Framing Effects: Dynamics and Task Domains

ORGANIZATIONAL BEHAVIOR AND HUMAN DECISION PROCESSES Vol. 68, No. 2, November, pp. 145–157, 1996 ARTICLE NO. 0095 Framing Effects: Dynamics and Task Domains X. T. WANG University of South Dakota able models of human decision making. q 1996 Academic The author examines the mechanisms and dynamics Press, Inc. of framing effects in risky choices across three distinct task domains (i.e., life–death, public property, and personal money). The choice outcomes of the problems INTRODUCTION presented in each of the three task domains had a bi- nary structure of a sure thing vs a gamble of equal Since the seminal, pioneering studies by Kahneman expected value; the outcomes differed in their framing and Tversky (e.g., Kahneman & Tversky, 1979; Tver- conditions and the expected values, raging from 6000, sky & Kahneman, 1981), framing effects have for both 600, 60, to 6, numerically. It was hypothesized that sub- theoretical and practical reasons received much re- jects would become more risk seeking, if the sure out- search attention from cognitive psychologists, decision come was below their aspiration level (the minimum scientists, and economists. Framing effects often refer requirement). As predicted, more subjects preferred the gamble when facing the life–death choice prob- to the changes in risk preferences as a result of how lems than facing the counterpart problems presented choices are described, or framed. Over the years, fram- in the other two task domains. Subjects’ risk prefer- ing effects have been extended to a wide variety of tasks ence varied categorically along the group size dimen- and procedures (e.g., Bazerman, Magliozzi, & Neale, sion in the life–death domain but changed more lin- 1985; Kramer, 1989; Meyerowitz & Chaiken, 1987; early over the expected value dimension in the mone- Neale, Huber, & Northcraft, 1987; Qualls & Puto, 1989; tary domain. Framing effects were observed in 7 of Travis, Phillippi, & Tonn, 1989) and have been found 13 pairs of problems, showing a positive frame–risk in different kinds of respondents, including university aversion and negative frame–risk seeking relation- faculty members, students, physicians, and financial ship. In addition, two types of framing effects were planners (e.g., McNeil, Pauker, Sox, & Tversky, 1982; theoretically defined and empirically identified. A bi- Roszkowski & Snelbecker, 1990; Tversky & Kahne- directional framing effect involves a reversal in risk man, 1981). These effects appear to be a general and preference, and occurs when a decision maker’s risk preference is ambiguous or weak. Four bidirectional persistent choice phenomena. effects were observed; in each case a majority of sub- However, converging evidence demonstrates that the jects preferred the sure outcome under a positive occurrence of a framing effect depends on many task, frame but the gamble under a negative frame. In con- content, and context variables inherent in choice prob- trast, a unidirectional framing effect refers to a prefer- lems, which themselves may involve distinct psycholog- ence shift due to the framing of choice outcomes: A ical mechanisms. Fischhoff (1983), for example, found majority of subjects preferred one choice outcome (ei- it hard to predict when certain frames would be used ther the sure thing or the gamble) under both framing by a decision maker. Fagley and Miller (1987) found no conditions, with positive frame augmented the prefer- framing effect in their subjects’ responses to a decision ence for the sure thing and negative frame augmented problem involving lives threatened by cancer. Schnei- the preference for the gamble. These findings revealed der (1992) found that framing effects are unstable and some dynamic regularities of framing effects and vary with the probability structure of choice problems posed implications for developing predictive and test- in different task scenarios. Many researchers have noted the erratic nature of framing effects and explored different factors that may I thank Roger Schvaneveldt, Gilbert French, and two anonymous determine their occurrence. Empirical studies have reviewers for helpful comments on an earlier draft of this paper. Address correspondence and reprint requests to the author at Psy- shown that the psychological mechanisms of framing chology Department, University of South Dakota, Vermillion, SD effects are sensitive to various social and cognitive vari- 57069. E-mail: via Internet to [email protected]. ables. These include the amount of information avail- 145 0749-5978/96 $18.00 Copyright q 1996 by Academic Press, Inc. All rights of reproduction in any form reserved. / a709$$2654 10-10-96 22:11:07 obhas AP: OBHDP 146 WANG able to a decision maker (Levin, Johnson, Russo, & versatile model to explain various framing effects. They Deldin, 1985; Shoorman, Mayer, Douglas, & Hetrick, pointed out the many confusions raised by comparing 1994), the justification required for a choice (Miller & choice phenomena that may involve different framing Fagley, 1991), the decision maker’s perspective change mechanisms. For this reason, framing effects should from his/her own money to his/her clients’ money (Rosz- have a clearer definition. According to them, there are kowski & Snelbecker, 1990), the relationship between at least three types of framing effects, each associated hypothetical decision recipients and a decision maker with its own kind of framing: (1) risky choice framing (Wang & Johnston, 1995), and the size of a social group affecting a decision maker’s willingness to take a risk; for which a decision problem is described (Bohm & (2) attribute framing affecting the encoding and evalu- Lind, 1992; Wang & Johnston, 1995). ation of object or event characteristics; and (3) action These findings call attention to the dynamic features framing affecting the persuasiveness of a communica- of framing effects and the production rules that control tion. They argued a need for different perceptual or their presence and absence. In some contexts, framing cognitive processes to explain the distinct types of effects appear robust and sizable. In others, the effects framing effect. appear highly variable and erratic. It is therefore im- In this study the author examines the dynamics of portant to know the antecedent conditions that deter- different types of framing effects in risky choice and mine their appearance and disappearance. the effects of task domains on the risk preference of A classical example of framing effects is Tversky and decision makers. Kahneman’s (1981) Asian disease problem. In their study, subjects were asked to choose between a sure Risk Preferences in Different Task Domains outcome that led to a certain survival of one third of Apparently, framing effects are not limited to specific 600 hypothetical patients (i.e., 200 people) and a risky decision tasks. However, framing effects and people’s probabilistic outcome, a one-third probability that all risk preferences do vary as a function of task domains 600 people would survive and a two-thirds probability (e.g., Fagley & Miller, 1987; Schneider, 1992; Schnei- that no one would survive. Tversky and Kahneman der & Lopes, 1986; Wagenaar, Keren, & Lichtenstein, found a pronounced reversal in risk preference as a 1988; Wang, 1996a; Wang & Johnston, 1995). result of how the choice outcomes were framed. Most A question of interest in this study is how the task of their subjects (72%) favored the sure thing when the domains in which a problem is described influence the choice outcomes were framed in terms of lives saved risk preference in human choices. Given the same prob- whereas most of the subjects (78%) in another group ability and payoff structure, a content and context-free favored the gamble (the probabilistic outcome) when utility model would predict the same risk preference the same choice outcomes were framed in terms of lives across different task domains. In other words, such lost. models would predict a similar risk preference pattern A framing effect, such as the one found in the Asian independent of whether the required choice is between disease problem, is often explained using Kahneman saving precious paintings or saving the same number and Tversky’s prospect theory (1979). Accordingly, peo- of human lives with the same probability. The present ple code the possible choice outcomes as gains and study examined this issue by using choice problems losses, and tend to be risk averse when choosing among that shared the same probability structure but were prospects seen as gains but risk seeking when choosing in three distinct task domains (i.e., life–death, public among prospects seen as losses. Thus, when choice op- property, and personal money). The choice outcomes of tions are framed positively, a decision maker tends to the problems presented in each of the three task do- perceive them as gains and becomes more risk averse. mains had a binary structure of a sure thing vs a gam- In contrast, when the same choice options are framed ble of equal expected value and differed in their fram- negatively, a decision maker tends to perceive them as ing conditions and the expected values. losses and becomes more risk seeking. Risk sensitivity can be considered an adaptation to Alternative hypotheses of framing effects have also different environmental problems (e.g., Real, 1991; been proposed (e.g., Frisch, 1993; Reyna & Brainerd, Real & Caraco, 1986; Wang, 1996a; Wang, 1996b). De- 1991; Schneider, 1992). These new hypotheses explore pending on the nature of a task, decision makers may possible mechanisms of framing effects beyond those have different minimum requirements in different task of prospect theory. domains. The minimum requirement for a decision task Recently, Schneider, Levin, and Gaeth (1995) ad- then could be psychologically translated into a decision dressed the limitations of using prospect theory as a maker’s aspiration level. When the mean expected / a709$$2654 10-10-96 22:11:07 obhas AP: OBHDP  FRAMING EFFECTS: DYNAMICS AND TASK DOMAINS 147 value of choice options is above the minimum require- more susceptible to the dichotic effects of positive frame ment, a rational choice would be risk- and variance- and negative frame. Therefore, the occurrence and ab- averse to avoid possible failures. In contrast, when the sence of framing effects would vary categorically as a mean expected value is below the minimum require- function of the perceived group contexts (e.g., large ment, a decision maker should be more risk- and vari- group, small group, and family group) rather than a ance-seeking to maximize the probability of achieving linear function of the group size. However, it is not the goal. It is therefore expected that given two choice clear how subjects would code the numerical numbers outcomes, one sure thing and one gamble of equal ex- of expected value (i.e., 6000, 600, 60, and 6) in the pected value, a decision maker would prefer the sure public property and monetary domains. Although it is thing if its expected value is above his/her aspiration possible that subjects classify the problems categori- level but prefer the gamble if the expected value of cally (e.g., large money versus small money) the cut the sure thing is below the task-determined aspiration point for such calcification may be more variable from level. individual to individual. Particularly, in the monetary In different task domains, however, the minimum domain, subjects’ risk perference may be more linearly requirement or aspiration level of a decision maker related to the stated expected values. may differ. While a sure outcome of saving one-third of $6 may be preferable to a gamble of saving all $6 with Bidirectional and Unidirectional Framing Effects a one-third probability for a person, a sure outcome In a recent study, Wang and Johnston (1995) used of saving one-third of 6 family members may become a life–death decision paradigm similar to the Asian unacceptable for the same person. As a result, the per- disease problem. In this study, they argued that saving, son would be more likely to choose the alternative prob- on average, one-third of group may have distinct adap- abilistic outcome that has a one-third probability to tive consequences depending on the size of the social save all 6 family members. It is therefore expected that group, and thus the risk attitude and framing effects compared to the counterpart problems presented in the may vary as a function of a systematic manipulation public property and monetary domains, the life–death of this variable. Framing effects appeared, disap- choice problems would evoke more risk-seeking peared, and reappeared in a markedly different form choices. as the life–death problem was described in a large Moreover, in a task domain, manipulations along a group, a small group, and a family social context, re- risk-sensitive value dimension, say the total amount spectively. In a hypothetical large group context with of money involved in a choice problem, may result in either 6000 or 600 people involved, subjects’ risk prefer- different risk preferences. If so, the risk proneness re- ence indistinguishably reversed from predominantly flected by the percentage of subjects choosing either risk averse when the choice outcomes were framed in the sure thing or the gamble at each point of a selected terms of lives saved to predominantly risk seeking value dimension would be different. In the present when the same outcomes were framed in terms of lives study, the numerical numbers of the expected values lost. However, when the hypothetical patients were de- used in each of the three task domains were 6000, 600, scribed in a small group or family context, no reversal 60, and 6. Along the expected value dimension in a in risk preference was found. The subjects were unam- certain task domain, a fuzzy area of ambiguous risk biguously risk seeking in order to save all the group preference may appear at a location dependent on the members. In addition, when the hypothetical patients selected values as well as the task, content, and context were described as subjects’ own family members, the of the problem. This difference in the risk preference subjects, although clearly being risk seeking, became may affect people’s susceptibility to the framing of significantly more risk seeking if the choice outcomes choice outcomes. were framed negatively in terms lives lost. The percent- Based on our previous findings (Wang & Johnston, age of subjects choosing the gamble over the sure thing 1995), it was predicted that the manipulation along the increased from 72% under positive framing to 94% un- group size dimension (e.g., the total number of lives der negative framing. The extreme risk seeking in this described in the life–death problems) would yield a case seems to have been elicited by proposed choice group context-specific risk preference pattern. In a outcomes that were both objectively negative and nega- small group or a family context, subjects may hold a tively worded or framed. This is a different framing risk attitude that ‘‘we all live or die together,’’ and tend effect. In this case, the predominant choice preference to be risk seeking under both framing conditions. In is unidirectionally risk seeking under both framing contrast, in a large group situation, they may become conditions. / a709$$2654 10-10-96 22:11:07 obhas AP: OBHDP 148 WANG It appears that framing effects take two distinct This distinction of the two types of framing effects forms. One type of framing effect involves preference can be used as an experimental probe for exploring reversal from predominantly risk averse to predomi- distinct cognitive mechanisms governing the risk atti- nantly risk seeking or vise versa, due to the dichotic tude in different social situations. From this viewpoint, effect of the framing of the choice outcomes. This bidi- the bidirectional framing effects in large group context, rectional framing effect (denoted B) is characterized by no framing effect in small group context, and the unidi- predominant risk-averse choices under positive fram- rectional framing effect in family context, found in our ing and predominant risk-seeking choices under nega- previous study (Wang & Johnston, 1995), reflect differ- tive framing.1 A second type of framing effect, unidirec- ent underlying decision mechanisms. Of equal impor- tional framing effect, involves no preference reversal tance, the current definition of bidirectional and unidi- but a shift to a more extreme risk preference. If the rectional framing effects provides useful constraints for predominant preference is unidirectionally risk averse making experimental predictions. For example, based under both framing conditions, it is even more risk on the proposed relationship between the framing ef- averse under positive frame than under negative fects and task-determined risk preference, certain risk frame. Similarly, if the predominant preference is uni- preference patterns (e.g., Ura-Urs-B; Urs-Ura-B; B- directionally risk seeking under both framing condi- Ura-Urs; B-Urs-Ura; Ura-B-Ura; or Urs-B-Urs) would tions, it is even more risk seeking under negative frame not occur along any selected value dimensions. These than under positive frame. Therefore, there are two constraints also serve as a set of criteria for testing possible forms of unidirectional effect, one augments and falsifying the proposed hypothesis. Although the the risk-averse preference (denoted Ura) and the other selected range on a value dimension may not cover all augments the risk-seeking preference (denoted Urs). the effective points, the empirical results can be ana- This view suggests that bidirectional framing effects lyzed on the basis of these theoretical constraints. That may result from the lack of clarity in choice prefer- is, if an unexpected pattern of framing effects is ob- ences. A decision maker with an ambiguous or ambiva- served, the present hypothesis would be falsified. lent risk preference may actively search for more infor- In sum, the experimental predictions of the present mation besides the task, content, and context variables study include: (1) there would be a task-domain effect embedded in a decision problem. In this condition, the on risk preference with a higher percentage of subjects decision maker’s risk preference may rely on not only choosing the gamble when the choice problems were the choice options themselves but also the way in which presented in the life–death task domain; (2) framing these choice options are worded, phrased, or framed. effects would appear and disappear in different man- Both positive and negative frames thus may work effec- ners along the selected value dimension in different tively but bidirectionally toward the opposite riskiness task domains; (3) the changes in subjects’ risk prone- direction. ness along the selected value dimension would appear On the other hand, the framing of choices may also to be more categorical in the life–death task domain lead to unidirectional effects. When the risk preference and more linear in the monetary domain; (4) a positive is clear, a decision maker would resist a framing manip- frame would tend to increase risk-averse (the sure ulation if it is inconsistent with the existing task-deter- thing) choices and a negative frame would tend to in- mined preference’s direction. However, a framing ma- crease the risk-seeking (the gamble) choices; and (5) nipulation consistent with an existing risk preference the observed framing effects would be identified as ei- may augment that preference. Positive framing there- ther bidirectional or unidirectional, and their occur- fore could intensify risk-averse preferences whereas rence would follow a predicted pattern. negative framing could magnify risk-seeking prefer- ences. Depending on the momentum of an existing pref- METHOD erence, the augmenting effect may be negligible or sig- nificant: The bigger the momentum, the larger the Experimental Materials effect. The experimental paradigms of this study involved three distinct problem domains: human lives, public 1 It needs to be noted that the reversal in risk preferences found property (museum paintings), and dollars of personal in an experiment with a between-subject design only reflects a con- money. Within each domain, the expected values of trast in subjects’ risk preference between two sampling groups under two different framing conditions. It does not necessarily mean that choice outcomes were manipulated at four numerical a majority of individual subjects would reverse their risk preference levels: 6000, 600, 60, and 6. These numbers repre- under the two different framing conditions. sented (1) the number of people threatened by a fatal / a709$$2654 10-10-96 22:11:07 obhas AP: OBHDP  FRAMING EFFECTS: DYNAMICS AND TASK DOMAINS 149 TABLE 1 Group Differences in Risk Preference for the Positively and Negatively Framed Outcomes in the Life–Death Problems Experimental Total Framing of the Choice of the Framing group (N) outcome sure thing effect x2 statistics P6000life 31 Lives saved 61.3% Bidirectional x2 Å 3.73 N6000life 30 Lives lost 33.7% p õ .05 P600life 31 Lives saved 58.1% Bidirectional x2 Å 8.23 N600life 34 Lives lost 23.5% p õ .004 P60life 33 Lives saved 42.4% None Not significant N60life 30 Lives lost 33.3% P6life 30 Lives saved 33.3% None Not significant N6life 33 Lives lost 24.4% P6rlife 33 Lives saved 33.3% Unidirectional x2 Å 5.52 N6rlife 31 Lives lost 9.7% (Risk seeking) p õ .02 Note. P denotes positive framing; N denotes negative framing; 6000, 600, 60, and 6 are the number of lives at risk; r denotes that the hypothetical patients in the life–death decision problem were described as the subjects’ relatives; life denotes the life–death problem. The overall risk-averse choice Å 35.4%. disease, (2) the number of museum paintings exposed the gamble under the positive frame than under the to chemical pollution, or (3) the amount of money at negative frame. If a unidirectional effect is risk-seeking risk due to a bankruptcy. In the life–death domain, augmenting, then more than 50% of subjects would group size six was used both in a small group context choose the gamble over the sure thing under both in which the hypothetical patients were six anonymous frames, but a significantly higher percentage of sub- people and in a family context in which the six hypo- jects would choose the gamble under the negative thetical patients were described as the subjects’ close frame than under the positive frame. relatives. The choice problems presented to each subject group Subjects and Procedure was framed either positively or negatively. A total of 26 choice problems were used in this study including The subjects were 902 undergraduate students en- five pairs of positively and negatively framed life– rolled in introductory psychology courses who agreed death problems, four pairs of museum paintings prob- to participate for extra course credit. The average age lems, and four pairs of personal money problems. of the subjects was 20.3 years. Subjects were randomly Examples of the choice problems presented within assigned to one of 26 experimental (subject) groups. each task domain appear in the Appendix. Subjects were instructed that there were no right or For the current between-subject design, a 50–50 risk wrong answers and were asked to choose anonymously preference point (i.e., equal percentages of subjects fa- between two effectively identical choice options: a sure voring the sure thing and the gamble in a binary deci- outcome versus a gamble of equal expected value. sion situation) can be considered a rough estimate of Of the total subjects, 316 (207 females and 109 the risk neutrality and therefore used as an operational males) were randomly assigned to 10 groups, each re- reference point to classify the two proposed types of ceiving one version of the five pairs of positively and framing effect. A bidirectional framing effect would be negatively framed life–death problems. Another 327 featured by a higher than 50% of subjects choosing the subjects (196 females and 131 males) were assigned to sure thing under positive framing and a lower than the eight groups that received the four pairs of the 50% of subjects choosing the sure thing under negative public property (museum paintings) problems. The re- framing. On the other hand, in the case of a unidirec- maining 259 subjects (166 females and 93 males) re- tional framing effect, the two percentage data points ceived the four pairs of choice problems presented in obtained under positive frame and negative frame the monetary domain. would locate on the same side of the 50% reference point. If a unidirectional effect is risk-aversion aug- RESULTS AND DISCUSSION menting, more than 50% of subjects would prefer the sure thing under both frames, but a significantly higher The choice percentages, sample sizes, and x2 statis- percentage of subjects would prefer the sure thing over tics for framing effects obtained from the 10 groups / a709$$2654 10-10-96 22:11:07 obhas AP: OBHDP 150 WANG TABLE 2 Group Differences in Risk Preference for the Positively and Negatively Framed Outcomes in the Museum Paintings Problems Experimental Total Framing of the Choice of the group (N) outcome sure thing Framing effect x2 statistics P6000painting 38 Paintings saved 81.6% Unidirectional x2 Å 4.17 N6000painting 38 Paintings lost 60.5% (Risk aversion) p õ .04 P600painting 39 Paintings saved 69.2% None Not significant N600painting 41 Paintings lost 65.9% P60painting 45 Paintings saved 75.6% Bidirectional x2 Å 8.45 N60painting 40 Paintings lost 45.0% p õ .004 P6painting 45 Paintings saved 62.2% Bidirectional x2 Å 4.67 N6painting 41 Paintings lost 39.0% p õ .03 Note. P denotes positive framing; N denotes negative framing; 6000, 600, 60, and 6 are the number of paintings at risk; paintings denotes the museum paintings problem. The overall risk-averse choice Å 62.4%. receiving the life–death problems, eight groups receiv- prominent when the selected group size was within a ing the museum paintings problems, and eight groups common range of human kith-and-kin groups. In a receiving the monetary problems are presented in Ta- small group or a familial context, the sure outcome of ble 1, Table 2, and Table 3, respectively. saving one-third of group members appears to fall be- low the subjects’ aspiration level (their minimum re- Task-Domain Effects on Risk Preference quirement). As a result, they chose the riskier probabi- The overall percentages of subjects choosing the sure listic outcome. Under these conditions, revealed by an- thing across all pairs of framing groups were 35.4% for other study (Wang, 1996b), even when the choice of the life–death problems, 62.4% for the museum paintings sure thing could save two-thirds of the hypothetical problems, and 64.9% for the personal money problems. group members, a substantial proportion of subjects The first percentage differed greatly from the latter still preferred the probabilistic outcome that had a two percentages, x2 (2, N Å 902) Å 65.86, p õ .00001. lower expected mean value of saving only one-third of There was no significant difference, however, in the the group members. overall risk preference between the data from the mu- The significantly higher percentage of subjects choos- seum paintings problems and those from the personal ing the sure thing when receiving the problems pre- money problems. As predicted, subjects were signifi- sented in either the public property (the museum paint- cantly more risk seeking when dealing with the choice ings) domain or monetary domain suggests that the problems in the life–death domain. This is particularly subjects had a lower aspiration level regarding the TABLE 3 Group Differences in Risk Preference for the Positively and Negatively Framed Outcomes in the Personal Money Problem Experimental Total Framing of the Choice of the group (N) outcome sure thing Framing effect x2 statistics P6000money 36 Money saved 91.7% Unidirectional x2 Å 6.65 N6000money 30 Money lost 66.7% (Risk aversion) p õ .01 P600money 32 Money saved 78.1% None Not significant N600money 31 Money lost 64.5% P60money 35 Money saved 57.1% None Not significant N60money 33 Money lost 48.5% P6money 31 Money saved 51.6% None Not significant N6money 31 Money lost 58.1% Note. P denotes positive framing; N denotes negative framing; 6000, 600, 60, and 6 are the number of dollars at risk; money denotes the personal money problem. The overall risk-averse choice Å 64.9%. / a709$$2654 10-10-96 22:11:07 obhas AP: OBHDP  FRAMING EFFECTS: DYNAMICS AND TASK DOMAINS 151 (Urs) in the family group context: a B-B-no-no-Urs pat- tern (for the theoretical and operational definitions of B, Ura, and Urs framing effects, see the relevant dis- cussions in the Introduction and Method parts of the paper). The framing effect found in the family group context was different from the two bidirectional framing effects found in the two large group contexts. For the two bidi- rectional effects, the choice percentage of the sure thing under the positive frame was above the 50% reference point but below 50% under the negative frame. How- ever, the unidirectional framing effect found in the fam- ily context was featured by a significant preference shift from risk seeking under the positive frame to more risk seeking under the negative frame. That is, the negative framing accentuated the task-determined ex- isting risk-seeking preference to a more extreme de- gree. Museum paintings problem. x2 analysis showed FIG. 1. Percentages of the risk averse choice as a function of the that the risk preferences between the eight subject number of hypothetical lives at risk plotted logarithmically; r denotes groups given the museum paintings problems were sig- the problem in which the patients were described as the subjects’ nificantly different, x2 (7, N Å 327) Å 25.35, p õ .0007. relatives. A significant overall framing effect was found, x2 (1, N Å 327) Å 13.14, p õ .0003. As Table 2 shows, framing effects appeared and dis- minimum proportion of the total value that had to be appeared along the manipulated expected value dimen- rescued. sion (i.e., the total number of paintings at risk) (see also Fig. 2). Both bidirectional as well as unidirectional Bidirectional and Unidirectional Framing Effects in Each of the Three Task Domains The patterns of framing effects emerging from the three task domains were different. Life–death problem. There was a significant differ- ence in subjects’ choices over the 10 experimental groups, x2 (9, N Å 327) Å 31.17, p õ .0003. The overall framing effect was also significant, x2 (1, N Å 316) Å 14.31, p õ .0002. A graphic representation of the dynamic pattern of the observed framing effects is shown in Fig. 1. Both bidirectional and unidirectional framing effects ap- peared. The 50–50 choice distribution point, repre- sented by the dotted line in the figure, was used as the reference for risk neutrality to classify the two types of framing effect. Also as summarized in Table 1, as the social group context changed from large group with either 6000 or 600 people to smaller group with either 60 or 6 people, and to the family group with 6 hypotheti- cal relatives, bidirectional framing effects (B) were found in the two large group contexts, followed by no FIG. 2. Percentages of the risk averse choice as a function of the framing effect (no) in the two smaller group contexts, number of hypothetical museum paintings at risk plotted logarithmi- and a risk-seeking augmenting unidirectional effect cally. / a709$$2654 10-10-96 22:11:07 obhas AP: OBHDP 152 WANG framing or negative framing condition separately. The only significant linear effect found was in the monetary domain and under the positive framing condition, x2 (3, N Å 134) Å 17.30, p õ .0006. As the expected values increased from 6 to 6000, the number of subjects choos- ing the sure thing under the positive framing linearly increased from 52 to 92%. Effects of Frames on Risk Preference Although the patterns of framing effects were task specific, there evidently was a dynamic regularity with regard to the occurrence of framing effects across all three task domains. All the seven framing effects found in the 13 pairs of choice problems showed a positive frame–risk aversion and/or negative frame–risk seek- ing pattern. GENERAL DISCUSSION FIG. 3. Percentages of the risk averse choice as a function of the The seemingly erratic pattern of framing effects may amount of hypothetical money at risk plotted logarithmically. have its dynamic regularities. The basic notion of the current hypothesis is that the primary determinants of a decision maker’s risk preference are the task, content, framing effects were identified. Unlike the categorical and contexts variables inherent in a decision problem. pattern of B-B-no-no found in the life–death domain, These factors determine the aspiration level of a deci- the framing effects showed a Ura-no-B-B pattern. See sion maker and his/her risk attitude. A decision maker Table 2 for the x2 statistics and p values. will tend to be risk seeking when facing a loss than a gain. In the case of a gain, the mean expected value of Personal money problem. An overall analysis choice outcomes is more likely to be above the aspira- showed that subjects’ choice preferences in the eight sub- tion level (the minimum requirement) of a decision ject groups differed markedly, x2 (7, N Å 259) Å 25.33, maker whereas, in the case of a loss, the mean expected p õ .0007. An overall framing effect, although much value is more likely to be below the reference level. smaller than those found with the life–death problems Positive frame of choice outcomes emphasizes the and the museum paintings problems, was statistically ‘‘gain’’ aspect of a prospect whereas negative frame em- significant, x2 (1, N Å 259) Å 4.44, p õ .04. phasizes the ‘‘loss’’ aspect of the prospect. In other As summarized in Table 3 and illustrated in Fig. 3, words, positive frame tends to increase the perceived of the four selected levels along a monetary dimension surplus and decrease the perceived deficit between the (i.e., the total amount of money at risk), only a unidirec- aspiration level and the expected mean value of choice tional framing effect was found (see Table 3 for x2 sta- outcomes. On the other hand, negative frame tends to tistic and p value). The pattern was Ura-no-no-no, sug- increase the perceived deficit and decrease perceived gesting that the subjects’ risk preference was quite surplus between the aspiration level of a decision clear and stable when dealing with personal money maker and the expected mean value of choice outcomes. type of problems with the total amount of money rang- However, whether or not the framing will have an ef- ing from $6000 to $6. fect on the risk preference of a decision maker depends on the direction and intensity of the risk preference Changes in Risk Preferences along the Expected Value elicited by task, content, and context variables in a Dimension decision problem. To examine whether there is a linear increase in risk The framing of choice outcomes may induce two dif- preference for the sure thing as the numerical number ferent types of effects in risky choices. When a decision of the expected value in each task domain increased maker’s risk preference is ambiguous or weak, s/he from 6 to 6000, a series of six linearity tests were con- may become susceptible to both positive and negative ducted for each of the three tasks under either positive frames of choices. The positive frame tends to pull a / a709$$2654 10-10-96 22:11:07 obhas AP: OBHDP  FRAMING EFFECTS: DYNAMICS AND TASK DOMAINS 153 choice in the risk-averse direction and the negative tional framing effect involved a preference shift, either frame in the risk-seeking direction (i.e., a bidirectional risk averse or risk seeking, due to the framing of choice effect). However, when the risk preference is unambig- outcomes. In the case of risk-aversion augmenting ef- uous, a decision maker will be more immune from the fect, more than 50% of subjects preferred the sure thing framing manipulations. Therefore, the framing of under both positive and negative framing conditions choices will have either no effect or a unidirectional but even more of them preferred the sure thing under effect that augments the decision maker’s existing pref- positive frame. When the augmenting effect was in the erence. direction of risk seeking, on the other hand, a majority To examine the dynamic functions of risk preference of subjects chose the gamble over the sure thing of and framing effects in different social task domains, equal expected value under both framing conditions, choice problems sharing the same formal probability but the choice of the gamble became even more domi- structure and expected values were presented in differ- nant among subjects under negative frame. ent task domains under either positive or negative It appears that the framing of choices as a weak framing condition. editing factor in choice making processes shows its ef- 1. The risk preference data revealed a task-specific fects when (1) the risk preference of a decision maker pattern. Subjects were significantly more risk seeking is weak, or (2) it can accentuate an existing risk prefer- when facing life–death choice problems than facing ence. The framing of choices does not determine risk their counterpart problems presented in a public prop- preference; it only regulates an existing risk attitude erty or personal money domain. This finding is in that is determined primarily by the task, content, and agreement with the hypothesis that the subjects’ aspi- context variables inherent in choice problems. ration level (the minimum requirement) for the life– Although the author did not directly test it, Frisch death problems was higher than that for the public (1993) expressed a similar idea, that if subjects do not property and personal money problems. really care whether they choose a sure outcome or a 2. Framing effects appeared and disappeared as a gamble, then minor variations in wording or phrasing result of experimental manipulations along the se- would greatly influence their choices. On this view, one lected expected value dimension. would expect that the proportions of subjects showing 3. The patterns of risk preference obtained from the the framing effect would increase as the strength of three task domains were different. In the life–death preference within each frame decreases. domain, a group-context specific pattern emerged. Bidi- Several investigators have argued that the classical rectional framing effect appeared in the large group framing effects (bidirectional effects in the present context (i.e., 6000life and 600life groups) and disap- definition) occur only when ambiguity about a choice peared in the small group context (i.e., 60life and 6life problem is high. For example, the framing effects occur groups), and a risk-seeking augmenting unidirectional when people are ambiguous in their experiences of con- framing effect occurred in the family context (i.e., 6rlife sequences of a decision (e.g., Frisch, 1993; Hoch & Ha, groups). In addition, the risk preference patterns were 1986), when consumers are unfamiliar with the prod- indistinguishable within the large group context or ucts they have to choose among (e.g., Bettman & Sujan, within the small group context. 1987), or when the information about the consequences Although in none of the three task domains, the risk of choice is incomplete or limited (e.g., Schoorman, preference data revealed a complete linear trend over Mayer, Douglas, & Hetrick, 1994). the manipulated value dimension under both framing In a recent study, Reyna and Brainerd (1991) pre- conditions, the monetary problems elicited more lin- sented a modified version of the Asian disease problem early correlated changes in risk preference under posi- (Tversky & Kahneman, 1981) to their subjects with tive framing as the expected values increased from 6, either positively framed or negatively framed choice 60, 600, to 6000. outcomes. They manipulated the explicitness of the 4. All the observed framing effects showed a positive problem by replacing the numerical information of ex- frame–risk aversion and negative frame–risk seeking pected payoff values and/or probabilities of choice out- relationship. comes with qualitative information such as ‘‘some,’’ 5. Two types of framing effects were theoretically ‘‘many,’’ ‘‘few,’’ or ‘‘higher.’’ What is interesting here is defined and empirically identified. In the case of a bidi- that the classical (bidirectional) framing effects were rectional framing effect, a majority of the subjects pre- not only observed in the subjects’ choices but also ap- ferred the sure outcome under a positive frame but the peared to be sensitive to the amount of numerical infor- gamble under a negative frame. In contrast, a unidirec- mation replaced by the qualitative information. The / a709$$2654 10-10-96 22:11:07 obhas AP: OBHDP 154 WANG amplitude of the framing effect peaked when all of the of the distinction between framing effect and what is numerical information was replaced, presumably when called reflection effect (Fagley, 1993). Reflection and the problem was least explicit and most ambiguous to framing effects are two distinct phenomena. In the lat- the subjects. ter, changes in risk preferences are caused by phrasing It is also worth mentioning that the choice pattern the same choice outcomes as though they are gains from the five pairs of framing groups receiving the life– versus phrasing them as though they are losses. The death problems replicated our previous findings dis- reflection effect is a decision bias, but not a cognitive cussed earlier in this paper (Wang & Johnston, 1995). illusion. It refers to having opposite preferences for the This validity check lends additional support to the no- positive versus negative prospects (i.e., whether the tion that social group size and perceived social contexts outcomes are net gains or net losses). Reflection effects are powerful contextual variables affecting human do not involve the same choice outcomes described as choices. though they are gains or losses; reflection effects in- The observed choice preferences in the present study volve real gains and loses. From this perspective, the showed a clear task domain-specific pattern. Both nor- framing effect (the bidirectional effect by the present mative theories such as expected utility theory and de- definition) is an irrational choice phenomenon but the scriptive theories such as prospect theory are silent reflection effect is not. about the specific roles of the task, content, and context Second, the task-specific choice pattern observed in of a decision problem in regulating the risk preference the current results suggests that framing effects should in human choices. Research has repeatedly demon- be examined separately from the contextual effects that strated that given the identical formal probability have often resulted from presenting the same choice structure of choice prospects, changes in a cover story options in different scenarios or cover stories. Contex- about the context in which a decision problem occurs tual description in a cover story is not a decoration for often resulted in significant reversal or shift in respon- an abstract problem but a real condition that triggers dents’ risk proneness (e.g., Schneider, 1992; Wagenaar, different risk attitudes. Although the expected values Keren, & Lichtenstein, 1988; Wang, 1996a; Wang & of two choice problems may be the same, the opposite Johnston, 1995). risk preferences could be produced by changing the so- Recently some researchers have proposed that future cial context in which the problem is described. This research should consider content and context of deci- contextual effect is not a result of a cognitive illusion, sion or reasoning problems not as intervening or deco- and therefore it should be clearly distinguished from a rative variables but as primary and defining factors of bidirectional framing effect. Unlike presenting a glass human reasoning and judgement. From this perspec- as ‘‘half full’’ versus ‘‘half empty,’’ contextual differ- tive, the content and contextual variables can be used ences specified in different task domains are meaning- either for evaluating human judgment or as a research ful antecedent factors that determine choice prefer- probe to uncover the underlying psychological mecha- ences. nisms (e.g., Cosmides & Tooby, 1992; Gigerenzer, 1996; Third, the current findings call attention to the ne- Gigerenzer & Hug, 1992, Lopes & Oden, 1991). cessity to classify framing effects by their forms: the bidirectional effects involving a reversal in risk prefer- Several Conceptual Distinctions Regarding Framing ences versus the unidirectional framing effects involv- Effects ing a preference shift. The notion of framing effects involves quite different Framing effects are commonly considered as one of connotations. The word framing was used to refer to a the most severe violations of normative utility axioms, variety of situations including any changes in a deci- and therefore a strong indication of irrationality. Fram- sion context, in complex mental accounting of pros- ing effects violate the description invariance axiom of pects, or in simple wording of choice options or out- utility theory of rational choices, which states that dif- comes. In this article, the term is restricted to the posi- ferent descriptions of the same choices should yield the tive and negative ways of presenting choice outcomes. same preference order. In a framing condition, whether In other words, the framing of decisions is defined in the change in phrasing the same decision prospect re- terms of how a problem is described to a decision maker sults in a significant reversal in risk preference appears but not in terms of how that person mentally repre- to be a key aspect for determining the rationality of sents the problem. the choices. The classical, bidirectional framing effects To facilitate discussions of framing effects, several often led to an irrational reversal in risk preferences points need to be clarified. First, one needs to be aware under different framing conditions. However, in the / a709$$2654 10-10-96 22:11:07 obhas AP: OBHDP  FRAMING EFFECTS: DYNAMICS AND TASK DOMAINS 155 case of the unidirectional framing effect, there is no Imagine that 6000 (600, 60, 6, 6r) people are infected risk preference reversal. The unidirectional effect was by a fatal disease. Two alternative medical plans to manifested only with strong risk preference and may treat the disease have been proposed. Assume that the be considered a kind of confirmation bias. exact scientific estimates of the consequences of the From this viewpoint, bidirectional framing effects re- plans are as follows: flect the susceptibility to an irrational preference rever- If plan A is adopted, 2000 (200, 20, 2, 2r) people will sal elicited by positive framing in one direction and by be saved. negative framing of the same outcomes in the opposite If plan B is adopted, there is a one-third probability direction. However, the unidirectional framing effects that all the people will be saved, and two-thirds proba- reflect an immunity from the irrational reversal in risk bility that none of them will be saved. preferences. While being resistant to a framing manip- Which of the two plans would you favor? ulation that is inconsistent with an existing risk prefer- 2. Negatively framed problems: ence, a decision maker is sensitive to framing that con- Imagine that 6000 (600, 60, 6, 6r) people are infected firms an existing risk preference. The finding of no by a fatal disease. Two alternative medical plans to framing effect in small group context but a significant treat the disease have been proposed. Assume that the unidirectional effect of risk-seeking augmentation in exact scientific estimates of the consequences of the family context suggests that subjects’ risk proneness plans are as follows: was stronger in the family context and became more If plan A is adopted, 4000 (400, 40, 4, 4r) people likely to be intensified by the negative framing of the will die. choice outcomes. It appears that unidirectional aug- If plan B is adopted, there is a one-third probability menting effects tend to occur when an existing risk that none of them will die, and two-thirds probability preference is strong. that all the people will die. Of 13 pairs of framing problems used in this study, Which of the two plans would you favor? seven showed framing effects. However, only four were Note. Each item in a parenthesis represents the num- identified as bidirectional effects according to the stan- ber of hypothetical patients used for a different subject dards proposed in the present study. The finding sug- group. The subscript r means that the hypothetical pa- gests that without a pronounced reversal in risk prefer- tients in the corresponding questionnaire were de- ences, a framing effect may not be an irrational viola- scribed as close relatives of the subject. tion of description invariance principle, but a confirmation bias produced by framing the choices in a II. Versions of the museum paintings problem. way that is consistent with the existing risk preference. 1. Positively framed problems: As Payne, Bettman, and Johnson (1992) pointed out, Imagine that 6000 (600, 60, 6) pieces of precious a complete theory of framing has proven to be difficult paintings in a world-famous museum are accidentally to formalize, although progress has been made in iden- exposed to a disastrous chemical pollution. Two alter- tifying important elements of framing. Further re- native plans to rescue these art treasures have been search is needed to develop a better theoretical founda- proposed. Assume that the exact estimates of the conse- tion for the conditions under which the framing of deci- quences of the plans made by scientists are as follows: sions will have either greater or less impact on decision If plan A is adopted, 2000 (200, 20, 2) pieces will be making. In addition, further research on framing ef- saved from the chemical pollution. fects should not only focus on demonstrating the effects If plan B is adopted, there is a one-third probability of decision framing but also attempt to clarify the ante- that all the paintings will be saved, and two-thirds cedent conditions that elicit framing effects, to search probability that none of these paintings will be saved. for underlying mechanisms, and to construct falsifiable Which of the two plans would you favor? models that can predict the riskiness direction of a 2. Negatively framed problems: framing effect and explain the absence as well as the Imagine that 6000 (600, 60, 6) pieces of precious presence of a framing effect. paintings in a world-famous museum are accidentally exposed to a disastrous chemical pollution. Two alter- native plans to rescue these art treasures have been APPENDIX proposed. Assume that the exact estimates of the conse- Choice Problems Used in the Study quences of the plans made by scientists are as follows: If plan A is adopted, 4000 (400, 40, 4) pieces will be I. Versions of the life–death problem. destroyed by the chemical pollution. 1. Positively framed problems: If plan B is adopted, there is a one-third probability / a709$$2654 10-10-96 22:11:07 obhas AP: OBHDP 156 WANG that none of these paintings will be destroyed, and two- Fischhoff, B. (1983). Predicting frames. Journal of Experimental Psy- thirds probability that all of these paintings will be chology: Learning, Memory, and Cognition, 9, 103–116. destroyed. Frisch, D. (1993). Reasons for framing effects. Organizational Behav- ior and Human Decision Processes, 54, 399–429. Which of the two plans would you favor? Gigerenzer, G. (1996). Rationality: Why social context matters. In Note. Each item in a parenthesis represents the num- P. B. Baltes & U. Staudinger (Eds.), Interactive minds: Life-span perspectives on the social foundation of cognition (pp. 319–346). ber of the paintings used for a different subject group. Cambridge, UK: Cambridge Univ. Press. III. Versions of the personal money problem. Gigerenzer, G., & Hug, K. (1992). Domain-specific reasoning: Social 1. Positively framed problems: contracts, cheating, and perspective change. Cognition, 43, 127– 171. Imagine that you brought $6000 (600, 60, 6) worth Hoch, S. J., & Ha, Y.-W. (1986). Consumer learning: Advertising and of stock from a company that has just filed a claim for the ambiguity of product experience. The Journal of Consumer bankruptcy recently. The company now provides you Research, 13, 221–233. with two alternatives to recover some of your money. Kahneman, D., & Tversky, A. (1979). Prospect theory. Econometrica, If you choose alternative A, you will save $2000 (200, 47, 263–292. 20, 2) of your money. Kramer, R. M. (1989). Windows of vulnerability or cognitive illu- If you choose alternative B, you will take part in sions? Cognitive processes and the nuclear arms race. Journal of a random drawing procedure with exactly a one-third Experimental Social Psychology, 25, 79–100. probability of saving all of your money, and two-thirds Levin, I. P., Johnson, R. D., Russo, C. P., & Deldin, P. J. (1985). Framing effects in judgment tasks with varying amounts of infor- probability of saving none of your money. mation. Organizational Behavior and Human Decision Processes, Which of the two alternatives would you favor? 36, 362–377. 2. Negatively framed problems: Lopes, L. L., & Oden, G. C. (1991). The rationality of intelligence. Imagine that you brought $6000 (600, 60, 6) worth In E. Eells & T. Maruszewski (Eds.), Rationality and reasoning: of stock from a company that has just filed a claim for Essays in honor of L. J. Cohen. Amsterdam: Rodopi. bankruptcy recently. The company now provides you McNeil, B. J., Pauker, S. G., Sox, H. C., Jr., & Tversky, A. (1982). with two alternatives to recover some of your money. On the elicitation of preferences for alternative therapies. New England Journal of Medicine, 306, 1259–1262. If you choose alternative A, you will lose $4000 (400, Meyerowitz, B. E., & Chaiken, S. (1987). The effect of message fram- 40, 4) of your money. ing on breast self-examination attitudes, intentions, and behavior. If you choose alternative B, you will take part in a Journal of Personality and Social Psychology, 52, 500–510. random drawing procedure with exactly a two-thirds Miller, P. M., & Fagley, N. S. (1991). The effects of framing, problem probability of losing all of your money, and one-third variations, and providing rationale on choice. Personality and So- probability of not losing any of your money. cial Psychology Bulletin, 17, 517–522. Which of the two alternatives would you favor? Neale, M. A., Huber, V. L., & Northcraft, G. B. (1987). The framing of negotiations: Contextual versus task frames. Organizational Be- Note. Each item in a parenthesis represents the havioral Behavior and Human Decision Processes, 39, 228–241. amount of money used for a different subject group. Payne, J. W., Bettman, J. R., & Johnson, E. J. (1992). Behavioral decision research: A constructive processing perspective. Annual REFERENCES Review of Psychology, 43, 87–131. Qualls, W. J., & Puto, C. P. (1989). Organizational climate and deci- Bazerman, M. H., Magliozzi, T., & Neale, M. A. (1985). Integrative sion framing: An integrated approach to analyzing industrial buy- bargaining in a competitive market. Organizational Behavior and ing decisions. Journal of Marketing Research, 26, 179–192. Human Decision Processes, 35, 294–313. Real, L. (1991). Animal choice behavior and the evolution of cognitive Bettman, J. R., & Sujan, M. (1987). Effects of framing on evaluation architecture. Science, 253, 980–986. of comparable and noncomparable alternatives by expert and nov- Real, L., & Caraco, T. (1986). Risk and foraging in stochastic environ- ice consumers. Journal of Consumer Research, 14, 141–154. ments. Annual Review of Ecology and Systematics, 17, 371–390. Bohm, P., & Lind, H. (1992). A note on the robustness of a classical Reyna, V. F., & Brainerd, C. J. (1991). Fuzzy-trace theory and fram- framing result. Journal of Economic Psychology, 13, 355–361. ing effects in choice: Gist extraction, truncation, and conversion. Cosmides, L., & Tooby, J. (1992). Cognitive adaptations for social Journal of Behavioral Decision Making, 4, 249–262. exchange. In J. H. Barkow, L. Cosmides, & J. Tooby (Eds.), The Roszkowski, M. J., & Snelbecker, G. E. (1990). Effects of ‘‘framing’’ adapted mind: Evolutionary psychology and the generation of cul- on measures of risk tolerance: Financial planners are not immune. ture (pp. 163–228). New York: Oxford Univ. Press. Journal of Behavioral Economics, 19, 237–246. Fagley, N. S. (1993). A note concerning reflection effects versus fram- Schneider, S. L. (1992). Framing and conflict: Aspiration level contin- ing effects. Psychological Bulletin, 113, 451–452. gency, the status quo, and current theories of risky choice. Journal Fagley, N. S., & Miller, P. M. (1987). The effects of decision framing of Experimental Psychology: Learning, Memory, and Cognition, 18, on choice of risky vs certain options. Organizational Behavior and 1040–1057. Human Decision Processes, 39, 264–277. Schneider, S. L., Levin, I. P., & Gaeth, G. J. (1995, November). The / a709$$2654 10-10-96 22:11:07 obhas AP: OBHDP  FRAMING EFFECTS: DYNAMICS AND TASK DOMAINS 157 three faces of framing: Risk preference, evaluation, and persuasion. Wagenaar, W. A., Keren, G., & Lichtenstein, S. (1988). Islanders and Paper presented at the 36th annual meeting of the Psychonomic hostages: Deep and surface structures of decision problems. Acta Society, Los Angeles, CA. Psychologica, 67, 175–189. Schneider, S. L., & Lopes, L. L. (1986). Reflection in preferences Wahlund, R. (1989). Att fatta beslut under osäkerhet och risk. Stock- under risk: Who and when may suggest why. Journal of Experi- holm: Norstedts Förlag. mental Psychology: Human Perception and Performance, 12, 535– Wang, X. T. (1996a). Domain-specific rationality in human choices: 548. Violations of utility axioms and social contexts. Cognition, 60, 31– Shoorman, F. D., Mayer, R. C., Douglas, C. A., & Hetrick, C. T. (1994). 63. Escalation of commitment and the framing effect: An empirical Wang, X. T. (1996b). Evolutionary hypotheses of risk-sensitive investigation. Journal of Applied Social Psychology, 24, 509–528. choice: Age differences and perspective change. Ethology and Soci- Travis, C. B., Phillippi, R. H., & Tonn, B. E. (1989). Judgment heuris- obiology, 17, 1–15. tics and medical decisions. Patient Education and Counseling, 13, Wang, X. T., & Johnston, V. S. (1995). Perceived social context and 211–220. risk preference: A reexamination of framing effects in a life–death Tversky, A., & Kahneman, D. (1981). The framing of decisions and decision problem. Journal of Behavioral Decision Making, 8, 279– the psychology of choice. Science, 211, 453–458. 293. Received: March 25, 1996 / a709$$2654 10-10-96 22:11:07 obhas AP: OBHDP