Looking at Both Sides of the Coin: Mixed Representation Moderates Attributeframing Bias in Written and Auditory Messages

Transcription

1 Applied Cognitive Psychology, Appl. Cognit. Psychol. 30: (2016) Published online 12 January 2016 in Wiley Online Library (wileyonlinelibrary.com) DOI: /acp.3203 Looking at Both Sides of the Coin: Mixed Representation Moderates Attributeframing Bias in Written and Auditory Messages HAMUTAL KREINER* and EYAL GAMLIEL Behavioural Science Department, Ruppin Academic Centre, Emek Hefer, Israel Summary: Objects and events are often evaluated more favourably when presented in a positive frame than when presented in the complementary negative framing. Recent studies show that this attribute-framing bias can be moderated when both positive and negative frames are represented in the message. Most attribute-framing studies used written messages, although important messages are often conveyed auditorily. Unlike written messages, recipients cannot reread auditory messages and have to rely on their memory when evaluating them; consequently, the moderating effect of mixed representation may depend on memory constraints. The current study compared the framing bias in single-attribute versus mixed-attribute representations in written and auditory messages. In both written and auditory messages, single-attribute representation yielded substantial framing bias whereas mixed-attribute representation moderated the bias. The results are discussed in terms of the role of memory and attention in the attribute-framing bias. Theoretical and practical implications are considered, and future research is suggested. Copyright 2016 John Wiley & Sons, Ltd. INTRODUCTION Following Tversky and Kahneman (1981), many studies have examined framing biases in various contexts, demonstrating that the mere presentation of the same glass as half full or half empty affects judgement and decision-making (Keren, 2010). These studies have demonstrated the systematic reversals of preferences that people exhibit in alternate framings of problems, contingencies or outcomes. These reversals have important practical implications, but perhaps the most important question in terms of applied implications is how the framing bias can be moderated. In the current paper, we investigated the moderating effect of using mixed representation of both the negative and the positive aspects of the critical attribute. Critically, most of the research on framing bias, including the few studies that have investigated the moderating effect of mixed representation, used written messages, disregarding the fact that everyday messages are often communicated orally. Unlike written messages, messages presented auditorily cannot be reread; hence, memory and attention processes may operate in a different way when making decisions or judgments about information provided auditorily. These differences may be more salient when mixed representations are used, because they contain more information, making it more difficult to retain in memory. Thus, the current study focused on the moderating effect of mixed representations of data and examined in what ways this moderation is affected by the medium of presentation: auditory versus written. Single versus mixed representations of the critical attribute Reviewing the abundant research on framing biases, Levin, Schneider, and Gaeth (1998) discerned different types of framing messages. The current study focuses on attributeframing messages, in which a message refers to an object *Correspondence to: Hamutal Kreiner, Behavioural Science Department, Ruppin Academic Centre, Emek Hefer, Israel. hautalk@ruppin.ac.il or event by presenting either the positive outcome of a central attribute or the complementary negative outcome of the same attribute. For example, Levin and Gaeth (1988) showed that people evaluated the quality of ground beef as better when it was labelled as 75% lean (positive framing) relative to 25% fat (negative framing). Such framing shows a robust bias, with positive framing typically resulting in more favourable evaluations of objects or events, relative to negative framing (for reviews, see Levin et al., 1998; Piñon & Gambara, 2005). For ethical or other considerations, in some contexts, information should be presented in a balanced way that refrains from biassing recipients judgement and decisionmaking. For example, medical authorities need to provide patients and their families with information about the success and failure rates of medical interventions in order to enable them to make informed and unbiased decisions. In order to prevent the bias, it is important to underpin the processes that generate it. Kahneman (2011) proposed the What You See Is All There Is principle to explain general biases in judgement and decision-making, including attribute framing. According to this principle, framing bias results from relying on available information and disregarding complementary information, such that when presented with a 90% success rate, recipients tend to ignore the complementary (albeit not presented) 10% failure rate and vice versa. Following this principle, it is only reasonable to hypothesise that by representing both the negative and the positive aspects of the message, namely using the mixed representation, the framing bias will be moderated. Nevertheless, very few framing studies have examined the effect of using mixed representation (e.g. Bigman, Cappella, & Hornik, 2010). In an attribute-framing study that asked participants to evaluate a human Papilloma Virus (HPV) vaccine and rate their intention to recommend it to their acquaintances, Bigman et al. (2010) examined whether a mixed representation of the critical attribute produced framing bias. The experiment included four framing conditions of written messages: the vaccine was described as follows: (i) effective against HPV strains that cause 70% of cervical cancers (positive framing); (ii) Copyright 2016 John Wiley & Sons, Ltd.

2 Looking at both sides of the coin 333 ineffective against HPV strains that cause 30% of cervical cancers (negative framing); (iii) mixed representation with either the positive preceding the negative; or (iv) mixed representation with the order reversed. The findings of this study were inconclusive because the typical bias in the nonmixed conditions was not replicated in all parameters. Consequently, the attenuation of the bias in the mixed conditions could be interpreted as associated with different factors not necessarily related to the mixed representation. In a different line of research, focusing on risky-choice framing, several studies compared single versus mixed representations (O Connor et al., 1985; O Connor, 1989; Kühberger, 1995; Kühberger & Gradl, 2013; Kühberger & Tanner, 2010; Peters & Levin, 2008). Unlike attribute framing where the framing is typically based on representing a single aspect of the critical attribute, in risky-choice problems, such as the Asian disease, participants are typically asked to rate their preferences on a scale that presents a single outcome (200 out of 600 people saved) on one end and a mixed representation (one-third probability that 600 people will be saved and a two-thirds probability that no people will be saved) on the other end. The framing is manipulated by presenting the outcomes either in positive (people saved) or in negative (people die) terms. Despite identical expected values, most participants prefer the risk-free alternative when the problem is framed positively and the risky alternative when framed negatively. Theoretical accounts of riskychoice framing bias, such as prospect theory, focus on the effect of framing on the amount of risk that people are willing to accept. Recent studies, however, have focused on the comparison between the single-outcome representation, used to describe the risk-free option, and the mixed-outcome representation, used to describe the risky alternative. For example, Kühberger and Gradl (2013) presented participants with choice and/or evaluation tasks of four different options: riskfree options represented as a single outcome in either positive or negative framing, or risky options represented as mixed outcome in probabilities of positive or negative framing. Their findings revealed a framing bias when comparing the positive and negative riskless options, with this bias being eliminated when comparing the positive and negative risky options. Kühberger and Gradl (2013) related the valence consistent shift found in the riskless options to the fact that they were described in terms of a single attribute (e.g. people saved or people dying), whereas the risky options are described in terms of multiple outcomes (saved and not saved; dying and not dying). Building on these findings, we hypothesised that in attribute-framing scenarios, mixed-outcome representation would moderate attributeframing bias relative to single-attribute representation. More recently, Gamliel and Kreiner (2013) examined how attribute-framing bias is affected by mixed representation of the critical attribute. Their findings showed that mixed representation of both positive and negative outcomes of the critical attribute moderated attribute-framing bias, although it did not eliminate it. This pattern was found for written messages, when the data relating to positive and negative outcomes of the critical attribute were visually presented, either in text or in figures. Interestingly, these results suggested that even with mixed representation, the attribute framing was not totally eliminated. Surprisingly, merely presenting one frame before the other was sufficient to yield some framing bias, suggesting that in mixed representation, the order of presentation may be crucial. These findings were consistent with findings from other types of judgement and decisionmaking tasks, suggesting that the order in which information is communicated to decision-makers influences preference construction (e.g. Krosnick, Miller, & Tichy, 2004; for a review, see Weber & Johnson, 2009). The mixed representation seems to have an advantage in moderating attribute-framing bias. However, the mixed representation may have a drawback, in that it conveys more information and, consequently, it may increase memory load. Memory constraints have been assumed to be one of the underlying factors that drive the heuristic thinking associated with various biases in judgments and decision-making. For example, the representativeness and the availability heuristics were directly related to particular memory mechanisms (e.g. Tversky & Kahneman, 1974). Various memory mechanisms may be involved in judgement and decision-making, but most relevant to the present study is working memory. The findings about the relationship between working memory and biases in judgement and decision-making are inconsistent. While some studies have failed to show a correlation between working-memory span and judgments (Whitney, Rinehart, & Hinson, 2008), others have shown that, under certain conditions, such a correlation can be demonstrated. For example, Corbin, McElroy, and Black (2010) showed that higher working-memory capacity predicted greater bias in risky-choice framing problems. Most relevant to us, however, is Dougherty and Hunter s (2003) study, where participants were asked to make judgments under different time-constraint conditions. They found that when there was no time constraint, the correlation between working-memory span and judgments was not significant, but when the time allotted for making judgments was constrained, the correlation was significant. Thus, it seems that working memory affects judgments only in the context of a time constraint. The effects of working memory under time-constraint conditions may be particularly important when recipients need to evaluate information presented auditorily, because information can be maintained in memory only for a limited time. Attribute framing: mode of presentation While in daily interaction information is often communicated orally, most studies investigating attribute-framing bias have presented recipients with a written message. Recently, however, a few framing studies that used auditory presentation suggested that the mode of presentation may play a role (Seta, McCormick, Gallagher, McElroy, & Seta, 2010; McCormick & Seta, 2012a, 2012b). These studies focused on auditory presentation of vignettes as a means for selective enhancement of hemispheric differences in information processing. Their findings build upon prior work (McElroy & Seta, 2004) and show that in both goal framing (McCormick & Seta, 2012a) and attribute framing (Seta et al., 2010; McCormick & Seta, 2012b), the speaker s voice frequency may affect the amplitude of the framing bias. Low-frequency voices, assumed to enhance right-hemisphere processing,

3 334 H. Kreiner and E. Gamliel were associated with enhanced framing bias, compared with high-frequency voices. While both attribute-framing studies (Seta et al., 2010; McCormick & Seta, 2012b) replicated previous findings showing significant attribute-framing bias for low-frequency voices, they did not find such reliable bias for high-frequency voices. These findings demonstrate that attribute framing is replicated with auditory presentation of information, yet they also suggest that the mode of presentation may affect the magnitude of the framing bias. Considering that everyday information is very often communicated orally, the dearth of studies using auditory messages is surprising. Investigating the differences between auditory and written messages seems particularly warranted in the context of applied research aimed to moderate the framing bias by using mixed representations of the critical attribute. In written messages, recipients can reread and reconsider the two aspects of the critical attribute for as long as they want, unless time constraints are set, as in Dougherty and Hunter s (2003) study. By contrast, when a message is presented orally, recipients ability to reconsider both frames is subject to memory and attention capacity. The current study attempts to address this lacuna in the literature. The current study When a written message is presented, the information is available to the recipients for rereading, allowing them to focus attention on one part of the message or the other, interchangeably. Hence, when presented with a mixed representation of a written message, the recipient can consider and reconsider both aspects of the critical attribute in a manner that seems to moderate attribute-framing bias (Gamliel & Kreiner, 2013). In contrast, when a message is presented auditorily, information decays quickly, and, as it is only available for a very short time, it may be more susceptible to working-memory constraints. Consequently, the recipients ability to shift their focus of attention and consider both aspects of the critical attribute is constrained. Thus, the current study was designed to examine the effects of single versus mixed representations of the critical attribute on framing bias when information is presented auditorily. Building on previous findings, we hypothesised that single representation of the critical attribute would result in a typical bias, with higher evaluations for positive compared with negative framing. We further hypothesised that mixed representation of both positive and negative aspects of the critical attribute would moderate the attribute-framing bias. Previous findings have suggested that the order of presentation in a mixed representation affected the direction of the bias, such that the first aspect presented determined the bias (Gamliel & Kreiner, 2013). It is not clear, however, whether similar results should be expected with auditory presentation of messages. On the one hand, it may be that, as in written messages, attention would be focused on the first aspect presented, leading to primacy bias. On the other hand, because of memory constraints, the first aspect presented may decay or be overridden by the second aspect, inevitably leaving the latter presented aspect as the focus of attention, thus leading to recency bias. EXPERIMENT 1 Method Participants Participants were 146 undergraduate students (M = 23.4; SD = 1.6; 87% were women) in an Israeli college, participating in the study as a course requirement. Design and procedure The experiment consisted of a between-subject design that included two framing conditions (positive versus negative) two critical attribute representation conditions (single outcome versus mixed outcome). The single-outcome condition presented either positive or negative information to participants. In the mixed condition, both positive and negative information was presented, and their order was manipulated, such that in one condition, the positive frame preceded the negative frame, and in the other condition, the negative frame was presented prior to the positive frame. Each participant was randomly assigned to either the positive-framing or the negative-framing condition of the single-outcome representation condition (n = 37 and n = 35, respectively) or to one of the mixed-representation conditions (n = 37 for positive first and n = 37 for negative). The experiment was conducted in Hebrew in a research laboratory, using computerised questionnaires, and monitored by an experimenter. Attribute-framing vignettes were recorded and presented auditorily to the participants, using headphones. Materials To allow comparison of our findings with the results obtained by Gamliel and Kreiner (2013), we used an identical vignette. Table 1 presents the message recorded for each of the four experimental groups (the same female voice was used for all messages; both single-outcome messages lasted 13 seconds, and both mixed-outcome messages lasted 16 seconds). After listening to the recorded message, Table 1. The text presented auditorily in the four conditions of experiment 2 Attribute representation Positive (only/first) framing Negative (only/first) framing Single outcome Mixed You know a driving instructor, 85% of whose students pass their driving test the first time they take it. You know a driving instructor, 85% of whose students pass their driving test the first time they take it, while 15% fail. You know a driving instructor, 15% of whose students fail their driving test the first time they take it. You know a driving instructor, 15% of whose students fail their driving test the first time they take it, while 85% pass.

4 Looking at both sides of the coin 335 participants were asked to rate their recommendation regarding the following question: would you recommend this driving instructor to your acquaintance? The answers were given on a 7-point scale, ranging from 1 (I definitely will not) to 2 (I will not), 3 (I probably will not), 4 (maybe I will, and maybe I will not), 5 (I probably will), 6 (I will) and 7 (I definitely will). The recommendation intention question was identical for all participants in the four experimental conditions, and it appeared on the same screen in which the auditory message was presented, allowing participants to answer it once they heard the message. Results and discussion Figure 1 presents the means of participants recommendation intentions as a function of attribute representation and framing. The results show that in the single-outcome condition, participants exhibit higher recommendation intentions in the positive-framing versus negative-framing conditions. Interestingly, in the mixed condition, a recency effect was found, indicating that participants exhibited higher recommendation intentions when the negative outcome preceded and the positive outcome followed. The attribute-framing effect sizes were medium large and statistically significant for both the single-outcome [Cohen s d = 1.14; t(70) = 4.00, p <.001] and the mixed-outcome [d = 0.61; t(72) = 2.71, p =.008] conditions. A two-way analysis of variance (ANOVA) predicting the recommendation intentions from attribute representation and framing revealed a significant interaction, F(1, 142) = 23.30, p <.001, η 2 =.14; the attribute-framing effect was not significant, F(1, 142) = 1.98, p =.16, η 2 =.014; and the effect of attribute representation was significant, F(1, 142) = 5.78, p =.017, η 2 =.039. These findings show large attribute-framing effect size when a single attribute is presented in auditory messages, replicating previous findings in auditory messages (Seta et al., 2010; McCormick & Seta, 2012a, 2012b), as well as previous findings obtained using written messages (for detailed review, see Levin et al., 1998) or figures (Gamliel & Kreiner, 2013). Moreover, mixed-outcome representation in auditory messages moderated attribute-framing bias in our study, similar to previous findings obtained when either written messages (e.g. Bigman et al., 2010) or figures were used (Gamliel & Kreiner, 2013). Critically, however, previous research using written messages with mixed-outcome representation showed a primacy effect, in which the first frame determined the valence of the bias, whereas the current study, using auditory messages, showed a recency effect, in which the last frame determined the valence of the bias. Unlike in written messages, information in auditory messages decays quickly and may be more susceptible to workingmemory constraints. Consequently, old information may be overridden by new information or simply decay, while the most recent information is more likely to be retained in memory when judgments are made. According to this account, the framing bias is generated by the unbalanced representation of the information in working memory, because of the stronger memory traces of recent information. To further examine this interpretation, we conducted experiment 2 that included a change-detection task as a memory check to examine the differences in memory traces between single and mixed representations of the outcomes and between messages presented in written or auditory form. Change-detection methods require the participants to detect a modification in visual (Rensink, 2002) or text (Sanford, 2002) stimuli and have been shown to be sensitive to semantic depth of processing, working-memory constraints and focus of attention. In addition, although experiment 1 suggested that mixed representations in auditory messages may yield somewhat different results than in written messages, the difference was revealed in an indirect comparison with previous findings (Bigman et al., 2010; Gamliel & Kreiner, 2013) that differed in various methodological factors, such as different samples and different scenarios. Experiment 2 was designed to allow a direct comparison between written and auditory messages in single-framing and mixed-framing conditions. Finally, experiment 2 included two different scenarios; each scenario was evaluated using several rating items, in order to allow better generalisation and more reliable measurements. EXPERIMENT 2 Figure 1. The means of recommendation intentions in experiment 1 as a function of the framing and the attribute representation (error bars represent one standard error) Method Participants Four-hundred participants recruited through Amazon Mechanical Turk (MTurk) (M age = 32.66; SD age = 10.95; 65.8% were male) completed the study, for a compensation of $0.50. Participants were prescreened to include only US residents, aged 18 years or above, and were required to have

5 336 H. Kreiner and E. Gamliel completed at least 500 previous tasks on MTurk with a success rate of at least 95%. Design and procedure The experiment comprised a between-subject design that manipulated two presentation modes (written versus auditory messages) two framing (positive versus negative) two critical attribute representation (single outcome versus mixed outcome) conditions. Each participant received two different vignettes that presented two different stories in the same experimental condition. The study was presented as dealing with judgments and evaluations, and each participant was presented with two vignettes in one of the eight conditions. Materials The first vignette Lasagna was adopted from the study of Levin, Gaeth, Evangelista, Albaum, and Schreiber (2001). The following sentence was presented to all participants, either written or auditory (the same female voice was used for all auditory messages; the message lasted 12 seconds): lasagna is a typical Italian dish. The main ingredients needed for preparing it are pasta, tomato sauce, ground beef and mozzarella cheese. To make sure that participants were listening to the auditory messages, participants in all conditions were asked to summarise this information briefly in their own words. Afterwards, they were presented with the following written sentence: In the following screens, you will be presented with some information about lasagna and asked to answer several questions. According to their condition, participants were presented with the following text, in either written or auditory form (both single-outcome messages lasted 13 seconds, and both mixed-outcome messages lasted 15 seconds): assume that you are inviting a special friend to dinner and that you are making your favourite lasagna dish with ground beef. You are in the store, and you are offered beef that is (80% lean/20% fat; 80% lean and 20% fat; 20% fat and 80% lean). Following this message, participants were asked three 7-point scale questions: how would you describe the beef? (1 = greasy and 7 = greaseless); how would they evaluate the beef? (1 = low quality and 7 = high quality); would you buy the beef? (1 = certainly not and 7 = certainly). Participants were then presented with a written changedetection question, in which the text they either read or heard was presented, replacing the 80% lean with 70% lean and/or the 20% fat with 30% fat. All participants were asked whether this description is identical to the description they read or heard before (yes or no) and how confident they were about their response (0 100%). The second vignette was adopted from the study of Levin (1987), presenting participants with a description of a basketball player. The following sentence was presented to all participants, in either written or auditory form (the message lasted 10 seconds): basketball is a popular sport in the USA. The National Basketball Association league is considered the best in the world. All participants were asked to briefly summarise this information in writing in their own words. Afterwards, all were presented with the following written sentence: in the following screens, you will be presented with some information about a basketball player and asked to answer several questions. According to their condition, participants were presented with the following text, in either written or auditory form (both single-outcome messages lasted 13 seconds, and both mixed-outcome messages lasted 15 seconds): assume that you are the coach of a basketball team during an important game. Tom makes 70% of his free throws (single-positive) / misses 30% of his free throws (single-negative) / makes 70% and misses 30% of his free throws (mixed-positive first) / misses 30% and makes 70% of his free throws (mixed-negative first). In the following screen, participants were asked three 7-point scale questions: how would you describe Tom s probability to make a free throw? (1 = low and 7 = high); how would you evaluate Tom as a free-throw shooter? (1 = poor and 7 = excellent); assume that your team was given a free throw. Would you choose Tom to shoot it? (1 = certainly not and 7 = certainly). Participants were then presented with a written change-detection question, in which the text they read or heard was presented, replacing the 70% with 60% and/or the 30% with 40%. All participants were asked whether this description is identical to the description they read or heard before (yes or no) and how confident they were about their response (0 100%). All participants were first presented with the lasagna story, followed by the basketball player story. Results and discussion All participants supplied answers that revealed they had understood the two vignettes they read or heard. Examining each of the six questions (three questions for each of two vignettes) separately revealed a similar pattern of results with respect to the three independent variables. As the six questions had relatively high internal reliability (α = 0.74), we averaged them to one index of response. Figure 2 presents the means of participants responses as a function of presentation mode, attribute representation and framing. Participants exhibited higher responses in the positive framing relative to the negative framing in all four conditions. The attribute-framing effect sizes were large and statistically significant in the single-outcome conditions for both written [Cohen s d = 0.93; t(108) = 4.87, p <.001] and auditory [d = 1.18; t(87) = 5.56, p <.001] messages. In contrast, only marginal and nonstatistically significant attribute-framing effect sizes were found for the mixedoutcome conditions for both written [Cohen s d = 0.16; t(104) = 0.83, p =.41] and auditory [d = 0.09; t(87) = 0.45, p =.65] messages. A three-way ANOVA predicting the response index from presentation mode, attribute representation and framing revealed that the three-way interaction was not significant, F(1, 392) = 1.24, p =.27, η 2 =.003; the two-way interactions between the framing and the presentation mode and between attribute representation and presentation mode were also nonsignificant, F(1, 392) = 0.63, 3.21, p =.43,.07, η 2 =.002,.008, respectively. The two-way interaction between framing and attribute representation was significant, F(1, 392) = 22.00, p <.001, η 2 =.053, indicating differential framing effect in the two attribute representation conditions: single versus mixed. The presentation mode did not have a significant effect, F(1, 142) = 1.11, p =.29, η 2 =.003. In

6 Looking at both sides of the coin 337 Figure 2. The mean response in experiment 2 as a function of presentation mode, framing and attribute representation (error bars represent one standard error) contrast, both attribute framing [F(1, 142) = 35.39, p <.001, η 2 =.083] and attribute representation [F(1, 142) = 9.29, p =.002, η 2 =.023] had significant effects. The ANOVA results confirmed the pattern revealed in Figure 2: framing affected the responses only in the singleoutcome representation, with the effect being similar for both written and auditory messages. In contrast, framing did not affect the responses in the mixed-outcome representation neither for written nor for auditory messages. Note that the response index in all four conditions of the mixed-outcome representation was much closer to the values of the positive than the negative conditions of the single-outcome representation. In order to examine the hypothesis that there were differences in memory traces in the different experimental conditions, we examined participants responses to the change-detection questions and their confidence ratings. The answers to the questions following each of the two vignettes revealed similar patterns with respect to the experimental conditions, and these were averaged to form an index of correct answers to the change-detection questions and an index of the confidence ratings. Figure 3 presents the proportion of correct change detection, and Figure 4 presents the mean confidence ratings as a function of the experimental conditions. Taken together, these figures show that in all conditions, more than 90% of the answers were correct and the mean confidence ratings exceeded 90%, suggesting a ceiling effect. Two three-way ANOVAs predicted the two indices from presentation mode, attribute representation and framing. The three-way interactions, all two-way interactions and all the three main effects were small and were not statistically significant, F(1, 392) = , p =.13.95, η 2 = General discussion This study has several important findings. First, the results replicate previous research indicating that single representation of the critical attribute results in substantial attributeframing bias when presented auditorily (Seta et al., 2010; McCormick & Seta, 2012a, 2012b) and that the magnitude of attribute-framing bias is similar for auditory and written messages. Most importantly, however, the results of both experiments show that in mixed-outcome representation, attribute framing is moderated compared with single-outcome representation. These findings are consistent with previous findings that showed similar moderation in written messages (e.g. Bigman et al., 2010) as well as when the information was presented in the form of figures (Gamliel & Kreiner, 2013). Although the results of the two experiments revealed moderated attribute framing in mixed-outcome compared with single-outcome representations, the pattern of the results was somewhat different: whereas the framing bias was only moderated in the mixed-outcome conditions of experiment 1, it was completely eliminated in the corresponding conditions of experiment 2. A possible explanation for these differences may be related to differences in participants memory and attention in the two experiments. At the beginning of experiment 2, participants were required to briefly summarise the first message they encountered. This was performed to make sure that they had listened to the auditory message (to avoid technical problems related to the auditory peripherals). However, participants may have inferred that this was a memory study and may have made an intentional effort to remember the information presented

7 338 H. Kreiner and E. Gamliel Figure 3. The mean correct change detection in experiment 2 as a function of presentation mode, framing and attribute representation (error bars represent one standard error) Figure 4. The mean confidence rating for the change detection in experiment 2 as a function of presentation mode, framing and attribute representation (error bars represent one standard error)

8 Looking at both sides of the coin 339 to them; this participant inference might have been reinforced by the change-detection question presented after the first scenario. These methodological issues can explain the very high detection rates and confidence ratings in all experimental conditions, which are higher than the rates reported in previous studies that used text-change detection (e.g. Sanford, 2002, reported a detection rate of between 15% and 50%). Hence, it is possible that the different pattern of results in experiments 1 and 2 is related to participants intentional efforts to remember and even rehearse the messages in experiment 2, compared with experiment 1. If this interpretation is correct, it suggests that when participants invest a cognitive effort in attending and remembering both positive and negative aspects of the message, the framing bias is totally eliminated. However, when they do not make such an effort, mixed representation only moderates the framing bias, as in experiment 1. This interpretation is consistent with Kahneman s (2011) principle, explaining the attribute framing, but suggests that it is What You Remember Is All There Is rather than What You See Is All There Is. The possible role of memory and attention in mediating the moderation of attribute-framing effect in mixed-outcome representation is tentative and requires further investigation as the two experiments differ with respect to several methodological aspects, such as the language used Hebrew in experiment 1 versus English in experiment 2 and the laboratory setting of experiment 1, in which several participants answered the questionnaire simultaneously, versus the individual setting in experiment 2, in which each participant completed the questionnaires at the time and place of their choosing. Future research is needed to examine the possibility that memory and attention operate in a different fashion in response to written versus auditory messages in attribute-framing tasks. Such research could shed more light on the conditions in which mixed-outcome representation eliminates attribute framing altogether. Our findings are consistent with previous findings of studies that investigated the risky-choice framing bias using the Asian disease and similar problems (Kühberger, 1995; Kühberger & Gradl, 2013; Kühberger & Tanner, 2010; Peters & Levin, 2008). Consistent with our findings, the typical findings of these studies show a framing bias in the single-outcome (risk free) conditions, which was eliminated in the mixed-outcome (risky) conditions. Thus, although the previous studies were aimed at examining biases related to risk judgments and evaluations, they found a moderating effect of mixed-outcome versus single-outcome representations, consistent with the findings of the current study that relate to attribute framing. The similar moderating effects suggest the possibility of a common moderating mechanism. Kühberger and Gradl (2013) have also suggested that the mechanisms underlying the framing bias in risky-choice and attribute-framing problems are similar: our data suggest that the risky-choice framing effect is at least in part a misnomer: its source is a difference in evaluation rather than choice, and its content can be any attribute, not just risk (p. 116). The possibility of moderating attribute-framing bias and even eliminating it using mixed-outcome representation has important practical implications, both for decision-makers and for policymakers. The findings clearly show that attribute-framing bias in auditory presentation is similar to the bias shown in written messages when a single-attribute representation is used. Because auditory presentation is ubiquitous in everyday communication, it is important to draw recipients attention to this bias. Moreover, in attempting to prevent attribute-framing bias and in striving to present information in a more balanced manner, policymakers sometimes employ a mixed representation, in which both positive and negative outcomes are represented. The current findings suggest that in order for mixed representation to eliminate the bias, both aspects need to be remembered and attended to. Policymakers should be aware of these findings when they wish to present information in a neutral way (such as when providing medical information) or in a manner that promotes a particular outcome (such as endorsing health behaviours). Finally, decisionmakers are not always aware of the cognitive processes that enhance biases in judgments and decision-making; it may be helpful to inform them about the effect of memory and attention constraints on framing bias in auditory presentation of information. Importantly, using written messages is not a magic solution, as written messages using mixed representation may show similar attenuated, but not eliminated, bias (Gamliel & Kreiner, 2013) when recipients do not make a deliberate effort to remember the information. Future research is needed to investigate more deeply the different order effects found in written versus auditory messages and to reveal the cognitive mechanisms underlying them. With respect to auditory presentation, it would be interesting to investigate the possible effects of prosodic characteristics of vocal presentation. In other contexts (Sanford, Sanford, Molle, & Emmott, 2006), it has been shown that prosodic stress may focus attention on particular information in the text. Similar effects may be hypothesised for attributeframing messages, particularly when they include mixed representation of the critical attribute. Thus, prosodic stress that focuses the recipients attention on one aspect or another in a mixed-representation scenario is hypothesised to affect the bias towards the stressed frame. ACKNOWLEDGEMENTS The authors declare that no potential conflicts of interests with respect to the authorship and/or publication of this article exist. The authors have received no financial support for the research and/or authorship of this article. The authors wish to thank Lior Ginat and Tomer Savir for their assistance in collecting the data used in this research, and to Hedy Gertler and Nitzan Kapach for their assistance in recording the auditory messages. REFERENCES Bigman, C. A., Cappella, J. N., & Hornik, R. C. (2010). Effective or ineffective: Attribute framing and the human papillomavirus (HPV) vaccine. Patient Education and Counseling, 81, S70 S76. DOI: /j. pec

9 340 H. Kreiner and E. Gamliel Corbin, J., McElroy, T., & Black, C. (2010). Memory reflected in our decisions: Higher working memory capacity predicts greater bias in risky choice. Judgment and Decision Making, 5(2), Dougherty, M. R., & Hunter, J. (2003). Probability judgment and subadditivity: The role of working memory capacity and constraining retrieval. Memory & Cognition, 31(6), Gamliel, E., & Kreiner, H. (2013). Is a picture worth a thousand words? The interaction of visual display and attribute representation in attenuating framing bias. Judgment and Decision Making, 8(4), Kahneman, D. (2011). Thinking, fast and slow. New York: Macmillan. Keren, G. (2010). On the definition and possible underpinnings of framing effects: A brief review and a critical evaluation. In G. Keren (Ed.), Perspectives on framing (pp. 3 33). New York: Psychology Press. Krosnick, J. A., Miller, J. M., & Tichy, M. P. (2004). An unrecognized need for ballot reform: Effects of candidate name order. In A. N. Crigler, M. R. Just, & E. J. McCaffery (Eds.), Rethinking the vote: The politics and prospects of American election reform (pp ). New York: Oxford University Press. Kühberger, A. (1995). The framing of decisions: A new look at old problems. Organizational Behavior and Human Decision Processes, 62(2), DOI: /obhd Kühberger, A., & Gradl, P. (2013). Choice, rating, and ranking: Framing effects with different response modes. Journal of Behavioral Decision Making, 26(2), DOI: /bdm.764. Kühberger, A., & Tanner, C. (2010). Risky choice framing: Task versions and a comparison of prospect theory and fuzzy-trace theory. Journal of Behavioral Decision Making, 23(3), DOI: /bdm.656. Levin, I. P. (1987). Associative effects of information framing. Bulletin of the Psychonomic Society, 25, Levin, I. P., & Gaeth, G. J. (1988). How consumers are affected by the framing of attribute information before and after consuming the product. Journal of Consumer Research, 15(3), Levin, I. P., Gaeth, G. J., Evangelista, F., Albaum, G., & Schreiber, J. (2001). How positive and negative frames influence the decisions of persons in the United States and Australia. Asia Pacific Journal of Marketing and Logistics, 13(2), Levin, I. P., Schneider, S. L., & Gaeth, G. J. (1998). All frames are not created equal: A typology and critical analysis of framing effects. Organizational Behavior and Human Decision Processes, 76(2), McCormick, M., & Seta, J. J. (2012a). Lateralized goal framing: How selective presentation impacts message effectiveness. Journal of Health Psychology, 17(8), DOI: / McCormick, M., & Seta, J. J. (2012b). A new method for selectively enhancing hemisphere processing: Voice frequency amplification influences the strength of attribute framing. Laterality: Asymmetries of Body, Brain and Cognition, 17(6), DOI: / X McElroy, T., & Seta, J. J. (2004). On the other hand am I rational? Hemispheric activation and the framing effect. Brain and Cognition, 55(3), O Connor, A. M. (1989). Effects of framing and level of probability on patients preferences for cancer chemotherapy. Journal of Clinical Epidemiology, 42(2), O Connor, A. M., Boyd, N. F., Tritchler, D. L., Kriukov, Y., Sutherland, H., & Till, J. E. (1985). Eliciting preferences for alternative cancer drug treatments. The influence of framing, medium, and rater variables. Medical Decision Making: An International Journal of the Society for Medical Decision Making, 5(4), Peters, E., & Levin, I. P. (2008). Dissecting the risky-choice framing effect: Numeracy as an individual-difference factor in weighting risky and riskless options. Judgment and Decision Making, 3(6), Piñon, A., & Gambara, H. (2005). A meta-analytic review of framing effect: Risky, attribute and goal framing. Psicothema, 17(2), Rensink, R. A. (2002). Change detection. Annual Review of Psychology, 53(1), Sanford, A. J. (2002). Context, attention and depth of processing during interpretation. Mind & Language, 17(1-2), Sanford, A. J., Sanford, A. J., Molle, J., & Emmott, C. (2006). Shallow processing and attention capture in written and spoken discourse. Discourse Processes, 42(2), Seta, J. J., McCormick, M., Gallagher, P., McElroy, T., & Seta, C. E. (2010). Voice frequency impacts hemispheric processing of attribute frames. Journal of Experimental Social Psychology, 46(6), DOI: /j.jesp Tversky, A., & Kahneman, D. (1981). The framing of decisions and the psychology of choice. Science, 211(4481), DOI: /science Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), DOI: / science Weber, E. U., & Johnson, E. J. (2009). Mindful judgment and decision making. Annual Review of Psychology, 60, Whitney, P., Rinehart, C. A., & Hinson, J. M. (2008). Framing effects under cognitive load: The role of working memory in risky decisions. Psychonomic Bulletin & Review, 15(6),