1983 APA Award Addresses

Choices, Values, and FramesDaniel Kahneman University of British Columbia

Amos Tversky Stanford University

ABSTRACT: We discuss the cognitive and the psy-chophysical determinants of choice in risky and risk-less contexts. The psychophysics of value induce riskaversion in the domain of gains and risk seeking inthe domain of losses. The psychophysics of chanceinduce overweighting of sure things and of improbableevents, relative to events of moderate probability. De-cision problems can be described or framed in multipleways that give rise to different preferences, contraryto the invariance criterion of rational choice. The pro-cess of mental accounting, in which people organizethe outcomes of transactions, explains some anomaliesof consumer behavior. In particular, the acceptabilityof an option can depend on whether a negative outcomeis evaluated as a cost or as an uncompensated loss.The relation between decision values and experiencevalues is discussed.

Making decisions is like speaking prose—people doit all the time, knowingly or unknowingly. It is hardlysurprising, then, that the topic of decision making isshared by many disciplines, from mathematics andstatistics, through economics and political science, tosociology and psychology. The study of decisions ad-dresses both normative and descriptive questions. Thenormative analysis is concerned with the nature ofrationality and the logic of decision making. The de-scriptive analysis, in contrast, is concerned with peo-ple's beliefs and preferences as they are, not as theyshould be. The tension between normative and de-scriptive considerations characterizes much of thestudy of judgment and choice.

Analyses of decision making commonly distin-guish risky and riskless choices. The paradigmaticexample of decision under risk is the acceptability ofa gamble that yields monetary outcomes with specifiedprobabilities. A typical riskless decision concerns theacceptability of a transaction in which a good or aservice is exchanged for money or labor. In the firstpart of this article we present an analysis of the cog-nitive and psychophysical factors that determine thevalue of risky prospects. In the second part we extendthis analysis to transactions and trades.

Risky ChoiceRisky choices, such as whether or not to take anumbrella and whether or not to go to war, are madewithout advance knowledge of their consequences.Because the consequences of such actions depend onuncertain events such as the weather or the opponent'sresolve, the choice of an act may be construed as theacceptance of a gamble that can yield various out-comes with different probabilities. It is therefore nat-ural that the study of decision making under risk hasfocused on choices between simple gambles withmonetary outcomes and specified probabilities, in thehope that these simple problems will reveal basic at-titudes toward risk and value.

We shall sketch an approach to risky choice thatderives many of its hypotheses from a psychophysicalanalysis of responses to money and to probability.The psychophysical approach to decision making canbe traced to a remarkable essay that Daniel Bernoullipublished in 1738 (Bernoulli 1738/1954) in whichhe attempted to explain why people are generallyaverse to risk and why risk aversion decreases withincreasing wealth. To illustrate risk aversion and Ber-noulli's analysis, consider the choice between a pros-pect that offers an 85% chance to win $1000 (with a15% chance to win nothing) and the alternative ofreceiving $800 for sure. A large majority of peopleprefer the sure thing over the gamble, although thegamble has higher (mathematical) expectation. Theexpectation of a monetary gamble is a weighted av-erage, where each possible outcome is weighted byits probability of occurrence. The expectation of thegamble in this example is .85 X $1000 + .15 X$0 = $850, which exceeds the expectation of $800associated with the sure thing. The preference for thesure gain is an instance of risk aversion. In general,a preference for a sure outcome over a gamble thathas higher or equal expectation is called risk averse,and the rejection of a sure thing in favor of a gambleof lower or equal expectation is called risk seeking.

Bernoulli suggested that people do not evaluateprospects by the expectation of their monetary out-comes, but rather by the expectation of the subjective

April 1984 • American PsychologistCopyright 1984 by the American Psychological Association, Inc.Vol. 39, No. 4, 341-350

341

value of these outcomes. The subjective value of a gamble is again a weighted average, but now it is the subjective value of each outcome that is weighted by its probability. To explain risk aversion within this framework, Bernoulli proposed that subjective value, or utility, is a concave function of money. In such a function, the difference between the utilities of $200 and $ l 00, for example, is greater than the utility dif­ference between $1,200 and $ 1 , l 00. It follows from concavity that the subjective value attached to a gain of $800 is more than 80% of the value of a gain of $1,000. Consequently, the concavity of the utility function entails a risk averse preference for a sure gain of $800 over an 80% chance to win $1,000, although the two prospects have the same monetary expectation.

It is customary in decision analysis to describe the outcomes of decisions in terms of total wealth. For example, an offer to bet $20 on the toss of a fair coin is represented as a choice between an individual's current wealth W and an even chance to move to W + $20 or to W – $20. This representation appears psychologically unrealistic: People do not normally think of relatively small outcomes in terms of states of wealth but rather in terms of gains, losses, and neutral outcomes (such as the maintenance of the status quo). If the effective carriers of subjective value are changes of wealth rather than ultimate states of wealth, as we propose, the psychophysical analysis of outcomes should be applied to gains and losses rather than to total assets. This assumption plays a central role in a treatment of risky choice that we called prospect theory (Kahneman & Tversky, 1979). In­trospection as well as psychophysical measurements suggest that subjective value is a concave function of the size of a gain. The same generalization applies to losses as well. The difference in subjective value be­tween a loss of$200 and a loss of$ l 00 appears greater than the difference in subjective value between a loss of $1,200 and a loss of$ l , 100. When the value func­tions for gains and for losses are pieced together, we obtain an S-shaped function of the type displayed in Figure l .

This article was originally presented as a Distinguished Scientific Contributions Award address at the meeting of the American Psy­chological Association, Anaheim, California, August I 983. This work was supported by grant NR 197-058 from the U.S. Office of Naval Research.

Award addresses, submitted by award recipients, are published as received except for minor editorial changes designed to maintain American Psychologist format. This reflects a policy ofrecognizing distinguished award recipients by eliminating the usual editorial review process to provide a forum consistent with that employed in delivering the award address.

Requests for reprints should be sent to Daniel Kahneman, Department of Psychology, The University of British Columbia, Vancouver, BC, Canada.

342

Figure 1 A Hypothetical Value Function

VALUE

The value function shown in Figure 1 is (a) de­fined on gains and losses rather than on total wealth, (b) concave in the domain of gains and convex in thedomain oflosses, and ( c) considerably steeper for lossesthan for gains. The last property, which we label lossaversion, expresses the intuition that a loss of $X ismore aversive than a gain of $X is attractive. Lossaversion explains people's reluctance to bet on a faircoin for equal stakes: The attractiveness of the possiblegain is not nearly sufficient to compensate for theaversiveness of the possible loss. For example, mostrespondents in a sample of undergraduates refusedto stake $10 on the toss of a coin if they stood towin less than $30.

The assumption of risk aversion has played a central role in economic theory. However, just as the concavity of the value of gains entails risk aversion, the convexity of the value oflosses entails risk seeking. Indeed, risk seeking in losses is a robust effect, par­ticularly when the probabilities ofloss are substantial. Consider, for example, a situation in which an in­dividual is forced to choose between an 85% chance to lose $1,000 (with a 15% chance to lose nothing) and a sure loss of $800. A large majority of people express a preference for the gamble over the sure loss. This is a risk seeking choice because the expectation of the gamble (-$850) is inferior to the expectation of the sure loss (-$800). Risk seeking in the domain of losses has been confirmed by several investigators (Fishburn & Kochenberger, 1979; Hershey & Schoe­maker, 1980; Payne, Laughhunn, & Crum, 1980; Slovic, Fischhoff, & Lichtenstein, 1982). It has also been observed with nonmonetary outcomes, such as hours of pain (£raker & Sox, 1981) and loss of human lives (Fischhoff, 1983; Tversky, 1977; Tversky &

April 1984 • American Psychologist

Kahneman, 1981). Is it wrong to be risk averse inthe domain of gains and risk seeking in the domainof losses? These preferences conform to compellingintuitions about the subjective value of gains andlosses, and the presumption is that people should beentitled to their own values. However, we shall seethat an S-shaped value function has implications thatare normatively unacceptable.

To address the normative issue we turn frompsychology to decision theory. Modern decision theorycan be said to begin with the pioneering work of vonNeumann and Morgenstern (1947), who laid downseveral qualitative principles, or axioms, that shouldgovern the preferences of a rational decision maker.Their axioms included transitivity (if A is preferredto B and B is preferred to C, then A is preferred toC), and substitution (if A is preferred to B, then aneven chance to get A or C is preferred to an evenchance to get B or C), along with other conditions ofa more technical nature. The normative and the de-scriptive status of the axioms of rational choice havebeen the subject of extensive discussions. In particular,there is convincing evidence that people do not alwaysobey the substitution axiom, and considerable dis-agreement exists about the normative merit of thisaxiom (e.g., Allais & Hagen, 1979). However, all anal-yses of rational choice incorporate two principles:dominance and invariance. Dominance demands thatif prospect A is at least as good as prospect B in everyrespect and better than B in at least one respect, thenA should be preferred to B. Invariance requires thatthe preference order between prospects should notdepend on the manner in which they are described.In particular, two versions of a choice problem thatare recognized to be equivalent when shown togethershould elicit the same preference even when shownseparately. We now show that the requirement of in-variance, however elementary and innocuous it mayseem, cannot generally be satisfied.Framing of Outcomes

Risky prospects are characterized by their possibleoutcomes and by the probabilities of these outcomes.The same option, however, can be framed or describedin different ways (Tversky & Kahneman, 1981). Forexample, the possible outcomes of a gamble can beframed either as gains and losses relative to the statusquo or as asset positions that incorporate initialwealth. Invariance requires that such changes in thedescription of outcomes should not alter the pref-erence order. The following pair of problems illustratesa violation of this requirement. The total number ofrespondents in each problem is denoted by N, andthe percentage who chose each option is indicated inparentheses.

Problem 1 (N = 152): Imagine that the U.S. is preparingfor the outbreak of an unusual Asian disease, which is

expected to kill 600 people. Two alternative programs tocombat the disease have been proposed. Assume that theexact scientific estimates of the consequences of the pro-grams are as follows:

If Program A is adopted, 200 people will be saved. (72%)

If Program B is adopted, there is a one-third probabilitythat 600 people will be saved and a two-thirds probabilitythat no people will be saved. (28%)

Which of the two programs would you favor?

The formulation of Problem 1 implicitly adoptsas a reference point a state of affairs in which thedisease is allowed to take its toll of 600 lives. Theoutcomes of the programs include the reference stateand two possible gains, measured by the number oflives saved. As expected, preferences are risk averse:A clear majority of respondents prefer saving 200lives for sure over a gamble that offers a one-thirdchance of saving 600 lives. Now consider anotherproblem in which the same cover story is followedby a different description of the prospects associatedwith the two programs:

Problem 2 (N = 155): If Program C is adopted, 400 peoplewill die. (22%)

If Program D is adopted, there is a one-third probabilitythat nobody will die and a two-thirds probability that 600people will die. (78%)

It is easy to verify that options C and D in Prob-lem 2 are undistinguishable in real terms from optionsA and B in Problem 1, respectively. The second ver-sion, however, assumes a reference state in which noone dies of the disease. The best outcome is themaintenance of this state and the alternatives are lossesmeasured by the number of people that will die ofthe disease. People who evaluate options in these termsare expected to show a risk seeking preference forthe gamble (option D) over the sure loss of 400 lives,Indeed, there is more risk seeking in the second versionof the problem than there is risk aversion in the first.

The failure of invariance is both pervasive androbust. It is as common among sophisticated re-spondents as among naive ones, and it is not elimi-nated even when the same respondents answer bothquestions within a few minutes. Respondents con-fronted with their conflicting answers are typicallypuzzled. Even after rereading the problems, they stillwish to be risk averse in the "lives saved" version;they wish to be risk seeking in the "lives lost" version;and they also wish to obey invariance and give con-sistent answers in the two versions. In their stubbornappeal, framing effects resemble perceptual illusionsmore than computational errors.

The following pair of problems elicits preferencesthat violate the dominance requirement of rationalchoice.

April 1984 • American Psychologist 343

Problem 3 (N = 86): Choose between:

E. 25% chance to win $240 and75% chance to lose $760

F. 25% chance to win $250 and75% chance to lose $750

(0%)

(100%)

It is easy to see that F dominates E. Indeed, allrespondents chose accordingly.

Problem 4 (N = 150): Imagine that you face the followingpair of concurrent decisions. First examine both decisions,then indicate the options you prefer.

Decision (i) Choose between:A. a sure gain of $240B. 25% chance to gain $1000 and

75% chance to gain nothing

Decision (ii) Choose between:C. a sure loss of $750D. 75% chance to lose $1000 and

25% chance to lose nothing

(84%)

(16%)

(13%)

(87%)

As expected from the previous analysis, a largemajority of subjects made a risk averse choice for thesure gain over the positive gamble in the first decision,and an even larger majority of subjects made a riskseeking choice for the gamble over the sure loss inthe second decision. In fact, 73% of the respondentschose A and D and only 3% chose B and C. The samepattern of results was observed in a modified versionof the problem, with reduced stakes, in which un-dergraduates selected gambles that they would actuallyplay.

Because the subjects considered the two decisionsin Problem 4 simultaneously, they expressed in effecta preference for A and D over B and C. The preferredconjunction, however, is actually dominated by therejected one. Adding the sure gain of $240 (optionA) to option D yields 25% chance to win $240 and75% to lose $760. This is precisely option E in Prob-lem 3, Similarly, adding the sure loss of $750 (optionC) to option B yields a 25% chance to win $250 and75% chance to lose $750. This is precisely option Fin Problem 3. Thus, the susceptibility to framing andthe S-shaped value function produce a violation ofdominance in a set of concurrent decisions.

The moral of these results is disturbing: In vari-ance is normatively essential, intuitively compelling,and psychologically unfeasible. Indeed, we conceiveonly two ways of guaranteeing invariance. The firstis to adopt a procedure that will transform equivalentversions of any problem into the same canonical rep-resentation. This is the rationale for the standard ad-monition to students of business, that they shouldconsider each decision problem in terms of total assetsrather than in terms of gains or losses (Schlaifer, 1959).Such a representation would avoid the violations ofinvariance illustrated in the previous problems, butthe advice is easier to give than to follow. Except in

the context of possible ruin, it is more natural toconsider financial outcomes as gains and losses ratherthan as states of wealth. Furthermore, a canonicalrepresentation of risky prospects requires a com-pounding of all outcomes of concurrent decisions (e.g.,Problem 4) that exceeds the capabilities of intuitivecomputation even in simple problems. Achieving acanonical representation is even more difficult in othercontexts such as safety, health, or quality of life.Should we advise people to evaluate the consequenceof a public health policy (e.g., Problems 1 and 2) interms of overall mortality, mortality due to diseases,or the number of deaths associated with the particulardisease under study?

Another approach that could guarantee invari-ance is the evaluation of options in terms of theiractuarial rather than their psychological consequences.The actuarial criterion has some appeal in the contextof human lives, but it is clearly inadequate for financialchoices, as has been generally recognized at least sinceBernoulli, and it is entirely inapplicable to outcomesthat lack an objective metric. We conclude that frameinvariance cannot be expected to hold and that asense of confidence in a particular choice does notensure that the same choice would be made in anotherframe. It is therefore good practice to test the ro-bustness of preferences by deliberate attempts toframe a decision problem in more than one way(Fischhoff, Slovic, & Lichtenstein, 1980).

The Psychophysics of Chances

Our discussion so far has assumed a Bernoullian ex-pectation rule according to which the value, or utility,of an uncertain prospect is obtained by adding theutilities of the possible outcomes, each weighted byits probability. To examine this assumption, let usagain consult psychophysical intuitions. Setting thevalue of the status quo at zero, imagine a cash gift,say of $300, and assign it a value of one. Now imaginethat you are only given a ticket to a lottery that hasa single prize of $300. How does the value of theticket vary as a function of the probability of winningthe prize? Barring utility for gambling, the value ofsuch a prospect must vary between zero (when thechance of winning is nil) and one (when winning$300 is a certainty).

Intuition suggests that the value of the ticket isnot a linear function of the probability of winning,as entailed by the expectation rule. In particular, anincrease from 0% to 5% appears to have a larger effectthan an increase from 30% to 35%, which also appearssmaller than an increase from 95% to 100%. Theseconsiderations suggest a category-boundary effect: Achange from impossibility to possibility or from pos-sibility to certainty has a bigger impact than a com-parable change in the middle of the scale. This hy-pothesis is incorporated into the curve displayed in

344 April 1984 • American Psychologist

Figure 2, which plots the weight attached to an eventas a function of its stated numerical probability. Themost salient feature of Figure 2 is that decision weightsare regressive with respect to stated probabilities. Ex-cept near the endpoints, an increase of .05 in theprobability of winning increases the value of the pros-pect by less than 5% of the value of the prize. Wenext investigate the implications of these psycho-physical hypotheses for preferences among risky op-tions.

In Figure 2, decision weights are lower than thecorresponding probabilities over most of the range.Underweighting of moderate and high probabilitiesrelative to sure things contributes to risk aversion ingains by reducing the attractiveness of positive gam-bles. The same effect also contributes to risk seekingin losses by attenuating the aversiveness of negativegambles. Low probabilities, however, are over-weighted, and very low probabilities are either over-weighted quite grossly or neglected altogether, makingthe decision weights highly unstable in that region.The overweighting of low probabilities reverses thepattern described above: It enhances the value of longshots and amplifies the aversiveness of a small chanceof a severe loss. Consequently, people are often riskseeking in dealing with improbable gains and riskaverse in dealing with unlikely losses. Thus, the char-acteristics of decision weights contribute to the at-tractiveness of both lottery tickets and insurance pol-icies.

The nonlinearity of decision weights inevitablyleads to violations of invariance, as illustrated in thefollowing pair of problems:

Problem 5 (N = 85): Consider the following two-stage game.In the first stage, there is a 75% chance to end the gamewithout winning anything and a 25% chance to move intothe second stage. If you reach the second stage you have achoice between:

A. a sure win of $30B. 80% chance to win $45

(74%)(26%)

Your choice must be made before the game starts, i.e.,before the outcome of the first stage is known. Please indicatethe option you prefer.

Problem 6 (N = 81): Which of the following options doyou prefer?

C. 25% chance to win $30D. 20% chance to win $45

(42%)(58%)

Because there is one chance in four to move intothe second stage in Problem 5, prospect A offers a.25 probability of winning $30, and prospect B offers.25 X .80 = .20 probability of winning $45. Problems5 and 6 are therefore identical in terms of probabilitiesand outcomes. However, the preferences are not thesame in the two versions: A clear majority favors thehigher chance to win the smaller amount in Problem

Figure 2A Hypothetical Weighting Function

£l.O

e>UJ

.5

oUJQ 0 .5 1.0

STATED PROBABILITY: p

5, whereas the majority goes the other way in Problem6. This violation of invariance has been confirmedwith both real and hypothetical monetary payoffs (thepresent results are with real money), with humanlives as outcomes, and with a nonsequential repre-sentation of the chance process.

We attribute the failure of invariance to the in-teraction of two factors: the framing of probabilitiesand the nonlinearity of decision weights. More spe-cifically, we propose that in Problem 5 people ignorethe first phase, which yields the same outcome re-gardless of the decision that is made, and focus theirattention on what happens if they do reach the secondstage of the game. In that case, of course, they facea sure gain if they choose option A and an 80% chanceof winning if they prefer to gamble. Indeed, people'schoices in the sequential version are practically iden-tical to the choices they make between a sure gainof $30 and an 85% chance to win $45. Because asure thing is overweighted in comparison with eventsof moderate or high probability (see Figure 2) theoption that may lead to a gain of $30 is more attractivein the sequential version. We call this phenomenonthe pseudo-certainty effect because an event that isactually uncertain is weighted as if it were certain.

A closely related phenomenon can be demon-strated at the low end of the probability range. Supposeyou are undecided whether or not to purchase earth-quake insurance because the premium is quite high.As you hesitate, your friendly insurance agent comes

April 1984 • American Psychologist 345

forth with an alternative offer: "For half the regularpremium you can be fully covered if the quake occurson an odd day of the month. This is a good dealbecause for half the price you are covered for morethan half the days." Why do most people find suchprobabilistic insurance distinctly unattractive? Figure2 suggests an answer. Starting anywhere in the regionof low probabilities, the impact on the decision weightof a reduction of probability from p to p/2 is con-siderably smaller than the effect of a reduction fromp/2 to 0. Reducing the risk by half, then, is not worthhalf the premium.

The aversion to probabilistic insurance is sig-nificant for three reasons. First, it undermines theclassical explanation of insurance in terms of a con-cave utility function. According to expected utilitytheory, probabilistic insurance should be definitelypreferred to normal insurance when the latter is justacceptable (see Kahneman & Tversky, 1979). Second,probabilistic insurance represents many forms ofprotective action, such as having a medical checkup,buying new tires, or installing a burglar alarm system.Such actions typically reduce the probability of somehazard without eliminating it altogether. Third, theacceptability of insurance can be manipulated by theframing of the contingencies. An insurance policythat covers fire but not flood, for example, could beevaluated either as full protection against a specificrisk, (e.g., fire) or as a reduction in the overall prob-ability of property loss. Figure 2 suggests that peoplegreatly undervalue a reduction in the probability ofa hazard in comparison to the complete eliminationof that hazard. Hence, insurance should appear moreattractive when it is framed as the elimination of riskthan when it is described as a reduction of risk. Indeed,Slovic, Fischhoff, and Lichtenstein (1982) showed thata hypothetical vaccine that reduces the probabilityof contracting a disease from 20% to 10% is less at-tractive if it is described as effective in half of thecases than if it is presented as fully effective againstone of two exclusive and equally probable virus strainstliat produce identical symptoms.

Formulation Effects

So far we have discussed framing as a tool to dem-onstrate failures of invariance. We now turn attentionto the processes that control the framing of outcomesand events. The public health problem illustrates aformulation effect in which a change of wording from"lives saved" to "lives lost" induced a marked shiftof preference from risk aversion to risk seeking. Ev-idently, the subjects adopted the descriptions of theoutcomes as given in the question and evaluated theoutcomes accordingly as gains or losses. Another for-mulation effect was reported by McNeil, Pauker, Sox,and Tversky (1982). They found that preferences ofphysicians and patients between hypothetical therapies

for lung cancer varied markedly when their probableoutcomes were described in terms of mortality orsurvival. Surgery, unlike radiation therapy, entails arisk of death during treatment. As a consequence,the surgery option was relatively less attractive whenthe statistics of treatment outcomes were describedin terms of mortality rather than in terms of survival.

A physician, and perhaps a presidential advisoras well, could influence the decision made by thepatient or by the President, without distorting or sup-pressing information, merely by the framing of out-comes and contingencies. Formulation effects can oc-cur fortuitously, without anyone being aware of theimpact of the frame on the ultimate decision. Theycan also be exploited deliberately to manipulate therelative attractiveness of options. For example, Thaler(1980) noted that lobbyists for the credit card industryinsisted that any price difference between cash andcredit purchases be labeled a cash discount ratherthan a credit card surcharge. The two labels framethe price difference as a gain or as a loss by implicitlydesignating either the lower or the higher price asnormal. Because losses loom larger than gains, con-sumers are less likely to accept a surcharge than toforego a discount. As is to be expected, attempts toinfluence framing are common in the marketplaceand in the political arena.

The evaluation of outcomes is susceptible to for-mulation effects because of the nonlinearity of thevalue function and the tendency of people to evaluateoptions in relation to the reference point that is sug-gested or implied by the statement of the problem.It is worthy of note that in other contexts peopleautomatically transform equivalent messages into thesame representation. Studies of language compre-hension indicate that people quickly recede much ofwhat they hear into an abstract representation thatno longer distinguishes whether the idea was expressedin an active or in a passive form and no longer dis-criminates what was actually said from what was im-plied, presupposed, or implicated (Clark & Clark,1977). Unfortunately, the mental machinery that per-forms these operations silently and effortlessly is notadequate to perform the task of receding the twoversions of the public health problem or the mortality-survival statistics into a common abstract form.

Transactions and TradesOur analysis of framing and of value can be extendedto choices between multiattribute options, such asthe acceptability of a transaction or a trade. We pro-pose that, in order to evaluate a multiattribute option,a person sets up a mental account that specifies theadvantages and the disadvantages associated with theoption, relative to a multiattribute reference state.The overall value of an option is given by the balanceof its advantages and its disadvantages in relation to

346 April 1984 • American Psychologist

the reference state. Thus, an option is acceptable ifthe value of its advantages exceeds the value of itsdisadvantages. This analysis assumes psychological—but not physical—separability of advantages and dis-advantages. The model does not constrain the mannerin which separate attributes are combined to formoverall measures of advantage and of disadvantage,but it imposes on these measures assumptions of con-cavity and of loss aversion.

Our analysis of mental accounting owes a largedebt to the stimulating work of Richard Thaler (1980,in press), who showed the relevance of this processto consumer behavior. The following problem, basedon examples of Savage (1954) and Thaler (1980), in-troduces some of the rules that govern the constructionof mental accounts and illustrates the extension ofthe concavity of value to the acceptability of trans-actions.

Problem 7: Imagine that you are about to purchase a jacketfor $125 and a calculator for $15. The calculator salesmaninforms you that the calculator you wish to buy is on salefor $ 10 at the other branch of the store, located 20 minutesdrive away. Would you make a trip to the other store?

This problem is concerned with the acceptability ofan option that combines a disadvantage of incon-venience with a financial advantage that can be framedas a minimal, topical, or comprehensive account. Theminimal account includes only the differences be-tween the two options and disregards the features thatthey share. In the minimal account, the advantageassociated with driving to the other store is framedas a gain of $5. A topical account relates the con-sequences of possible choices to a reference level thatis determined by the context within which the decisionarises. In the preceding problem, the relevant topicis the purchase of the calculator, and the benefit ofthe trip is therefore framed as a reduction of theprice, from $15 to $10. Because the potential savingis associated only with the calculator, the price of thejacket is not included in the topical account. Theprice of the jacket, as well as other expenses, couldwell be included in a more comprehensive accountin which the saving would be evaluated in relationto, say, monthly expenses.

The formulation of the preceding problem ap-pears neutral with respect to the adoption of a min-imal, topical, or comprehensive account. We suggest,however, that people will spontaneously frame deci-sions in terms of topical accounts that, in the contextof decision making, play a role analogous to that of"good forms" in perception and of basic-level cate-gories in cognition. Topical organization, in con-junction with the concavity of value, entails that thewillingness to travel to the other store for a saving of$5 on a calculator should be inversely related to theprice of the calculator and should be independent of

the price of the jacket. To test this prediction, weconstructed another version of the problem in whichthe prices of the two items were interchanged. Theprice of the calculator was given as $125 in the firststore and $120 in the other branch, and the price ofthe jacket was set at $ 15. As predicted, the proportionsof respondents who said they would make the tripdiffered sharply in the two problems. The resultsshowed that 68% of the respondents (N = 88) werewilling to drive to the other branch to save $5 on a$15 calculator, but only 29% of 93 respondents werewilling to make the same trip to save $5 on a $125calculator. This finding supports the notion of topicalorganization of accounts, since the two versions areidentical both in terms of a minimal and a compre-hensive account.

The significance of topical accounts for consumerbehavior is confirmed by the observation that thestandard deviation of the prices that different storesin a city quote for the same product is roughly pro-portional to the average price of that product (Pratt,Wise, & Zeckhauser, 1979). Since the dispersion ofprices is surely controlled by shoppers' efforts to findthe best buy, these results suggest that consumershardly exert more effort to save $ 15 on a $ 150 pur-chase than to save $5 on a $50 purchase.

The topical organization of mental accountsleads people to evaluate gains and losses in relativerather than in absolute terms, resulting in large vari-ations in the rate at which money is exchanged forother things, such as the number of phone calls madeto find a good buy or the willingness to drive a longdistance to get one. Most consumers will find it easierto buy a car stereo system or a Persian rug, respec-tively, in the context of buying a car or a house thanseparately. These observations, of course, run counterto the standard rational theory of consumer behavior,which assumes invariance and does not recognize theeffects of mental accounting.

The following problems illustrate another ex-ample of mental accounting in which the posting ofa cost to an account is controlled by topical orga-nization:Problem 8 (N = 200): Imagine that you have decided tosee a play and paid the admission price of $10 per ticket.As you enter the theater, you discover that you have lostthe ticket. The seat was not marked, and the ticket cannotbe recovered.

Would you pay $10 for another ticket?Yes (46%) No (54%)

Problem 9 (N = 183): Imagine that you have decided tosee a play where admission is $10 per ticket. As you enterthe theater, you discover that you have lost a $10 bill.

Would you still pay $10 for a ticket for the play?Yes (88%) No (12%)

The difference between the responses to the two prob-

April 1984 • American Psychologist 347

lems is intriguing. Why are so many people unwillingto spend $10 after having lost a ticket, if they wouldreadily spend that sum after losing an equivalentamount of cash? We attribute the difference to thetopical organization of mental accounts. Going to thetheater is normally viewed as a transaction in whichthe cost of the ticket is exchanged for the experienceof seeing the play. Buying a second ticket increasesthe cost of seeing the play to a level that many re-spondents apparently find unacceptable. In contrast,the loss of the cash is not posted to the account ofthe play, and it affects the purchase of a ticket onlyby making the individual feel slightly less affluent.

An interesting effect was observed when the twoversions of the problem were presented to the samesubjects. The willingness to replace a lost ticket in-creased significantly when that problem followed thelost-cash version. In contrast, the willingness to buya ticket after losing cash was not affected by priorpresentation of the other problem. The juxtapositionof the two problems apparently enabled the subjectsto realize that it makes sense to think of the lost ticketas lost cash, but not vice versa.

The normative status of the effects of mentalaccounting is questionable. Unlike earlier examples,such as the public health problem, in which the twoversions differed only in form, it can be argued thatthe alternative versions of the calculator and ticketproblems differ also in substance. In particular, itmay be more pleasurable to save $5 on a $ 15 purchasethan on a larger purchase, and it may be more an-noying to pay twice for the same ticket than to lose$10 in cash. Regret, frustration, and self-satisfactioncan also be affected by framing (Kahneman & Tver-sky, 1982). If such secondary consequences are con-sidered legitimate, then the observed preferences donot violate the criterion of invariance and cannotreadily be ruled out as inconsistent or erroneous. Onthe other hand, secondary consequences may changeupon reflection. The satisfaction of saving $5 on a$ 15 item can be marred if the consumer discoversthat she would not have exerted the same effort tosave $10 on a $200 purchase. We do not wish torecommend that any two decision problems that havethe same primary consequences should be resolvedin the same way. We propose, however, that systematicexamination of alternative framings offers a usefulreflective device that can help decision makers assessthe values that should be attached to the primary andsecondary consequences of their choices.

Losses and Costs

Many decision problems take the form of a choicebetween retaining the status quo and accepting analternative to it, which is advantageous in some re-spects and disadvantageous in others. The analysis ofvalue that was applied earlier to unidimensional risky

prospects can be extended to this case by assumingthat the status quo defines the reference level for allattributes. The advantages of alternative options willthen be evaluated as gains and their disadvantages aslosses. Because losses loom larger than gains, the de-cision maker will be biased in favor of retaining thestatus quo.

Thaler (1980) coined the term "endowment ef-fect" to describe the reluctance of people to part fromassets that belong to their endowment. When it ismore painful to give up an asset than it is pleasurableto obtain it, buying prices will be significantly lowerthan selling prices. That is, the highest price that anindividual will pay to acquire an asset will be smallerthan the minimal compensation that would inducethe same individual to give up that asset, once ac-quired. Thaler discussed some examples of the en-dowment effect in the behavior of consumers andentrepreneurs. Several studies have reported substan-tial discrepancies between buying and selling pricesin both hypothetical and real transactions (Gregory,1983; Hammack & Brown, 1974; Knetsch & Sinden,in press). These results have been presented as chal-lenges to standard economic theory, in which buyingand selling prices coincide except for transaction costsand effects of wealth. We also observed reluctance totrade in a study of choices between hypothetical jobsthat differed in weekly salary (S) and in the temper-ature (T) of the workplace. Our respondents wereasked to imagine that they held a particular position(S,, T|) and were offered the option of moving to adifferent position (S2, T2), which was better in onerespect and worse in another. We found that mostsubjects who were assigned to (S1} T,) did not wishto move to (S2, T2), and that most subjects who wereassigned to the latter position did not wish to moveto the former. Evidently, the same difference in payor in working conditions looms larger as a disadvan-tage than as an advantage.

In general, loss aversion favors stability overchange. Imagine two hedonically identical twins whofind two alternative environments equally attractive.Imagine further that by force of circumstance thetwins are separated and placed in the two environ-ments. As soon as they adopt their new states as ref-erence points and evaluate the advantages and dis-advantages of each other's environments accordingly,the twins will no longer be indifferent between thetwo states, and both will prefer to stay where theyhappen to be. Thus, the instability of preferencesproduces a preference for stability. In addition to fa-voring stability over change, the combination of ad-aptation and loss aversion provides limited protectionagainst regret and envy by reducing the attractivenessof foregone alternatives and of others' endowments.

Loss aversion and the consequent endowmenteffect are unlikely to play a significant role in routine

348 April 1984 • American Psychologist

economic exchanges. The owner of a store, for ex-ample, does not experience money paid to suppliersas losses and money received from customers as gains.Instead, the merchant adds costs and revenues oversome period of time and only evaluates the balance.Matching debits and credits are effectively cancelledprior to evaluation. Payments made by consumersare also not evaluated as losses but as alternative pur-chases. In accord with standard economic analysis,money is naturally viewed as a proxy for the goodsand services that it could buy. This mode of evaluationis made explicit when an individual has in mind aparticular alternative, such as "I can either buy a newcamera or a new tent." In this analysis, a person willbuy a camera if its subjective value exceeds the valueof retaining the money it would cost.

There are cases in which a disadvantage can beframed either as a cost or as a loss. In particular, thepurchase of insurance can also be framed as a choicebetween a sure loss and the risk of a greater loss. Insuch cases the cost-loss discrepancy can lead to fail-ures of invariance. Consider, for example, the choicebetween a sure loss of $50 and a 25% chance to lose$200. Slovic, Fischhoff, and Lichtenstein (1982) re-ported that 80% of their subjects expressed a risk-seeking preference for the gamble over the sure loss.However, only 35% of subjects refused to pay $50 forinsurance against a 25% risk of losing $200. Similarresults were also reported by Schoemaker and Kun-reuther (1979) and by Hershey and Schoemaker(1980). We suggest that the same amount of moneythat was framed as an uncompensated loss in the firstproblem was framed as the cost of protection in thesecond. The modal preference was reversed in thetwo problems because losses are more aversive thancosts.

We have observed a similar effect in the positivedomain, as illustrated by the following pair of prob-lems:

Problem 10: Would you accept a gamble that offers a 10%chance to win $95 and a 90% chance to lose $5?

Problem 11: Would you pay $5 to participate in a lotterythat offers a 10% chance to win $100 and a 90% chanceto win nothing?

A total of 132 undergraduates answered the two ques-tions, which were separated by a short filler problem.The order of the questions was reversed for half therespondents. Although it is easily confirmed that thetwo problems offer objectively identical options, 55of the respondents expressed different preferences inthe two versions. Among them, 42 rejected the gamblein Problem 10 but accepted the equivalent lottery inProblem 11. The effectiveness of this seemingly in-consequential manipulation illustrates both the cost-loss discrepancy and the power of framing. Thinking

of the $5 as a payment makes the venture more ac-ceptable than thinking of the same amount as a loss.

The preceding analysis implies that an individ-ual's subjective state can be improved by framingnegative outcomes as costs rather than as losses. Thepossibility of such psychological manipulations mayexplain a paradoxical form of behavior that could belabeled the dead-loss effect. Thaler (1980) discussedthe example of a man who develops tennis elbow soonafter paying the membership fee in a tennis club andcontinues to play in agony to avoid wasting his in-vestment. Assuming that the individual would notplay if he had not paid the membership fee, the ques-tion arises: How can playing in agony improve theindividual's lot? Playing in pain, we suggest, maintains

' the evaluation of the membership fee as a cost. If theindividual were to stop playing, he would be forcedto recognize the fee as a dead loss, which may bemore aversive than playing in pain.

Concluding RemarksThe concepts of utility and value are commonly usedin two distinct senses: (a) experience value, the degreeof pleasure or pain, satisfaction or anguish in theactual experience of an outcome; and (b) decisionvalue, the contribution of an anticipated outcome tothe overall attractiveness or aversiveness of an optionin a choice. The distinction is rarely explicit in de-cision theory because it is tacitly assumed that decisionvalues and experience values coincide. This assump-tion is part of the conception of an idealized decisionmaker who is able to predict future experiences withperfect accuracy and evaluate options accordingly.For ordinary decision makers, however, the corre-spondence of decision values between experience val-ues is far from perfect (March, 1978). Some factorsthat affect experience are not easily anticipated, andsome factors that affect decisions do not have a com-parable impact on the experience of outcomes.

In contrast to the large amount of research ondecision making, there has been relatively little sys-tematic exploration of the psychophysics that relatehedonic experience to objective states. The most basicproblem of hedonic psychophysics is the determi-nation of the level of adaptation or aspiration thatseparates positive from negative outcomes. The he-donic reference point is largely determined by theobjective status quo, but it is also affected by expec-tations and social comparisons. An objective im-provement can be experienced as a loss, for example,when an employee receives a smaller raise than ev-eryone else in the office. The experience of pleasureor pain associated with a change of state is also crit-ically dependent on the dynamics of hedonic adap-tation. Brickman & Campbell's (1971) concept of thehedonic treadmill suggests the radical hypothesis that

April 1984 • American Psychologist 349

rapid adaptation will cause the effects of any objectiveimprovement to be short-lived. The complexity andsubtlety of hedonic experience make it difficult forthe decision maker to anticipate the actual experiencethat outcomes will produce. Many a person who or-dered a meal when ravenously hungry has admittedto a big mistake when the fifth course arrived on thetable. The common mismatch of decision values andexperience values introduces an additional elementof uncertainty in many decision problems.

The prevalence of framing effects and violationsof invariance further complicates the relation betweendecision values and experience values. The framingof outcomes often induces decision values that haveno counterpart in actual experience. For example,the framing of outcomes of therapies for lung cancerin terms of mortality or survival is unlikely to affectexperience, although it can have a pronounced influ-ence on choice. In other cases, however, the framingof decisions affects not only decision but experienceas well. For example, the framing of an expenditureas an uncompensated loss or as the price of insurancecan probably influence the experience of that out-come. In such cases, the evaluation of outcomes inthe context of decisions not only anticipates experi-ence but also molds it.

REFERENCES

Allais, M., & Hagen, O. (Eds.). (1979). Expected utility hypothesesand the Allais paradox. Hingham, MA: D. Reidel Publishing.

Bernoulli, D. (1954). Exposition of a new theory on the measurementof risk. Econometrica 22, 23-36. (Original work published 1738)

Brickman, P., & Campbell, D. T. (1971). Hedonic relativism andplanning the good society. In M. H. Appley (Ed.), Adaptation-level theory: A symposium (pp. 287-302). New York: AcademicPress.

Clark, H. H., & Clark, E. V. (1977). Psychology and language. NewYork: Harcourt Brace Jovanovich.

Erakar, S. E., & Sox, H. C. (1981). Assessment of patients' pref-erences for therapeutic outcomes. Medical Decision Making, 1,29-39.

Fischhoff, B. (1983). Predicting frames. Journal of ExperimentalPsychology: Learning, Memory and Cognition. 9, 103-116.

Fischhoff, B., Slovic, P., & Lichtenstein, S. (1980). Knowing whatyou want: Measuring labile values. In T. Wallsten (Ed.), Cognitiveprocesses in choice and decision behavior (pp. 117-141). Hillsdale,NJ: Erlbaum.

Fishburn, P. C., & Kochenberger, G. A. (1979). Two-piece von

Neumann-Morgenstern utility functions. Decision Sciences, 10,503-518.

Gregory, R. (1983). Measures of consumer's surplus: Reasons forthe disparity in observed values. Unpublished manuscript, KeeneState College, Keene, NH.

Hammack, J., & Brown, G. M., Jr. (1974). Waterfowl and wetlands:Toward bioeconomic analysis. Baltimore: Johns Hopkins Uni-versity Press.

Hershey, J. C., & Schoemaker, P. J. H. (1980). Risk taking andproblem context in the domain of losses:' An expected-utilityanalysis. Journal of Risk and Insurance, 47, 111-132.

Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysisof decision under risk. Econometrica, 47, 263-291.

Kahneman, D., & Tversky, A. (1982). The simulation heuristic.In D. Kahneman, P. Slovic, & A. Tversky (Eds.), Judgment underuncertainty: Heuristics and biases (pp. 201-208). New York:Cambridge University Press.

Knetsch, J., & Sinden, J. (in press). Willingness to pay and com-pensation demanded: Experimental evidence of an unexpecteddisparity in measures of value. Quarterly Journal of Economics.

March, J. G. (1978). Bounded rationality, ambiguity, and the en-gineering of choice. Bell Journal of Economics, 9, 587-608.

McNeil, B., Pauker, S., Sox, H., Jr., & Tversky, A. (1982). On theelicitation of preferences for alternative therapies. New EnglandJournal of Medicine, 306, 1259-1262.

Payne, J. W., Laughhunn, D. J., & Crum, R. (1980). Translationof gambles and aspiration level effects in risky choice behavior.Management Science, 26, 1039-1060.

Pratt, J. W., Wise, D., & Zeckhauser, R. (1979). Price differencesin almost competitive markets. Quarterly Journal of Economics,93. 189-211.

Savage, L. J. (1954). The foundation of statistics. New York: Wiley.Schlaifer, R. (1959). Probability and statistics for business decisions.

New York: McGraw-Hill.Schoemaker, P. J. H., & Kunreuther, H. C. (1979). An experimental

study of insurance decisions. Journal of Risk and Insurance, 46,603-618.

Slovic, P., Fischhoff, B., & Lichtenstein, S. (1982). Response mode,framing, and information-processing effects in risk assessment.In R. Hogarth (Ed.), New directions for methodology of socialand behavioral science: Question framing and response consistency(pp. 21-36). San Francisco: Jossey-Bass.

Thaler, R. (1980). Toward a positive theory of consumer choice.Journal of Economic Behavior and Organization, 1, 39-60.

Thaler, R. (in press). Using mental accounting in a theory of con-sumer behavior. Journal of Marketing.

Tversky, A. (1977). On the elicitation of preferences: Descriptiveand prescriptive considerations. In D. Bell, R. L. Kenney, & H.Raiffa (Eds.), Conflicting objectives in decisions. InternationalSeries on Applied Systems Analysis (pp. 209-222). New York:Wiley.

Tversky, A., & Kahneman, D. (1981). The framing of decisionsand the psychology of choice. Science, 211, 453-458.

von Neumann, J., & Morgenstern, O. (1947). Theory of games andeconomic behavior (2nd ed.). Princeton: Princeton UniversityPress.

350 April 1984 • American Psychologist