of our past experienced utility (Kahne- man 1994; and Kahneman, Peter Wak- ker, and Rakesh Sarin 1997). As Kahne- man (1994, p. 21) puts it, "These considerations suggest an explicit distinction between two notions of utility. The experienced utility of an outcome is the measure of the hedonic experience of that outcome. . . . The decision util- ity of an outcome . . . is the weight as- signed to that outcome in a decision." The realization that decision and expe- rienced utility may be systematically different cuts to the core of our models of choice. It also cuts to the core of our methods of research, requiring us to formulate ways of inferring and eliciting preferences that go beyond a "revealed preference" method to attempt to infer people's hedonic experiences, through such methods as self reports of satisfac- tion and even psychological measurements. How do people misperceive their utilities? One pattern is that we tend to underestimate how quickly and how fully we will adjust to changes, not fore- seeing that our reference points will change. In a classic study, Philip Brick- man, Dan Coates, and Ronnie Janoff- Bulman (1978) interviewed both lottery winners (with average winnings of about $479,545) and a control group; they found virtually no difference in rated happiness of lottery non-winners and winners. While such interview evidence is inconclusive, the researchers controlled for alternative explanations (such as selection bias or biased presen- tation by interviewers). Two effects seemed to explain why lottery winners would be less happy than the winners had presuinably anticipated. First, mun- dane experiences become less satisfying by contrast to the "peak" experience of winning the lottery. Second, we become habituated to our circumstances: Along the lines of the material presented in Section 2, eventually the main carriers of utility become not the absolute levels of consumption, but departures from our (new) reference 1evel.28 "Brickman, Coates, and Janoff-Bulinan (1978) also found inore equivocal evidence for these ef- fects by interviewing 29 paraplegics and quadri- People do not anticipate the degree of such adaptation, and hence exagger- ate expected changes in utility caused by changes in their lives." This suggests that the "decision-utilityn aversion people have to losses is not consonant with "experienced utility." This realiza- tion, in turn, calls for a reexamination of the first topic of Section 2: Are loss aversion. the endowment effect. and other reference effects rational or irra- tional? If people experience losses rela- tive to a status quo as quite unpleasant, then loss-averse behavior is rational, be- cause people are correctly anticipating and avoiding unpleasant sensations. And, the remembered "loss" of an owned mug may carry over time, or in anv event be substantial relative to the long-term utility consequences of own- ing the mug. Yet loss aversion often seems to be a judgmental bias: In decisions with sig- nificant long-run consequences, people should put less weight than they do on their initial experience of losses. Indeed, some researchers invoke loss aversion more as an irrational rule of thumb than as a rational utility func- tion. Shlomo Benartzi and Thaler (1995) argue that the equity-premium puzzle can be explained by investors' aversion to short-term financial losses, plegics about their happiness before their accidents, their current hap iness, and their expected future happiness ParapLgics felt less well off on average than they felt before, and rated their hap- piness as lower than did lottery winners or the control group. Moreover, they did not currently enjoy mundane activities more than lottery win- ners or the control group. However, accident vic- tims put more em hasis on, and took a more tie view of, nunane pleasures-the ii rated Eh past pleasure and anticipated future p easure from such mundane activities slightly higher than either the lottery winners or the control group. 29111 the simple model of reference-point ad- justment discussed in Section 2, this can be trans- lated as saying that people systenlatically underes- tinlate the parameter a. even though they will not be spending their investment in the short term. Camerer et al. (1997) argue that New York taxi drivers decide when to quit driving for the day by a rule of thumb that says they should make sure to match their usual take for the day. In some more extreme examples of loss aversion it is hard to believe that the "transition utility" can rationally rank high relative to long-term utility. For instance, Thaler (1980) cornpared sub- jects' willingness to pay for a cure for a disease that leads to a quick and pain- less death with probability ,001 versus the minimum price you would accept to voluntarily subject yourself to the same disease. Subjects often required an or- der of magnitude more money to expose themselves to the disease than they would pay for a cure. People charge heavy premiums for losses relative to their status quo, even when it is hard to imagine that any experienced "tran- sition utility" is significant relative to long-term utility consequences-here, whether or not you live beyond a week. Another example of how people misperceive utility consequences of their choices is Richard Herrnstein and Prelec's (1992b) theory of melioration. Based on a mass of evidence gath- ered from people and other animals, they argue that people tend to make current choices according to which choice directly yields the highest utility, without taking into account the choice's effect on the utilities from future choices. That is, people often ignore "internalitiesn-the effects a current choice has on the utilities of later choices. Say you eat at one of two restaurants every night, either Blondie's or Fat Slice. You enjoy Fat Slice more, but because you also enjoy vari- ety, your utility each evening is as fol- lows: Utility from Fat Slice = 7 if you ate at Blon d~e'slast night Utility from Fat Slzce = 5 if you ate at Fat Sllce last night Utility froin Blor~die's= 4 if you ate at Fat Shce last night Utility from Blondie's = 3 if you ate at Blon die's last night On any given day, no matter your re- cent eating pattern, you get higher util- ity from eating at Fat Slice than at Blondie's. Yet your utility-maximizing consumption program is to alternate be- tween Fat Slice and Blondie's (thus al- ternating between payoffs of 7 and 4, for an average of 5.5) rather than eating all the time at Fat Slice (thus getting a payoff of 5 each period). Yet, because at each moment we tend to ask, "Which will yield me more pleasure-Fat Slice or Blondie's?," we may eat too often at Fat Slice. Of course, we do often train ourselves, or learn over time, to take into account internalities, but the evidence suggests that we take too little account of the global utility effects of individual choices. Herrnstein and Prelec (199213, pp. 257-58) argue that even when peo- ple do seem aware of the shape of their global utility functions, they may not properly maximize those preferences because they take an overly "piecemeal" approach.30 A major way people predict utility they will derive from future experiences is to recollect utility from comparable past experiences. While we might pre- sume that people accurately recollect their utility from familiar experiences, research on the endowment effect hints that this presumption may not be accu- rate: If we systematically lnisperceive the long-run consequences of giving up ,JOSee Herrnstein and Prelec 11992. , o. L 236) for a nice discussion of how econoinists glide over the distinction between global and piecemeal inaxiini- zation. See also Fehr and Peter Zych (1994) for an experimental exploratioll of this distinction. minor consumer items such as mugs, we may not have learned to assess correctly the utility consequences of even our everyday choices. Additional research even more dramatically demonstrates systematic differences between people's experienced utility of episodes and their recollections of tllose Gpisodes. Several recent experiments compare recollected utility to experienced utility for episodes extended over time, by collecting periodic hedonic reports by subjects of their current well-being. In evaluating the overall utility from such an extended episode, one must formulate cri- teria for adding up flows of experienced well-being. Kahneman (1994) posits that an uncontroversial criterion for comparing episodes is temporal monotonicity-that adding moments of pain to an otherwise unchanged experience decreases overall well-being, and that adding episodes of pleasure increases overall well-being. Kahneman (1994) argues that experiments suggest biases in how people's own retrospective evaluations of episodes compare to their experienced well-being. First, in evaluating past episodes, people tend to remember the extremes of pain and pleasure more than the average. Sec- ond, when an "episode" is well-defined (e.g., a vacation), people tend to put too much weight on the end of the episode (e.g., the last night of the vacation) in assessing their overall experience of the episode. Finally, we tend to neglect the duration of an episode. In assessing the dissatisfaction of an extremely unpleas- ant medical procedure (colonoscopy), for instance, patients seem to all but ne- glect the duration of the procedure- which ranged from 4 to 69 minutes. Of course, one must carefully consider the nain and pleasure associated with an ;pisode and after the actual epi- and of pain, for instance, are clearly important influences on long-run utility, just as anticipation and recollection of a vaca- tion are very significant in evaluating the overall well-being associated wit11 vacations. Such an interpretation of most of the experimental evidence, however, seems tenuous. The fact that we don't always correctly predict experienced utility is ob- viously important- for welfare implications of choice, and it prescribes caution in reliance on revealed-prefer- ence-based welfare economics. But there may be important behavioral im- plications of a related phenomenon whereby people misperceive their future behavior. Loewenstein and Daniel Adler (1995) performed an experiment based on the endowment-effect experi- ments of Kahneman, Knetsch, and Thaler (1991) discussed in Section 2. Some subjects were first asked to "imagine that we gave you a mug exactly like the one you can see, and that we gave you the opportunity to keep it or trade it for some money." All sub- jects were then given a mug, and their minimal selling prices were elicited. Be- fore receiving the mugs, subjects on av- erage predicted their own minimal sell- ing price at $3.73. Once they had the mugs, however, their actual minimal selling price averaged $4.89. That is, subjects systematically underestimated the endowment effect, and behaved sig- nificantly differently than they had pre- dicted about themselves moments earlier. (Such a procedure underestimates the true degree of misperception, because people don't like to contradict re- cently expressed predictions of their own behavior. Indeed, subjects who had made no prediction averaged a selling price of $5.62.) C. Elicitation Effects People often lack stable preferences that are robust to different ways of elic- iting those preferences.31 The most prominent set of research that points to such an interpretation of choice behav- ior concerns framing effects: Two logi- cally equivalent (but not transparently equivalent) statements of a problem lead decision makers to choose differ- ent options. An important and predict- able influence of framing on choice re- lates to loss aversion and diminishing sensitivity, as outlined in Section 2 above. Because losses resonate with people more than gains, a frame that highlights the losses associated with a choice makes that choice less attractive. Similarly, a frame that exploits dirnin- ishing sensitivity by making losses appear small relative to the scales involved makes that choice more attrac- tive. Tversky and Kahneman (1986, pp. S254-55) give the following example of framing effects, taken from a study of medical decisions by Barbara McNeil et a]. (1982): Respondents were given statistical informa- tion about the outcomes of two treatments of lung cancer. The same statistics were pre-sented to some respondents in terms of mor- tality rates and to others in terms of sun,ival rates. The respondents then indicated their preferred treatment. The information was presented [exactly] as follows. Problem 1(Survival frame) Surgery: Of 100 people having surgery 90 live through the post-operative period, 68 are alive at the end of the first year and 34 are alive at the end of five years. Radiation Therapy: Of 100 people having ra- diation therapy all live through the treatment, 77 are alive at the end of one year and 22 are alive at the end of five years Problem 1(Mortality frame) Surgery: Of 100 people having surgery 10 die during surgery or the post-operative period, 31 The hypersensitivity of references" to the " method of eliciting those pregrences is a key is- sue in the emerging debate on "contingent valu- ation"; see, e g., Diamond and Jerry Hausman (1994). 32 die by the end of the first year and 66 die by the end of five years. Radiation Therapy: Of 100 people having ra- diation therapy, none die during treatment, 23 die by the end of one year and 78 die by the end of five years. The inconsequential difference in formula-tion produced a marked effect. The overall percentage of respondents who favored radia- tion therapy rose from 18% in the survival frame (N = 247) to 44% in the mortality frame (N = 336). The advantage of radiation therapy over surgery evidently looins larger when stated as a reduction of the risk of in- mediate death from 10% to 0% rather than as an increase from 90% to 100% in the rate of survival. The framing effect was not smaller for experienced physicians or for statistically sophisticated business students than for a group of clinic patients. This question is hypothetical, but similar framing effects were found in choices over lotteries with small mone- tary stakes, and Tversky and Kahneman (1986) cite some important real-world examples of framing effects. For instance, people react differently to firms charging different prices for different services (or the same service at differ- ent times) depending on whether the lower price is called a discount or the higher price is called a surcharge. Simi- larly, Thomas Schelling (1981) noticed huge differences in his students' atti- tudes toward tax deductions for children depending on how the deductions were framed. Money illusion provides perhaps the best example of the impor- tance of framing effects for economics. Kahneman, Knetsch, and Thaler (1986a) provide survey evidence that people are very attentive to nominal rather than real changes in wages and prices in assessing the fairness of firm behavior. A nominal wage increase of 5 percent in a period of 12 percent infla- tion offends people's sense of fairness less than a 7 percent decrease in a time of no inflation. More generally, people react more to decreases in real wages when they are also nominal decreases, and react negatively to nominal price increases even if they represent no in- crease in real prices (Shafir, Diamond, and Tversky (1997). Framing effects can often be viewed as heuristic errors-people are boundedly rational, and the presentation of a choice may draw our attention to differ- ent aspects of a problem, leading us to make mistakes in pursuing our true, un- derlying preferences. As such, framing effects to some extent are a topic for Section 3. But sometimes framing ef- fects cut more deeply to economists' model of choice: More than confusing people in pursuit of stable underlying preferences, the "frames" may in fact partially determine a person's preferences. Related phenomena even more strongly call into doubt the view that choices reflect stable, well-defined preferences. Preference reversals have been studied widely by economists and psychologists over the years: When confronted with certain pairs of gambles with roughly the same expected value, people often choose one of the pair over the other, while pricing the other more highly. To use an example from Tversky and Thaler (1990), consider an H bet that with 8/9 chance yields $4 and with 1/9 chance yields $0, and an L bet with a 1/9 chance to win $40 and 8/9 chance of $0. Most subjects choose the H bet over the L bet when asked to choose between the two. But when asked to state the lowest price at which they would be willing to sell each gamble, most subjects put a higher price on the L bet. More generally, people choose bets with a high chance of winning small amounts, but put a higher price on bets with a low chance of winning big amounts; economic theory predicts these two different elicitation procedures should yield the same preferences. Itamar Simonson and Tversky (1992) provide examples of context effects, where the addition of a new option to a menu of choices may actually increase the proportion of consumers who choose one of the existing options. For example, the proportion of consumers who chose a particular model of micro- wave oven increased when a second, more expensive model was added to their choice set. (Subjects were first asked to look at a catalogue containing the prices and descriptions of all the relevant choices from which their even- tual choice sets were drawn, so the re- sults seem unlikely to be due to any in- formation revealed by the choice sets.) As another example, Simonson and Tversky (1992) ran an experiment that illustrates that elicited subjects' prefer- ence for an elegant Cross pen versus re- ceiving $6. While only 36 percent of subjects choosing only between the Cross pen and the $6 chose the Cross pen, 46 percent of subjects who were also given the choice of a less attractive pen chose the Cross pen. In both these examples, the addition of an option that compared unfavorably (as more expensive or lower quality) to an existing op- tion enhanced the perceived attractive- ness of the existing option. While people are often unaware that the menu of choices influences their de- cisions, Simonson and Tversky note that at other times decision makers explicitly rationalize their choices with references to their choice sets. For instance, people may state explicitly that a given choice is a compromise between two other choices. Indeed, such findings suggest an alternative to the utility-maximization framework that may help explain framing effects, preference reversals, and context effects: People may make choices in part by asking them- selves whether they have a "reason" to choose one option over another (Shafir, Simonson, and Tversky 1993). D. Time-Variant Preferences People have a taste for immediate gratification. We procrastinate on tasks such as mowing the lawn that involve immediate costs and delayed rewards and do soon things such as seeing a movie that involve immediate rewards and delayed costs. Economists traditionally model such tastes by assuming that people discount streams of utility over time exponentially. An important qualitative feature of exponential discounting is that it implies that a per-son's intertemporal preferences are time-consistent: A person feels the same about a given intertemporal trade- off no matter when she is asked. Casual observation, introspection, and psychological research all suggest that the assumption of time-consistency is importantly wrong. Our short term tendency to pursue immediate gratifica- tion is inconsistent with our long term preferences. While today we feel that it is best that we not overeat tomorrow, tomorrow we tend to overeat; while to- day we feel we should write a referee report tomorrow, tomorrow we tend to put it off. More generally, when consid- ering tradeoffs between two future mo- ments, we give stronger relative weight to the earlier moment as it gets closer. Kris Kirby and Herrnstein (1995), for instance, asked subjects to state their preferences among a series of pairs, in each case choosing between a smaller, earlier reward and a larger, later reward. Subjects were (truthfully) told that one of their choices would be im- plemented. In two experiments with monetary rewards, 23 of 24 subjects "consistently reversed their choices from the smaller, earlier reward to the later, larger reward as the delay to both rewards increased." Both the monetary stakes and the delays were substantial- subjects received an average of about $21.50, with an average delay of about 2% weeks.3" Hence, a person's preferences today over her future delays in rewards are different than her future preferences over those same delays, so that prefer- ences are not time consistent. Formal models of such time-oariant preferences have been developed.33 Edmund Phelps and Robert Pollak (1968) capture the taste for immediate gratifica- tion with a simple two-parameter model that slightly modifies exponential discounting. Let ut be the instantaneous utility a person gets in period t. Then her intertemporal preferences at time t, Ut, can be represented by the following utility function, where both j3 and 6 lie between 0 and 1: For all t, T Uf(ut,ut+I, ...,ut)r(6)t.ut+ P.C(Zi)T. u,. t=t+l The parameter 6 determines how "time-consistently patient" a person is, just as in exponential discounting. If j3 = 1, then these preferences are simply exponential discounting. But for P < 1, these preferences capture in a parsimo- nious way the type of time-inconsistent preferences so widely observed. To see how these preferences capture the pref- erence for immediate gratification, sup- pose that you had a choice between do- ing ten hours of an unpleasant task on '32These numbers are calculated from the data presented by Kirby and Herrnstein (1995, p. 85- 86). Other psychological research showing ences are not time-consistent includes Sf::-?; Chung and Herrnstein (1967), George Ainslie (1991), Ainslie and Herrnstein (19811, Thaler (1981), and Loewenstein and Prelec (1992). a:, For papers on tilne.inconsistPnt discounting, see, e.g., Robert Strotz (1955), Steven Goldlnan (1979, 1980), Schelling (1978), Thaler and Hersh Shefrin (1981), David Laibson (1994, 1997), and O'Donoghue and Rabin (1997a. 1997b). April 14, versus spending eleven hours to complete the same task on April 15. Assume that your instantaneous disutil- ity from doing work is simply the num- ber of hours of work- ut(lO)=-10 and ut(ll) =-I1 for all t. Suppose that 6 = 1, but that j3 = .8 for a one-day delay: You are willing to suffer a given loss in utility tomorrow for a gain in utility to- day that is 80 percent as large. Suppose that April 14 has arrived and you are considering whether or not to work. You can experience a disutility of -10 by working today, or experience a discounted utility of .a,(-11) = -8.8 by delaying the work until tomorrow. You will, therefore, delay work. Contrast this with what your decision would be if, instead of choosing when to work on April 14, you are told by your boss that you must decide on February 1. Because from February 1 you discount both dates by j3,you will choose to work 10 hours on April 14 rather than 11 hours on Anril 15. From the Februarv 1 point of view, you find procrastinating in April an undesirable thing. For the exact same problem, your choice on February 1 is different than your choice on April 14. Irrespective of its specific prediction, exponential discounting would predict that your choice would be the same whether you made that choice on February 1 or April 14. This example seeins well-calibrated: On April 14, inost of us are apt to put off the work until April 15, even if it means a little more work. Absent a substantive difference between the two dates, virtu- ally no one would choose the delay if asked On To examine dynamic choice given time-variant preferences given these preferences, for each point in time, a person is modeled as a separate "agent" chooses her current behavior to lnaximize her current preferences, whereas each of her future selves, with her own preferences, will choose her future behavior to maximize her preferences. On one level, this idea of multiple selves-that a single human does not have unified preferences that are stable over time-is a radical departure from the utility-maximization framework. But because this conceptu- alization of intertemporal choice uses a familiar tool-dynamic game theory-it is ready-made for adoption by economists interested in improving the behavioral realism of our models. The behavior predicted by models of time-variant preferences often differs dramatically from the behavior predicted by the exponential model. The most notorious examples are efforts at self control: Because you may not like the way you will behave in the future, you may scheme to manipulate your fu- ture options. Consider again the work example. Instead of your boss telling you that you must choose on February 1 when to work, suppose now she gives you three options: You commit to do the task on April 14; you commit to do the task on April 15; or you wait until April 14 and then choose on which day to do the task. Which would you choose? The advantage of waiting is manifest: By not precluding either of your options, if there are any uncertainties that may be resolved between now and April, the flexibility you have retained may be valuable. Yet we sometimes engage in behavior precisely to restrict our own future flexibility. If there were few uncertainties, you might want to commit on February 1 to the April 14 date. Given your current pref- erence to do the task earlier, you wish to restrict your future self from procras- tinating. More generally, researchers have explored many self-commitment devices we employ to limit our future choices. Such self-commitment devices include alcohol clinics and fat farms from which you cannot check out, not owning a television, contributing to a "Christmas Club" from which you are not allowed to withdraw money until Christmas, or buying only small pack- ages of enticing foods so that you won't overeat when you get home. More sub- tly, you may try to control yourself through a variety of internal "rules" (e.g., never drink alcohol), even if you have no external mechanisms of self-control. Attempts to control our own future behavior indicate an awareness that we may not behave as we would wish to be- have. This raises the question of how aware people are of their time-inconsis- tency. You may have expectations about your propensity to misbehave, or you may nai'vely believe that your preferences in the future will match your cur- rent preferences. If today you prefer not to overeat tomorrow, you may nai'vely believe that you will feel the same way when facing an enticing bowl of ice cream tomorrow. If on February 1 you prefer less work on April 14 to more work on April 15,you may believe you'll feel the same way in April. Strotz (1955) labels people who are fully aware of their future self-control problems as sophisticated, and people who are fully unaware that they will have a self-control problem as nayve. While some degree of sophistication is implied by the existence of some of the self-commitment devices illustrated above, it does appear that people underestimate the degree to which their future behavior will not match their current preferences over future behav- ior. This accords with the evidence dis- cussed earlier, that people often incor- rectly predict their own future preferences: As with predicting the ef- fects of changes in reference points, here too knowing your future prefer- ences means that you know your prefer- ences won't accord with your current preferences. For example, people may repeatedly not have the "will power" to forego tempting foods or quit smoking while predicting that tomorrow they will have this will power. While behav- ioral evidence that calibrates the degree of sophistication seems sparse, Loewen- stein (1996, pp. 281-82) reaches the conclusion that people may be nai've in- directly from psychological findings such as the evidence of people mispre- dicting changes in utility. Whether they are sophisticated or nai've, people's time-inconsistent propensity for immediate gratification is important in a variety of econolnic realms. As investigated by several researchers (see, e.g., Thaler and Shefrin 1981; and Laibson 1997), such prefer- ences may be important to savings be- havior because the benefits of current consumption are immediate, whereas the increased future consumption that saving allows is delayed. Self-control problems are also clearly important in the demand for addictive goods and fatty foods. Similarly, the role of self- control in purchasing decisions is well known a~nong marketing experts (Stephen Hoch and Loewenstein 1991). Naughty goods are sold in small pack- ages because people tend to avoid large packages of such goods to prevent over- consumption. 5. Conclusion Over the years, economists have prof- fered many reasons for downplaying the relevance of behavioral research challenging our habitual assumptions. Claims abound that evidence inconsis- tent with our traditional model of hu- man behavior can be neglected because the evidence derives from observations of people insufficiently motivated to be- have themselves according to economic assumptions, or because it fails to bear sufficiently great burdens of proof, or because the implied behavior is unlikely to matter in the types of (market) set- tings that economists care about. Be- cause behavioral research is so often as- sessed in light of such arguments, it is common when presenting psychological findings to discuss broad methodologi- cal objections and attempt to rebut them. I refrain from doing so. It is my strong impression that many of the ar- guments invoked against the reality or relevance of behavioral research derive from unfamiliarity with the details of this research. Hence, my hope and guess is that as economists become more familiar with this research, such arguments will dissipate. And as the ag- gressive uncuriosity shown in the past toward behavioral research continues to diminish, we can look forward to focus- ing entirely on its substance. While none of the broad-stroke argu- ments for inattention to psychological research are compelling, obviously not all psychological research will be both confirmed by field data and proven to be of great economic importance. In- deed, abandoning the view that hypotheses departing from rationality, self-interest, or other habitual assump- tions are methodologically illicit can free us to evaluate these hypotheses with the same rigorous standards that our discipline, at its best, applies else- where. We can confront plausible hy- potheses about human behavior with both healthy skepticism and genuine cu- riosity, empirically test their validity, and carefully draw out their economic implications. And, as we apply these rig- orous standards, we can then begin to treat claims that (say) investors irration- ally infer too much from short-term performance, or that employees feel re- sentful when mistreated, as presumptively plausible and presumptively rele- vant hypotheses worth keeping in mind in our economic analysis. REFERENCES ADA~IS, J. STACY. "Wage Inequities, Productivity and Work Quality," Ind. Relat., Oct. 1963, 3, pp 9-16. AINSLIE, GEORGE W. "Derivation of 'Rational' Economic Behavior from Hyperbolic Discount Curves," Anzer. Econ. Rev., May 1991, 81(2), ARD J: "Preference Reversal and Delayed Rein- forcement," Aninzal Learning & Behaoior, Nov. 1981, 9(4), pp 476-82. AKEKLOF,GEORGE A. "Labor Contracts as Partial Gift Exchange," Quart. J. Econ., Nov. 1982, 97(4), pp 543-69. "Gift Exchange and Efficiency-Wage The- ory: Four Views," Anzer. Econ. Rev., May 1984, 74(2), pp. 79-83. AKERLOIT,GEORGEA. AND YELLEN, JANET L. "The Fair Wage-Effort Hypothesis and Unem- ployment," Quart. J. Econ., May 1990, 105(2), pp 255-83. ANDERSON,CRAIG A,; LEPPER, MARK R. AND ROSS, LEE. "Perseverance of Social Theories: The Role of Explanation in the Persistence of Discredited Information," J. Personality G Social Psychology, Dec. 1980, 39(1-6), pp. 1037- 49. ANDREONI,JA~IES. "Cooperation in Public-Goods Experiments: Kindness or Confusion?" Amer. Econ. Rev., Sept. 1995, 85(4), pp. 891-904. ANDKEONI,JAMESAND MILLER, JOHN H. "Giv- ing According to GARP: An Experimental Study of Rationality and Altruism." Working Paper No. 9601. Social Systems Research Institute, U. Wisconsin at Madison, 1996. BAUIIANN,ANDREA0.;DEBER, RAISA B. AND THOIIPSON, GAIL G. "Overconfidence Among Physicians and Nurses: The 'Micro-Certainty, Macro-Uncertainty' Phenomenon," Social Science G Medicine, 1991, 32(2), pp. 167-74. BENAKTZI,SHLOMOAND THALER, RICHARD H. "Myopic Loss Aversion and the Equity Premium Puzzle," Quart. J. Econ., Feb. 1995, 110(1), pp. 73-92. BEKSCHEID,ELLEN; BOYE, DAVID AND WALSTEK, ELAINE. "Retaliation as a Means of Re- storing Equity," J. Personality G Social Psychol- ogy, 1968, 10(4), p 370-76. BLOUNT, SALLY. ''Wfen Social Outcomes Aren't Fair: The Effect of Causal Attributions on Pref- erences," Organizational Behauior 6-Hunzan Decision Processes, Aug. 1995, 63(2), pp. 131- 44. Bon.bri\N, DAVID; MINEHART, DEBORAH AND RABIN, MATTHEW "Loss Aversion in a Consumption-Savings Model." Mimeo, U. Califor- nia at Berkeley, 1997. BKICKMAN,PHILIP; COATES, DAN AND JANOFFBULMAN,RONNIE."Lottery Winners and Acci- dent Victims: Is Happiness Relative?" J. Personality & Social Psychology, Aug. 1978, 36(8), pp. 917-27. BRUNER,JEROME S. AND POTTER, MARY C. "In- terference in Visual Recognition," Science, Apr. 24, 1964, 144(3617), pp. 424-25. CAMEKEK,COLIN F. "Individual Decision Mak- " in Handbook of experimental econonzics. ,: JOHN KAOEL AND ALVIN E. RUTH, Princeton: Princeton U. Press, 1995, pp, 587- 703. CAMERER,COLIN ET AL. "Labor Supply of New York City Cabdrivers: One Day at a Time," Quart. J. Econ., May 1997,112(2), pp. 407-41. CAMERER,COLIN F. AND THALEK,RICHARDH. "Ultimatums, Dictators and Manners," J. Econ. Perspectives, Spring 1995, 9(2), pp. 209-19. CHAP~IAN, AND CHAPMAN,LOREN J. JEAN P. "Genesis of Popular But Erroneous Psychodiag- nostic Observations," J. Abnomzal Psychology, 1967, 72(3), pp. 193-204. "Illusory Correlation as an Obstacle to the Use of Valid Psychodiagnostic Signs," J. Abnornzal Psychology, 1969, 74(3), pp. 271-80. "Test Results Are What You Think They Are," Psychology Today, Nov. 1971, 5(6), pp. 18-22. CHAKNESS,GARY. "Attribution and Reciprocity in a Simulated Labor Market: An Ex erimen- tal Investigation." Mimeo, U Calitrnia at Berkeley, 1996. CHRISTENSEN-SZ,%L,%NSKI,JAY J. AND WILLHAM, CYNTHIA."The Hindsight Bias: A Meta-Aualy- sis," Organizational Behauior 6-Human Deci- sion Processes, Feb. 1991, 48(1), pp. 147-68. CHUNG,SHIN-HOAND HERRNSTEIN,RICHARDJ."Choice and Delay of Reinforcement," J. Experimental Analysis of Behavior, 1967, 10(1), pp 67-74. CONLISK, JOHN. "Why Bounded Rationality?" J. Econ. Lit., June 1996, 34(2), pp. 669-700. CROSON, RACHEL T. A. "Expectations in Volun- tary Contribution Mechanisms." Working Paper No. 95-03-02, Department of Operations and Information Management, The Wharton School, U. Penn., 1995. DAKLEY,JOHNM. AND GROSS,PAGETH. "A Hy- pothesis-Confirming Bias in Labeling Effects," J. Personality G Social Psychology, Jan. 1983, 44(1), pp 20-33. D,%wEs, ROBYN M. AND THALEK, RICHARD H. "Anomalies: Cooperation," J. Econ. Perspectiues, Summer 1988,2(3), pp. 187-97. DIAMOND,PETER A. AND HAUSMAN,JERRY A. "Contingent Valuation: Is Some Number Better than No Number?" J. Econ. Perspectiues, Fall 1994, 8(4), pp. 45-64. DUESENBERRY,J.~MESS. Inconze, savin and the theory of consui~zer behaoior. Cambriige: Har- vard U. Press, 1949. EASTEKLIN, RICHARD A. "Does Economic Growth Improve the Human Lot? Some Em- pirical Evidence," in Nations and households in ccononzic growth: Essays in honor of ~Moses Abramovitz. Eds.: PAUL A. DAVID AND MELVIN V7. REDER. New York: Academic Press, 1974, pp. 89-125. EPSTEIN,LARRYG. AND ZIN, STANLEY E. "'First- Order' Risk Aversion and the Equity Premium Puzzle," J. Monet. Econ., Dec. 1990, 26(3), pp. 387-407. FEHR, ERNST AND F.\LK, ARMIN. "Wage Rigidi- ties in a Competitive Incomplete Contract Mar- ket: An Experimental Investi ation " Mimeo, Institute for Empirical Rerearc1 in Economics, U. Zurich, 1996. FEHR, ERNST; KIKCHLER, ERICHAND WEICHBOLD, ANDREAS."When Social Forces Remove the Impact of Competition: An Experimental Investigation." Mimeo, Department of Econom- ics, U. Technology, Vienna, 1994. FEHR, ERNST; KIKCHSTEIGEK, GEORG AND RIEDL, AKNO. "Does Fairness Prevent Market Clearing? An Experimental Investigation," Quart.J. Econ., May 1993, 108(2), p 437-59. FEHR, ERNST .AND ZYCH, PETER "TI; Power of Temptation: Irrationally Myopic Excess Consumption in an Addiction Experiment." Mimeo, Institute for Empirical Research in Economics, U. Zurich, 1994. FISCHHOFF, BARUCH. "Hindsight Is Not Equal To Foresight: The Effect of Outcome Knowl- edge on Judgment Under Uncertainty," J. Experimental Psychology: Hunzan Perception 6Performance, Aug. 1975,104(1), pp. 288-99. FISCHHOFF,BARUCH;SLOi'IC, PAUL AND LICHTENSTEIN, SARAH. "Knowing with Certainty: The Appropriateness of Extreme Confidence," J. Experimental Psychology: Human Perception 6-Perfoi-nzance, Nov. 1977, 3(4), pp. 552-64. FISKE, SUSAN T. AND TAYLOR,SHELLEYE. Social cognition. New York: McGraw-Hill, 1991. FKANCIOSI,ROBEKT ET AL. "Fairness: Effect on Temporary and Equilibrium Prices in Posted- Offer Markets," Econ. ]., July 1995, 105(431), pp 938-50. FRANK,ROBERT H. Choosing the right pond: Hu- man behavior and the quest for status. Oxford: Oxford U. Press, 1985. . "Frames of Reference and the Oualitv of Llfe," Anzer Econ Re0 , May 1989, ?9(2),'pp 80-85 "Local Status, Fairness, and Wage Com- pression Revisited." Mimeo, Department of Economics, Cornell U., 1994. FRANK,ROBEKTH. AND HUTCIIENS,ROBERTM. "Wages, Seniority, and the Demand for Rising Consulnption Profiles," J. Econ. Behav. Org., Aug. 1993, 21(3), pp. 251-76. GEANAKOPLOS,JOHN; PEAKCE, DAVID AND STACCHETTI, ENNIO. "Psychological Games and Sequential Rationality," Games Econ. Bc- hau., Mar. 1989, 1(I),pp. 60-79. GILO171CH, T~okl.4~; VALLONE, ROBEKT AND TVEKSKY, AMOS. "The Hot Hand in Basketball: On the Misperception of Random Sequences," Cognitiue Psychology, July 1985, 17(3), pp 295-314. GOLD~IAN, STEVEN M. "Intertemporally Inconsis- tent Preferences and the Rate of Consumption," Econornetrica, May 1979, 47(3), pp. 621- 26. "Consistent Plans," Rev. Econ. Stud., Apr. 1980, 47(3), pp. 533-37. GOK~NSON,RICHARD E. AND BERKOIYITZ, LEONARD."Reciprocity and Responsibility Re- actions to Prior Help," J. Personality 6-Social Psychology, 1966, 3(2), pp. 227-32. GRIFFIN, DALE AND TVERSKY, AMOS. "The Weighing of Evidence and the Determinants of Confidence," Cognitive Psychology, July 1992, 24(3), pp 411-35. GOTH, WERNER; SCHMITTBERGER, ROLF AND SCHIV.\KZE, BEKND. "An Experimental Analysis of Ultimatum Bargaining," J. Econ. Behau. Or- gan., Dec. 1982, 344), pp. 367-88. GUTH, WERNER AND TIETZ, REINHARD. "Ultimatum Bargaining Behavior: A Survey and Com- parison of Experimental Results," J. Econ. Psych., Sept. 1990,11(3), pp. 417-49. HA~IILTON,DAVID L. AND ROSE, TERRENCE L. "Illusory Correlation and the Maintenance of Stereot ic Beliefs," J. Prersonality 6-Social Psycho$y, Nor 1980, 39(o), pp 832-45. HARLESS, DALE W. AND CAXIERER,COLIN F. "The Predictive Utility of Generalized Expected Utility Theories," Econornetrica, Nov. 1994, 62(6), pp 1251-89. HARTMAN,RAYX'IONDS.; DOANE, MICHAEL ANDWOO, CHI-KEUNG. "Consumer Rationality and the Status Ouo," Quart. 7. Econ., Feb. 1991, sight: Biased Judgments of past Events After the Outcomes Are Known," Psychology Bullc- tin, May 1990, 107(3), pp. 311-27. HELSON, HARRY. Adaptation leuel theory: An ex- perinzental ancl systematic approach to behav- ior. New York: Harper & Row, 1964. HERRNSTEIN, RICHARD J. AND PRELEC, DRAZEN. "Melioration," in. Choice over tirnc. Eds.: GEORGE LOEIYENSTEIN JOIIN EL- AND STER. New York: Russell Sage Foundation, 1992, pp 235-63. HOCII, STEPIIEN J. AND LOEI\,ENSTEIN,GEORGE F. "Time-Inconsistent Preferences and Consumer Self-Control," J. Cons. Res., Mar. 1991, 17(4), pp 492-507. JENNINGS,DENNIS,L.; A~IABILE,TERESA M. AND ROSS, LEE. "Informal Covariation Assess- ment: Data-based versus Theory-based Judg- ments," in DANIEL Ki\IfNEhI.\N, PAUL SLOTTC, AND AMOS TITERSKY, eds. 1982, pp. 211-30. KAGEL, JOHN H.; KIhI, CHUNG AND MOSEK, DONALD."'Fairness' in Ultimatum Games with As mmetric Information and Asymmetric Pay- ofz,'' Games Econ. Behav, Mar 1996, J;(l), pp 100-10. KAHN, ARNOLD AND TICE, TIIOMAS E. "Return- ing a Favor and Retaliating Harm: The Effects of Stated Intentions and Actual Behavior," J. Experimental Social Psychology, Jan. 1973, 9(1), pp 43-56. KAIINEMAN, "New Challenges to the Ra- DANIEL. tionality Assumption." J. Inst. Theoretical Econ., Mar. 1994, 150(1), pp. 18-36. KAHNEWN, DANIEL; KNETSCH, JACK L. AND THALER,RICHARDH. "Fairness as a Constraint on Profit Seeking: Entitlements in the Market," Anzer. Econ. Rev., Sept. 1986a, 76(4), pp. 728- 41. , "Fairness and the Assumptions of Econom- ics,"J. Bus., Oct. 1986b, 59(4), pp. S285-300. -. "Ex erimental Tests of the Endowment Effect an% the Coase Theorem," J. Polit Econ., Dec. 1990, 98(6), p 1325-48. _-"Anomalies: T%e Endowment Effect, Loss Aversion, and Status Quo Bias," J. Econ. Per- spectives, Winter 1991, 5(1), pp. 193-206. KAHNEMAN, TVERDANIEL; SLOVIC, PAUL AND SKY, AMOS, eds. Judgment under uncertainty: Heuristics and biases. Cambridge: Cambridge U. Press, 1982. KAHNE~IAN,DANIELAND TI-ERSKY,AMOS. "On the Psychology of Prediction," Psychology Rev., July 1973, 80(4), p 237 51 -"Prospect T\eoh:Tn hna~ysis of Decision under Risk," Econometrica, Mar. 1979, 47(2), pp. 263-91. "Subjective Probability: A Judgment of Representativeness," 1982a in DANIEL KAHNE- ~IAN,PAUL. SLOVIC, AND AMOS TVERSKY, eds. 1982, pp 32-47. "On the Study of Statistical Intuitions," 19821, in DANIEL KAIINEMAN, PAUL SLOVIC, AND AMOS TVEKSKY, eds. 1982, pp. 493-508. KAHNEMAN,DANIEL; WAKKER, PETER P. AND SARIN, RAKESH. "Back to Bentham? Explora- tions of Experienced Utility," Quart. J. Econ., May 1997,112(2), pp. 375-406. KEKEN, GIDEON. "Facing Uncertainty in the Game of Bridge: A Calibration Study," Organizational Behavior 6-Hunzan Decision Processes, Feb. 1987, 39(1), pp. 98-114. "On the Ability of Monitoring Non-Veridi- cal Perce tions and Uncertain Knowledge: Some Caliiration Studies," Acta Psychologica, May 1988, 67(2), pp. 95-119. KESER, CLAUDIA. "Voluntary Contributions to a Public Good When Partial Contribution Is a Dominant Strategy," Econ. Letters, Mar. 1996, 50(3), pp. 359-66. KIRBY, KRIS N. AND HERRNSTEIN,RICHARD J. "Preference Reversals Due to Myo ic Discounting of Delayed Reward," Psychofogy Sci- ence, Mar. 1995, 6(2), pp. 83-89. KNETSCH, JICK. "The Endowment Effect and Evidence of Nonreversible Indifference Curves," Amer. Econ. Rev., Dec. 1989, 79(5), pp 1277-84. KNETSCH,JACK L. IND SINDEN, JOHN A. ''Will- ingness to Pay and Compensation Demanded: Experimental Evidence of an Unexpected Dis- parity in Measures of Value," Quart. J. Econ., Aug. 1984, 99(3), pp. 507-21. KREBS, DENNIS L. "Altruism-An Examination of the Concept and a Review of the Literature," Psychology Bulletin, 1970, 73(4), pp. 258-302. -. "Prosocial Behavior, Equity, and Justice," in Equity and justice in social behavior. E~S.:~ERALDGREENBERGAND RONALD L. COHEN. New York: Academic Press, 1982, pp. 261-308. LAIBSON,DAVID. "Essays in Hyperbolic Discount- ing." Mimeo, M.I.T., 1994. "Golden Eggs and Hyperbolic Discount- ing," Quart. J. Econ., May 1997, 112(2), pp. 443-78. LOEIVENSTEIN,GEORGE. "Out of Control: Vis- ceral Influences on Behavior," Organizational Behavior G Hunzan Decision Processes, Mar. 1996, 65(3), pp. 272-92. LOEWENSTEIN,GEORGEAND ADLEK, DANIEL. "A Bias in the Prediction of Tastes," Econ. J., July 1995, 105(431), pp. 929-37. LOE~~ENSTEIN,GEORGEAND PRELEC, DRAZEN. "Anomalies in Intertemporal Choice: Evidence and an Interpretation," Quart. J. Econ., May 1992, 107(2), pp. 573-97. LOEWENSTEIN, GEORGE AND SICHER~IAN, NACHUX'I."Do Workers Prefer Increasing Wage Profiles?" J. Lab. Econ., Jan. 1991, 9(1), pp. 67- 84. LORD, CIIAKLES. G.; ROSS, LEE AND LEPPER, MARK R. "Biased Assimilation and Attitude Po- larization: The Effects of Prior Theories on Subsequently Considered Evidence," J. Person- ality 6-Social Psychology, Nov. 1979, 37(11), pp. 2098-2109. MAHAJAN,JAYASIIREE."The Overconfidence Ef- fect in Marketing Management Predictions," J. A4arketing Res, hug 1992, 29(3), ~p. 329:42: MCNEIL, BARBAKA J. ET AL. "On t e Elicitation of Preferences For Alternative Theravies." Net0 England J Medlctne, May 27, 198i, 306, pp 1259-62 ---~ MEHRA, RAJNISH AND PRESCOTT, EDIVARD C. "The Equity Premium: A Puzzle," J. Monet. Econ., Mar. 1985, 15(2), pp. 145-61. NISBETT, RICH.4RD E. AND ROSS, LEE. Human inference: Strategies and shortconzings of social judgnzent. Englewood Cliffs, NJ: Prentice Hall, 1980. O'DONO~;HUE, "DOTEDAND RABIN, MATTHEW ing It Now or Later." Working Paper No. 97- 253. U. California at Berkeley, 1997a. -. "Incentives for Procrastinators." Discussion Paper No. 1181. Northwestern U., CMSEMS, 1997b. OSKA~IP, "Overconfidence in Case-Study STUART. Judgments." J. Consulting Psychology, 1965, 29(3), pp 261-65. PAESE, PAUL \'\'. IND KINNALY,MARYELLEN. "Peer Input and Revised Judgment: Exploring the Effects of (Un)Biased Confidence," J. Applied Social Psychology, Dec. 1993. 23(23), pp. 1989-2011. PEKKINS,DAVID N. The nzind's best work. Cambridge: Harvard U. Press, 1981. PIIELPS, EDMUND S, AND PoL,L.\K, ROBERT A. "On Second-Best National Saving and Game- Equilibrium Growth," Rev. Econ. Stud.. Apr. 1968, 35(2), pp. 185-99. PLOIJS, SCOTT. "Biases in the Assimilation of Technolo ical Breakdowns: Do Accidents Make US safer?' 1,Applied Social ~sychology, ~uly 1991,21(13), pp. 1058-82. RABIN, MATTHETI'. "Incorporatinu Fairness into Game Theory and ~conornicsb'Amer. Econ. Reu., Dec. 1993, 83(5), pp. 1281-1302. "Risk Aversion and Diminishing Marginal Utility: A Calibration Theorem." Mimeo, U. California at Berkeley, 1997. R~RIN, JOEL. "First Inl- M.\TTHEIT' AND SCHK.~~;, pressions Matter: A Model of Confirmatory Bias." Working Paper No. 97-250. U. California at Berkeley, 1997. ROSS, LEE ET AL. "Social Explanation and Social Expectation: Effects of Real and HJ othetical Explanations 011 Subjective Likelihoo?" J Posotzality 6-Social Psychology, Nov. 1977, 35(11), pp. 817-29. ROTH, ALVIN E. AND MGRNIGHAN, J. KEITH. "The Role of Information in Bargaining: An Ex- perimental Study," Econonzetrica, Sept. 1982, 50(5), pp. 1123-42. RYDRK,H.\KL E., JK. AND HEAL, GEOFFREY M. "Optimal Growth with Interte~nporally Depen- dent Preferences," Rev. Econ. Sttld., Jan. 1973, 40(1), pp 1-33. SALLY,DAI'ID. "Conversation and Cooperation in Social Dilemmas: A Meta-Analysis of Ex eri lnents from 1958 to 1992,' Rafiona/ify an8so: ciety, Jan. 1995, 7(1), pp. 58-92. SAMGELSON, AND ZECKH.\ITSEK, RICH- TI:ILLIA\I ARD. "Status Quo Bias in Decision Making," J. Risk Uncertaiizty, Mar. 1988, 1(1), pp. 7- 59 SCHELLIUC:,THO\I-\S C "Egonomlcs, or the Ait of Self-Management," Anzer Econ Reu , May 1978, 68(2), pp 290-94 ---"Econom~c Reason~ng and the Eth~cs of Pollcy,' Publrc Interest, Sprlng 1981, 63, pp 77-61 . . SEG.~L,UZI AND SPII'AK, Av1.i. "First Order ver- sus Second Order Risk Aversion,"J. Econ. The- ory, June 1990, 51 (I), pp. 111-25. SH,iFIK, ELDAR; DIA\IOND, PETER AND TT'ERSKY, A\IOS. "Money Illusion," Qtiart. J. Econ., May 1997,112(2), pp. 341-74. SH.AFI~I,EI_.II.AK: AND TVEKSI\IOKSON,IT.A~I.AK SKY, Ah'fOS. "Reason-Based Choice," Cognition, Special Issue: Reasolli~lg and Decision Making. 0ct.-Nov. 1993, 49(1-2), pp. 11-36. SI-IAFIK,ELDA~IAKD TI'EKSKY,A~IOS."Thinlting Through Uncertainty: Nonconsequential Reasoning and Choice," Cog~zitiue Psyclzology, Oct. 1992. 24(4), pp. 449-74. SIIE,~,JOIIN. "Union Contracts and the LifeCycle/Permanent-Illcome Hypothesis,'' Airier. Econ. Reu., Mar. 1995a, 85(1), pp. 186-200. 'Myop~a, Llqu~d~ty Constra~nts, and Ag- gregate Consumption A S~mple Test, J Money, Crerllf, Banklng, Aug 199513, 27(3), pp 798-805 Sr~onsou ITIMI\K A~DTIERSK~ AMOS "Cholce In Context Tradeoff Contrast and Ex- tremeness Ave~sion," J ~VarLettng Res , Aug 1992,29(3), pp 281-95 SLOIIC, PALL IUD LICHTEUSTEIU SARAH 'Compar~son of Bayeslan and Regress~on Ap- proaches to the Stud) of Information Piocess- lng In Judgment ' Organlzatlonal Behnvzor G Human Performance, No\ 1971, 6(6), pp 649744 SMITH, AD,\\I An lnqulry rnto the nature and causes of the uealth of natlons Repl~nted Rok H CI\IPBELLAUD AUDKETI eds S S~IUUEK, Oxford Clarelldon Press, [I7761 1976 ST\U(,OK CHARLES 'Stereotype Accesslb~l~ty and Infornlatlon Processing Personal~tyb Soczal Psychology Bulletzn Dec 1988, 14(4), pp 694708 ~ STKOTZ,ROHEIIT H. "klyopia and Illcoilsistency in Dynamic Utility Maximization," Reu. Econ. Stnd., 1955, 23(3), pp. 165-80. THALEK,RICH.4KD H. "Toward a Positive Theory of Consuiner Choice," J. Econ. Behac. Organ., Mar. 1980, 1(1),PP 39-60. "Some Ein1)irical Evidence of Dynamic Inconsistency," E'con Letters, 1981, dl, pp. 201-07. "Mental Accounting and Consumer Choice." Marketing Sci., Summer 1985. 4(3), pp 199-214. "Anoinalies: The Ultitnatuln Game," J. Econ. Perspectives, Fall 1988, 2(4), pp. 195 206. THALEK,RICHARDH. AND SHEFRIIN,HEKSH M. "An Eco~lomic Theory of Self-Control," J. Polif. Econ., Apr. 1981, 89(2), pp. 392-406. TR.-\IN, KENNETH E.; MCFADDEN, DANIEL L. AND GOETT, ANDKETY A. "Consumer Attitudes and Voluntary Rate Schedules for Public Utili- ties," Reu. Ecoiz. Statisf., Aug. 1987, 69(3), pp. 383-91. T17EKSKY, AhlOS AND GILOI.ICH, THOI\.I\S. "The Cold Facts about the 'Hot Hand' in Basketball," Chance, 1989a, 2(1), pp. 16-21. "The Hot Hand: Statistical Reality or Cognitive Illusion," Chance, 1989b, 2(4), pp. 31-34. TVERSKY,A\IOS :\NU K,\HNEhl:\N, D,\NIEL. "Be- lief in the Law of Small Numbers," Psychology Btllletin, hug. 1971, 76(2), pp. 105-10. "Availability: A Heuristic for Judging Fre- quency and Probability," Cognitice Psychology, Sept. 197.3, 5(2), pp. 207-32. "Judument Under Uncertainty: Heuristics and ~iase?' Science, Sept. 1974, 185(4157), pp. 1124-31. - "Judgments of and by Representatlveness," In D~IUIEL \U, K~IIUE~IPAITLSLOIIL, ~IUIIAMOS TIEKSKY, eds 1982, pp 84-98 "Rational Choice and the Framing of De- cisions," J. Bus., Oct. 1986, 59(4, Part 2), pp. S251-78. "Loss Aversion in Riskless Choice: A Ref- erence-Dependent Model," Quart. ]. Econ., Nov. 1991, 106(4), pp. 1039-61. TVEKSKY,AXIOS AND THALEK, RICHARD H. "Anomalies: Preference Reversals," J. Econ. Perspecti~es, Spring 1990, 4(2), pp. 201-11. WILSOK, TIMOTIIY D. .\KD LAFLEUK,SUZANNE J. "Knowing What You'll Do: Effects of Analyz- ing Reasons on Self-Prediction," J. Personality 6-Social Psychology, Jan. 1995, 68(1), pp. 21- 35. WILSON, TIX'~OTHPD. AND SCI-IOOLEK, JONATHAN W. "Thinking Too Much: Introspec- tion Can Reduce the Quality of Preferences and Decisions," J. Personality 6-Social Psychology, Feb. 1991, 60(2), pp. 181-92. YAAKI, MAYA E. AND BAK-HILLEL,MENAHEXI. "On Dividing Justly," Social Choice Welfare, 1984, 1(1),pp. 1-24. YATES,J. F~IANK.udgment and decision making. Englewood Clif /s, NJ: Prentice-Hall, 1990.

C. Reciprocity and Attribution

The previous subsection considered evidence about social preferences defined over the allocations of goods. Psy- chological evidence indicates, however, that social preferences are not merely a function of consumption levels, or even changes in consumption levels. Rather, social preferences over other people's consumption depend on the behavior, motivations, and intentions of those other people. The same people who are altruistic toward deserving people are often indifferent to the plight of unde- serving people, and motivated to hurt those whom they believe to have misbe- haved. If somebody is being nice to you or others, you are inclined to be nice to him; if somebody is being mean to you or others, you are inclined to be mean to him.

This "reciprocal" nature of preferences manifests itself in the distinction between simple altruism, as outlined earlier, and reciprocal altruism. Consider the question of why people conserve water during a drought. Clearly they perceive that conservation contributes to the general good, which at a small cost is something they eagerly do. First note that, because the marginal social value of water is greater the less water there is, there are dimin- ishing social benefits of conservation: If other people conserve, it is less urgent for you to do so; if other people don't conserve, it is more urgent for you to do so. If you were a simple altruist, therefore, learning that others were not conserving would cause you to in- tensify your conservation efforts. This prediction is inconsistent with intuition and empirical evidence: People are more inclined to conserve water if they think other people are conserving, not if they think others are splurging. Peo- ple reciprocate the lack of public spirit- edness in others-they don't counteract it.

orr I-clntcd topics, I flinrik Holry Anrorr, Geol-gc Al<erlo.)r, Jnr1ti.s A~l(/reorii, Stcce~l Blntt, Gnry Chol-rlec.s, Erlrlie Delcel, Petrr Din~)iorrrl, Jo~i Elvter, Erik Ey~ter, Ermst Fehr. Dnuirl I. Leci~le. George Locice~,stcirl. Bob JlncCoirrl, Jorric.~ Alonlgo~~lery,Tni-Lorii Mrii, Ted O'Dor,ogliric. Drorcrl Prclcc. llyn Segol, Eldnr Shqfir. Ge~rc Sriiolcr~sky, Joel Sobcl, Ar)io.s Tcer.sliy, foirr nnollyJilolrs referees. nrlrl e.spcciolly Colirl Cn~)~erer, Dn~triy Knh~,e~)lnn, of thi.9

Comments
  • Recommend Us