8/2/2019 311E-C02 Tversky Kahneman 1981

http://slidepdf.com/reader/full/311e-c02-tversky-kahneman-1981 1/7

The Framing of Decisions and the Psychology of ChoiceAuthor(s): Amos Tversky and Daniel KahnemanReviewed work(s):Source: Science, New Series, Vol. 211, No. 4481 (Jan. 30, 1981), pp. 453-458Published by: American Association for the Advancement of ScienceStable URL: http://www.jstor.org/stable/1685855 .

Accessed: 22/01/2012 23:10

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of 

content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms

of scholarship. For more information about JSTOR, please contact support@jstor.org.

 American Association for the Advancement of Science is collaborating with JSTOR to digitize, preserve and

extend access to Science.

8/2/2019 311E-C02 Tversky Kahneman 1981

http://slidepdf.com/reader/full/311e-c02-tversky-kahneman-1981 2/7

The majority choice in this problem isrisk averse: the prospect of certainlysaving 200 lives is more attractive han arisky prospect of equal expected value,that is, a one-in-three chance of saving600 lives.

A second group of respondents wasgiven the cover story of problem 1 with adifferent formulation of the alternative

programs,as follows:Problem2tN= 155]:

If Program C is adopted 400 people will die.[22 percent]

If Program D is adopted there is 1/3 probabil-ity that nobody will die, and 2/3 probabili-ty that 600 people will die. [78 percent]

Which of the two programs would you favor?

The majority choice in problem 2 isrisk taking: the certain death of 400people is less acceptablethan the two-in-three chance that 600 will die. The pref-erences in problems 1 and 2 illustrate acommonpattern:choices involvinggains

are often risk averse and choices in-volving losses are often risk taking.However, it is easy to see that the twoproblems are effectively identical. Theonly differencebetween them is that theoutcomes are described in problem 1 bythe numberof lives saved and in problem2 by the numberof lives lost. The changeis accompanied by a pronvunced shiftfrom risk aversion to risk taking. Wehave observed this reversal in severalgroupsof respondents, includinguniver-sity faculty and physicians. Inconsistentresponses to problems 1 and 2 arise fromthe conjunctionof a framingeffect withcontradictoryattitudes toward risks in-volving gains and losses. We turn nowto an analysis of these attitudes.

The Evaluation f Prospects

The major theory of decision-makingunder risk is the expected utility model.This model is based on a set of axioms,for example, transitivityof preferences,which provide criteria for the rationalityof choices. The choices of an individualwho conforms to the axioms can be de-

scribed n terms of the utilitiesof variousoutcomes for that individual.The utilityof a risky prospect is equal to the ex-pected utility of its outcomes, obtainedby weighting he utility of each possibleoutcome by its probability. When facedwith a choice, a rationaldecision-makerwill prefer the prospect that offers thehighest expected utility (1, 2).

Dr. Tversky s a professorof psychology at Stan-ford University,Stanford,California 4305, and Dr.Kahneman s a professorof psychology at the Uni-versity of British Columbia, Vancouver, CanadaV6T 1W5.

Explanations and predictions ofpeople's choices, in everyday ife as wellas in the social sciences, are often found-ed on the assumptionof humanrational-ity. The definitionof rationalityhas beenmuchdebated,but there is generalagree-ment that rationalchoices should satisfysome elementary requirements of con-sistency and coherence. In this article

tionalchoice requires hat the preferencebetween options should not reverse withchanges of frame. Because of imperfec-tions of humanperceptionand decision,however, changes of perspective oftenreverse the relative apparent ize of ob-jects and the relative desirabilityof op-tions.

We have obtained systematic rever-

Summary.The psychologicalprinciples hatgovern he perception f decision prob-lems and the evaluationof probabilities nd outcomes produce predictable hifts ofpreferencewhen the same problem s framed n differentways. Reversals of prefer-ence are demonstrated n choices regardingmonetaryoutcomes, both hypotheticaland real, and in questions pertaining o the loss of human ives. The effects of frameson preferences are compared to the effects of perspectives on perceptualappear-ance. The dependence of preferences on the formulation f decision problems s asignificant oncern for the theory of rational hoice.

we describe decision problems n whichpeople systematically violate the re-quirements of consistency and coher-ence, and we trace these violations o thepsychologicalprinciples that govern theperceptionof decision problemsand theevaluationof options.

A decision problem is defined by theacts or options among which one mustchoose, the possible outcomes or con-sequences of these acts, and the contin-gencies or conditional probabilities hatrelateoutcomes to acts. We use the term"decision frame" to refer to the deci-sion-maker's onception of the acts, out-cornes, and contingencies associatedwith a particular hoice. The frame hat a

decision-maker dopts is controlledpart-ly by the formulationof the problemandpartlyby the norms, habits, and personalcharacteristics f the decision-maker.

It is often possible to framea given de-cision problem in more than one way.Alternative rames for a decision problenzmay be compared o alternativeper-spectives on a visual scene. Veridicalperception requires that the perceivedrelativeheight of two neighboringmoun-tains, say, should not reverse withchanges of vantage point. Similarly, ra-

SCIENCE, VOL. 211, 30 JANUARY 1981

sals of preference by variations in theframing of acts, contingencies, or out-comes. These effects have been ob-served in a variety of problems and inthe choices of differentgroupsof respon-dents. Here we present selected illustra-tions of preference reversals, with dataobtainedfrom students at StanfordUni-versity and at the University of BritishColumbiawho answered brief question-naires in a classroom setting. The totalnumberof respondents or each problemis denoted by N, and the percentagewho chose each option is indicated inbrackets.

The effect of variations in framing sillustrated n problems 1 and 2.

Problem1 [N = 152]: magine hat the 1] S.is preparing or the outbreak of an unusualAsian disease, which is expected to kill 600people. Two alternativeprograms o combatthe disease have been proposed.Assume thatthg exact scientific estimate of the con-sequences of the programsare as follows:

If ProgramA is adopted, 200 people will besaved. [72 percent]

If ProgramB is adopted, there s 1/3probabil-ity that 600 people will be saved, and2/3 probability that no people will besaved. [28 percent]

Whichof the two programswouldyou favor?

0036-8075/81/0130-0453$01.50/0 Copyright t 1981 AAAS 453

The Framing of Decisionsand the

Psychologyof Choice-

Amos Tversky and Daniel Kahneman

8/2/2019 311E-C02 Tversky Kahneman 1981

http://slidepdf.com/reader/full/311e-c02-tversky-kahneman-1981 3/7

As will be illustratedbelow, people ex-hibitpatternsof preferencewhichappearincompatiblewith expected utilitytheo-ry. We have presented elsewhere (3) adescriptivemodel, called prospecttheo-ry, which modifies expected utilitytheo-ry so as to accommodate hese observa-tions. We distinguish two phases in thechoice process: an initialphase in which

acts, outcomes, and contingencies areframed,and a subsequentphase of eval-uation(4). Forsimplicity, we restrict heformal reatmentof the theory to choicesinvolving stated numerical probabilitiesand quantitative utcomes, suchas mon-ey, time, or numberof lives.

Consider a prospect that yields out-come x with probability p, outcome ywith probabilityq, and the status quowith probability 1 - p - q. Accordingto prospect theory, there are values v(.)associated with outcomes, and decisionweights w(.) associated with probabili-ties, such that the overall value of theprospect equalsw(p) v(x) + w(q) v(y). Aslightly differe'nt quation shouldbe ap-plied if all outcomes of a prospect are onthe same side of the zero point (5).

In prospect theory, outcomes are ex-pressed as positive or negative devia-tions (gains or losses) from a neutralref-erence outcome, which is assigneda val-ue of zero. Although subjective valuesdiffer among individualsand attributes,we propose that the value function iscommonly S-shaped,concave above thereference point and convex below it, asillustrated n Fig. 1. For example,the dif-

ference in subjective value betweengains of $10 and $20 is greater than thesubjective difference between gains of$110 and $120. The same relation be-tween value differences holds for thecorresponding osses. Another'propertyof the value function is that the responseto losses is more extreme than the re-sponse to gains. The displeasureassoci-ated with losinga sum of money is gener-ally greater hanthe pleasureassociatedwith winningthe same amount,as is re-flected in people's reluctance to acceptfairbets on a toss of a coin. Several stud-ies of decision (3, 6) and judgment (7)

have confirmedthese properties of thevalue function(8).

The second major departureof pros-pect theory from the expected utilitymodel involves the treatment of proba-bilities. In expected utility theory theutilityof an uncertainoutcome s weight-ed by its probability; n prospecttheorythe vafue of an uncertainoutcome s mul-tiplied by a decision weight w(p), whichis a monotonicfunction of p but is not aprobability. The weighting function v

454

Value The Framing of Acts

XGains

Problem [N = 150]: magine hat you facethe following pair of concurrentdecisions.First examine both decisions, then indicatethe options you prefer.

Decision(i). Choose between:A. a sure gain of $240 [84 percent]B. 25%chance to gain$1000, and

75% hance o gain nothing 16percent]

Decision(ii). Choose between:C. a sure loss of $750 [13 percent]D. 75%chance to lose $1000, and

25% hance o losenothing 87percent]

The majoritychoice in decision (i) isrisk averse: a riskless prospect is pre-ferred to a risky prospect of equal orgreaterexpected value. In contrast, themajority hoice in decision(ii) is risktak-ing: a risky prospect is preferred to ariskless prospect of equal expected val-ue. This pattern of risk aversion inchoices involving gains and risk seekingin choices involving osses is attributableto the propertiesof v and w. Because thevaluefunction s S-shaped, the value as-sociated with a gain of $240 is greaterthan 24 percent of the value associatedwith a gain of $1000, and the (negative)value associated with a loss of $750 issmaller han75 percentof the value asso-ciated with a loss of $1000. Thus theshape of the value functioncontributesto risk aversion n decision(i) and to riskseeking in decision (ii). Moreover, theunderweightingof moderate and highprobabilitiescontributes to the relativeattractivenessof the sure gain in (i) andto the relative aversiveness of the sure

loss in (ii). The same analysis applies toproblems 1 and 2.Because (i) and (ii) were presented o-

gether, the respondentshad in effect tochoose one prospect from the set: A andC,BandC,AandD, BandD. Themostcommon pattern (A and D) was chosenby 73 percent of respondents,while theleast popular pattern (B and C) waschosenby only 3 percentof respondents.However, the combination of B andC is definitely superior to the combina-tion A and D, as is readilyseen in prob-lem4.

Problem4 [N = 86]. Choose between:

A & D. 25%chance to win $240, and75% chance to lose $760. [0 per-cent]

B & C. 25%chance to win $250, and75% chance to lose $750. [100 per-cent]

When the prospects were combinedandthe dominanceof the second optionbecame obvious, all respondents chosethe superior option. The popularity ofthe inferioroption in problem 3 impliesthatthis problemwas framedas a pairof

SCIENCE, VOL. 211

Losses

Fig. 1. A hypotheticalvalue function.

has the following properties.First, im-possible events are discarded, that is,w(0) = O,and the scale is normalizedsothat w(1) = 1, but the function s not wellbehaved near the endpoints. Second,for low probabilities w@) >p, butw@) + w(1 - p) c 1. Thus low proba-bilities are overweighted, moderate andhigh probabilities are underweighted,and the latter effect is morepronouncedthan the former. Third, w(pq)lv(p) <

w(pqr)lv(pr) for all O< p, q, r ' 1. Thatis, for any fixed probabilityratioq, theratio of decision weights is closer tounity when the probabilities are lowthan when they are high, for example,w( 1 /v( 2) > w( 4)/v( 8) A hypotheticalweighting unction which satisfies theseproperties s shown in Fig. 2. The majorqualitativepropertiesof decisionweightscan be extended to cases in which theprobabilities f outcomes are subjective-ly assessed rather than explicitlygiven.In these situations, however, decision

weights may also be affected by othercharacteristicsof an event, such as am-biguityor vagueness (9).

Prospect theory, and the scales illus-trated n Figs. 1 and 2, should be viewedas an approximate, ncomplete,and sim-plified description of the evaluation ofrisky prospects. Although he propertiesof v andv summarizea common patternof choice, they are not universal: thepreferences of some individualsare notwell described by an S-shaped valuefunctionand a consistent set of decisionweights. The simultaneousmeasurementof values and decision weights involves

seriousexperimentaland statisticaldiffi-culties (10).

If v and v were linear throughout, hepreferenceorder between optionswouldbe independentof the framing of acts,outcomes, or contingencies. Because ofthe characteristicnonlinearities f v andv, however, different ramescan lead todifferent choices. The following threesections describe reversalsof preferencecaused by variations in the framing ofacts, contingencies,and outcomes.

8/2/2019 311E-C02 Tversky Kahneman 1981

http://slidepdf.com/reader/full/311e-c02-tversky-kahneman-1981 4/7

separate choices. The respondents ap-parently ailed to entertain hepossibilitythat the conjunction of two seeminglyreasonablechoices could lead to an un-tenableresult.

The violations of dominanceobservedin problem 3 do not disappear in thepresence of monetaryincentives. A dif-ferent group of respondents who an-

swered a modifiedversion of problem3,with realpayoffs,produceda similarpat-tern of choices (11). Otherauthors havealso reported hat violations of the rulesof rationalchoice, originallyobserved inhypotheticalquestions, were not elimi-nate,dby payoffs(12).

We suspect that manyconcurrentde-cisions in the real worldareframed nde-pendently,andthat the preferenceorderwouldoften be reversed if the decisionswere combined. The respondents inproblem3 failed to combineoptions, al-though the integration was relativelysimple and was encouraged by instruc-tions (13). The complexity of practicalproblemsof concurrentdecisions, suehas portfolio selection, would preventpeople from integratingoptions withoutcomputational ids, even if they were in-clined to do so.

TheFraming f Contingencies

The followingtriple of problems llus-trates he framingof contingencies.Eachproblem was presented to a differentgroup of respondents. Each group was

told that one participant in ten, pre-selected at random, would actually beplayingfor money. Charsce vents wererealized, in the respondents' presence,by drawinga single ball from a bag con-taininga knownproportion f balls of thewinningcolor, and the winnerswere paidimmediately.

Problem 5 [N = 77]: Which of the followingoptions do you prefer7

A. a sure win of $30 [78 percent]B. 80Wohance to win $45 [22 percent]

Problem 6 [N = 85]: Consider the followingtwo stage game. In the first stage, there is a75% chance to end the game without winning

anything, and a 25%chance to move into thesecond stage. If you reach the second stageyou have a choice between:

C. a sure win of $30 [74 percent]D. 80Wohance to win $45 [26 percent]

Your choice must be made before the gamestart:s, i.e., before the outcome of the firststage is known. Please indicate the option youprefer.

Problem 7 [N = 8i]: Which of the followingoptioIls do you prefer?

E. 25(Hohance to win $30 [42 percent]F. 20to chance to win $45 [58 percent]

30 JANUARY 1981

1.0 ,8

C2 #' /

e - wS /

y w#/. - w /

3; 0.5 - ,w'/

t ,' /

X _ ,,/- ':t

43

C]o t^ l ' l l

0 0.5 1.0

Stated probability: p

Fig. 2. A hypotheticalweighting unction.

Let us examine the structureof theseproblems. First, note that problems 6and 7 are identical n termsof probabili-ties and outcomes, because prospect Coffersa .25 chance to win $30 and pros-pect D offers a probability of .25 x.80 = .20 to win $45. Consistency there-fore requires that the same choice bemadein problems 6 and 7. Second, notethatproblem6 differs romproblem5 on-ly by the introduction of a preliminarystage. If the second stage of the game isreached, hen problem6 reduces to prob-lem5; if the gameends at the first stage,thedecision does not affectthe outcome.Hence there seems to be no reason tomake a different choice in problems 5and6. By this logicalanalysis,problem6is equivalent to problem 7 on the onehand and problem 5 on the other. Theparticipants,however, responded simi-larlyto problems 5 and 6 but differently

to problem7. This patternof responsesexhibits two phenomena of choice: thecertainty tffect and the pseudocertaintyeffect.

The contrast between problems 5 and7 illustrates a phenomenon discoveredby Allais (i4), whichwe havelabeled thecertaintyeffect: a reductionof the proba-bilityof an outcomeby aconstantfactorhas more impact when the outcome wasinitiallycertain than when it was merelyprobable.Prospecttheory attributes hiseffect to the propertiesof w. It is easy toverify, by applying he equationof pros-pect theory to problems 5 and 7, that

people for whom the value ratio v(30)/v(45) lies between the weight ratiosw(.20)/X(.25)and w(.80)/X(1.0)will pre-fer A to B and F to E, contrary to ex-pected utility theory. Prospect theorydoes not predicta reversalof preferencefor every individual in problems 5 and7. It only requires hat an ihdividualwhohasno preferencebetween A and B pre-fer F to E. For group data, the theorypredicts the observed directional shiftof preferencebetween thetwoproblems.

The first stage of problem6 yields thesame outcome (no gain) for both acts.Consequently,we propose, people eval-uate the options conditionally,as if thesecond stage had been reached. In thisframing,of course, problem6 reduces toproblem5. More generally, we suggestthat a decision problem s evaluatedCOh-

ditionally when (i) there is a state in

which all acts yield the same outcome,such as failingto reach the second stageof the game in problem 6, and (ii) thestated probabilities of other outcomesareconditionalon the nonoccurrenceofthis state.

The strikingdiscrepancybetweerltheresponses to problems6 and 7, whichareidentical in outcomes and probabilities,couid be described as a pseudocertaintyeffect. The prospect yielding$30 is rela-tively more attractivein problem 6 thanin problem7, as if it had the advantageofcertainty. The sense of certainty associ-ated with option C is illusory, however,since the gain is in fact contingent onreaching the second stage of the game(15).

We have observed the certaintyeffectin several sets of problems, with out-comes ranging romvacationtrips to theloss of human ives. In the negative do-main,certaintyexaggerates he aversive-ness of losses that are certainrelativetolosses that are merely probable. In aquestiondealing with the response to anepidemic, for example, most respond-entsfound iia sure loss of 75 lives" moreaversive than ii80% chance to lose 100

lives" butpreferred ; lOSo chance to lose75 lives" over ii8% chance to lose 100lives," contrary o expected utilitytheo-ry.

We also obtainedthe pseudocertaintyeffect in several studies where the de-scription of the decision problems fa-vored conditional evaluation. Pseudo-certaintycan be inducedeither by a se-quential ormulation,as in problem6, orby the introduction of causal contin-gencies. In another version of the epi-demic problem, for instance, respond-ents were toldthat risk to life existed on-ly in the event (probability io) that the

diseasewas carriedby a particular irus.Two alternativeprogramswere said tvyield i'a sure loss of 75 lives" or ii80%chance to lose 100 lives" if the criticalviruswas involved, and no loss of life inthe event (probability 90) that the dis-ease was carriedby anothervirus. In ef-fect, the respondents were asked tochoose between 10 percent chance oflosing 75 lives and 8 percent chance oflosing 100 lives, but their preferenceswere the same as when the choice was

455

8/2/2019 311E-C02 Tversky Kahneman 1981

http://slidepdf.com/reader/full/311e-c02-tversky-kahneman-1981 5/7

between a sure loss of 75 lives and 80percent chance of losing 100 lives. Aconditional framing was evidentlyadopted n which the contingencyof thenoncriticalvirus was eliminated, givingrise to a pseudocertainty ffect. The cer-taintyeffect reveals attitudes owardriskthat are inconsistent with the axioms ofrational choice, whereas the pseudo-

certaintyefFectviolates the more funda-mental requirement that preferencesshould be independent of problem de-scription.

Many significantdecisions concern ac-tions that reduce or eliminate he proba-bility of a hazard, at some cost. Theshape of 7r n the range of low probabili-ties suggests that a protective actionwhich reduces the probabilityof a harmfrom 1 percent to zero, say, will be val-ued more highly than an action that re-duces the probabilityof the same harmfrom 2 percent to 1 percent. Indeed,probabilistic insurance, which reduces

the probabilityof loss by half, is judgedto be worth less than half the price ofregular nsurance hat eliminates he riskaltogether 3).

It is often possible to frame protectiveaction in either conditional or uncon-ditional orm. For example, an insurancepolicy that covers fire but not flood couldbe evaluated either as full protectionagainst he specific risk of fire or as a re-duction in the overall probability ofproperty loss. The preceding analysissuggests that insurance should appearmore attractive when it is presented asthe eliminationof risk than when it is de-scribed as a reduction of risk. P. Slovic,B. Fischhoff, and S. Lichtenstein, n anunpublishedstudy, found that a hypo-thetical vaccine which reduces the prob-ability of contractinga disease from .20to .10 is less attractive f it is describedaseffective n half the cases than f it is pre-sented as fully effective against one oftwo (exclusive and equiprobable)virusstrains hat produce denticalsymptoms.In accord with the present analysis ofpseudocertainty, he respondentsvaluedfull protection against an identified vi-rus more than probabilistic protection

againstthe disease.The preceding discussion highlightsthe sharpcontrastbetween lay responsesto the reduction and the eliminationofrisk. Because no form of protective ac-tion can cover all rtsks to humanwelfare,all insurance s essentially probabilistic:it reduces but does not eliminate risk.The probabilisticnature of insurance iscommonIymasked by formulations hatemphasize the completeness of pro-tection against identified harms, but thesense of security that such formulations

456

provide s an illusion of conditional ram-ing. It appears that insurance s boughtas protection against worry, not onlyagainst risk, and that worry can be ma-nipulated by the labeling of outcomesand by the framingof contingencies.It isnot easy to determine whether peoplevalue the eliminationof risk too much orthe reductionof risk too little. The con-

trastingattitudes o the two formsof pro-tective action, however, are difficult ojustify on normativegrounds(16).

TheFraming f Outcomes

Outcomesare commonly perceived aspositive or negative in relation o a refer-ence outcome that is judged neutral.Variations of the reference point canthereforedeterminewhethera given out-come is evaluated as a gain or as a loss.Because the value function is generallyconcave for gains, convex for losses, and

steeper or losses than for gains, shifts ofreference can change the value dif-ference between outcomes and therebyreverse the preference order betweenoptions (6). Problems 1 and 2 illustrateda preferencereversal induced by a shiftof reference that transformedgains intolosses.

For another example, consider a per-son who has spent an afternoon at therace track, has already lost $140, and isconsideringa $10 bet on a 15 1 long shotin the last race. This decision can beframed in two ways, which correspondto two natural reference points. If thestatusquo is the referencepoint, the out-comes of the bet are framedas a gain of$140 and a loss of $10. On the otherhand, t may be more natural o view thepresent state as a loss of $140, for thebetting day, and accordingly frame thelast bet as a chance to return o the refer-ence point or to increase the loss to $150.Prospect theory implies that the latterframe will produce more risk seekingthan the former. Hence, people who donot adjust their reference point as theylose are expected to take bets that theywould normally find unacceptable. This

analysis is supportedby the observationthat bets on long shots are most popularon the last race of the day (17).

Because the value function is steeperfor losses than for gains, a differencebe-tween options will loom largerwhen it isframed as a disadvantageof one optionrather han as an advantageof the otheroption. An interesting example of suchan effect in a riskless context has beennoted by Thaler (18). In a debate on aproposal to pass to the consumer someof the costs associated with the process-

ing of credit-cardpurchases, representa-tives of the credit-card industry re-quested that the price difference be la-beled a cash discount rather than acredit-card urcharge.The two labels in-duce difFerenteferencepointsby implic-itly designatingas normal reference thehigheror the lower of the two prices. Be-cause losses loom larger han gains, con-

sumers are less willing to accept a sur-charge han to forego a discount. A simi-lar effect has been observed inexperimental studies of insurance: theproportion f respondentswho preferreda sure loss to a largerprobable oss wassignificantly greater when the formerwas called an insurance premium (19,20).

These observationshighlight he labil-ity of reference outcomes, as well astheir role in decision-making. n the ex-amples discussed so far, the neutralref-erence point was identifiedby the label-ing of outcomes. A diversity of factors

determine the reference outcome ineveryday life. The reference outcome isusuallya state to which one has adapted;it is sometimes set by social norms andexpectations; it sometimes correspondsto a level of aspiration, which may ormay not be realistic.

We have dealt so far with elementaryoutcomes, such as gains or losses in asingleattribute. n many situations,how-ever, an action gives rise to a compoundoutcome, which oins a series of changesin a single attribute, such as a sequenceof monetarygains and losses, or a set ofconcurrent hanges in several attributes.To describe the framingand evaluationof compoundoutcomes, we use the no-tion of a psychological account, definedas an outcome frame which specifies (i)the set of elementaryoutcomes that areevaluated jointly and the manner inwhich they are combinedand (ii) a refer-ence outcome that is considered neutralor normal. In the account that is set upfor the purchase of a car, for example,the cost of the purchase s not treatedasa loss nor is the car viewed as a gift.Rather, the transaction as a whole isevaluated as positive, negative, or neu-

tral, depending on such factors as theperformanceof the car and the price ofsimilarcars in the market. A closely re-lated treatmenthas been offeredby Tha-ler (18).

We propose that people generallyevaluate acts in terms of a minimalac-count, which includes only the directconsequences of the act. The minimalaccount associated with the decision toaccept a gamble, for example, includesthe money won or lost in thatgambleandexcludes other assets or the outcome of

SCIENCE, VOL. 211

8/2/2019 311E-C02 Tversky Kahneman 1981

http://slidepdf.com/reader/full/311e-c02-tversky-kahneman-1981 6/7

previous gambles. People commonlyadopt minimal accounts because thismode of framing i) simplifiesevaluationand reduces cognitive strain, (ii) refiectsthe intuition that consequences shouldbe causally linked to acts, and (iii)matches the properties of hedonic expe-rience, which is more sensitive to desir-able and undesirable changes than to

steady states.There are situations, however-, inwhich the outcomes of an act affect thebalance in an account that was pre-viously set up by a related act. In thesecases, the decision at hand may be eval-uated in terms of a more inclusive ac-count, as in the case of the bettor whoviews the last race in the context of ear-lier losses. More generally, a sunk-costeffect arises when a decision is referredto an existing account in which the cur-rent balance is negative. Because of thenonlinearities f the evaluation process,the minimal account and a more in-

clusive one often lead to differentchoices.

Problems 8 and 9 illustrate anotherclass of situations in which an existingaccount affects a decision:

Problem 8 [N= 183]: Imagine that youhave decidedto see a play where admission s$10 per ticket. As you enter the theater youdiscover that you have lost a $10 bill.

Wouldyou still pay $10 for a ticket for theplay?Yes [88percent] No [12 percent]

Problem 9 [N = 200]: Imagine that youhave decided o see a play and paid he admis-sion price of $10 per ticket. As you enter the

theater you discover that you have lost theticket. The seat was not marked nd the ticketcannotbe recovered.

Wouldyou pay $10 for another icket?Yes [46percent] No [54 percent]

The markeddifferencebetween the re-sponses to problems 8 and 9 is an effectof psychological accounting. We pro-pose that the purchase of a new ticket inproblem9 is entered in the account thatwas set up by the purchaseof the originalticket. In terms of this account, the ex-pense required o see the show is $20, acost which many of our respondentsap-

parently ound excessive. In problem 8,on the other hand, the loss of $10 is notlinked specifically o the ticket purchaseand its effect on the decision is accord-ingly slight.

The following problem, based on ex-amples by Savage (2, p. 103) and Thaler(18), further llustrates the effect of em-

beddingan option in differentaccounts.Two versions of this problem were pre-sented to different groups of subjects.One group (N = 93) was given the val-ues that appear n parentheses, and the

30 JANUARY 1981

other group (N = 88) the values shownin brackets.

Problem 10: Imagine that you are about topurchase a jacket for ($125) [$15], and a calcu-lator for ($ 15) [$ 125] The calculator salesmaninforms you that the calculator you wish tobuy is on sale for ($10) [$120] at the otherbranch of the store, located 20 minutes driveaway. Would you make the trip to the otherstore?

The response to the two versions ofproblem 10 were markedlydifferent:68percent of the respondents were willingto make an extra trip to save $5 on a $15calculator;only 29 percent were willingto exert the same effortwhen the price ofthe calculator was $125. Evidently therespondentsdo not frame problem 10 inthe minimalaccount, which involves on-ly a benefit of $5 and a cost of some in-convenience. Instead, they evaluate thepotential saving in a more inclusive ac-count, which includes the purchase ofthe calculatorbut not of the jacket. By

the curvatureof v, a discount of $5 has agreater mpact when the price of the cal-culator s low than when it is high.

A closely related observationhas beenreportedby Pratt, Wise, and Zeckhauser(21), who found that the variability f theprices at which a given product s sold bydifferent tores is roughlyproportional othe mean price of that product.The samepatternwas observed for both frequentlyand infrequentlypurchased tems. Over-all, a ratio of 2: 1 in the meanprice of twoproducts is associated with a ratio of1.86:1 in the standard deviation of therespective quoted prices. If the effortthat consumersexert to save each dollaron a purchase, for instance by a phonecall, were independentof price, the dis-persionof quoted prices should be aboutthe same for all products. In contrast,the data of Prattet al. (21 are consistentwith the hypothesis that consumershardlyexert more effort to save $15 on a$150 purchase than to save $5 on a $50purchase (18). Many readers will recog-nize the temporarydevaluationof moneywhich facilitates extra spending and re-duces the significanceof small discountsin the context of a large expenditure,

such as buying a house or a car. Thisparadoxical variation in the value ofmoney is incompatiblewith the standardanalysis of consumer behavior.

Discussion

In this article we have presented a se-ries of demonstrations in which seem-ingly inconsequentialchanges in the for-muiationof choice problemscaused sig-nificant shifts of preference. The in-

consistencies were traced to the inter-action of two sets of factors: variationsin the framingof acts, contingencies,andoutcomes, and the characteristic non-linearities of values and decisionweights. The demonstrated effects arelarge' and systematic, although by nomeans universal. They occur when theoutcomes concern the loss of human

lives as well as in choices about money;they are not restricted to hypotheticalquestionsand arenot eliminatedby mon-etary incentives.

Earlier we comparedthe dependenceof preferenceson frames to the depen-dence of perceptualappearanceon per-spective. If while traveling n a mountainrange you notice that the apparentrela-tive height of mountainpeaks varies withyour vantage point, you will concludethat some impressionsof relative heightmust be erroneous, even when you haveno access to the correct answer. Similar-ly, one may discover that the relative at-

tractiveness of options varies when thesame decision problem is framed in dif-ferent ways. Such a discovery will nor-mally lead the decision-maker o recon-siderthe originalpreferences,even whenthere is no simple way to resolve the in-consistency. The susceptibility to per-spective effects is of special concern inthe domain of decision-makingbecauseof the absence of objective standardssuch as the true height of mountains.

The metaphorof changingperspectivecan be applied to other phenomena ofchoice, in addition o the framingeffectswith which we have been concernedhere(19). The problemof self-control s natu-rally construed n these terms. The storyof Ulysses' request to be bound to themast of the ship in anticipationof the ir-resistible emptationof the Sirens' call iso'ftenused as a paradigmcase (22). Inthis example of precommitment,an ac-tion taken n thepresentrenders nopera-tive an anticipated uture preference. Anunusualfeature of the problemof inter-temporal conflict is that the agent whoviews a problemfrom a particular em-poral perspective is also aware of theconflictingviews that futureperspectives

will offer. In most other situations, deci-sion-makersare not normally aware ofthe potential effects of differentdecision'frames on theirpreferences.

The perspective metaphor highlightsthe following aspects of the psychologyof choice. Individualswho face a deci-sion problem and have a definite prefer-ence (i) mighthave a differentpreferencein a different ramingof the same prob-lem, (ii) are normallyunawareof alterna-tive frames and of their potential effectson the relative attractivenessof options,

4s7

8/2/2019 311E-C02 Tversky Kahneman 1981

http://slidepdf.com/reader/full/311e-c02-tversky-kahneman-1981 7/7

(iii) would wish their preferencesto beindependentof frame, but (iv) are oftenuncertainhow to resolve detectedincon-;sistencies (23). In some cases (such asproblems3 and 4 and perhapsproblems8and 9) the advantage of one frame be-comes evident once the competingframesarecompared,but in othercases-(problems1 and 2 andproblems6 and 7)

it is not obvious which preferencesshouldbe abandoned.These observationsdo not imply that

preferencereversals, or other errors ofchoice orjudgment (24), are necessarilyirrational.Like other intellectual imita-tions, discussedby Simon (25)under heheading of ';bounded rationality," thepractice of acting on the most readilyavailableframecan sometimesbe justi-fiedby referenceto the mentaleffort re-quired o explore alternative ramesandavoid potential inconsistencies. How-ever, we propose that the details of thephenomenadescribed in this article are

better explainedby prospecttheory andby an analysis of framingthan by adhoc appeals to the notion of cost ofthinking.

The present work has been concernedprimarilywith the descriptive questionof how decisions are made, but the psy-chology of choice is aiso relevantto thenormative question of how decisionsought to be made. In orderto avoid thedifficultproblemof justifyingvalues, themodern theory of rational choice hasadopted he coherenceof specificprefer-ences as the sole criterionof rationality.This approach enjoins the decision-nzaker o resolve inconsistenciesbut of-fers no guidaxlce n how to do so. It im-plicitly assumes that the decisionmakerwho carefully answers the questio';Whatdo I really want?"will eventuallyachieve coherentpreferences.However,the susceptibilityof preforences o varia-tions of framingraises doubt about thefeasibility and adequacy of the coher-ence criterion.

Consistency is only one aspect of thelaynotionof rationalbehavior.As notedby March(?6), the commonconceptionof rationalityalso requires that prefer-

ences or utilitiesfor particular utcomesshould be predictive of the experiencesof satisfactionor displeasureassociatedwith theiroccurrence.Thus, a mancouldbe judged irratioIlaleither because hispreferencesarecontradictory r becausehis desires and aversions do not reflecthis pleasures and pains. The-predictivecriterionof rationalitycan be appliedtoresolve inconsistent preferences and toimprove he qualityof decisions. A pre-

dictive orientationencouragesthe deci-sion-makero focus on futureexperienceand to ask " What will I feel then?"rather han"What do I want now?" Theformer question, when answered withcare, can be the moreusefulguide indif-ficult decisions. In particular,predictiveconsiderationsmay be appliedto selectthe decision frame that best represents

the hedonicexperience of outcomes.Furthercomplexities arise in the nor-mative analysis because the framingofan action sometimes affects the actualexperience of its outomes. For ex-ample, framing outcomes in terms ofoverallwealth or welfare ratherthan intermsof specificgainsand osses may at-tenuate one's emotional response to anoccasional loss. Similarly, the experi-ence of a changefor the worse mayvaryif the changeis framed as an uncompen-sated loss or as a cost incurred toachieve some benefit. The framing ofacts and outcomes can also reflect the

acceptanceor rejectionof responsibilityfor particular onsequences, and the de-liberatemanipulation f framing s com-monly used as an instxumentof self-control (22 . When framing influencesthe experience of consequences, theadoptionof a decision frame is an ethi-cally significantact.

References ndNotes

1. J. VonNeumann nd 0. Morgenstern, heory fGames and Economic Behavior (PrincetonUniv. Press, Princeton,N.J., 1947);H. Raiffa,DecisionAnalysis:Lectures on Choices UnderUncertainty Addison-Wesley,Reading,Masst,1968);P. Fishburn,UtilityTheory or DecisionMaking Wiley, New York, 1970).

2. L. J. Savage, The Foundationsof Statistics(Wiley, New York, 1954).

3. D. Kahnemanand A. Tversky,Econometrica47, 263 (19793.

4. The framing hase ncludesvarious ditingoper-ations hat areapplied o simplifyprospects, orexampleby comblning ventsor outcomesor bydiscarding egligiblecomponents 3).

5. Efp + q = 1 andeitherx > y > Oorx < y < O,the equation in the text is replaced byv@) + 7r@) [v(x) - v(y)l, so that decisionweightsare not applied o sure outcomes.

6. P. Fishburn ndG. Kochenberger, ecisionSci.10, 503 (1979);D. J. Laughhunn, . W. Payne,R. Crum,Manage. Sci., in press;J. W. Payne,D. J. Laughhunn,R. Crum,bid., in press;S. A.Erakerand H. C. Sox, Med. Decision MakingFin press.In the last studyseveralhundred linicpatients made hypothetical choices betweendrug therapies or severe headaches,hyperten-sion, and chest pain. Most patientswere riskaverse when the outcomes were described aspositive for example,reducedpainor increasedlife expectancy)and risk takingwhen the out-comes were described as negative (increasedpain or reduced ife expectancy).No significantdifferenceswere found between patients whoactually sufferedfrom the ailmentsdescribedand patientswho did not.

7. E. Galanterand P. Pliner, in Sensation andMeasurement,H. R. Moskowitzet al., Eds.(Reidel,Dordrecht,1974),pp 65-76.

8. The extension of the proposed value functiorl tomultiattribute options, with or without risk, de-serves careful analysis.- In particular, indif-ference curves between dimensions of loss maybe concave upward, even when the value func-tions for the separate losses are both convexbecause of marked subadditivity between di-mensions.

9. D. Ellsberg,Q.J. Econ. 7S, 643 (1961);W. Fell-ner, Probability and Profit-A Study of Eco-nomic Behavior Along Bayesian Lines (Irwin,Homewood,Ill., 1965).

10. The scalingof v a,nd7rby paircomparisons e-quiresa largenumberof observations.The pro-cedureof pricing ambles s moreconvenient orscalingpurposes,but t is subject o a severean-choringbias: the orderingof gamblWesy theircash equivalenfs diverges systematically romthe preferenceorder observed in direct com-parisons S. Lichtensteinand P. Slovic,J. Exp.Psychol. 89, 46 (1971)].

11. A newgroupof respondents N - 126)was pre-sentedwith a modifiedversionof problem3, in

which the outcomes were redllcedby a factorof 50. The participantswere informed hat thegambleswouldactually be playedby tossing apair of fair coins, that one participantn tenwouldbe selectedat randomo play thegamblesof his or her choice.To ensurea positivereturnfor the entireset, a thirddecision,yieldingonlypositive outcomes, was added. These payoffconditions did not alter the patternof prefer-ences observed n'the hypotheticalproblem:67percentof respondents hose prospectA and 86percentchose prospectD. Thedominated om-bination f A and D was chosenby 60 percentofrespondents,and only 6 percent favored thedominant'combinationf B and C.

12. S. Lichtenstein nd P. Slovic,J. Exp. Psychol.101, 16 (1973);D. M. Gretherand C. R. Plott,Am. Econ. Rev. 69, 623 (1979) I. I,ieblichandA. Lieblich,Percept. Mot. skilis 29, 467 (1969)D. M. Grether,Social Science Working PaperAto 245 (CaliforniaInstitute of Technology,Pasadena, 979).

13. Other demonstrationsof a reluctance to in-

tegrateconcurrent ptions have been reported:P. Slovicand S. Lichtenstein, . Exp. Psychol.78, 646 (1968), J. W. Payne and M. L. Braun-stein, bid. 87, 13 (1971).

14. M. Allais, Econometrica 21, 503 (1953); K.McCrimmonnd S. Larsson, n Expected Util-ity Hypotheses and the Allais Paradox, M. All-ais and O. Hagan,'Eds. (Reidel, Dordrecht,1979).

15. Another group of respondents N= 205) waspresentedwith all three problems, n differentorders,'withoutmonetarypayoffs.'Theointfre-quencydistribution f choicesin problems5, 6,and7 was as follows:ACE, 22; ACF, 65;ADE,4; ADF, 20; BCE, 7; BCF, 18; BDE, 17;BDF,52.These dataconfirmn a within-subjectesignthe analysisof conditional valuationproposedin the text. More han 75 percentof respondentsmade compatiblechoices (AC or BD) in prob-lems 5 'and , and ess thanhalfmadecompatiblechoices n problems6 and 7 (CEor DF) or 5 and7 (AE or BF).The elimination f payoffs n thesequestionsreducedrisk aversionbut did not sub-

stantially alter the effects of certainty andpsoudocertainty

16. For further discussion of rationality n pro-tective action see H. KunreutherDisaster In-surance Protection: Public Poiicy Lessons(Wiley, New York, 1978).

17. W. H. McGlothlin,Am J. Psychol. 69, 604(1956)

18. R. Thaler,J.Econ. Behav. Organ. 1, 39 (1980).19. B. Fischhoff,P. Slovic, S. Lichtenstein,nCog-

nitive Processes in Choice and Decision Behav-ior, T. Wallsten,Ed. (Erlbaum,Hillsdale,N.J.,1980).

2Q.J. C. Hersheyand P. J. H. Schoemaker, . Risk.

lnsur., ln press.21. J. Pratt,A. Wise, R. Zeckhauser,Q. J. Econ.

93, 189 (1979).22. R. H. Strotz,Rev. Econ. Stud. 23, 165 1955);G.

Ainslie,Psychol. Bull. 82, 463(1975);J. Elster,Ulysses and the Sirens: Studies in Rationalityand Irrationality (CambridgeUniv. Press, Lon-don, 1979);R.Thalerand H. M.Shifrin, . PolitItcon., ln press.

23. P. Slovic and A. Tversky,Behav. Sci. 19, 368(1974).

24. A. Tversky and D. Kahneman,Science 18S,1124(1974); P. Slovic, B. Fischhoff,S. Lich-tenstein,Annu. Rev. Psychol. ''28,1 (1977);R.Nisbettand L. lAoss,Human Inference: Strate-gies and Shortcomings of Social Judgment(Prentice-Hall,Englewood Cliffs, N.J., 1980);H. Einhornand R'.Hogarth,Annu. Rev. Psy-chol. 32, 53 (1981).

25. H. A. Simon,Q.J. Econ. 69, 99 (1955); sychol.Rev. 63, 129 (1956).

26. J. March, Bell J. Econ. 9, 587 (1978).27. This work was supported by the Office of Naval

Research under contract NQ0014-79-C-0077 toStanford University;

SCIENCE,VOL. 21158