Emotional Dynamics in Trust Behavior: Supplemental 1
Supplemental Materials
Trust Against all Odds? Emotional Dynamics in Trust Behavior
by T. Schlösser et al., 2015, Decision
Notes on Methods and Measures
Self-Assessment Manikin (Study 1)
The Self-Assessment Manikin (SAM) asks respondents how they feel along three emotional dimensions: valence (positive versus negative), arousal (calm versus aroused), and dominance (not in control versus in control), each assessed by a non-verbal 5-point scale. For a visual depiction of the SAM, see Figure 1 (p. 760) in Schlösser, Dunning, & Fetchenhauer (2013). The SAM allows for consistent measurement across cultures (Morris, 1995) and enables the researcher to capture much of the variation and range in human emotional experience in just a few measures (Mehrabian & Russell, 1974; Russell & Mehrabian, 1977)—important in our case because we to ask about two immediate emotion scenarios plus four anticipated ones. As several researchers have established, it is also a valid measure of emotion (Arcos, Verdejo-García, Peralta-Ramírez, Sánchez-Barrera, & Pérez-García, 2005; Bradley, Codispoti, Cuthbert, & Lan, 2001; Güntekin & Basar, 2007; Hillmann, Rosengren & Smith, 2004; Lang, Bradley, & Cuthbert, 1997).
Specific Emotions Measured in Study 2
We conducted a survey of the literature on trust games and risky decisions to identify relevant emotions we should assess. For immediate emotions, we identified 13 specific emotions that we distilled into 5 different dimensions. Among those dimensions, we naturally featured social agitation (guilt, anxious, remorse, tense), emotions that are associated with fulfilling social duties (Higgins, 1987) and which in our previous work has found to be a strong and reliable predictor of trust (Dunning, Anderson, Schlösser, Ehlebracht, & Fetchenhauer, 2014; see also De Hooge, Nelissen, Breugelmans, & Zeelenberg, 2011). We also assessed fear (fearful, afraid), which has been implicated in perceptions of risk (Kugler, Connolly, & Ordóñez, 2012; Nelissen, Dijker, de Vries, 2007) and also regret on its own (Martinez & Zeelenberg, 2015; Zeelenberg et al., 1998). We also measured “approach” emotions such as interest (excited, interested) and their opposite in the form of indifference (indifferent, bored, calm, relaxed)—both of which have been associated with desire in consumer and political choice (Druckman & McDermott, 2008; Mano, 1994; Mellers, 2000) and decision conflict (Elliot & Devine, 1994).
For anticipated emotions, participants rated how they would feel along 9 different emotional items, representing 5 different dimensions. We featured betrayal as one dimension (betrayed, anger) because of its obvious relevance to the notion of betrayal aversion one finds in the economic literature on trust (Bohnet & Zeckhauer, 2004; Kugler et al., 2012). We also assessed contentment (pleased, content, joy, delight) (Haselhuhn & Mellers, 2005; Mano, 1994; Mellers, Ritov, & Schwarz, 1999), regret(Bell, 1982; Loomes & Sudgen, 1982; Martinez & Zeelenberg, 2015) and disappointment (Bell, 1985; Loomes & Sudgen, 1982;Wu, 1999; Zeelenberg et al., 1998), as well as guilt (De Hooge, Nelissen, Breugelmans, & Zeelenberg, 2011; Ketelaar & Au, 2003; Nelissen et al., 2007; Pelligra, 2011; Tangney, Stuewig, & Mashek, 2007) as separate constructs.
Emotional Dynamics in Trust Behavior: Supplemental 1
Table S1
Cluster Descriptions and Associated Trust Rates for Each Immediate and Anticipated Emotional Scenario (Study 1)
Average Rating forEmotional Scenario / % of Participants / Valence / Arousal / Dominance / % Trust
Immediate Emotion
A Trusts / Cluster 1 / 56.7 / .12 / .17 / .11 / 65.5
Cluster 2 / 43.3 / -.28 / .71 / -.35 / 47.6*
A Keeps / Cluster 1 / 38.8 / .59 / -.58 / .43 / 26.3
Cluster 2 / 61.2 / -.06 / -.37 / -.18 / 76.7**
Anticipated Emotions
A Trusts & Trust Honored / Cluster 1 / 24.7 / .50 / -.21 / .31 / 45.8
Cluster 2 / 75.3 / 1.0 / -.19 / .58 / 61.6
A Trusts & Trust Violated / Cluster 1 / 29.6 / -.33 / -.29 / .03 / 62.1
Cluster 2 / 38.8 / -.63 / .62 / -.07 / 63.2
Cluster 3 / 31.6 / -.97 / .74 / -.76 / 45.2
A Keeps & Trust Honored / Cluster 1 / 54.1 / .14 / -.32 / .13 / 52.8
Cluster 2 / 45.9 / -.49 / .21 / -.24 / 62.2
A Keeps & Trust Violated / Cluster 1 / 43.9 / .19 / -.15 / -.05 / 67.4
Cluster 2 / 56.1 / .86 / -.32 / .60 / 49.1*
Clusters differ within a scenario on trust rates, *p < .10, **p < .05
Table S2Summary of Binary-logistic Regression Analysis Predicting Risk-Seeking from Game Type (Trust vs. Coin Flip) and Specific Immediate Emotions (Study 2)
Model 1 / Model 2 / Model 3 / Model 4 / Model 5 / Model 6
Game type / social agitation / fear / regret / indifference / interest
Variable / B / Odds Ratio / B / Odds Ratio / B / Odds Ratio / B / Odds Ratio / B / Odds Ratio / B / Odds Ratio
Immediate Emotions
A keeps / 1.15*** / 3.17 / .42 / 1.52 / .85*** / 2.34 / .34 / 1.40 / -.15 / .87
A risks / -1.39*** / .25 / -.66*** / .52 / -.75** / .47 / -1.49*** / .23 / .29 / 1.34
Game / trust game vs. coin flip / 1.39*** / 4.02 / -.48 / .62 / 1.21** / 3.35 / 1.27** / 3.56 / .88 / 2.41 / 1.59*** / 4.91
constant / -.76 / .47 / 1.51 / 4.50 / .61 / 1.83 / -.99 / .37 / 4.78* / 118.57 / -1.64 / .19
Nagelkerke's R² / .13 / .43 / .29 / .37 / .31 / .15
Significant effect on the decision to risk/trust: * p<.10 ** p<.05 *** p<.01
Increase in predicitive power (ΔR): Model 1 to Model 2: sig. , to Model 3: sig. , to Model 4: sig., to Model 5: sig., to Model 6:n.s.
Table S3
Summary of Binary-logistic Regression Analysis Predicting Risk-Seeking from Game Type (Trust vs. Coin Flip) and Specific Anticipated Emotions (Study 2)
Model 1 / Model 2 / Model 3 / Model 4 / Model 5 / Model 6
Game type / betrayal / contentment / regret / disappointment / guilt
Variable / B / Odds Ratio / B / Odds Ratio / B / Odds Ratio / B / Odds Ratio / B / Odds Ratio / B / Odds Ratio
Anticipated Outcome
Risk & Lose / -.58† / .56 / .91 / 2.48 / -.56** / .57 / -.35 / .70 / -1.05*** / .35
Keep & Lose / .18 / 1.12 / -.66† / .52 / .51† / 1.66 / -.56 / .57 / 1.54** / 4.68
Risk & Gain / a / 1.18** / 3.26 / .00 / 1.00 / -.56 / .57 / .56 / 1.75
Keep & Gain / .11 / 1.11 / -.62† / .54 / .56 / 1.75 / .35 / 1.42 / .46† / 1.58
Game / trust game vs. coin flips / 1.39*** / 4.02 / 1.93*** / 6.88 / .98 / 2.66 / 1.43** / 4.19 / 1.88*** / 6.58 / .17 / 1.19
constant / -.76 / .47 / .09 / 1.09 / -3.40 / .03 / -1.45 / .24 / .64 / 1.90 / -1.63 / .20
Nagelkerke's R² / .13 / .21 / .27 / .30 / .24 / .46
Significant effect on the decision to risk/trust: † p<.10 ** p<.05 *** p<.01, a= excluded, constant value with one exception
Increase in predicitive power (ΔR): Model 1 to Model 2: n.s., to Model 3: marg. sig. (p=.095) , to Model 4: sig., to Model 5: n.s., to Model 6: sig.
Emotional Dynamics in Trust Behavior: Supplemental 1
The Trust Game and Inequality Aversion
Did participants trust in our experiments because to forgo trust was to guarantee an unequal result between them and their “partner” (Engelmann & Strobel, 2004)? We do not wish to dismiss inequality aversion as a motive that sometimes underlies decision-making in economic games or exchange relationships.
However, the specific question under study is whether inequality aversion is the motive or a primary motive underlying high rates of trust. Our past work, and the work of others, rule out inequality aversion as a primary motive underlying trust. It may, perhaps, underlie the decisions of scattered individuals, of course, but we find it fails to explain why so many people choose to trust in our paradigm.
In looking over the literature, we think it is important to distinguish decisions that guarantee an equal division of reward versus those that only make it possible. To our eye, inequality aversion seems to evaporate once one can no longer guarantee an equal division of reward.
Such a situation involves risky dictator games, in which a participant can keep their money or gamble it to potentially increase what they and another participant receive. Bolton and Ockenfels (2010) presented participants with several variations of risky dictator games. Fetchenhauer & Dunning (2012) reanalyzed the data from those studies in a direct discussion of inequality aversion in economic exchange (see pp. 539-540). They looked at two different cases.
Case I. In some versions of the risky dictator game, the participant and partner received the same amount if participant decided to gamble. However, if the participant chose not to gamble, the other person in some versions of the game continued to receive the sure amount the participant did. In other versions, the partner instead received nothing. According to inequality aversion account, participants should be more likely to bet in the latter case (other person received nothing, an unequal result) than in the former (other person receives the same amount of money). However, betting rates in both conditions were roughly equal (46% v. 48%, respectively).
Case II. In other conditions of Bolton and Ockenfels, if the participant bet, either the participant or their partner won all the money, with nothing for the other person. However, because the bet presented the participant a 50-50 chance of winning, one could say the decision to bet provided an equal outcome for the participant and their partner. Again, Bolton and Ockenfels varied what happened if the participant chose not to bet. Either the partner received nothing or received an equal amount. According to an inequality aversion explanation, this should lead participants to bet more in the former circumstance, but this was not significantly the case, 52% vs. 38%, respectively, p= .17.
In short, in decisions introducing uncertainty of outcome, inequality aversion appears to evaporate as a motive, although no specific study contrasting certain from uncertain outcomes has been formally conducted.
But what about the trust game itself?
Xiao and Bicchieri (2010) found that the amount a participant sent to a partner in a trust game did not depend on whether the endowments of both were equal or unequal to start. In an equal condition both were endowed with the equivalent of $8.In an unequal condition the participant’s endowment was doubled ($16) compared to the partner ($8). The percentage of participants sending money ($2, to be multiplied to $6) in a binary choice trust game did not significantly differ between both treatments (60.5% vs. 64.7%, respectively, p= .72). If inequality aversion were a major motive for the trust decision, one would expect more participants to send money (i.e., trust) in the unequal condition. Brülhart and Usunier (2012) as well as Anderson,Mellor, and Milyo (2006) also investigated the role of inequality in static two person games. The latter found that unequal endowments had only small and non-systematic effects on trust behavior. The former did not find any effect.
Our previous work has shown that trust games elicit risk-seeking that significantly outstripsrisk-seeking seen in similar situations that should evoke inequality aversion to an equally strong degree. Dunning et al. (2014, Study 5) compared a binary trust game to a “coin flip” game that presented similar outcomes to participant and partner. In the trust game, participants decided whether to hand $5 to a stranger, who would be then given a choice of keeping $20 or splitting $10 back to the participant. The coin flip contained the same payoffs, except that the partner had to determine their decision by flipping a coin.
Although each game presents the exact same outcomes (and thus the exact same pressures toward gaining equal outcomes by gambling), participants gambled their money at a significantly higher rate in the trust game (67%) than they did in the coin flip game (44%), p = .033. To recapitulate, both trust and coin flip games offered equivalent opportunities to manufacture equal outcomes via gambling, but people “gambled” more in the trust game. Indeed, participants in the trust game decision gambled more frequently than they did in a later decision whether to risk $5 on the flip of a coin to possibly win $10, with no other person involved (p = .009 by sign test)). Participants in the coin flip game condition did not (p = .167). This difference arose even though people expected they were more likely to have their “trust” reciprocated in coin flip game than in the trust game.
In addition, in Dunning et al (2014, Study 6), participants were given three options about what to do in a trust game. They could keep their $5, give it to another person who would then decide whether to return $10, or force their partner to make their decision via flipping a coin. According to an inequality aversion account, participants should be indifferent between having the other person freely decide versus forcing that person to flip a coin. Both decisions create the possibility of an equal distribution of reward.
Actually, though, one could argue that inequality aversion should prompt participants to prefer having the partner flip a coin. Under a coin flip, the odds are 50-50 of an equal distribution. Giving their partner a free choice instead, provides only a 38% chance of an equal outcome, given participants’ rather pessimistic expectations that their partner would choose selflessly.
However, participants preferred to let the other person freely decide over forcing a coin flip (54% to 22%), with an additional 24% deciding to keep the money.
In short, people choose to trust more than they choose to gamble on others in situations that present the exact same opportunities for equal outcomes.
Thus, if one wants to continue to adhere to an inequality aversion account of trust, one has to explain why inequality pressures would “turn on” only when the partner has a free choice rather than must follow the dictates of a coin. That would involve creating an inequality aversion account that somehow paradoxically ignores the primary importance of the outcomes in play, and whether they are equal versus unequal.
In the present manuscript, there are two additional findings that are difficult for an inequality aversion account to explain. The first is that participants brought their own money to the experiment in Study 1. The only way to insure equality is to refuse to bet, yet a majority did went ahead and bet—a finding we have replicated elsewhere in which participants bring their own money to the trust game (Schlösser et al., in press, Social Cognition).
Also, in Study 2, the extended coin flip should exert inequality aversion pressures that increase gambling relative to the regular coin flip condition. That is, in the regular coin flip, there is no other person to worry about. However, in the extended coin flip, participants’ decisions influence the outcomes of another person. Deciding not to flip the coin produces unequal outcomes for self versus other ($5 for self versus $0 for another person). That should cause participants to gamble at a higher rate, yet they fail to do so: 31% gamble in the extended coin flip vs. 32% in the regular coin flip.
We have replicated this lack of a difference between regular and extended coin flips elsewhere (Schlösser et al., in press, Social Cognition). In Study 1 of that article, only 28% of participants gambled in the extended coin flip versus 25% for a regular coin flip (rate of gambling in a trust game: 57%). In Study 2, in which gambling rates were higher across the board, the extended versus regular coin flip difference was 60% vs. 50%, ns, with rates of taking a gamble in the trust game being 89%. In both studies, participants’ expectations that they would receive $10 in the trust game and extended coin flip were roughly the same (i.e., 50%).
Again, people trust to a higher extent in the trust game than they do in a coin flip game that presents roughly the same opportunities to produce equal outcomes. Yet, gambling rates differ, suggesting that outcome equality is not a prime mover of trust behavior.