Quick viewing(Text Mode)

The Framing of Decisions and the Psychology of Choice

The Framing of Decisions and the Psychology of Choice

The majority choice in this problem is averse: the prospect of certainly saving 200 lives is more attractive than a risky prospect of equal expected value, that is, a one-in-three chance of saving 600 lives. The Framing of Decisions and the A second group of respondents was given the cover story ofproblem 1 with a different formulation of the alternative of Choice programs, as follows: and Problem 2 [N = 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- Explanations and predictions of tional choice requires that the ty that 600 people will die. [78 percent] people's choices, in everyday life as well between options should not reverse with Which ofthe two programs would you favor? as in the social sciences, are often found- changes of frame. Because of imperfec- ed on the assumption of rational- tions of human and decision, The majority choice in problem 2 is ity. The definition of has been however, changes of perspective often risk taking: the certain death of 400 much debated, but there is general agree- reverse the relative apparent size of ob- people is less acceptable than the two-in- ment that rational choices should satisfy jects and the relative desirability of op- three chance that 600 will die. The pref- some elementary requirements of con- tions. erences in problems 1 and 2 illustrate a sistency and coherence. In this article We have obtained systematic rever- common pattern: choices involving gains are often risk averse and choices in- volving losses are often risk taking. Summary. The psychological principles that govern the perception of decision prob- However, it is easy to see that the two lems and the evaluation of probabilities and outcomes produce predictable shifts of problems are effectively identical. The preference when the same problem is framed in different ways. Reversals of prefer- only difference between them is that the ence are demonstrated in choices regarding monetary outcomes, both hypothetical outcomes are described in problem 1 by on May 27, 2010 and real, and in questions pertaining to the loss of human lives. The effects of frames the number of lives saved and in problem on are compared to the effects of perspectives on perceptual appear- 2 by the number of lives lost. The change ance. The dependence of preferences on the formulation of decision problems is a is accompanied by a pronounced shift significant concern for the theory of rational choice. from to risk taking. We have observed this reversal in several groups of respondents, including univer- we describe decision problems in which sals of preference by variations in the sity faculty and physicians. Inconsistent

people systematically violate the re- framing of acts, contingencies, or out- responses to problems 1 and 2 arise from www.sciencemag.org quirements of consistency and coher- comes. These effects have been ob- the conjunction of a framing effect with ence, and we trace these violations to the served in a variety of problems and in contradictory attitudes toward in- psychological principles that govern the the choices of different groups of respon- volving gains and losses. We turn now perception of decision problems and the dents. Here we present selected illustra- to an analysis of these attitudes. evaluation of options. tions of preference reversals, with data A decision problem is defined by the obtained from students at Stanford Uni- acts or options among which one must versity and at the University of British The Evaluation of Prospects choose, the possible outcomes or con- Columbia who answered brief question- Downloaded from sequences of these acts, and the contin- naires in a classroom setting. The total The major theory of decision-making gencies or conditional probabilities that number of respondents for each problem under risk is the expected model. relate outcomes to acts. We use the term is denoted by N, and the percentage This model is based on a set of axioms, "'decision frame" to refer to the deci- who chose each option is indicated in for example, transitivity of preferences, sion-maker's conception of the acts, out- brackets. which provide criteria for the rationality comes, and contingencies associated The effect of variations in framing is of choices. The choices of an individual with a particular choice. The frame that a illustrated in problems 1 and 2. who conforms to the axioms can be de- decision-maker adopts is controlled part- scribed in terms of the of various Problem 1 [N = 152]: Imagine that the U.S. outcomes for that individual. The ly by the formulation of the problem and is preparing for the outbreak of an unusual utility partly by the norms, habits, and personal Asian disease, which is expected to kill 600 of a risky prospect is equal to the ex- characteristics of the decision-maker. people. Two alternative programs to combat pected utility of its outcomes, obtained It is often possible to frame a given de- the disease have been proposed. Assume that by weighting the utility of each possible cision problem in more than one way. the exact scientific estimate of the con- by its probability. When faced Alternative frames for a decision prob- sequences of the programs are as follows: with a choice, a rational decision-maker If Program A is adopted, 200 people will be will the prospect that offers the lem may be compared to alternative per- saved. [72 percent] prefer spectives on a visual scene. Veridical highest expected utility (1, 2). that the If Program B is adopted, there is 1/3 probabil- perception requires *perceived ity that 600 people will be saved, and Dr. Tversky is a professor of psychology at Stan- relative height of two neighboring moun- 2/3 probability that no people will be ford University, Stanford, 94305, and Dr. Kahneman is a professor of psychology at the Uni- tains, say, should not reverse with saved. [28 percent] versity of British Columbia, Vancouver, Canada changes of vantage point. Similarly, ra- Which ofthe two programs would you favor? V6T IW5. SCIENCE, VOL. 211, 30 JANUARY 1981 0036-8075/81/0130-0453$01.50/0 Copyright © 1981 AAAS 453 As will be illustrated below, people ex- Value The Framing of Acts hibit patterns of preference which appear incompatible with expected utility theo- Problem 3 [N = 150]: Imagine that you face the following pair of concurrent decisions. ry. We have presented elsewhere (3) a First examine both decisions, then indicate descriptive model, called prospect theo- the options you prefer. ry, which modifies expected utility theo- Losses Decision (i). Choose between: ry so as to accommodate these observa- A. a sure gain of $240 [84 percent] tions. We distinguish two phases in the B. 25% chance to gain $1000, and choice process: an initial phase in which 75% chance to gain nothing [16 percent] acts, outcomes, and contingencies are Decision (ii). Choose between: C. a sure loss of $750 [13 percent] framed, and a subsequent phase of eval- D. 75% chance to lose $1000, and uation (4). For simplicity, we restrict the 25% chance to lose nothing [87 percent] formal treatment of the theory to choices Fig. 1. A hypothetical value function. involving stated numerical probabilities The majority choice in decision (i) is and quantitative outcomes, such as mon- has the following properties. First, im- risk averse: a riskless prospect is pre- ey, time, or number of lives. possible events are discarded, that is, ferred to a risky prospect of equal or Consider a prospect that yields out- *0) = 0, and the scale is normalized so greater expected value. In contrast, the come x with probability p, outcome y that r( 1) = 1, but the function is not well majority choice in decision (ii) is risk tak- with probability q, and the status quo behaved near the endpoints. Second, ing: a risky prospect is preferred to a with probability 1 - p - q. According for low probabilities 7r(p) > p, but riskless prospect of equal expected val- to , there are values v(.) r(p) + 7r(1 - p) < 1. Thus low proba- ue. This pattern of risk aversion in associated with outcomes, and decision bilities are overweighted, moderate and choices involving gains and risk seeking weights 7r(.) associated with probabili- high probabilities are underweighted, in choices involving losses is attributable ties, such that the overall value of the and the latter effect is more pronounced to the properties of v and 7r. Because the prospect equals 7r(p) v(x) + ir(q) v(y). A than the former. Third, 7r(pq)/ir(p) < value function is S-shaped, the value as- slightly different equation should be ap- 7r(pqr)/7r(pr) for all 0 < p, q, r ' 1. That sociated with a gain of $240 is greater plied if all outcomes of a prospect are on is, for any fixed probability ratio q, the than 24 percent of the value associated the same side of the zero point (5). ratio of decision weights is closer to with a gain of $1000, and the (negative)

In prospect theory, outcomes are ex- unity when the probabilities are low value associated with a loss of $750 is on May 27, 2010 pressed as positive or negative devia- than when they are high, for example, smaller than 75 percent ofthe value asso- tions (gains or losses) from a neutral ref- ir(.1)/7r(.2) > 7r(.4)/7r(.8). A hypothetical ciated with a loss of $1000. Thus the erence outcome, which is assigned a val- weighting function which satisfies these shape of the value function contributes ue of zero. Although subjective values properties is shown in Fig. 2. The major to risk aversion in decision (i) and to risk differ among individuals and attributes, qualitative properties of decision weights seeking in decision (ii). Moreover, the we propose that the value function is can be extended to cases in which the underweighting of moderate and high commonly S-shaped, concave above the probabilities of outcomes are subjective- probabilities contributes to the relative reference point and convex below it, as ly assessed rather than explicitly given. attractiveness of the sure gain in (i) and illustrated in Fig. 1. For example, the dif- In these situations, however, decision to the relative aversiveness of the sure www.sciencemag.org ference in subjective value between weights may also be affected by other loss in (ii). The same analysis applies to gains of $10 and $20 is greater than the characteristics of an event, such as am- problems 1 and 2. subjective difference between gains of biguity or vagueness (9). Because (i) and (ii) were presented to- $110 and $120. The same relation be- Prospect theory, and the scales illus- gether, the respondents had in effect to tween value differences holds for the trated in Figs. 1 and 2, should be viewed choose one prospect from the set: A and corresponding losses. Another property as an approximate, incomplete, and sim- C,BandC, AandD, BandD. The most of the value function is that the response plified description of the evaluation of common pattern (A and D) was chosen Downloaded from to losses is more extreme than the re- risky prospects. Although the properties by 73 percent of respondents, while the sponse to gains. The displeasure associ- of v and Xr summarize a common pattern least popular pattern (B and C) was ated with losing a sum of money is gener- of choice, they are not universal: the chosen by only 3 percent of respondents. ally greater than the associated preferences of some individuals are not However, the combination of B and with winning the same amount, as is re- well described by an S-shaped value C is definitely superior to the combina- flected in people's reluctance to accept function and a consistent set of decision tion A and D, as is readily seen in prob- fair bets on a toss of a coin. Several stud- weights. The simultaneous measurement lem 4. ies of decision (3, 6) and judgment (7) of values and decision weights involves Problem 4 [N = 86]. Choose between: have confirmed these properties of the serious experimental and statistical diffi- A & D. 25% chance to win $240, and value function (8). culties (10). 75.% chance to lose $760. [O per- The second major departure of pros- If Xr and v were linear throughout, the cent] pect theory from the expected utility preference order between options would B & C. 25% chance to win $250, and model involves the treatment of proba- be independent of the framing of acts, 75% chance to lose $750. [100 per- cent] bilities. In expected utility theory the outcomes, or contingencies. Because of utility of an uncertain outcome is weight- the characteristic nonlinearities of ir and When the prospects were combined ed by its probability; in prospect theory v, however, different frames can lead to and the dominance of the second option the value of an uncertain outcome is mul- different choices. The following three became obvious, all respondents chose tiplied by a decision weight 7r(p), which sections describe reversals of preference the superior option. The popularity of is a monotonic function of p but is not a caused by variations in the framing of the inferior option in problem 3 implies probability. The weighting function ir acts, contingencies, and outcomes. that this problem was framed as a pair of 454 SCIENCE, VOL. 211 separate choices. The respondents ap- 1.0 The first stage of problem 6 yields t. parently failed to entertain the possibility same outcome (no gain) for both act that the conjunction of two seemingly Consequently, we propose, people eval reasonable choices could lead to an un- uate the options conditionally, as if the tenable result. second stage had been reached. In this The violations of dominance observed 005 framing, of course, problem 6 reduces to in problem 3 do not disappear in the problem 5. More generally, we suggest 0 presence of monetary incentives. A dif- that a decision problem is evaluated con- CD ferent group of respondents who an- ditionally when (i) there is a state in swered a modified version of problem 3, which all acts yield the same outcome, with real payoffs, produced a similar pat- such as failing to reach the second stage tern of choices (11). Other authors have 0 0.5 1.0 of the game in problem 6, and (ii) the Stated p also reported that violations of the rules probability: stated probabilities of other outcomes of rational choice, originally observed in Fig. 2. A hypothetical weighting function. are conditional on the nonoccurrence of hypothetical questions, were not elimi- this state. nated by payoffs (12). The striking discrepancy between the We suspect that many concurrent de- Let us examine the structure of these responses to problems 6 and 7, which are cisions in the real world are framed inde- problems. First, note that problems 6 identical in outcomes and probabilities, pendently, and that the preference order and 7 are identical in terms of probabili- could be described as a pseudocertainty would often be reversed if the decisions ties and outcomes, because prospect C effect. The prospect yielding $30 is rela- were combined. The respondents in offers a .25 chance to win $30 and pros- tively more attractive in problem 6 than problem 3 failed to combine options, al- pect D offers a probability of .25 x in problem 7, as if it had the advantage of though the integration was relatively .80 = .20 to win $45. Consistency there- certainty. The sense of certainty associ- simple and was encouraged by instruc- fore requires that the same choice be ated with option C is illusory, however, tions (13). The complexity of practical made in problems 6 and 7. Second, note since the gain is in fact contingent on problems of concurrent decisions, such that problem 6 differs from problem 5 on- reaching the second stage of the game as portfolio selection, would prevent ly by the introduction of a preliminary (15). people from integrating options without stage. If the second stage of the game is We have observed the certainty effect computational aids, even if they were in- reached, then problem 6 reduces to prob- in several sets of problems, with out- on May 27, 2010 clined to do so. lem 5; if the game ends at the first stage, comes ranging from vacation trips to the the decision does not affect the outcome. loss of human lives. In the negative do- Hence there seems to be no reason to main, certainty exaggerates the aversive- The Framing of Contingencies make a different choice in problems 5 ness of losses that are certain relative to and 6. By this logical analysis, problem 6 losses that are merely probable. In a The following triple of problems illus- is equivalent to problem 7 on the one question dealing with the response to an trates the framing of contingencies. Each hand and problem 5 on the other. The epidemic, for example, most respond- problem was presented to a different participants, however, responded simi- ents found "a sure loss of75 lives" more group of respondents. Each group was larly to problems 5 and 6 but differently aversive than "80%o chance to lose 100 www.sciencemag.org told that one participant in ten, pre- to problem 7. This pattern of responses lives" but preferred "10%o chance to lose selected at random, would actually be exhibits two phenomena of choice: the 75 lives" over "8% chance to lose 100 playing for money. Chance events were certainty effect and the pseudocertainty lives," contrary to expected utility theo- realized, in the respondents' presence, effect. ry. by drawing a single ball from a bag con- The contrast between problems 5 and We also obtained the pseudocertainty taining a known proportion ofballs ofthe 7 illustrates a phenomenon discovered effect in several studies where the de-

winning color, and the winners were paid by Allais (14), which we have labeled the scription of the decision problems fa- Downloaded from immediately. certainty effect: a reduction of the proba- vored conditional evaluation. Pseudo- bility of an outcome by a constant factor certainty can be induced either by a se- Problem 5 [N = 77]: Which ofthe following has more impact when the outcome was quential formulation, as in problem 6, or options do you prefer? initially certain than when it was merely by the introduction of causal contin- A. a sure win of $30 [78 percent] B. 80%o chance to win $45 [22 percent] probable. Prospect theory attributes this gencies. In another version of the epi- effect to the properties of ir. It is easy to demic problem, for instance, respond- Problem 6 [N = 85]: Consider the following verify, by applying the equation of pros- ents were told that risk to life existed on- two-stage game. In the first stage, there is a 75% chance to end the game without winning pect theory to problems 5 and 7, that ly in the event (probability .10) that the anything, and a 25% chance to move into the people for whom the value ratio v(30)/ disease was carried by a particular virus. second stage. If you reach the second stage v(45) lies between the weight ratios Two alternative programs were said to you have a choice between: 7r(.20)/ir(.25) and ir(.80)/ir(1.0) will pre- yield "a sure loss of 75 lives" or "80% C. a sure win of $30 [74 percent] fer A to B and F to E, contrary to ex- chance to lose 100 lives" if the critical D. 8%o chance to win $45 [26 percent] pected utility theory. Prospect theory virus was involved, and no loss of life in Your choice must be made before the game starts, i.e., before the outcome of the first does not predict a reversal of preference the event (probability .90) that the dis- stage is known. Please indicate the option you for every individual in problems 5 and ease was carried by another virus. In ef- prefer. 7. It only requires that an individual who fect, the respondents were asked to has no between A and B pre- choose between 10 chance of Problem 7 [N = 81]: Which of the following preference percent options do you prefer? fer F to E. For group data, the theory losing 75 lives and 8 percent chance of E. 25% chance to win $30 [42 percent] predicts the observed directional shift losing 100 lives, but their preferences

F. 20%o chance to win $45 [58 percent] ofpreference between the two problems . were the same as when the choice was 30 JANUARY 1981 455 -ween a sure loss of 75 lives and 80 provide is an illusion of conditional fram- ing of credit-card purchases, representa- rcent chance of losing 100 lives. A ing. It appears that insurance is bought tives of the credit-card industry re- inditional framing was evidently as protection against worry, not only quested that the price difference be la- .-dopted in which the contingency of the against risk, and that worry can be ma- beled a cash discount rather than a noncritical virus was eliminated, giving nipulated by the labeling of outcomes credit-card surcharge. The two labels in- rise to a pseudocertainty effect. The cer- and by the framing of contingencies. It is duce different reference points by implic- tainty effect reveals attitudes toward risk not easy to determine whether people itly designating as normal reference the that are inconsistent with the axioms of value the elimination of risk too much or higher or the lower of the two prices. Be- rational choice, whereas the pseudo- the reduction of risk too little. The con- cause losses loom larger than gains, con- certainty effect violates the more funda- trasting attitudes to the two forms ofpro- sumers are less willing to accept a sur- mental requirement that preferences tective action, however, are difficult to charge than to forego a discount. A simi- should be independent of problem de- justify on normative grounds (16). lar effect has been observed in scription. experimental studies of insurance: the Many significant decisions concern ac- proportion of respondents who preferred tions that reduce or eliminate the proba- The Framing of Outcomes a sure loss to a larger probable loss was bility of a hazard, at some cost. The significantly greater when the former shape of Xr in the range of low probabili- Outcomes are commonly perceived as was called an insurance premium (19, ties suggests that a protective action positive or negative in relation to a refer- 20). which reduces the probability of a harm ence outcome that is judged neutral. These observations highlight the labil- from 1 percent to zero, say, will be val- Variations of the reference point can ity of reference outcomes, as well as ued more highly than an action that re- therefore determine whether a given out- their role in decision-making. In the ex- duces the probability of the same harm come is evaluated as a gain or as a loss. amples discussed so far, the neutral ref- from 2 percent to 1 percent. Indeed, Because the value function is generally erence point was identified by the label- probabilistic insurance, which reduces concave for gains, convex for losses, and ing of outcomes. A diversity of factors the probability of loss by half, is judged steeper for losses than for gains, shifts of determine the reference outcome in to be worth less than half the price of reference can change the value dif- everyday life. The reference outcome is regular insurance that eliminates the risk ference between outcomes and thereby usually a state to which one has adapted; altogether (3). reverse the preference order between it is sometimes set by social norms and It is often possible to frame protective options (6). Problems 1 and 2 illustrated expectations; it sometimes corresponds action in either conditional or uncon- a preference reversal induced by a shift to a level of aspiration, which may or on May 27, 2010 ditional form. For example, an insurance of reference that transformed gains into may not be realistic. policy that covers fire but not flood could losses. We have dealt so far with elementary be evaluated either as full protection For another example, consider a per- outcomes, such as gains or losses in a against the specific risk of fire or as a re- son who has spent an afternoon at the single attribute. In many situations, how- duction in the overall probability of race track, has already lost $140, and is ever, an action gives rise to a compound property loss. The preceding analysis considering a $10 bet on a 15:1 long shot outcome, which joins a series of changes suggests that insurance should appear in the last race. This decision can be in a single attribute, such as a sequence

more attractive when it is presented as framed in two ways, which correspond of monetary gains and losses, or a set of www.sciencemag.org the elimination of risk than when it is de- to two natural reference points. If the concurrent changes in several attributes. scribed as a reduction of risk. P. Slovic, status quo is the reference point, the out- To describe the framing and evaluation B. Fischhoff, and S. Lichtenstein, in an comes of the bet are framed as a gain of of compound outcomes, we use the no- unpublished study, found that a hypo- $140 and a loss of $10. On the other tion of a psychological account, defined thetical vaccine which reduces the prob- hand, it may be more natural to view the as an outcome frame which specifies (i) ability of contracting a disease from .20 present state as a loss of $140, for the the set of elementary outcomes that are to .10 is less attractive if it is described as betting day, and accordingly frame the evaluated jointly and the manner in effective in half the cases than if it is pre- last bet as a chance to return to the refer- which they are combined and (ii) a refer- Downloaded from sented as fully effective against one of ence point or to increase the loss to $150. ence outcome that is considered neutral two (exclusive and equiprobable) virus Prospect theory implies that the latter or normal. In the account that is set up strains that produce identical symptoms. frame will produce more risk seeking for the purchase of a car, for example, In accord with the present analysis of than the former. Hence, people who do the cost of the purchase is not treated as pseudocertainty, the respondents valued not adjust their reference point as they a loss nor is the car viewed as a gift. full protection against an identified vi- lose are expected to take bets that they Rather, the transaction as a whole is rus more than probabilistic protection would normally find unacceptable. This evaluated as positive, negative, or neu- against the disease. analysis is supported by the observation tral, depending on such factors as the The preceding discussion highlights that bets on long shots are most popular performance of the car and the price of the sharp contrast between lay responses on the last race of the day (17). similar cars in the . A closely re- to the reduction and the elimination of Because the value function is steeper lated treatment has been offered by Tha- risk. Because no form of protective ac- for losses than for gains, a difference be- ler (18). tion can cover all risks to human welfare, tween options will loom larger when it is We propose that people generally all insurance is essentially probabilistic: framed as a disadvantage of one option evaluate acts in terms of a minimal ac- it reduces but does not eliminate risk. rather than as an advantage of the other count, which includes only the direct The probabilistic nature of insurance is option. An interesting example of such consequences of the act. The minimal commonly masked by formulations that an effect in a riskless context has been account associated with the decision to emphasize the completeness of pro- noted by Thaler (18). In a debate on a accept a gamble, for example, includes tection against identified harms, but the proposal to pass to the consumer some the money won or lost in that gamble and sense of security that such formulations of the costs associated with the process- excludes other assets or the outcome of 456 SCIENCE, VOL. 211 previous gambles. People commonly other group (N = 88) the values shown consistencies were traced to the inter- adopt minimal accounts because this in brackets. action of two sets of factors: variations mode of framing (i) simplifies evaluation in the framing of acts, contingencies, and Problem 10: Imagine that you are about to and reduces cognitive strain, (ii) reflects purchase ajacket for ($125) [$15], and a calcu- outcomes, and the characteristic non- the that consequences should lator for ($15) [$125]. The calculator salesman linearities of values and decision be causally linked to acts, and (iii) informs you that the calculator you wish to weights. The demonstrated effects are matches the properties of hedonic expe- buy is on sale for ($10) [$1201 at the other large and systematic, although by no rience, which is more sensitive to desir- branch of the store, located 20 minutes drive means universal. They occur when the away. Would you make the trip to the other able and undesirable changes than to store? outcomes concern the loss of human steady states. lives as well as in choices about money; There are situations, however, in The response to the two versions of they are not restricted to hypothetical which the outcomes of an act affect the problem 10 were markedly different: 68 questions and are not eliminated by mon- balance in an account that was pre- percent of the respondents were willing etary incentives. viously set up by a related act. In these to make an extra trip to save $5 on a $15 Earlier we compared the dependence cases, the decision at hand may be eval- calculator; only 29 percent were willing of preferences on frames to the depen- uated in terms of a more inclusive ac- to exert the same effort when the price of dence of perceptual appearance on per- count, as in the case of the bettor who the calculator was $125. Evidently the spective. If while traveling in a mountain views the last race in the context of ear- respondents do not frame problem 10 in range you notice that the apparent rela- lier losses. More generally, a sunk-cost the minimal account, which involves on- tive height of mountain peaks varies with effect arises when a decision is referred ly a benefit of $5 and a cost of some in- your vantage point, you will conclude to an existing account in which the cur- convenience. Instead, they evaluate the that some impressions of relative height rent balance is negative. Because of the potential saving in a more inclusive ac- must be erroneous, even when you have nonlinearities of the evaluation process, count, which includes the purchase of no access to the correct answer. Similar- the minimal account and a more in- the calculator but not of the jacket. By ly, one may discover that the relative at- clusive one often lead to different the curvature of v, a discount of $5 has a tractiveness of options varies when the choices. greater impact when the price of the cal- same decision problem is framed in dif- Problems 8 and 9 illustrate another culator is low than when it is high. ferent ways. Such a discovery will nor- class of situations in which an existing A closely related observation has been mally lead the decision-maker to recon- account affects a decision: reported by Pratt, Wise, and Zeckhauser sider the original preferences, even when (21), who found that the variability of the there is no simple way to resolve the in- on May 27, 2010 Problem 8 [N = 183]: Imagine that you have decided to see a play where admission is prices at which a given product is sold by consistency. The susceptibility to per- $10 per ticket. As you enter the theater you different stores is roughly proportional to spective effects is of special concern in discover that you have lost a $10 bill. the mean price of that product. The same the domain of decision-making because Would you still pay $10 for a ticket for the pattern was observed for both frequently of the absence of objective standards play? and infrequently purchased items. Over- such as the true height of mountains. Yes [88 percent] No [12 percent] all, a ratio of 2: 1 in the mean price of two The metaphor of changing perspective Problem 9 [N = 200]: Imagine that you products is associated with a ratio of can be applied to other phenomena of have decided to see a play and paid the admis- 1.86:1 in the standard deviation of the choice, in addition to the framing effects www.sciencemag.org sion price of $10 per ticket. As you enter the respective quoted prices. If the effort with which we have been concerned here theater you discover that you have lost the that consumers exert to save each dollar (19). The problem of self-control is natu- ticket. The seat was not marked and the ticket cannot be recovered. on a purchase, for instance by a phone rally construed in these terms. The story Would you pay $10 for another ticket? call, were independent of price, the dis- of Ulysses' request to be bound to the Yes [46 percent] No [54 percent] persion of quoted prices should be about mast of the ship in anticipation of the ir- the same for all products. In contrast, resistible temptation of the Sirens' call is The marked difference between the re- the data of Pratt et al. (21) are consistent often used as a paradigm case (22). In sponses to problems 8 and 9 is an effect with the hypothesis that consumers this example of precommitment, an ac- Downloaded from of psychological accounting. We pro- hardly exert more effort to save $15 on a tion taken in the present renders inopera- pose that the purchase of a new ticket in $150 purchase than to save $5 on a $50 tive an anticipated future preference. An problem 9 is entered in the account that purchase (18). Many readers will recog- unusual feature of the problem of inter- was set up by the purchase ofthe original nize the temporary devaluation of money temporal conflict is that the agent who ticket. In terms of this account, the ex- which facilitates extra spending and re- views a problem from a particular tem- pense required to see the show is $20, a duces the significance of small discounts poral perspective is also aware of the cost which many of our respondents ap- in the context of a large expenditure, confficting views that future perspectives parently found excessive. In problem 8, such as buying a house or a car. This will offer. In most other situations, deci- on the other hand, the loss of $10 is not paradoxical variation in the value of sion-makers are not normally aware of linked specifically to the ticket purchase money is incompatible with the standard the potential effects of different decision and its effect on the decision is accord- analysis of consumer behavior. frames on their preferences. ingly slight. The perspective metaphor highlights The following problem, based on ex- the following aspects of the psychology amples by Savage (2, p. 103) and Thaler Discussion of choice. Individuals who face a deci- (18), further illustrates the effect of em- sion problem and have a definite prefer- bedding an option in different accounts. In this article we have presented a se- ence (i) might have a different preference Two versions of this problem were pre- ries of demonstrations in which seem- in a different framing of the same prob- sented to different groups of subjects. ingly inconsequential changes in the for- lem, (ii) are normally unaware of alterna- One group (N = 93) was given the val- mulation of choice problems caused sig- tive frames and of their potential effects ues that appear in parentheses, and the nificant shifts of preference. The in- on the relative attractiveness of options, 30 JANUARY 1981 457 (iii) would wish their preferences to be dictive orientation encourages the deci- 9. D. Ellsberg, Q. J. Econ. 75, 643 (1961); W. Fell- ner, Probability and Profit-A Study of Eco- independent of frame, but (iv) are often sion-maker to focus on future experience nomic Behavior Along Bayesian Lines (Irwin, Homewood, III., 1965). uncertain how to resolve detected incon- and to ask "What will I feel then?" 10. The scaling of v and ir by pair comparisons re- sistencies (23). In some cases (such as rather than "What do I want now?" The quires a large number of observations. The pro- cedure ofpricing gambles is more convenient for problems 3 and 4 and perhaps problems 8 former question, when answered with scaling purposes, but it is subject to a severe an- and 9) the advantage of one frame be- care, can be the more useful guide in dif- choring : the ordering of gambles by their cash equivalents diverges systematically from comes evident once the competing ficult decisions. In particular, predictive the preference order observed in direct com- frames are compared, but in other cases considerations may be applied to select parisons [S. Lichtenstein and P. Slovic, J. Exp. Psychol. 89, 46 (1971)]. (problems 1 and 2 and problems 6 and 7) the decision frame that best represents 11. A new group of respondents (N = 126) was pre- sented with a modified version of problem 3, in it is not obvious which preferences the hedonic expenence of outcomes. which the outcomes were reduced by a factor should be abandoned. Further complexities arise in the nor- of 50. The participants were informed that the gambles would actually be played by tossing a These observations do not imply that mative analysis because the framing of pair of fair coins, that one participant in ten preference reversals, or other errors of an action sometimes affects the actual would be selected at random to play the gambles of his or her choice. To ensure a positive return choice or judgment (24), are necessarily experience of its outcomes. For ex- for the entire set, a third decision, yielding only irrational. Like other intellectual limita- ample, framing outcomes in terms of positive outcomes, was added. These payoff conditions did not alter the pattern of prefer- tions, discussed by Simon (25) under the overall wealth or welfare rather than in ences observed in the hypothetical problem: 67 percent of respondents chose prospect A and 86 heading of "," the terms of specific gains and losses may at- percent chose prospect D. The dominated com- practice of acting on the most readily tenuate one's emotional response to an bination of A and D was chosen by 60 percent of respondents, and only 6 percent favored the available frame can sometimes be justi- occasional loss. Similarly, the experi- dominant combination of B and C. fied by reference to the mental effort re- ence of a change for the worse may vary 12. S. Lichtenstein and P. Slovic, J. Exp. Psychol. 101, 16 (1973); D. M. Grether and C. R. Plott, quired to explore alternative frames and if the change is framed as an uncompen- Am. Econ. Rev. 69, 623 (1979); I. Lieblich and avoid potential inconsistencies. How- sated loss or as a cost incurred to A. Lieblich, Percept. Mot. Skills 29, 467 (1969); D. M. Grether, Working Paper ever, we propose that the details of the achieve some benefit. The framing of No. 245 (California Institute of Technology, Pasadena, 1979). phenomena described in this article are acts and outcomes can also reflect the 13. Other demonstrations of a reluctance to in- better explained by prospect theory and acceptance or rejection of responsibility tegrate concurrent options have been reported: P. Slovic and S. Lichtenstein, J. Exp. Psychol. by an analysis of framing than by ad for particular consequences, and the de- 78, 646 (1968); J. W. Payne and M. L. Braun- hoc appeals to the notion of cost of liberate manipulation of framing is com- stein, ibid. 87, 13 (1971). 14. M. Allais, 21, 503 (1953); K. thinking. monly used as an instrument of self- McCrimmon and S. Larsson, in Expected Util- The present work has been concerned control (22). When framing influences ity Hypotheses and the , M. All- ais and 0. Hagan, Eds. (Reidel, Dordrecht, on May 27, 2010 primarily with the descriptive question the experience of consequences, the 1979). how are but the of a an ethi- 15. Another group of respondents (N = 205) was of decisions made, psy- adoption decision frame is presented with all three problems, in different chology of choice is also relevant to the cally significant act. orders, without monetary payoffs. The joint fre- quency distribution of choices in problems 5, 6, normative question of how decisions and 7 was as follows: ACE, 22; ACF, 65; ADE, ought to be made. In order to avoid the References and Notes 4; ADF, 20; BCE, 7; BCF, 18; BDE, 17; BDF, 1. J. Von Neumann and 0. Morgenstern, Theory of 52. These data confirm in a within-subject design difficult problem ofjustifying values, the Games and Economic Behavior (Princeton the analysis of conditional evaluation proposed modern theory of rational choice has Univ. Press, Princeton, N.J., 1947); H. Raiffa, in the text. More than 75 percent of respondents Decision Analysis: Lectures on Choices Under made compatible choices (AC or BD) in prob- adopted the coherence of specific prefer- Uncertainty (Addison-Wesley, Reading, Mass., lems 5 and 6, and less than half made compatible 1968); P. Utility Theory for Decision choices in problems 6 and 7 (CE or DF) or 5 and ences as the sole criterion of rationality. Fishburn, www.sciencemag.org Making (Wiley, New York, 1970). 7 (AE or BF). The elimination ofpayoffs in these This approach enjoins the decision- 2. L. J. Savage, The Foundations of Statistics questions reduced risk aversion but did not sub- (Wiley, New York, 1954). stantially alter the effects of certainty and maker to resolve inconsistencies but of- 3. D. Kahneman and A. Tversky, Econometrica pseudocertainty. fers no guidance on how to do so. It im- 47, 263 (1979). 16. For further discussion of rationality in pro- 4. The framing phase includes various editing oper- tective action see H. Kunreuther, Disaster In- plicitly assumes that the decision-maker ations that are applied to simplify prospects, for surance Protection: Public Policy Lessons who carefully answers the question example by combining events or outcomes or by (Wiley, New York, 1978). discarding negligible components (3). 17. W. H. McGlothlin, Am. J. Psychol. 69, 604 "What do I really want?" will eventually 5. Ufp + q = I and eitherx > y > Oorx < y < 0, (1956). achieve coherent preferences. However, the equation in the text is replaced by 18. R. Thaler,J. Econ. Behav. Organ. 1, 39 (1980). v(y) + r*) [v(x) - v(y)], so that decision 19. B. Fischhoff, P. Slovic, S. Lichtenstein, in Cog- Downloaded from the susceptibility of preferences to varia- weights are not applied to sure outcomes. nitive Processes in Choice and Decision Behav- about the 6. P. Fishburn and G. Kochenberger, Decision Sci. ior, T. Wallsten, Ed. (Erlbaum, Hillsdale, N.J., tions of framing raises doubt 10, 503 (1979); D. J. Laughhunn, J. W. Payne, 1980). feasibility and adequacy of the coher- R. Crum, Manage. Sci., in press; J. W. Payne, 20. J. C. Hershey and P. J. H. Schoemaker, J. Risk D. J. Laughhunn, R. Crum, ibid., in press; S. A. Insur., in press. ence criterion. Eraker and H. C. Sox, Med. Decision Making, 21. J. Pratt, A. Wise, R. Zeckhauser, Q. J. Econ. Consistency is only one aspect of the in press. In the last study several hundred clinic 93, 189 (1979). patients made hypothetical choices between 22. R. H. Strotz, Rev. Econ. Stud. 23, 165 (1955); G. lay notion of rational behavior. As noted drug therapies for severe headaches, hyperten- Ainslie, Psychol. Bull. 82, 463 (1975); J. Elster, by March (26), the common conception sion, and chest pain. Most patients were risk Ulysses and the Sirens: Studies in Rationality averse when the outcomes were described as and (Cambridge Univ. Press, Lon- of rationality also requires that prefer- positive (for example, reduced pain or increased don, 1979); R. Thaler and H. M. Shifrin,J. Polit. ences or utilities for outcomes life expectancy) and risk taking when the out- Econ., in press. particular comes were described as negative (increased 23. P. Slovic and A. Tversky, Behav. Sci. 19, 368 should be predictive of the experiences pain or reduced life expectancy). No significant (1974). differences were found between patients who 24. A. Tversky and D. Kahneman, Science 185, of satisfaction or displeasure associated actually suffered from the ailments described 1124 (1974); P. Slovic, B. Fischhoff, S. Lich- with their occurrence. Thus, a man could and patients who did not. tenstein, Annu. Rev. P-sychol. 28, 1 (1977); R. 7. E. Galanter and P. Pliner, in Sensation and Nisbett and L. Ross, Human Itference: Strate- be judged irrational either because his Measurement, H. R. Moskowitz et al., Eds. gies and Shortcomings of Social Judgment preferences are contradictory or because (Reidel, Dordrecht, 1974), pp. 65-76. (Prentice-tIall, Englewood Cliffs, N.J., 1980); 8. The extension of the proposed value function to H. Einhorn and R. Hogarth, Annu. Rev. Psy- his desires and aversions do not reflect multiattribute options, with or without risk, de- chol. 32, 53 (1981). his and pains. The predictive serves careful analysis. In particular, indif- 25. H. A. Simon, Q. J. Econ. 69, 99 (1955);Psychol. ference curves between dimensions of loss may Rev. 63, 129 (1956). criterion of rationality can be applied to be concave upward, even when the value finc- 26. J. March, Bell J. Econ. 9, 587 (1978). tions for the separate losses are both convex, 27. This work was supported by the Office of Naval resolve inconsistent preferences and to because of marked subadditivity between di- Research under contract N00014-79-C-0077 to improve the quality of decisions. A pre- mensions. .

458 SCIENCE, VOL. 211

Top View