௠ Academy of Management Journal 2010, Vol. 53, No. 2, 411–431.

AFFECT AND THE FRAMING EFFECT WITHIN INDIVIDUALS OVER TIME: RISK TAKING IN A DYNAMIC INVESTMENT SIMULATION

MYEONG-GU SEO BRENT GOLDFARB University of Maryland

LISA FELDMAN BARRETT Boston College

We examined the role of affect (pleasant or unpleasant feelings) and decision frames (gains or losses) in risk taking in a 20-day stock investment simulation in which 101 participants rated their current feelings while making investment decisions. As pre- dicted, affect attenuated the relationships between decision frames and risk taking. After experiencing losses, individuals made more risky choices, in keeping with the framing effect. However, this tendency decreased and/or disappeared when loss was simultaneously experienced with either pleasant or unpleasant feelings. Similarly, individuals’ tendency to avoid risk after experiencing gains disappeared or even reversed when they simultaneously experienced pleasant feelings.

One of the most commonly cited deviations from only small to moderate size and that nearly a quarter rational decision making is what is commonly re- of the effects examined (28%) were either nonsignif- ferred to as the “framing effect”—that is, the ten- icant or in the direction opposite to prediction. dency for people to avoid risk when a decision is A growing body of research suggests that human framed in terms of potential gains—and to instead affect1 is an important individual-level factor that increase risk when a choice is framed in terms of influences both risk perception and risk choice (cf. potential losses (Kahneman & Tversky, 1979; Tver- Rottenstreich & Hsee, 2001; Shiv, Loewenstein, Be- sky & Kahneman, 1991, 1992). Widely accepted in chara, Damasio, & Damasio, 2005; Slovic, Finucane, the literature as a dominant descriptive theory of Peters, & MacGregor, 2002). This literature suggests irrational choices (see Hastie and Dawes [2001] and that deviations from the framing effect can be attrib- Bazerman [2006] for reviews), the framing effect uted to affective states. Yet past research has shown has been extensively applied to understanding complex and inconsistent patterns of affective influ- firm-level risk choices and supported in various ence on risky choice (cf. Au, Chan, Wang, & Vertin- empirical settings (e.g., Audia, Locke, & Smith, sky, 2003; see Isen [2000] for a review), making it 2000; Baum, Rowley, Shipilov, & Chuang, 2005; difficult to explain how and when such affective de- Chattopadhyay, Glick, & Huber, 2001; Fiegenbaum viations may occur. For example, positive affect has & Thomas, 1988; Greve, 1998; Jegers, 1991; Lant, promoted risk seeking in some studies (e.g., Au et al., Milliken, & Batra, 1992; Wiseman & Gomez-Mejia, 2003; Isen & Patrick, 1983) but risk aversion in others 1998). Yet, despite its status, the framing effect is not (e.g., Arkes, Herren, & Isen, 1988; Isen & Geva, 1987). robust at the individual level of analysis (Sitkin & Isen and her colleagues (e.g., Isen, Nygren, & Ashby, Pablo, 1992; Wiseman & Catanach, 1997). For exam- 1988; Nygren, Isen, Taylor, & Dulin, 1996) have sug- ple, in a meta-analysis of 136 empirical studies Ku¨h- gested that one factor responsible for these inconsis- berger (1998) concluded that the framing effect was of tent findings might be a decision’s situational con- text, which in turn influences how the decision is

This research was supported by a National Science Foundation (NSF DRMS #0215509, to Lisa Feldman Bar- 1 We use “affect” as a broad and general term referring rett) and a Boston College Dissertation Research Grant (to both to various affective states, including mood, which is Myeong-Gu Seo). We give our thanks to Jason Colquitt, a prolonged and diffused affective state associated with Cindy Stevens, Ken Smith, Debra Shapiro, Subra Tangi- no particular object, and to discrete emotions, such as rala, Gilad Chen, Jeffrey Furman, Gerard Hoberg, Zur anger and fear, which are intense prototypical affective Shapira, and the anonymous reviewers for their helpful experiences directed toward certain objects (cf. Forgas, comments and feedback on this article. 1995; Russell, 2003).

411 Copyright of the Academy of Management, all rights reserved. Contents may not be copied, emailed, posted to a listserv, or otherwise transmitted without the copyright holder’s express written permission. Users may print, download or email articles for individual use only. 412 Academy of Management Journal April framed. Through a series of experimental studies, is held constant at the neutral level. Then we de- they found that people in a positive feeling state velop specific hypotheses regarding how changes (compared with those in neutral affective states) were in individuals’ affective states may influence risk more risk averse only when a possible loss appeared taking by affecting these cognitive processes or by- real and salient. However, positive feelings led to passing them entirely. greater risk seeking in other conditions (e.g., Isen et al., 1988). These findings suggest that framing effects and affective influences on risk taking are highly in- THEORETICAL BACKGROUND terrelated; perhaps one effect cannot be precisely un- AND HYPOTHESES derstood without explicitly considering the other. In this study, we contribute to research on deci- Much research has been devoted to understand- sion making under risk and to prospect theory in ing how managers and employees make decisions particular (Tversky & Kahneman, 1991, 1992), by when faced with uncertainty (see Hastie and Dawes examining the relationships between framing, af- [2001] for a review). A central finding in this long fect, and risk taking. In so doing, we explore both stream of research is the framing effect: individuals the mediating and the moderating mechanisms tend to avoid risks when experiencing gains or through which a broad range of affective states— exceeding a reference point, and they seek risks both pleasant and unpleasant feelings—systemati- when facing losses or performing below a reference cally create deviations from predicted framing ef- point. The framing effect has been mainly under- fects. We also contribute to research on affect and stood from the perspective of prospect theory (cf. decision making by demonstrating that the deci- Ku¨ hberger, 1995). As in other cognitively based sion frames of gain and loss are important situa- decision-making theories (e.g., subjective utility the- tional factors that shift the effects of pleasant and ory), in prospect theory it is assumed that risky unpleasant feelings on risk taking into functionally choices result from two cognitive judgments: (1) a opposite directions by magnifying certain types of judgment about the utility (value) of a decision out- affective influences while inhibiting others. come, and (2) a judgment about the subjective prob- We provide support for our theory in an empiri- ability of that outcome. To generate predictions con- cal setting that closely mimics actual decision mak- sistent with the framing effect, prospect theory ing under risk. Subjects were recruited from invest- suggests that individuals frame decisions relative to a ment clubs and made investment decisions that reference point, so that the marginal utility decreases had significant financial consequences (payoffs as the decision outcome deviates from this reference ranged from $100 to $1,000). These payoffs were point. (Kahneman & Tversky, 1979; Tversky & Kah- likely to trigger a wide range of pleasant and un- neman, 1991, 1992). Prospect theory also explains pleasant feelings during decision making. This sit- that the utility is further weighted by the subjective uation contrasts sharply with those in past studies probability assessments that overweight low-proba- in which student participants have often made bility outcomes and underweight high-probability choices with small or hypothetical payoffs (cf. Ku¨h- outcomes. Such weighting may overwhelm the fram- berger, 1998; Ku¨ hberger, Schulte-Mecklenbeck, & ing effect when the probability of an outcome is very Perner, 2002), potentially inhibiting the role of af- small. This feature reconciles the theory with “one- fect. In this sense, we complement the findings of in-a-million” phenomena such as the purchase of previous experimental studies that have empha- negative expected–value lottery tickets or a greater sized internal validity (cf. Ku¨ hberger et al., 2002; fear of travel on airplanes than travel in cars. Rottenstreich & Hsee, 2001; Shneideman, 2008). In In general, scholars explain the framing effect addition, our longitudinal research design followed using a consequentialist assumption that people individuals over time, allowing us to examine dy- make decisions after weighing the risks and payoffs namic, within-person relationships among frames, associated with possible choices (Loewenstein, We- affect, and risk taking, thereby extending research ber, Hsee, & Welch, 2001). As illustrated in Figure 1, findings that are predominantly based on a be- one’s wealth position relative to a reference point tween-individual (cross-sectional) research design (i.e., perceived as a gain or a loss) is subject to cogni- (Ku¨ hberger, 1998). Since real-life decisions under tive processes (e.g., utility functions and probability risk are seldom made in isolation from previous weights), which in turn determine risk taking (arrow decisions, this research design also contributes to “a”). The implicit assumption here is that any factors the predictive relevance of our study. unspecified in these relationships remain constant or We begin our theory development by focusing on do not influence risk taking. If there is a third vari- the cognitive processes underlying the framing ef- able, affect, that mediates (arrows “b” and “c”) or fect, assuming that decision makers’ affective state moderates (arrow “d”) the relationship between deci- 2010 Seo, Goldfarb, and Barrett 413

FIGURE 1 Cognitive and Affective Processes Underlying the Framing Effect

Cognitive Processes Decision Frames: a Risk Gain or Loss Taking

d c b

Affect

sion frames and risk taking, the framing effect may be strong evidence that affect influences risk taking, suppressed or amplified. although it did not investigate how risk taking Prior to further discussing the possible mediating changed with the nature and degree of affect. and moderating roles of affect, we propose the main The literature suggests two possible mechanisms— predictions of the framing effect as baseline hypoth- mediation and moderation—through which affect eses2 while assuming that decision makers’ affective may influence risk taking. We explore both these states are held constant at the neutral level: mechanisms, and to this end, we focus on a funda- mental dimension of core affect, pleasant and un- Hypothesis 1a. Gain is negatively related to pleasant feelings (Barrett, 2006a, 2006b; Russell, risk taking. 2003; Russell & Barrett, 1999), as our unit of anal- Hypothesis 1b. Loss is positively related to risk ysis to understand decision makers’ affective expe- taking. riences. Past studies on affective structure (e.g., Barrett & Russell, 1998; Russell & Carroll, 1999; Research suggests that risk perceptions and asso- Watson, Wiese, Vaidya, & Tellegen, 1999) have sug- ciated choices may be influenced by affect experi- gested that pleasant and unpleasant feelings as well enced at the moment of decision making (e.g., Isen, as their cognitive and behavioral consequences are 2000; Loewenstein et al., 2001; Raghunathan & neither independent of each other (e.g., Barrett & Pham, 1999; Rottenstreich & Hsee, 2001; Shiv et al., Russell, 1998), nor complete bipolar opposites (e.g., 2005; Slovic et al., 2002). For example, Shiv and Erber & Erber, 1994; Forgas, 1991; Raghunathan & colleagues (2005) recently found that in an experi- Pham, 1999). Therefore, we treated them as nega- mental investment setting in which expected gains tively correlated but distinctive entities in both our were greater than expected losses, brain-damaged theory development and empirical investigation. patients who could not process affective informa- tion achieved better investment performance than control group patients. The control patients, who The Mediating Role of Affect in the could process affect normally, consistently avoided Framing Effect risk after either a gain or a loss. This study provided Core affect is characterized as the constant stream of transient alterations in an organism’s 2 We do not attempt to reconcile our theory to predict neurophysiological state that represents its imme- behavior regarding very low probability events. This allows diate relationship to the flow of changing events us to rely on prospect theory to generate the unambiguous (Barrett, 2006a; Russell, 2003; Russell & Barrett, predictions in Hypotheses 1a and 1b. It also sets a boundary 1999). Pleasant and unpleasant feelings, as the condition on our theoretical interpretation. For example, if most basic dimension of core affect, summarize one assumed that some subjects thought that gains and how well/badly, positively/negatively, or appeti- losses were very low probability events, equivalent, say, to winning the lottery, then the predictions of Hypotheses 1a tively/aversively one is doing in relation to his or and 1b would be reversed (Tversky & Kahneman, 1991). We her current environment. Thus, any change in the believe this boundary condition is reasonable given the environment that is meaningful to one’s goal or well- context of our study, and perhaps generally reasonable in being is likely to induce pleasant and/or unpleasant many real-world risk-taking situations. feelings, a positive change for pleasantness and a 414 Academy of Management Journal April negative change for unpleasantness (Frijda, 1986; less unpleasant feeling, both of which will increase Lazarus, 1991; Ortony, Clore, & Collins, 1988). risk taking. Similarly, pleasant and unpleasant feel- Because individual wealth is a fundamental ele- ings are likely to mediate the hypothesized positive ment of one’s personal environment, changes in relationship between loss and risk taking (Hypothesis one’s wealth position, framed as gains or losses, are 1b) because an increased loss will lead to more un- likely to induce a range of pleasant or unpleasant pleasant and/or less pleasant feelings, both of which feelings (arrow “b” in Figure 1). In particular, gain will decrease risk taking: may be positively related to pleasant feeling and Hypothesis 2a. Gain has a positive indirect negatively related to unpleasant feeling, whereas relationship with risk taking through the mech- loss is likely to increase unpleasant feeling and anism of pleasant feelings: gain is positively decrease pleasant feeling. This view is consistent related to pleasant feelings, and pleasant feel- with many anecdotal accounts telling how inves- ings are positively related to risk taking. tors often experience strong feelings while engag- ing in their day-to-day investment activities (e.g., Hypothesis 2b. Gain has a positive indirect Babin & Donovan, 2000; Lo, 2002). Once induced, relationship with risk taking through the mech- those pleasant and unpleasant feelings are likely to anism of unpleasant feelings: gain is negatively influence decision makers’ choices in several dis- related to unpleasant feelings, and unpleasant tinctive ways (cf. Loewenstein et al., 2001; Seo, feelings are negatively related to risk taking. Barrett, & Bartunek, 2004). Hypothesis 2c. Loss has a negative indirect An accumulating body of research suggests that relationship with risk taking through the mech- affect can directly influence individuals’ choices anism of unpleasant feelings: loss is positively by bypassing and/or overpowering cognitive related to unpleasant feelings, and unpleasant processes entirely (e.g., Berridge & Winkielman, feelings are negatively related to risk taking. 2003; Gray, 1999; LeDoux, 1996; Shah, Friedman, & Kruglanski, 2002). For example, Winkielman, Hypothesis 2d. Loss has a positive indirect re- Zajonc, and Schwarz (1997) produced affectively lationship with risk taking through the mech- charged responses in liking ratings and drinking anism of pleasant feelings: loss is negatively behavior through exposure to a stimulus presented related to pleasant feelings, and pleasant feel- for 1/250 of a second, an interval so short that there ings are positively related to risk taking. is no conscious recognition of the stimulus. Like- wise, it is possible that affect may bypass cognitive The Moderating Role of Affect in the assessment of risk and directly influence risk Framing Effect choices (arrow “c” in Figure 1) and thus mediate the effect of decision frames on risk taking. If this A considerable body of literature suggests that af- bypassing occurs, the approach or avoidance orien- fective states influence cognitive processes involved tations inherent in the hedonic quality of affect (cf. in decision making under risk (arrow “d” in Figure 1), Gray, 1999; Raghunathan & Pham, 1999; Shah et al., including memory (e.g., Erber, 1991; LeDoux, 1993; 2002) may cause decision makers experiencing Meyer, Gayle, Meeham, & Harman, 1990) and judg- pleasant feelings to ”approach” risky choices and ments (e.g., Johnson & Tversky, 1983; Meyer, Gas- pursue greater risk, whereas decision makers expe- chke, Braverman, & Evans, 1992; see Forgas [1995] riencing unpleasant feelings may ”avoid” risky and Schwarz and Clore [2003] for reviews). In pio- choices and take less risk (cf. Bargh & Chartrand, neering work, Isen and her colleagues (e.g., Isen & 1999; Cacioppo, Gardner, & Berntson, 1999; Frijda, Geva, 1987; Isen & Patrick, 1983; Isen et al., 1988) and 1986; Peters & Slovic, 2000). Supporting these ar- later others (e.g., Finucane, Alhakami, Slovic, & John- guments, Au and his colleagues (Au et al., 2003) son, 2000; Loewenstein et al., 2001; Rottenstreich & found in two foreign exchange trading experiments Hsee, 2000; Slovic et al., 2002) have suggested that that traders experiencing pleasant feelings placed affect experienced at the moment of decision making larger bets, but those with unpleasant feelings might influence cognitive processes associated with made more conservative choices. risk choices in two ways. First, affect may influence Accordingly, we hypothesize that pleasant and un- subjective probability judgments (e.g., Finucane et al., pleasant feelings mediate the relationships between 2000; Loewenstein et al., 2001; Rottenstreich & Hsee, decision frames and risk taking. Moreover, this me- 2000; Slovic et al., 2002). Scholars have found a gen- diation may work in a direction that is functionally eral effect known as mood congruence judgment (e.g., opposite to the hypothesized negative relationship Meyer et al., 1992), whereby individuals in pleasant between gains and risk taking (Hypothesis 1a) be- affective states tend to perceive positive events or cause a gain will lead to more pleasant feeling and/or outcomes as more likely (cf. Rottenstreich & Hsee, 2010 Seo, Goldfarb, and Barrett 415

2001; Wegener & Petty, 1996), whereas those in un- likely to shift the relative positions of the opposing pleasant affective states assign greater likelihood to effects. In a series of experimental studies, they found negative events or outcomes (e.g., Johnson & Tversky, that risk aversion dominated both when individuals 1983). These effects may occur because pleasant and experienced positive, as opposed to neutral, feelings, unpleasant feelings involve encoding and retrieving and when possible losses were real and salient. In positive and negative memories from the brain that contrast, risk taking became dominant when possible indirectly influence one’s judgment of the likelihood losses were not salient. These results suggest that of positive or negative events (cf. Erber, 1991; Le- decision frames may determine when we might ex- Doux, 1993; Meyer, Gayle, Meeham, & Harman, pect mood congruence or mood maintenance/repair 1990). In addition, pleasant and unpleasant feelings to dominate by making future gains or losses appear convey evaluative information regarding whether a more or less salient to decision makers. situation is safe or problematic, which may directly In particular, the mood congruence effect is likely influence judgments of the probability of positive or to be dominant when individuals’ current decision negative events (Schwarz, 1990; Schwarz & Clore, frame of gain or loss matches the valence of their 1983; 1988). Second, pleasant and unpleasant feel- current feelings, which may happen when pleasant ings may influence the utility judgments associated feelings are experienced after gains or, alternatively, with possible outcomes (e.g., Damasio, 1994; Finu- when unpleasant feelings are experienced after cane et al., 2000; Isen et al., 1988; Seo et al., 2004). For losses. To explain, individuals who experience a gain example, scholars have found two general effects of may also perceive future gains as more likely because pleasant and unpleasant feelings rooted in the two the possibility is more accessible in memory. More- distinctive motivational impetuses of seeking plea- over, they may simultaneously view future losses as sure and avoiding displeasure (Wegener & Petty, unlikely because these are perceived as less relevant 1996). The mood maintenance effect is a tendency to future events. When such individuals also experience maintain current pleasant feelings by avoiding situa- pleasant feelings, the mood congruence effect may tions that may take away those pleasant feelings (e.g., strengthen the salience of the future gains, but the Forgas, 1995; Isen, 2000). Formally, such behavior mood maintenance effect should be minimized be- implies increased utility associated with avoiding cause future losses appear remote. Conversely, the possible losses, and all else being equal, greater risk mood maintenance/repair effect should dominate aversion (Isen et al., 1988; Nygren et al., 1996). The when the current decision frame mismatches the va- mood repair effect is a tendency to behave in ways lence of the current feelings. This would happen that change current unpleasant feelings toward more when pleasant feelings are experienced after losses positive ones by actively seeking situations likely to or, alternatively, when unpleasant feelings are expe- bring about more pleasant feelings (Forgas, 1991, rienced after gains, for reasons external to the gains or 1995; Morris & Reilly, 1987). This implies increased losses. For example, a gain and unpleasant feelings utility of gains, and greater risk taking (Forgas, 1991; should prompt individuals to experience (1) a weak- Raghunathan & Pham, 1999). ened mood congruence effect because future losses Taken together, these effects suggest that affect are seen as less likely because of their recent experi- may moderate the relationship between framing ence of gain and (2) a strengthened mood repair effect and risk taking. However, the mood congruence because future gains are more salient and thus more effect on subjective probability may be functionally attractive. opposite to the mood maintenance and mood repair Below we consider four permutations—pleasant effects on utility judgments. For example, pleasant feeling amidst gains, pleasant feeling amidst losses, feeling can positively influence subjective proba- unpleasant feelings amidst losses, and unpleasant bility of possible gains and thus foster risk seeking feelings amidst gains—more systematically and through the mood congruence effect. Yet these summarize each with a testable hypothesis.3 same pleasant feelings may simultaneously pro- mote risk aversion by increasing the negative utility 3 of losses through the mood maintenance effect. We note that mood congruence effects occur in rela- Therefore, different risk behaviors may arise de- tion to both pleasant and unpleasant feelings, but only pending on whether mood congruence or mood for the mood-congruent (positive or negative) events. For example, pleasant feeling only influences the probabili- maintenance/repair is more dominant in a given ties of positive (mood-consistent) future events, not the situation (cf. Au et al., 2003; Isen, 2000). probabilities of negative (mood-inconsistent) events. In Isen and her colleagues (e.g., Isen & Geva, 1987; contrast, the mood maintenance effect is only relevant Isen & Patrick, 1983; Isen et al., 1988; Nygren et al., when individuals have pleasant feelings (one does not 1996) have provided important theoretical insights seek to maintain unpleasant moods) and only associated regarding how the decision frame of gain or loss is with possible negative (mood-inconsistent) events, 416 Academy of Management Journal April

Pleasant feelings in the midst of gains. When by magnifying probability estimates for losses, un- pleasant feelings are experienced in the midst of pleasant feeling experienced following a recent loss gains, the mood congruence effect leads individu- may moderate (attenuate) the usual risk-seeking effect als to estimate that a future gain is more likely. The of loss and prompt individuals to become more risk experienced gain makes the prospect of future averse. Thus, our next hypothesis is: gains more salient, magnifying the effect. In con- trast, although the mood maintenance effect may Hypothesis 3c. Unpleasant feelings attenuate increase the utility associated with losses, cur- the positive relationship between loss and risk rently experienced gain makes potential losses taking. seem less salient. This minimizes the mood main- Unpleasant feelings in the midst of gains. When tenance effect on the utility of losses. As a result, individuals have unpleasant feelings following a re- mainly owing to enhanced probability estimates for cent gain, the prospects of future gains become more future gains, pleasant feelings experienced follow- salient and real. This prominence amplifies the mood ing recent gains moderate (attenuate) the usual risk- repair effect and increases the utility of pursuing such aversion framing effects of gains and lead to greater salient and positive future events. Moreover, the re- risk taking. Thus, we hypothesize: cent experience of gains makes future losses seem Hypothesis 3a. Pleasant feelings attenuate the less salient. The mood congruence effect is de- negative relationship between gain and risk pressed, as individuals are less likely to focus on taking. nonsalient negative events and their probabilities. As a result, unpleasant feelings experienced in the midst Pleasant feelings in the midst of losses. When of gains may moderate (attenuate) the usual tendency pleasant feelings are experienced following recent for individuals to become risk averse following gains losses, future losses become salient, which in turn and instead prompt them to become risk seeking. may prompt individuals to avoid such salient nega- Therefore, we hypothesize: tive events and subsequently increase the utility as- sociated with losses (i.e., a mood maintenance effect). Hypothesis 3d. Unpleasant feelings attenuate Since a current loss position makes potential gains the negative relationship between gain and risk seem less salient in the future, however, the individ- taking. uals’ tendency to perceive future gains more likely is diminished (a mood congruence effect), as is the re- METHODS lated risk-seeking tendency. As a result, pleasant feel- ings experienced within a loss frame may moderate To explore the role of framing and affect in risk (attenuate) the typical risk-taking orientation taking, as part of a larger data collection (e.g., Seo & prompted by losses in such a way that individuals Barrett, 2007; Seo & Ilies, 2009), we developed and become more risk averse. Thus, we predict: ran an internet-based stock investment simulation combined with an experience sampling procedure Hypothesis 3b. Pleasant feelings attenuate the (e.g., Barrett, 1998; Barrett & Barrett, 2001; Feldman, positive relationship between loss and risk 1995). Experience sampling procedures, in which taking. thoughts and feelings are measured at the time they Unpleasant feeling in the midst of losses. Sim- are being experienced, minimize the memory biases ilarly, when individuals experience both unpleas- that are typically observed when using retrospective ant feelings and large losses, the salience of the self-report measures (e.g., Barrett, 1997). current negative events (the losses) amplifies the The stock investment simulation consisted of 20 mood congruence effect and accordingly increases investment sessions, 1 session for each day over 20 the effect of unpleasant feeling on the subjective consecutive business days. Each day, participants probability of losses. The recent loss also makes logged onto the stock investment simulation website future gains seem less salient, which in turn weak- and viewed market and stock information pertaining ens the mood repair effect and decreases the utility of to 12 anonymous stocks, checked their current invest- pursuing nonsalient future gains. As a result, mainly ment performance (which also determined their mon- etary reward after the simulation), and finally, made investment decisions about which and how many whereas the mood repair effect is activated only when shares of stocks to buy or sell for the day. All infor- individuals feel unpleasant (one does not seek to repair a mation was updated daily after the public market good mood) and only associated with possible positive closed, and one set of decisions was allowed each (mood-inconsistent) events. This argument leaves us to day. Importantly, just before making investment de- focus only on these four possible interactions. cisions, participants reported their current affect. 2010 Seo, Goldfarb, and Barrett 417

Each participant was rewarded at the end of the minded that the task was strictly individual based simulation conditional on full participation. Re- and that thus they must not discuss it with or get wards were distributed as a function of perfor- advice from any other person during the entire mance relative to the local stock index (expressed period of the simulation. They were then asked to as the difference in percentage between an individ- describe their current day, including the amount of ual’s current stock portfolio value and its initial time they had spent to catch up on the stock market value), and ranged from $100 to $1,000 ($100, less and their initial prediction of the day’s national than 3 percent; $150, between 2.99 and 1 percent; stock market condition. $200, between 99 and .99 percent; $250, between 1 The initial screen reported the daily stock market and 2.99 percent; $300, between 3 and 4.99 per- information, including daily changes and five-day cent; $350, between 5 and 6.99 percent; $400, 7 trends of three major national stock market indexes percent and above; $500, above 7 percent, and third (e.g., the Dow Jones Industrial Average, the NAS- place among the participants; $750, above 7 per- DAQ composite index, and the S&P 500 index), and cent, and second place; and $1,000, above 7 per- of the local market index, the composite index of cent, and first place). The local stock market index the 12 anonymous stocks selected for this study.5 was a function solely of the performance of the 12 The next page contained information for these 12 stocks in the national markets; hence, subjects had stocks. Participants saw generic individual stock no influence on local prices. names (e.g., stock A, B, and C) as well as the current price (normalized to $100.00 on day 1), daily price change (%), average price change for the past five Sample days, beta coefficient (measuring a stock’s volatility We recruited 118 private stock investors for the in relation to the market), one-year stock perfor- stock investment simulation from six investment mance (the percent change in stock price over the clubs4 located in the New England area. Their ages trailing 52 weeks), the price-earnings ratio (a ratio ranged from 18 to 74 years (mean ϭ 24.7, s.d. ϭ of stock price to its trailing 12-month earnings per 13.2). As is typical in most investment clubs, the share), and company size (sales volume). Then par- majority of them were male (86 of 118, or 80%). ticipants saw a report that summarized their cur- Their investment experience was 4.3 years on av- rent investment performance and expected reward. erage (s.d. ϭ 7.4), ranging from 0 to 50 years. A total In addition, they saw their current cash balance, of 108 participants (91.5%) completed the simula- the total value of their current stock portfolio, the tion task and generated 2,059 daily observations. number and average performance of the partici- Seven participants (totaling 126 daily observations) pants in the simulation, and the performance of the were dropped because of noncompliance with in- best performer in the simulation. structions; a further 57 daily observations were On the next page, participants rated the various eliminated because of reported interruptions dur- feelings that together comprised their current affec- ing the simulation; and 6 cases were eliminated tive state (core affective feelings of pleasantness and because of data transmission errors. In all, we activation). Moving to the next page, they reported analyzed 1,870 daily observations stemming from their subjective beliefs, aspirations, and goals for var- 101 participants. ious aspects of the investment simulation. On the subsequent page, participants made their own investment decisions on which stocks to sell and Procedures which to buy for the day. Each participant was ini- For 20 consecutive business days (four weeks), tially given hypothetical cash of $10,000 and allowed participants logged onto a website once between to invest all or a part of it in any of the 12 stocks in the 6:00 p.m. and 9:00 a.m. the next morning. Upon local market with no transaction costs and as long as logging in (each using a unique code name and the cash balance did not go below zero. All mathe- password), they were reminded that they should matical calculations required for investment decision avoid interruption as much as possible while par- ticipating in the simulation. They were also re- 5 The 12 stocks were randomly selected from the na- tional stock market on the basis of risk, profitability, indus- 4 An investment club is a group of private investors who tries, and company size. The local market index was meet regularly to exchange investment information, learn tracked for several months prior to the simulation and was specific principles and techniques regarding investing, highly correlated with the national market indexes (r Ͼ .8). and/or in some cases, make investments together. Size and During the 20-day period of the simulation, the local market specific activities vary across investment clubs. index rose 14 times and fell 6 times. 418 Academy of Management Journal April making were automatically and instantly performed local market, riskier portfolios contained a few by the simulation. For reference, the current national high-beta stocks, whereas a risk-free portfolio and local market and stock information that partici- would be fully diversified over the 12 stocks. This pants had seen in the previous pages also became latter case, in which the diversification score was available on a separate web page. zero, was without risk, as it mimicked the local Before logging out, participants saw their invest- market index, which in turn determined rewards. ment summary in a table and explained the reasons The compound measure captured this intuition behind their investment decisions for that day. and hence was our preferred measure of risk. More- Then they reported whether, when, and how long over, it was superior to an additive measure that any type of interruptions had occurred during the would have assigned a certain degree of risk to a tasks for the day. This process was repeated daily perfectly diversified portfolio because of the posi- for 20 business days. tive value of its averaged beta coefficient. Gain and loss. In the simulation, performance relative to the local market determined final re- Measurement wards and was prominently reported daily to each Dependent variable: Risk taking. Risk-taking participant. Thus, we expected this to be the pri- was measured in two ways. First, we measured the mary reference point around which individuals weighted averaged beta coefficient in a subject’s stock framed their decisions. Performance was measured portfolio. The beta coefficient of each stock, which as the difference between an individual’s return participants saw every day during the simulation pe- (Vt) relative to its initial position (V1) and the mar- riod, was a measure of the volatility of the stock price ket index return (mktt) relative to its initial position in relation to the stock market (cf. Bodie, Kane, & (mkt1), per Equation 1: Marcus, 2001).6 This is a well known parameter of a V Ϫ V mkt Ϫ mkt stock’s potential risk. Second, we measured the de- ϭ t 1 Ϫ t 1 Performance1 . (1) gree of diversification in individuals’ portfolios. Di- V1 mkt1 versification is a well-known financial strategy for avoiding risk (cf. Bodie et al., 2001). As our measure Given the expected differences between risk tak- of diversification, we used Herfindahl’s index, the ing in the realms of gains and losses, we measured sum of the squares of the weighted share of portfolio the variable gain (the maximum of performancet stocks (0 Ͻ index Ͻ 1). A lower score on the index and 0) and the variable loss by taking the absolute indicates less risk taking, and a higher score indicates value of the negative scores of performance and greater risk taking. setting all other (nonnegative) scores at 0 (the min-

In addition, we created a compound measure of imum of performancet and 0). risk taking as a product of the degree of diversifi- It was important to ensure that our measures of cation and the weighted averaged beta coefficient.7 gain and loss corresponded to the frames that par- The weighted averaged beta coefficient reflects the ticipants used for their decisions. In particular, al- level of volatility of the portfolio relative to the though a performance level below the local market national market, and the degree of diversification return (a positive loss score) incurred a reduction of dampens the effects of price changes of individual monetary reward from the initial endowment of stocks relative to the local market. Thus, more di- $200, it is possible that participants perceived it as versified portfolios contain less risk, even if they a small gain rather than a loss. have the same average beta. Since rewards were To assess this possibility, we examined two sets distributed according to performance relative to the of findings, both of which suggested that our mea- sures of gains and losses represented the decision frames adopted by most participants. First, gain 6 Participants could hold a part of or the entire amount scores significantly and positively predicted pleas- of their initial capital ($10,000) as cash without investing ant feelings such as satisfaction (␤ ϭ .20, p Ͻ .001) it in any of the 12 stocks and divest their purchased and excitement (␤ ϭ .18, p Ͻ .001), whereas loss stocks into cash at any time during the simulation. We scores significantly and negatively predicted un- treated the amount of cash (noninvested money) as the ␤ ϭ amount of money invested on a low-risk (zero-beta) stock pleasant feelings such as disappointment ( .17, Ͻ ␤ ϭ Ͻ in calculating the weighted average beta coefficient. p .001) and irritation ( .13, p .001). This 7 A factor analysis (principal component extraction finding suggests that our measures of gain and loss method) showed that these two risk parameters constituted were, on average, perceived as positive and nega- one factor that explained 65 percent of the total variance tive events. Most importantly, loss scores would be with a factor loading of .80, which allowed us to use either unlikely to predict unpleasant feelings if the major- an additive or multiplicative form of aggregation. ity of the participants perceived them as small 2010 Seo, Goldfarb, and Barrett 419 gains. Second, we asked the participants to indicate laxed”) that are positively valenced and at the same the degree to which they perceived their current time centered on neutral activation (␣ ϭ .82). Un- performance as either positive progress (0 ϭ “neu- pleasant feeling was similarly computed by averag- tral,” 1 ϭ “slightly good,” 2 ϭ “notably good,” 3 ϭ ing the scores of four items (“sad,” “disappointed,” “very good”) or negative progress (0 ϭ “neutral,” “depressed,” and “irritated”) that are negatively va- 1 ϭ “slightly bad,” 2 ϭ “notably bad,” 3 ϭ “very lenced and centered on neutral activation (␣ ϭ .83). bad”) toward their current goal (a reference point). Next, we measured the two anchors of the activa- When participants performed above the local mar- tion dimension, activated and deactivated feelings. ket return, in keeping with our measure of gain, We assessed activated feeling by averaging the most of them perceived their current performance scores of four items (“aroused,” “surprised,” “inter- as positive progress toward their goal (positive ϭ ested,” and “nervous”) that are high in activation 67%, neutral ϭ 23%, negative ϭ 10%, n ϭ 716). In and neutral in pleasantness (␣ ϭ .61), and deacti- contrast, when participants performed below the vated feeling by averaging the scores of four items local market return, in keeping with our measure of (“quiet,” “still,” “calm,” and “tired”) that are low in loss, the majority perceived their current perfor- activation and neutral in pleasantness (␣ ϭ .64).8 mance as negative progress towards their goal (neg- Other controls. Although performance relative to ative ϭ 76%, neutral ϭ 16%, positive ϭ 7%, n ϭ the local market was the most obvious frame of 1,062). This pattern would not have been observed reference provided to each participant in each ses- if the goal (reference point) of the majority of par- sion, several other features of the stock investment ticipants was far below the local market return. simulation suggested that participants might have This might have indicated that their current perfor- used alternative frames of reference. To explore the mance below the local market return (losses) could robustness of the results, we explicitly measured be perceived as positive progresses. and controlled for these alternative frames. Affect. Although we focus on basic feelings of First, given that an individual’s entire perfor- pleasure or displeasure, core affect simultaneously mance history was also reported, we considered consists of another fundamental dimension, called two possible alternative reference points: immedi- “activation” or “arousal,” which refers to the de- ate performance and its three-day moving average. gree to which an individual is feeling a sense of Immediate performance was measured as the dif- energy or mobilization (Barrett, 2006a, 2006b). To- ference between an individual’s current perfor- gether, the pleasantness dimension and the activa- mance (performancet) and his or her performance tion dimension map out a broad range of affective at the previous round (performancet Ϫ 1) for each experiences into a two-dimensional circular space, period t. Immediate gain was measured by taking commonly called the affective circumplex. For ex- the positive scores of immediate performance and ample, an individual who is feeling excited is feel- holding all other (nonpositive) scores constant at 0 ing pleasantness and activation, whereas a calm (i.e., the maximum between immediate perfor- individual is experiencing pleasantness and deac- mance and 0), and immediate loss was measured by tivation. To examine the effect of the pleasantness taking the absolute value of the negative scores of dimension on risk taking independent of the acti- immediate performance and setting all other (non- vation dimension, we measured and controlled for negative) scores at 0 (i.e., the minimum between the activation dimension. immediate performance and 0). Average three-day Drawing on conceptual and empirical examina- gain was the three-period moving average of imme- tion of the core affective structure by Barrett and diate gain. Average three-day loss was the three- Russell (1998), we selected 16 items, all rated on a period moving average of immediate loss. scale ranging from 0, “not at all,” to 4, “extremely Second, participants may have had different as- so,” that captured both the pleasantness and acti- piration levels regarding their performance in this vation dimensions of the affective circumplex and stock investment simulation, and these aspiration measured the four anchors of these two dimen- levels could be used as alternative frames of refer- sions, so that the measures of one dimension were relatively neutral in the other dimension (by sam- 8 ␣ pling items that are neutral in the other dimension The reliability statistics ( ) of the activated and de- activated feelings were relative low (.61 and .63), reflect- and/or selecting equal numbers of items rated high ing a general tendency toward low reliability in measur- and low on the other dimension). In particular, we ing the activation dimension in previous studies (cf. first measured the two anchors of the pleasantness Cropanzano, Weiss, Hale, & Reb, 2003). A part of the dimensions: pleasant and unpleasant feelings. reason could be that individuals tend to attend less to Pleasant feeling was the average score of four items activation and more to pleasantness when they report (“happy,” “satisfied,” “enthusiastic,” and “re- their affective states (e.g., Barrett, 2004; Feldman, 1995). 420 Academy of Management Journal April

ence for their investment decisions. To control for where yit is the risk taking of individual i in period this possibility, every day we asked participants to t; Zit is a matrix that contains measures for individ- report to which performance level, as measured by ual i at period t of performance (broken up into gain performance relative to the local market, they as- and loss), affective state (broken up into positive pired. Participants could choose one of these per- and negative components) and, depending on spec- Ϫ Ϫ Ϫ Ϫ Ϫ centages: 5, 4, 3, 2, 1, 0, 1, 3, 5, 10, 20, or ification, their interactions; dt is a period dummy ␥ 30. We first computed performance-to-aspiration (20 periods, the first omitted); i is an individual by subtracting the reported aspiration level from effect, that is, any individual characteristic, such as the current performance level. Then, we measured risk preference, that is fixed over the 20 periods; ␩ gain-to-aspiration by taking the positive scores of and t is an idiosyncratic error term assumed to be performance-to-aspiration and holding all other uncorrelated with the other variables. If this error (nonpositive) scores constant at 0, and loss-to-aspi- term was correlated with the covariates, then the ration by taking the absolute value of the negative coefficients of interest in the vector ␤ would be ␥ scores of performance-to-aspiration and holding all biased. Including the fixed effects parameter i may other (nonnegative) scores constant at 0. mitigate such problems. For example, if an individ- Third, cash rewards to the top performers were ual was predisposed to experiencing positive affect strongly and positively skewed. Hence, some indi- by underlying personality traits and similarly pre- viduals may have evaluated their performance with disposed to risk-taking behavior, then a regression reference to the performance of the top performer. without fixed effects would incorrectly attribute The measure relative to best performer measures risk taking to affective state as opposed to underly- the absolute value of the difference between a par- ing traits. Use of a fixed effects methodology re- ticipant’s performance and the top performer’s per- moves this possibility. Similarly, by including a

formance. Since there was only one top performer series of period dummies (dt), we avoided possible in any given period, all except the top performer biases stemming from factors related to a specific experienced negative performance relative to this day (round) during the simulation period. potential reference point.9 In addition, as discussed below, we statistically RESULTS controlled for individual-level differences (e.g., gender, age, skills, personalities, etc.) and for any Table 1 presents the means and standard devia- period-related effects (e.g., weather, stock market tions of each variable and the within-person corre- movements, political events, increased learning lations among them. For most of the variables, there and comfort with the simulation itself, etc.). was substantial within-person variation (between 47 and 63 percent), which suggests that the data supported a fixed effects model. We present the Analysis fixed effects regression results in Table 2.10 All We used a fixed effects (within-subject) regres- models include period dummies (i.e., indicator sion methodology to predict risk-taking levels as a variables for 19 periods, the first omitted). function of individuals’ decision frames and affec- Prior to testing our hypotheses, we examined the tive states. The fixed effects methodology is appro- relationships between decision frames and affec- priate when there is heterogeneity associated with tive states by regressing pleasant and unpleasant a decision maker that remains constant over the feelings on the degrees of gain and loss. As reported period of analysis (Griliches & Mairesse, 1998). in models 2a and 2b in Table 2, the degree of gain This approach allowed us to control for any indi- was significantly and positively related to pleasant vidual-specific (fixed) effects that might be related feeling, whereas the degree of loss was significantly to risk-taking propensity, so all of the results re- and negatively related to pleasant feeling. Simi- ported reflect within-individual variation. larly, the degree of loss was positively and signifi- We estimated the following model: cantly related to unpleasant feeling, and the degree of gain was significantly and negatively related to un- y ϭ ␤ ϩ ␤ЈZ ϩ ␤ d ϩ ␥ ϩ ␩ , it 0 it 2 t i t pleasant feeling. These results generally supported our earlier argument that experience of gain or loss is an important antecedent of decision makers’ affective 9 Given the strong monetary incentives of our simula- tion, we also used actual (unrealized) rewards relative to states. Gain and loss together explained 8 and 6 per- the initial endowment of $200 as an alternative to the chosen frame of reference, precise performance relative to the local market. However, the results using the two 10 Our results are robust to a least-squares specifica- alternative measures were almost identical. tion and to hierarchical linear modeling (HLM). 2010 Seo, Goldfarb, and Barrett 421 cent of the within-person variances in pleasant and equation (model 2e), failing to meet a necessary unpleasant feelings, respectively. condition for mediation (Shrout & Bolger, 2002). Testing main effects. To test Hypotheses 1a and Using the path coefficients estimated above, we 1b, we regressed the single product measure of risk formally tested the mediation hypotheses with the taking to performance relative to the market index in Sobel (1982) test, which estimates the product of the form of gains and losses in model 2c. To obtain a the path coefficients comprising a given mediated baseline, so as to better interpret the within-person effect (e.g., “b” ϫ “c” in Figure 1) and thus allowed variance explained by model 2c, we first estimated a us to directly examine whether pleasant and un- model that included only the 19 period dummies (the pleasant feelings significantly mediated the influ- first is omitted) and found that they explained 2 per- ence of gain and loss on risk taking. The results cent of the within-person variance in risk-taking be- suggest that none of the hypothesized mediated havior. As reported in model 2c in Table 2, we found effects were significant. Therefore, none of the me- support for Hypothesis 1a; participants who experi- diation hypotheses were supported. enced more losses took greater risk in the simulation. Testing moderation effects. In model 2f, we re- However, gain did not significantly relate to risk tak- port our findings regarding the moderating effect of ing, and thus Hypothesis 1b was not supported. emotions on risk taking. The interactions between These frames explained approximately 7 percent of gain/loss frames and pleasant/unpleasant feelings the within-person variance in risk taking. were added. We included four separate interactions Testing mediation effects. To test the four me- as predicted by Hypotheses 3a, 3b, 3c and 3d. diation effects predicted by Hypotheses 2a, 2b, 2c, First, in keeping with Hypothesis 3a, pleasant and 2d, we first estimated the individual path co- feelings negatively and significantly attenuated the efficients that comprised the mediated effects in relationship between gain and risk taking; the ten- several regression models. As discussed above dency to avoid risk after a gain was weaker for (regarding models 2a and 2b ), all four path co- those experiencing more pleasant feelings. To get a efficients linking gain and loss decision frames sense of the nature of this interaction effect, we to affect were significant in the hypothesized direc- plotted this relationship in Figure 2a.11 As shown tions. However, as shown in model 2d, reporting regressions of risk taking on pleasant and unpleas- ant feelings, neither of these was significantly re- 11 We note that the minimum gain or loss was set at 0 lated to risk taking. These results remained nonsig- in these figures because one standard deviation less than nificant when both independent and mediating the mean was less than 0, which is not a possible score in variables were simultaneously entered into the the current measure of gain or loss.

TABLE 1 Means, Standard Deviations, Variance Proportions, and Correlationsa

Percent Variance

Variables Mean s.d. Within Between 12345678910111213

1. Risk taking 0.09 1.37 42% 58% 2. Gain 0.94 2.02 47 53 Ϫ.10 3. Loss 1.41 2.19 47 53 .08 Ϫ.46 4. Pleasantness 1.13 0.84 50 50 Ϫ.01 .23 Ϫ.34 5. Unpleasantness 0.79 0.80 44 56 .02 Ϫ.18 .23 Ϫ.48 6. Activation 0.97 0.71 56 44 Ϫ.02 .15 Ϫ.16 .40 .09 7. Deactivation 1.14 0.72 53 47 .02 Ϫ.04 Ϫ.05 .11 .25 .00 8. Relative to best 8.73 5.93 30 70 .04 .06 .32 Ϫ.26 .13 Ϫ.15 Ϫ.16 9. Gain to aspiration 0.18 0.68 44 56 Ϫ.14 .50 Ϫ.16 .07 Ϫ.03 .05 Ϫ.04 .19 10. Loss to aspiration 3.99 5.03 55 45 .10 Ϫ.42 .42 Ϫ.03 .13 Ϫ.02 .05 Ϫ.15 Ϫ.47 11. Immediate gain 0.43 0.92 23 77 .00 .29 Ϫ.17 .39 Ϫ.29 .19 Ϫ.07 Ϫ.03 .08 Ϫ.13 12. Immediate loss 0.45 0.82 26 74 .08 Ϫ.22 .42 Ϫ.35 .33 Ϫ.09 Ϫ.04 .14 Ϫ.09 .24 Ϫ.40 13. Average three-day gain 0.25 0.53 29 71 Ϫ.07 .49 Ϫ.29 .27 Ϫ.24 .15 Ϫ.06 .02 .21 Ϫ.29 .41 Ϫ.28 14. Average three-day loss 0.25 0.50 30 70 .06 Ϫ.25 .57 Ϫ.36 .28 Ϫ.14 Ϫ.05 .28 Ϫ.10 .25 Ϫ.24 .48 Ϫ.38

a n ϭ 1,870 (101 participants, 20 rounds). Means, standard deviations, and correlations were computed for each individual in all rounds and then averaged for averaged within-individual correlations. Correlation coefficients that are equal to or larger than .05 are significant at a .05 level. TABLE 2 Results of Fixed Effects Regression Analysesa

Model 2c Model 2d Model 2e Model 2f Model 2h: Model 2a: Model 2b: Model 2g: Diversification Variables Pleasantness Unpleasantness Risk-Taking: Mean Beta ؋ Diversification Index Mean Beta Index

Gain 0.06*** (5.79) Ϫ0.03** (Ϫ2.96) Ϫ0.003 (Ϫ1.34) Ϫ0.003 (Ϫ1.49) Ϫ0.02*** (Ϫ3.57) Ϫ0.11** (Ϫ2.80) Ϫ0.04*** (Ϫ4.93) Loss Ϫ0.05*** (Ϫ5.47) 0.07*** (6.86) 0.02*** (10.40) 0.02*** (10.60) 0.03*** (9.50) 0.27*** (9.49) 0.03*** (5.92) Pleasantness Ϫ0.004 (Ϫ0.82) 0.002 (0.42) 0.003 (0.46) 0.08 (1.46) 0.003 (0.31) Unpleasantness 0.001 (0.22) Ϫ0.01 (Ϫ1.13) 0.004 (0.67) 0.05 (0.83) Ϫ0.003 (Ϫ0.29) Gain ϫ pleasantness 0.01*** (3.76) 0.03 (1.49) 0.01*** (3.80) Loss ϫ pleasantness Ϫ0.01** (Ϫ3.09) Ϫ0.08*** (Ϫ4.04) Ϫ0.01 (Ϫ1.76) Loss ϫ unpleasantness Ϫ0.01*** (Ϫ3.57) Ϫ0.07*** (Ϫ4.44) 0.002 (0.77) Gain ϫ unpleasantness 0.001 (0.73) 0.01 (0.77) 0.002 (0.60)

R2 .08*** .06*** .09*** .02 .09*** .11*** .09*** .15***

a T-statistics are in parentheses. R2 reflects within-subject explained variance only. R2 significance estimates were based on F-statistics. The constant term was estimated but not reported. All regressions controlled for investment round. n ϭ 1,870 (101 participants, 20 rounds). ** p Ͻ .01 *** p Ͻ .001 2010 Seo, Goldfarb, and Barrett 423

FIGURE 2A Moderation Effect of Pleasantness on the Relationship between Gain and Risk Taking

in the figure, the framing effect of gain was strong Finally, and contrary to Hypothesis 3d, we did when participants experienced a neutral level of not find a significant interaction between unpleas- pleasantness (one standard deviation below the antness and gain in influencing risk taking. Thus, mean). However, as individuals had more pleasant Hypothesis 3d was not supported. feelings, their propensity for risk taking increased. We note that once we controlled for interactions When individuals were feeling very pleasant (i.e., between affect and gain/loss frames, the data sup- one or two standard deviations above the mean), ported Hypothesis 1a—that is, holding affective ex- the typical risk-aversion framing effect of gain was perience constant at the mean level, risk taking generally eliminated or reversed. decreased following gains. This indicates that the Second, and supporting Hypothesis 3b, pleasant lack of results in model 1c (the nonsignificant gain feelings significantly interacted with degree of loss coefficient) was observed because we had not con- and attenuated the relationship between loss and trolled for the affect of individuals who were also risk seeking. As individuals experienced more experiencing a gain. In addition, these interactions pleasant feelings despite losses, their propensity to explained 9 percent of the within-person variance make riskier choices diminished. Figure 2b illus- in risk taking, as opposed to the 7 percent that was trates the pattern of this interaction. For individu- explained by the framing effects alone.12 als feeling a neutral level of pleasantness (one stan- In models 2g and 2h in Table 2, we explored the dard deviation below the mean), a clear framing sensitivity of our results to our choice of risk mea- effect was observed (top line). However, this effect sure. In particular, we examined the effects of the diminished as they experienced a greater degree of independent variables on the beta index and the pleasantness. For individuals who experienced diversification index separately. In general, our strong (one standard deviation above the mean) findings mostly replicated those of model 2f, with pleasant feelings (bottom line), the loss effect on two exceptions. First, the moderation effects of af- risk seeking was substantially weakened. In support of Hypothesis 3c, unpleasant feelings significantly attenuated the relationship between 12 Importantly, the 2 percent increase in predictive loss and risk taking. This result suggests that the power was not a story of means, but rather a story of the propensity for taking greater risk after experiencing tails, since it was mainly driven by individuals who were a loss was weaker when individuals experienced simultaneously performing very well or poorly and were experiencing strong emotions. In our sample, such ex- more unpleasant feelings. Figure 2c shows the pat- treme cases—for example, the cases of over one standard tern of this interaction effect in more detail. For indi- deviation above the mean in both decision frames (gain viduals feeling neutral (one standard deviation below or loss) and affect (pleasant or unpleasant feeling)— the mean), risk taking increased as losses increased ranged only between 1.8 and 2.5 percent of the total (top line). However, this effect became weaker as cases, but the R2-statistic downplayed this fact, as it subjects experienced more unpleasant feelings. captures averages. 424 Academy of Management Journal April

FIGURE 2B Moderation Effect of Pleasantness on the Relationship between Loss and Risk Taking

FIGURE 2C Moderation Effect of Unpleasantness on the Relationship between Loss and Risk Taking

fect on the loss and risk-seeking relationship (Hy- explicitly took into account the activation dimen- potheses 3b and 3c) did not predict the level of sion of core affect. In an unreported regression diversification. Thus, this affective influence fol- model, we included measures of activated and de- lowing loss occurred mainly through the selection activated feelings as well as measures of their in- of less volatile stocks, rather than through in- teractions with gain and loss frames. Including creased diversification of a stock portfolio. Second, these measures had no effect on our results regard- the moderation effect of pleasant feeling on the gain ing the effects of pleasant and unpleasant feelings and risk aversion relationship (Hypothesis 3a) did on risk taking. However, the results were, in and of not predict the averaged beta; this effect occurred themselves, interesting. First, including the activa- mainly through diversifying a stock portfolio. tion and deactivation measures increased the ex- Testing activated and deactivated feelings. We planatory power of the model by an amount similar examined whether our results would change if we to that explained by the valence measures (2%). 2010 Seo, Goldfarb, and Barrett 425

Second, although activated and deactivated feel- suggests that these frames were not generally relevant ings had no direct effects, we found that risk taking to decision makers in this simulation. increased for deactivated individuals who were performing poorly. Unlike pleasant and unpleasant feelings, which generally attenuated the framing DISCUSSION effect, deactivated feeling magnified the framing effect of loss toward a greater degree of risk taking. The findings of this study first contribute to the Testing alternative reference points. Individu- literature on risk taking and the framing effect in als might evaluate performance using alternative particular (Tversky & Kahneman, 1991, 1992), in frames of reference, and moreover, the choice of which the role of affect has received relatively little frames might depend on an individual’s affective attention (cf. Rottenstreich & Hsee, 2001; Shiv et al., state (cf. Seo et al., 2004). If this situation existed, it 2005; Slovic et al., 2002). We found that affective would lead us to conflate the impact of affective experiences at the moment of decision making led state with the impact of alternative frames. In an participants to behave in ways that sometimes were unreported analysis, we conducted several robust- inconsistent with framing effects and generally miti- ness checks of our results.13 We explored four al- gated them. In particular, we found that when indi- ternative frames described in the measurement viduals who experienced large gains also had pleas- section: performance on the previous day, perfor- ant feelings, they were not risk averse but instead mance of a three-day moving average, performance were risk seeking. Conversely, when individuals si- relative to one’s aspiration level, and performance multaneously experienced pleasant feelings and large relative to the best performer. Alternative gain and losses, they became less risk seeking, and when such loss frames were computed on the basis of each individuals were feeling particularly pleasant, the alternative reference point. The influence of the framing effect of loss was mostly eliminated. More- alternative reference points and their interactions over, when individuals experienced a loss and felt with pleasant/unpleasant feelings were entered si- unpleasant, they became less risk seeking. multaneously with the default reference point and These results suggest that the framing effect can its interactions. be more completely understood only when deci- The main findings in model 2g in Table 2 were sion makers’ affective states are explicitly taken robust both statistically and qualitatively to consid- into account. Indeed, we found that the framing ering each of the four alternative reference points, effect only partially explained risk-taking behavior; and further, there was little evidence that these when we did not take decision makers’ affective alternative frames influenced decision making.14 experiences into account, only those participants In these robustness checks, we also developed and who experienced losses made riskier choices. This tested a series of models that allowed individuals to finding is consistent with the findings of Ku¨ hberg- switch between the default reference point (the local er’s (1998) meta-analysis, in which the framing ef- market return) and one of the four alternative refer- fect was not robust across empirical settings. Only ence points. We hypothesized that the switch was after accounting for the interactions between deci- related to affective states.15 We found that the switch- sion frames and decision makers’ affective experi- ing models did not fit the data well. Our failure to ences did we find general support for the predic- find robust effects for any of the alternative frames tions of prospect theory. In particular, when decision makers’ affective experiences and their interactions with decision frames were held constant 13 Results of these analyses are available upon request from the lead author. 14 One exception is that when we examined the differ- ␤ ϩ ⑀ Ͼ ence from the best performer as the reference point, the x1 1 1 if y* 0 y ϭ ͕ ␤ ϩ ⑀ Յ significance of the loss-unpleasantness interaction x2 2 2 if y* 0 dropped at the expense of the interaction between the ϭ ␥ ϩ ␩ alternative loss and pleasantness. This particular result y* z may suggest that sometimes and for some individuals, where y is the measure of risk, x1 is a vector of variables the top performance level became a relevant reference with one candidate reference point, x2 is an alternative point, in addition to local market performance. However, reference point, z is a vector of affect variables that in- when this occurred, pleasant feelings still interacted fluence which reference point is relevant for the decision with gain and loss in a predicted direction. maker, and y* is never observed and hence is latent. 15 This econometrical model is called a model with a Since a maximum-likelihood procedure did not reach a latent switching component (Porter, 1983), which is converged solution, we used a nonlinear least squares based on the following mathematical specifications: procedure in estimating the parameters. 426 Academy of Management Journal April at the mean level, greater gains led to decreased tant situational contexts that determine which of risk, whereas greater losses led to increased risk. the two conflicting affective influences dominates. Moreover, our results offer important theoretical For example, by providing a context in which po- implications for prospect theory (Kahneman & Tver- tential losses appear less salient than potential sky, 1979), in which the framing effect is mainly gains to decision makers, experiencing a gain may understood as a function of two cognitive properties: magnify the influence of pleasant feelings on their the subjective utilities and probabilities of decision probability estimates of future gains but simulta- outcomes. In particular, our finding that strong feel- neously inhibit their influence on the utility esti- ings generally attenuate the framing effect in the mates of potential losses. Consequently, pleasant realms of both gains and losses suggests several pos- feelings may promote risk-seeking propensity fol- sible mechanisms through which decision makers’ lowing gains. In contrast, pleasant feelings may foster affective states may shift the utility and probability a risk-averse tendency after an individual experi- functions. Specifically, in the realm of gains, pleasant ences a loss, because the decision frame of loss can feelings may reduce risk aversion by increasing the make potential losses appear more salient to decision subjective probabilities of future gains (through a makers and thus magnify the effect of pleasant feel- mood congruence effect). In the realm of losses, un- ings on the utility estimates of potential losses. pleasant feelings may reduce risk seeking by increas- Also as an extension to existing research, which ing the subjective probabilities of future losses (again has mainly focused on the effects of pleasant feel- through to a mood congruence effect). Alternatively, ings on risk taking, we simultaneously examined pleasant feelings may increase the subjective utilities the effects of unpleasant feelings on risk taking of further losses (through a mood maintenance effect) under varying conditions of gains and losses. How- and thus diminish the risk-seeking tendency. These ever, compared to the effects of pleasant feelings, findings are consistent with those of several previous the effects of unpleasant feelings were less appar- studies showing that affect changes risk taking by ent (only the interaction with loss was supported) shifting the probability and utility functions (e.g., and weaker in magnitude. One possible explana- Isen et al., 1988; Rottenstreich & Hsee, 2001). Future tion is that individuals may have a more differen- research is needed to examine precisely how such tiated understanding of unpleasant feelings (cf. shifts occur as a function of changes in pleasant and Fredrickson, 2001, 2003). Moreover, this may show unpleasant feelings in the realms of gains and losses. more complex patterns in how those differentiated Second, the results of this study contribute to the negative feelings (e.g., anger versus sadness or anx- literature on affect and risk taking. Scholars have iety) impact risk taking. For example, anxiety may discussed whether affective experiences influence promote risk aversion, whereas anger or sadness risk taking directly or indirectly, via decision mak- may foster risk taking (Lerner & Keltner, 2001; Ra- ers’ cognitive processes (cf. Forgas, 1995; Loewen- ghunathan & Pham, 1999). Since the present study stein et al., 2001; Seo et al., 2004). Contrary to some examined the effect of overall unpleasant affect on recent findings (e.g., Au et al., 2003; Shiv et al., risk taking, such complex patterns may have been 2005), this study failed to show a direct influence washed out. Future research is needed to determine of decision makers’ affective states on risk taking. whether this weaker support came from the nature Instead, our findings indicate that affect influences of the unpleasant feelings or from the strong main risk taking mostly via its effect on cognitive pro- effect of loss on risk taking. cesses, and more importantly, that affect (in this Finally, we found that the activation dimension in case, pleasant and unpleasant feelings) interacts core affect had as powerful an impact on risk taking with situational contexts (in this case, gains and as did the effects of the pleasant/unpleasant dimen- losses) in a complex manner. sion. In particular, our results suggest that individu- In particular, our findings support Isen and her als experiencing deactivated feelings (absence of en- colleagues’ (e.g., Isen & Geva, 1987; Isen, Nygren, & ergy) and simultaneously experiencing losses tend to Ashby, 1988; Isen & Patrick, 1983) argument that increase risk. Yet little past research has addressed the same pleasant feelings can influence the cogni- the role of the activation dimension in risk taking tive components underlying risky choice (e.g., (Cropanzano, Weiss, Hale, & Reb, 2003). This limita- probability or utility judgments) in functionally op- tion makes interpretation of our results challenging. posite directions toward risk, as evidenced by our Perhaps, just as unpleasant feelings motivate “repair- results that pleasant feelings led to increased risk ing” the unpleasantness of one’s current feelings taking in the realm of gains but decreased risk (mood repair effect), deactivated feelings may moti- taking in the realm of losses. Our results further vate repairing low energy by seeking more stimula- extend this theoretical argument by suggesting that tion, such as taking risks. This tendency may increase decision frames (gain or loss) may provide impor- when individuals experience greater losses, which 2010 Seo, Goldfarb, and Barrett 427 may signal that their current situation is problematic ments may have systematically affected our results, and thus requires them to have more energy to cope nor can we take into account the effect of choices with it. Clearly, more research is needed to explore with extremely high or low outcome probabilities, the role of activation in risk taking. which may have even overwhelmed the framing ef- fect itself (Tversky & Kahneman, 1991). Future stud- ies explicitly examining the role of decision makers’ Practical Implications subjective utilities and probabilities together with Our results directly speak to managers and employ- their affective states may not only reveal the under- ees making decisions in organizations and facing un- lying mechanisms through which decision makers’ certainty. Per framing effect predictions, as soon as decision frames and affective states interact with each decision makers perceive their current performance other in influencing risk taking, but also enable more going over or under a certain reference point, they precise examination of such effects by controlling for pursue lesser or greater degrees of risk. Our results the possible nonadditive weighting of subjective suggest that through an interaction with decision probabilities. frames, affective experiences generally mitigate these Second, our simulation is more generalizable to biases. Thus, assuming that decision frames will real-world settings, in particular to investment elicit suboptimal outcomes, a broad range of pleasant markets, than the simulations generally used in and unpleasant feelings may improve performance by prior research. However, this virtue is not costless, correcting cognitive biases that affect risk taking. This as it comes at the expense of internal validity. For argument is consistent with recent findings that af- example, our measure of risk taking is a combina- fective experiences (intensity) are functional for de- tion (product) of two major indicators, averaged cision-making performance (e.g., Seo & Barrett, 2007). beta and a diversification index. Individuals work- In particular, our findings suggest that the corrective ing in finance and investment frequently use these influences of affective experiences are stronger for indicators. The strength of this measure is that it pleasant feelings than for unpleasant feelings in both does not rely on participants’ risk perceptions, but gain and loss conditions. Interpreted broadly, our is based on concrete decision outcomes objectively study suggests that managerial practices that foster a observable in a complex and naturally occurring broad range of affective experiences at work, particu- setting. As a result, the practical significance of our larly pleasant feelings, may improve performance. measure of risk itself is likely to be high, in the Implementation of such practices requires reexamin- sense that the patterns of decision outcomes (risk ing and unlearning the dominant beliefs, norms, and taking) found here are also likely to be observed in languages in organizations that discourage experienc- similar, natural settings. However, the flip side is ing and expressing feelings and emotions (Ashforth & that, unlike in most other experimental studies, in Humphrey, 1995). which risk is assessed by survey items or a forced Despite this potentially functional role of affect choice between two hypothetical risky options, our in decision making, however, our results also sug- measure is subject to other factors besides individ- gest that extremely intense pleasant or unpleasant uals’ perceptions of risk, and thus our models ex- feelings may overcorrect and reverse the framing plain only relatively small percentages of its vari- effect. Moreover, our finding indicates that feeling ance (7 percent was explained by the framing an absence of physical energy (deactivation) may effects, and 2 percent by their interactions with magnify the framing effect of loss toward a greater affect). For example, some individuals may pur- degree of risk taking. Therefore, this study informs chase high-beta stocks simply because these stocks decision makers to be aware of such possible affec- have been rising, ignoring their beta components. tive biases in making their decisions in those situ- In addition, our correlational research design gen- ations or to avoid such situations entirely. erally makes it difficult to determine the causal relationships between the key variables. Future re- search is needed to increase the internal validity of Limitations and Future Research Directions these measures, perhaps through the use of both Our study has several limitations. First, we did not objective and subjective measures of risk. directly examine the underlying mechanisms leading Third, the remuneration structure for the partic- to the results of this study—how decision makers’ ipants in this study was designed to prevent them affective states may shift their subjective utility and from incurring losses. The worst performer still got probability functions in particular. We only inferred paid $100 (incurring a $100 loss from the initial such mechanisms from our theoretical development. baseline reward of $200). In addition, the expected Thus, for example, we cannot eliminate the possibil- gains for participants in this simulation (between ity that the nonadditive weighting of probability judg- $0 and $800 in cash remuneration) were much 428 Academy of Management Journal April greater than the expected losses (between $0 and ber of investigations to adopt an event-contingent $100). This imbalance between possible gains and experience sampling procedure in the study of losses in the remuneration structure may have bi- emotion (e.g., Barrett, 1998; Barrett & Barrett, 2001; ased the results. In particular, participants may Feldman, 1995). That is, participants’ affective have been more sensitive to rewards and less sen- states were measured at the moment they were sitive to possible losses, and thus have taken more being experienced during a meaningful event or risk than they would have taken if the remunera- task. This procedure was enabled by recent advances tion structure had been more balanced. Moreover, in internet technology that allowed many critical fea- the lack of real and serious losses in this study tures such as online measurement, dynamic session might have undermined both the psychological re- and identity control, and instant and dynamic data alism and the replicability of these results to other transfer, transformation, and presentation to both in- decision-making settings in which actual and seri- vestigators and participants. None of these would ous losses can occur. In principle, additional ex- have been possible using a traditional pen-and-paper perimental research might need to examine the key method (cf. Shneideman, 2008). We hope this meth- hypotheses of this study in a setting in which a odological advance stimulates future studies that ap- balanced range of both actual gains and losses is ply this method to investigation of affective experi- possible. However, since it may raise an ethical ence and its effects on decision making under risk. issue if experiments result in financial losses for participants, field studies are probably necessary to determine the importance of this limitation. Conclusion Fourth, as is typical in most investment clubs, This study provides strong evidence that both af- participants were predominantly male (80%) and fective experiences and decision frames are impor- young. In addition, the participants were not ran- tant factors influencing decision making under risk in domly selected but instead drawn voluntarily from an ambiguous, real-life setting. In particular, we six clubs, which violated the assumption of inde- found that in the realms of gains and losses, pleasant pendence among the respondents. The gender and feelings can completely eliminate the framing effect, age imbalance and the nonrandomness in the sam- whose operation is a central tenet of many theories, ple may have constrained the applicability of study including prospect theory. Moreover, we also found results to the general population. Additional exami- that unpleasant feelings can attenuate framing effects nation using a more gender-balanced, age-balanced, in the realms of gains. However, we found no direct and/or randomized sample would be of value. influence of affective experience on risk taking. Thus, Fifth, we did not find direct effects of pleasant our results suggest that decision makers’ feelings in- and unpleasant feelings on risk taking, in contrast fluence risky choices by interacting with situational to some previous studies of financial decision mak- factors, the decision frames of gain or loss in partic- ing (e.g., Au et al., 2003; Shiv et al., 2005). One ular. Surprisingly, we found that when such interac- possible explanation for this discrepancy is that, tions happen, feelings generally negate cognitive bi- unlike other studies using a between-individual ases associated with the framing effect. research design, we used a longitudinal panel data approach. Our results were obtained from the with- in-individual relationships between affect and risk REFERENCES taking. If the direct influence of affect on risk taking Arkes, H. R., Herren, L. T., & Isen, A. M. 1988. Role of operates through other individual-level factors, a possible loss in the influence of positive affect on relationship would disappear when those individ- risk preference. Organizational Behavior and Hu- ual factors are taken into account, as we did here. In man Decision Processes, 42: 181–193. keeping with this line of reasoning, in unreported Ashforth, B. E., & Humphrey, R. H. 1995. Emotion in the regression analyses in which we ignored individual workplace: A reappraisal. Human Relations, 48: 97– factors, we found that positive affect was associated 125. with greater risk taking, though we found no rela- Au, K., Chan, F., Wang, D., & Vertinsky, I. 2003. Mood in tionship between negative affect and risk taking. foreign exchange trading: Cognitive processes and When we analyzed each round separately, we performance. Organizational Behavior and Human found a similar association between positive affect Decision Processes, 91: 322–338. and risk taking in 3 of the 20 rounds. These results Audia, P. G., Locke, E. A., & Smith, K. G. 2000. The suggest that taking into account individual effects paradox of success: An archival and a laboratory is important. Nevertheless, more research is needed study of strategic persistence following radical envi- to determine the precise causes of this discrepancy. ronmental change. Academy of Management Jour- Finally, this is among the relatively small num- nal, 43: 837–853. 2010 Seo, Goldfarb, and Barrett 429

Babin, C. E., & Donovan, W. J. 2000. Investing secrets of the Damasio, A. R. 1994. Descartes’ error: Emotion, reason, masters: Applying classical investment ideas to to- and the human brain. New York: Avon. day’s turbulent markets. New York: McGraw Hill. Erber, R. 1991. Affective and semantic priming: Effect of Bargh, J. A., & Chartrand, T. L. 1999. The unbearable mood on category accessibility and inference. Journal automaticity of being. American Psychologist, 54: of Experimental Social Psychology, 27: 480–498. 462–279. Erber, R., & Erber, M. W. 1994. Beyond mood and social Barrett, L. F. 1997. The relationship among of momentary judgment: Mood incongruent recall and mood regu- emotional experiences, personality descriptions, lation. European Journal of Social Psychology, 2: and retrospective ratings of emotion. Personality 79–88. and Social Psychology Bulletin, 23: 1100–1110. Feldman, L. A. 1995. Valence focus and arousal focus: Barrett, L. F. 1998. Discrete emotions or dimensions? The Individual differences in the structure of emotional role of valence focus and arousal focus. Cognition experience. Journal of Personality and Social Psy- and Emotion, 12: 579–599. chology, 69: 153–166. Barrett, L. F. 2004. Feelings or words? Understanding the Fiegenbaum, A., & Thomas, H. 1988. Attitudes toward content in self- report ratings of experienced emo- risk and the risk-return paradox: Prospect theory tion. Journal of Personality and Social Psychology, explanations. Academy of Management Journal, 87: 266–281. 31: 85–106. Barrett, L. F. 2006a. Valence as a basic building block of Finucane, M. L., Alhakami, A., Slovic, P., & Johnson, S. emotional life. Journal of Research in Personality, M. 2000. The affective heuristic in judgments of risks 40: 35–55. and benefits. Journal of Behavioral Decision Mak- ing, 13: 1–17. Barrett, L. F. 2006b. Emotions as natural kinds? Perspec- tives on Psychological Science, 1: 28–58. Forgas, J. P. 1991. Mood effects on partner choice: Role of affect in social decisions. Journal of Personality Barrett, L. F., & Barrett, D. J. 2001. An introduction to and Social Psychology, 61: 708–720. computerized experience sampling in psychology. Forgas, J. P. 1995. Mood and judgment: The affective Social Science Computer Review, 19(2): 175–185. infusion model (AIM). Psychological Bulletin, 117: Barrett, L. F., & Russell, J. A. 1998. Independence and 39–66. bipolarity in the structure of current affect. Journal Fredrickson, B. L. 2001. The role of positive emotions in of Personality and Social Psychology, 74: 967–984. positive psychology: The broaden-and-build theory Baum, J. A., Rowley, T. J., Shipilov, A. V., & Chuang, Y. of positive emotions. American Psychologist, 56: 2005. Dancing with strangers: Aspiration perfor- 218–226. mance and the search for underwriting syndicate Fredrickson, B. L. 2003. The value of positive emotions: partners. Administrative Science Quarterly, 50: The emerging science of positive psychology is com- 536–575. ing to understand why it’s good to feel good. Amer- Bazerman, M. H. 2006. Judgment in managerial deci- ican Scientist, 91: 330–335. sion making (6th ed.). New York: Wiley. Frijda, N. H. 1986. The emotions. London: Cambridge Berridge, K. C., & Winkielman, P. 2003. What is an un- University Press. conscious emotion? (The case for unconscious lik- Gray, J. R. 1999. A bias toward short-term thinking in ing). Cognition and Emotion, 17: 181–211. threat-related negative emotional state. Personality Bodie, Z., Kane, A., & Marcus, A. 2001. Essentials of and Social Psychology Bulletin, 25: 65–75. investments. New York: McGraw-Hill. Greve, H. R. 1998. Performance, aspirations, and risky Brehm, J. W. 1999. The intensity of emotion. Personality organizational change. Administrative Science and Social Psychology Review, 3: 2–22. Quarterly, 43: 58–86. Cacioppo, J. T., Gardner, W. L., & Berntson, G. G. 1999. Griliches, Z., & Mairesse, J. 1998. Production functions: The affect system had parallel and integrative pro- The search for identification in practicing economet- cessing components: Form follows function. Journal rics. In Z. Griliches (Ed.), Essays in method and of Personality and Social Psychology, 76: 839–855. application: 232–411. Northampton, MA: Elgar. Chattopadhyay, P., Glick, W. H., & Huber, G. P. 2001. Hastie, R., & Dawes, R. M. 2001. Rational choice in an Organizational actions in response to threats and uncertain world: The psychology of judgment and opportunities. Academy of Management Journal, decision making. Thousand Oaks, CA: Sage. 44: 937–955. Isen, A. M. 2000. Positive affect and decision making. In Cropanzano, R., Weiss, H. M., Hale, J. M. S., & Reb. J. M. Lewis & J. M. Haviland-Jones (Eds.), Handbook of 2003. The structure of affect: Reconsidering the re- emotions (2nd ed.): 417–435. New York: Guilford lationship between negative and positive affectivity. Press. Journal of Management, 29: 831–857. Isen, A. M., & Geva, N. 1987. The influence of positive 430 Academy of Management Journal April

affect on acceptable level of risk and thoughts about Meyer, J. D., Gayle, M., Meeham, M. E., & Harman, A. K. losing: The person with large canoe has a large 1990. Toward a better specification of the mood- worry. Organizational Behavior and Human Deci- congruency effect in recall. Journal of Experimental sion Processes, 39: 145–154. Psychology, 26: 465–480. Isen, A. M., Nygren, T. E., & Ashby, F. G. 1988. Influence Morris, W. N., & Reilly, N. P. 1987. Toward the self- of positive affect on the subjective utility of gains regulation of mood: Theory and research. Motiva- and losses: It is just not worth the risk. Journal of tion and Emotion, 11: 215–249. Personality and Social Psychology, 55: 710–717. Nygren, T. E., Isen, A. M., Taylor, P. J., & Dulin, J. 1996. Isen, A. M., & Patrick, R. 1983. The effect of positive The influence of positive affect on the decision rule feelings on risk taking: When the chips are down. in risk situations: Focus on outcome (and avoidance Organizational Behavior and Human Decision of loss) rather than probability. Organizational Be- Processes, 31: 194–202. havior and Human Decision Processes, 66: 59–72. Jegers, M. 1991. Prospect theory and the risk-return rela- Ortony, A., Clore, G. L., & Collins, A. 1988. The cognitive tion: Some Belgian evidence. Academy of Manage- structure of emotions. New York: Cambridge Uni- ment Journal, 34: 215–225. versity Press. Johnson, E., & Tversky, A. 1983. Affect, generalization, Peters, E., & Slovic, P. 2000. The springs of action: Af- and the perception of risk. Journal of Personality fective and analytical information processing in and Social Psychology, 45: 20–31. choice. Personality and Social Psychology Bulle- Kahneman, D., & Tversky, A. 1979. Prospect theory: An tin, 26: 1465–1475. analysis of decision under risk. Econometrica, 47: Porter, R. H. 1983. A study of cartel stability: The joint 263–291. executive committee, 1880–1886. Bell Journal of Ku¨ hberger, A. 1995. The framing of decisions: A new Economics, 14: 301–314. look at old problems. Organizational Behavior and Raghunathan, R., & Pham, M. T. 1999. All negative moods Human Decision Processes 62: 230–240. are not equal: Motivational influences of anxiety and Ku¨ hberger, A. 1998. The influence of framing on risky sadness on decision making. Organizational Behav- decisions: A meta-analysis. Organizational Behav- ior and Human Decision Processes, 79: 56–77. ior and Human Decision Processes, 75: 23–55. Rottenstreich, Y., & Hsee, C. K. 2001. Money, kisses, and Ku¨ hberger, A., Schulte-Mecklenbeck, M., & Perner, J. electric shocks: On the affective psychology of risk. 2002. Framing decisions: Hypothetical and real. Or- Psychological Science, 12: 185–190. ganizational Behavior and Human Decision Pro- Russell, J. A. 2003. Core affect and the psychological cesses, 89: 1162–1175. construction of emotion. Psychological Review, Lant, T. K., Milliken, F. J., & Batra, B. 1992. The role of 110: 145–172. managerial learning and interpretation in strategic Russell, J. A., & Barrett, L. F. 1999. Core affect, prototyp- persistence and reorientation: An empirical explora- ical emotional episodes, and other things called tion. Strategic Management Journal, 13: 585–608. emotion: Dissecting the elephant. Journal of Person- Lazarus, R. S. 1991. Emotion and adaptation. New York: ality and Social Psychology, 76: 805–819. Oxford University Press. Russell, J. A., & Carroll, J. M. 1999. On the bipolarity of LeDoux, J. E. 1993. Emotional memory systems in the positive and negative affect. Psychological Bulletin, brain. Behavioral Brain Research, 58: 69–79. 125: 3–30. LeDoux, J. E. 1996. The emotional brain: The mysteri- Schwarz, N. 1990. Feelings as information: Informational ous underpinnings of emotional life. New York: and motivational functions of affective states. In Simon & Shuster. E. T. Higgins & R. M. Sorrentino (Eds.), Handbook of motivation and cognition: Foundations of social Lerner, J. S., & Keltner, D. 2001. Fear, anger, and risk. behavior: 527–561. New York: Guilford. Journal of Personality and Social Psychology, 81: 146–159. Schwarz, N., & Clore, G. L. 1983. Mood, misattribution and judgments of well-being: Informative and direc- Lo, A. W. 2002. The psychophysiology of real-time finan- tive functions of affective states. Journal of Person- cial risk processing. Journal of Cognitive Neuro- ality and Social Psychology, 45: 513–523. science, 14: 323–339. Loewenstein, G. F., Weber, E. U., Hsee, C. K., & Welch, N. Schwarz, N., & Clore, G. L. 1988. How do I feel about it? The 2001. Risk as feelings. Psychological Bulletin, 127: informative functions of affective states. In K. Fiedler & 267–286. J. P. Forgas (Eds.), Affect, cognition, and social behav- ior: 44–62. Toronto: Hogrefe International. Meyer, J. D., Gaschke, Y. N., Braverman, D. L., & Evans, T. W. 1992. Mood-congruent judgment is a general Schwarz, N., & Clore, G. L. 2003. Mood as information: 20 effect. Journal of Personality and Social Psychol- years later. Psychological Inquiry, 14: 296–303. ogy, 63: 119–132. Seo, M., & Barrett, L. F. 2007. Being emotional during de- 2010 Seo, Goldfarb, and Barrett 431

cision making—Good or bad? An empirical investiga- biological evidence. Journal of Personality & Social tion. Academy of Management Journal, 50: 923–940. Psychology, 76: 820–838. Seo, M., Barrett, L. F., & Bartunek, J. M. 2004. The role of Wegener, D. T., & Petty, R. E. 1996. Effects of mood on affective experience in work motivation. Academy persuasion processes: Enhancing, reducing, and bi- of Management Review, 29: 423–439. asing scrutiny of attitude-relevant information. In Seo, M., & Ilies, R. 2009. The role of self-efficacy, goal, L. L. Martin & A. Tesser (Eds.), Striving and feeling: and affect in dynamic motivational self-regulation. Interactions among goals, affect, and self- regula- tion: 329–362. Mahwah, NJ: Erlbaum. Organizational Behavior and Human Decision Processes, 109: 120–133. Winkielman, P., Zajonc, R. B., & Schwarz, N. 1997. Sub- liminal affective priming resists attributional inter- Shah, J. Y., Friedman, R., & Kruglanski, A. W. 2002. ventions. Cognition and Emotion, 11: 433–465. Forgetting all else: On the antecedents and conse- quences of goal shielding. Journal of Personality Wiseman, R. M., & Catanach, A. H. 1997. A longitudinal and Social Psychology, 83: 1261–1280. disaggregation of operational risk under changing reg- ulations: Evidence from the savings and loan industry. Shiv, B., Loewenstein, G., Bechara, A., Damasio, H., & Academy of Management Journal, 40: 799–830. Damasio, A. R. 2005. Investment behavior and the negative side of emotion. Psychological Science, 16: Wiseman, R. M., & Gomez-Mejia, L. R. 1998. A behavioral 435–439. agency model of managerial risk taking. Academy of Management Review, 23: 133–153. Shneideman, B. 2008. Science 2.0. Science, 319: 1349– 1350. Shrout, P. E., & Bolger, N. 2002. Mediation in experimen- tal and nonexperimental studies: New procedures Myeong-Gu Seo ([email protected]) is an assistant and recommendations. Psychological Methods, 7: professor in the R. H. Smith School of Business at the 422–445. University of Maryland. He received his Ph.D. in organ- Sitkin, S. B., & Pablo, A. L. 1992. Reconceptualizing the ization studies from Boston College. His research focuses determinants of risk behavior. Academy of Manage- on emotion in organizations, organizational change and ment Review, 17: 9–38. development, and institutional change. Slovic, P., Finucane, M., Peters, E., & MacGregor, D. G. Brent Goldfarb ([email protected]) is an asso- 2002. The affect heuristic. In G. D. Griffin & D. Kah- ciate professor of entrepreneurship and strategy at the neman (Eds.), Intuitive judgment: Heuristics and Robert H. Smith School of Business at the University of biases: 397–420. New York: Cambridge University Maryland. He studies how decision making under uncer- Press. tainty affects markets, particularly in new industries. He Sobel, M. E. 1982. Asymptotic confidence intervals for in- has also focused on how the production and exchange of direct effects in structural equation models. In S. Lein- technology differ from those of more traditional eco- hardt (Ed.), Sociological methodology: 290–312. nomic goods, and the implications of these differences Washington, DC: American Sociological Association. for firms and policy. Tversky, A., & Kahneman, D. 1991. Loss aversion in Lisa Feldman Barrett ([email protected]) is currently a riskless choice: A reference-dependent model. professor of psychology and the director of the Interdis- Quarterly Journal of Economics, 106: 1039–1061. ciplinary Affective Science Laboratory at Boston College, Tversky, A., & Kahneman, D. 1992. Advances in prospect as well as an assistant in research at Harvard Medical theory: Cumulative representation of uncertainty. School at Massachusetts General Hospital. Her major re- Journal of Risk and Uncertainty, 5: 297–323. search focus addresses the nature of emotion from social- psychological, psychophysiological, cognitive science, Watson, D., Wiese, D., Vaidya, J., Tellegen, A. 1999. The and neuroscience perspectives. two general activation systems of affect: Structural findings, evolutionary considerations, and psycho- Copyright of Academy of Management Journal is the property of Academy of Management and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.