The Effect of Emotions on Risk Perception : Experimental Evaluation of The Affective Tendencies Framework

The aim of this study was to assess the role of specific emotions on risk perception providing a more stringent experimental test of the Appraisal Tendencies Framework (ATF). Consistent with expectations, angry and happy participants made more optimistic risk estimates than participants who were made sad. As hypothesized by ATF, happiness and anger also led people to somewhat higher certainty appraisals than sadness. However, this change in perception did not mediate the impact of emotions on risk estimates. Taken together, our results provide the evidence for causal role of specific emotions in risk perception and contribute to literature showing that the effects of emotion on judgment are not solely due to the valence of the experienced emotion. However, they also suggest that the processes underlying emotion effects remain in need for further specifications.

Three decades of research attest that affect exerts a pervasive influence on the perception of risk (e.g., DeSteno, Petty, Wegener, & Rucker, 2000;Johnson & Tversky, 1983, see, for a review, Waters, 2009).For instance, in their influential study, Johnson and Tversky (1983) found that participants who previously read a story inducing a negative mood were less optimistic concerning the risks of various causes of death than participants from in a non-manipulated mood.In contrast, participants who had previously read a story inducing a positive mood were more optimistic.This work is of primary importance for human health as the perception of risk could be a major determinant of the engagement in risky behavior (e.g., binge drinking, smoking, unprotected sexual relationships, hazardous economic investment, risky driving, and so forth; e.g.Brewer, Chapman, Gibbons, Gerrad, & McCaul, & Weinstein, 2007;Ulleberg & Rundmo, 2003).
Despite its importance at both the theoretical and the applied levels, the role of affects in risk perception domain still remains unclear (Millstein, 2003).Three main explanations have been offered to account for these effects.The Corresponding author: dracesasa@hotmail.comassociative network theory (Bower, 1981) posits that moods activate congruent material in memory that is later used to interpret incoming information (Tversky & Kahneman, 1974).Another explanation, the affect-as-information model (see Schwarz & Clore, 2007), holds that people rely on their affective states as a source of information regarding the global status of their environment (Schwarz, 1990).According to this position, a positive affect informs people that their environment is safe whereas a negative affect signals a problematic environment.As a result, positive affect would promote optimism whereas negative affect would lead to pessimism.Although these explanations differ in the mediating mechanisms involved (mood as information versus mood as prime), they share a basic common feature: They emphasize the role of valence of affect in the production of the effects (Forgas, 1995).
This valence-based approach has been challenged by Lerner andKeltner (2000, 2001; see also Keltner, Ellsworth, & Edwards, 1993) who have generated an alternative emotion-specific approach to affect and judgment: the Appraisal-Tendency Framework (ATF).Drawing on cognitive appraisals models of emotion elicitation (e.g., Smith & Ellsworth, 1985), they hypothesized that the appraisal dimensions underlying the triggering of a specific emotion also influence the appraisal of future events.For instance, fear and anger differ markedly in appraisal dimensions of certainty and control.Whereas fear is defined by an appraisal pattern of low certainty and of situational control, anger is defined by an appraisal pattern of high certainty and of individual control (Smith & Ellsworth, 1985).Thus, these appraisals would lead individuals who experience anger to consider their environment as predictable and under control whereas those experiencing fear would assess the environment as being unpredictable and out of control.As a result, the former should consider to be less at risk (more optimistic) that the latter.To test this, Lerner and Keltner asked participants to estimate the number of annual fatalities due to 12 events that lead to a certain number of deaths each year in the United States (e.g., brain cancer, strokes, floods; Lerner & Keltner, 2000) or to estimate the likelihood that specific positive and negative events would occur in their own life compared to the lives of relevant peers (Lerner & Keltner, 2001).Consistent with ATF predictions, naturally fearful people made more pessimistic risk assessments than angry people.Moreover, in a follow-up study (Lerner and Keltner, 2001, Experiment 4), appraisals of control (but not of certainty) were shown to mediate the link from emotion condition to risk assessment.

THE PRESENT RESEARCH
As noted above, research suggests that emotions sharing the same valence influence risk assessment in different ways depending on the appraisals associated with them.However, this research seems to be characterized by two main limitations that we propose to address in the current research.First, previous work relied mainly on designs in which both the emotions and likelihood estimates were only measured (see, for an exception, Lerner and Keltner, 2001, Exp. 4).
For instance, Lerner and Keltner assessed participants' general disposition to experience specific emotions and measured the statistical relationships between these dispositions and participants' likelihood estimates.Such correlational designs can provide an assessment of the relationships between individual differences (dispositional emotions) and likelihood estimates.However, they do not allow an evaluation of the causal link between emotion and risk assessment since uncontrolled variables correlated with dispositional emotions (such as different life experiences) may predispose people to evaluate risk in different way.As a result, this theoretical explanation (in its causal aspects) is ultimately based on one experimental study in which anger was compared to fear.
Second, this specific design could have several consequences: It is unclear whether these effects could extend beyond these two affective states, and namely whether they can account for the differences observed in the literature between prototypical positive and negative affective states (i.e., happiness and sadness).Because Lerner and Keltner compared the effects of induced fear and anger, we do not know whether happiness would lead to the same effects as anger, as hypothesized by the AFT, or to diverging results as hypothesized by more traditional valence-based models.For instance, one might hypothesize that happiness would lead to different effects (more optimistic) than both anger and sadness, the effects of these negative emotions differing from each other, but to a lesser extent.
In order to address these concerns, we designed an experiment that follows a procedure similar to the one used by Lerner and Keltner (2000) with substantial changes.First, to ensure the causal role of emotions, we manipulated them directly.Second, our design includes three different emotions.Happiness and sadness were manipulated as prototypical instances of positive and negative emotions (e.g.Johnson and Tversky, 1983).A third emotion group was included in which participants were induced to feel angry.The specific features of this experimental design allow us to test directly the congruence explanation and the appraisal tendency explanations against each other.According to a congruence explanation, sad and angry participants should be less optimistic than happy ones.However, even though anger and sadness are both negative, anger is characterized by high personal control whereas sadness is characterized by low personal control and low certainty.Happiness, although of positive valence, is associated with appraisals of elevated certainty and personal control, as is anger (Smith & Ellsworth, 1985).Therefore, according to the appraisal tendency hypothesis, the appraisals of uncertainty and of situational control that define sadness should lead sad participants to make relatively pessimistic risk assessments, compared with angry and happy participants.

Method
Participants.Fifty-six psychology undergraduates (42 females, M age = 19.1 SD = 1.2) from Sarajevo University were randomly assigned in one of three emotion conditions: anger, sadness or the happiness.Participants were run individually.To dissociate the emotion induction from the likelihood estimate measures, we told participants that different researchers had pooled together their respective questionnaire packets.
Emotion inductions.The emotion induction procedure was similar as the one used by Lerner and Keltner (2001).Subjects were told that they were participating in an experiment investigating the capacity of people to imagine and remember past events.Participants were instructed to answer two open-ended questions as truthfully as possible and to provide as many details as possible.They were first asked to briefly describe three to five things that make them most angry vs. sad vs. happy.Then, they described in more details "the very situation that makes you, or has made you, the most angry (vs.sad vs. happy)."Participants were further told to write their description so that someone reading it might even get mad (vs.sad vs. happy, respectively) just from learning about the situation.

Measures
Emotions.Immediately after the induction, participants completed a commonly used emotion self-report form (see Gross & Levenson, 1995;Lerner & Keltner, 2001) in which they rated on the 4-point scale (1 = not at all, 4 = extremely) the extent to which they felt each of 16 separate emotion terms (amused, angry, anxious, disgusted, downhearted, engaged, fearful, frustrated, happy, joyful, interested, irritated, nervous, mad, repulsed, and sad).Composite measures of anger, sadness and happiness were computed by averaging participants' responses for the anger and mad items (r = .78,p <.05), downhearted and sad items (r = .65,p <.05) and happy and joyful items (r = .78,p <.05).
Appraisal measures of certainty and control.We created three self-report items for each of the two appraisal dimensions (Lerner & Keltner, 2001;Smith & Ellsworth, 1985).The control items assessed participants' views about the extent to which the events they described were under individual versus situational control (α = .50).The certainty items assessed the extent to which the event described was predictable and certain versus unpredictable and uncertain (α = .55).For each item, participants responded on a scale ranging from 1 (not at all) to 6 (very much).
Likelihood estimates.The questionnaire assessing likelihood estimates presented eight positive (e.g., "to receive statewide recognition in my profession", "to marry someone wealthy"), and eight negative events (e.g., to contract sexually transmitted disease, to be sued by someone) issued from previous research (see Drace, Desrichard, Shepperd & Hoorens, 2009;Drace, Ric, & Desrichard, 2010;Lerner & Keltner, 2001).For each event, participants were provided likelihood estimates on a scale ranging from 0% (not likely) to 100% (very likely).We reversed scores for negative items and then combined all items into one optimism score (α = .65),with higher scores indicate that participants displayed more optimistic view of their future.

Manipulation check
To test the discreteness of the induced emotions in each emotion measure we conducted one-way ANOVAs with two orthogonal contrasts: a planned comparison testing the model and a contrast testing the remaining variance (i.e., the only contrast that should not be significant if the model tested fit the data).
We then test the mediating role of appraisals in the relation between emotion inductions and likelihood estimates (Baron & Kenny, 1986).We report only the analysis of the mediating role of appraisals of certainty since we did not find the expected effects of emotion on appraisals of control.However, the effect of appraisal of certainty on likelihood estimates was far from significance (b = -0.01,t <1) and controlling the appraisals of certainty does not reduce the impact of emotions on likelihood estimates.1

DISCUSSION
The aim of this study was to replicate and extend the findings of Lerner and Keltner (2000) in a design that provides a more stringent test of the AFT.Our results are only partially consistent with the hypotheses derived from the AFT.At the judgmental level, participants who were made angry or happy were more optimistic that participants who were made sad.These findings are important because previous evidence for such pattern were only correlational.This experiment provides thus further evidence for a causal role of emotions in these effects.
The results are however at odds with other predictions derived from the AFT and namely those related to the underlying processes.As hypothesized, happiness and anger led people to be more certain about the events than sadness.However, this change in perception does not mediate the impact of emotions on likelihood estimates.Thus, these results question the processes that are hypothesized to underlie the effects of emotion on likelihood estimates.This absence of mediation is particularly important as our procedure closely followed the one used by Lerner and Keltner with highly similar materials and that, except mediation, all other results are consistent with the AFT hypotheses.In a nutshell, the emotion inductions appear to be effective and seem to have an impact both on likelihood estimates and on appraisals of certainty and of control.Importantly, if the appraisals of certainty are consistent with appraisal theories of emotion, this is not the case for control.In contrast with what could be expected, angry participants indeed found the situation more controllable than both happy and sad participants.Thus, it would suggest that the difference between happiness and sadness generally observed in the literature (e.g., Johnson & Tversky, 1983) could not be attributed to appraisal of control and certainty.
One could question the measurement of appraisals as the alphas were modest.However, we tested the mediation with each of the items that were used in this experiment to measure appraisals.None of the analyses tend to be consistent with the mediation-by-appraisals hypothesis.
However, it is worth noting that other mediational processes have been proposed.DeSteno and colleagues (2000) have proposed that the effects of emotions on likelihood estimates could be due to their use as informative cues about the state of the environment (e.g., Schwarz, 1990).According to this position, anger informs people that events associated with anger (e.g., violence) are likely to occur.In contrast, sadness should inform of the problematic nature of the environment, leading people to overestimate the likelihood of sad events.However, this first position is unlikely to account for our results as this position would predict that happy participants should evaluate the environment at less at risk than others: a prediction that is totally at odds with our results.
Finally, we know from previous research that emotions influence the way people process information (e.g., Schwarz & Clore, 2007).Sadness is generally associated with more systematic processing, analysis of argument quality rather than reliance on simple credibility heuristics (e.g., Bless, Bohner, Schwarz, & Strack, 1990;Bodenhausen, Sheppard, & Kramer, 1994), and judgments that are generally more accurate (Alloy & Abramson, 1979).Findings such as these indicate that sad persons may be generally less susceptible to common biases and shortcomings of human inference and judgment, presumably because of their tendency toward more extensive processing of judgment-relevant information.Therefore, it is possible that participants in sad condition provided more realistic (less biased) likelihood estimates than those in happiness and anger conditions.However, in the absence of objective standards for comparison, this reasoning remains speculative and should be tested in future research.
Thus, although our study contributes to the growing literature showing that dimensions of emotions other than valence may have as much (or more) impact as (than) valence (DeSteno et al. 2000;Keltner et al., 1993;Lerner & Keltner, 2000, 2001;Tiedens & Linton, 2001), we may acknowledge that the processes underlying these effects remain in need for further specifications.

Figure 1 .
Figure 1.Optimism scores for each emotion condition