The role of emotions in explaining non rational behavior in one-shot games

(1)

-Master Thesis-

The Role of Emotions in Explaining Non Rational

Behavior in One-Shot Games

Name: Fabienne Cantner Student number: 11084766 Master: Economics and Business

Track: Behavioral Economics and Game Theory Supervisor: Professor Jan Engelmann

Date: 12.08.2016

The emotional cost of selfishness! Text

(2)

Abstract

This paper analyzes the role of emotions within hypothetical four economic games (Dictator Game, Ultimatum Game, Trust Game, Prisoners Dilemma) and decisions. The purpose is to investigate whether positive and negative emotions depend on the type of the economic game, the fairness of the situation and the role individuals play within the game. Additionally, social value orientation was assessed to see whether distinctions between differently categorized subjects can be made. Subjects participated in four hypothetical situations in which decisions and responses of their partners were already predetermined. Participants had to imagine the particular situation as well as possible and were asked to indicate the strength of several emotions they might experience. To check for social value orientation, subjects had to answer the SVO. Results show that individuals experience significantly stronger positive emotions in fair scenarios and significantly higher negative emotions in unfair scenarios. Controlling for social value orientation does not reveal any noteworthy changes. Additional analyses (visual inspection) of all interrogated separate emotions (and not summarized to positive and negative ones) show interesting results where particular emotions appear to be dominant in specific situations.

(3)

Table of content

I. Introduction 4

II. Literature review 5

A. Emotions - behavioral dimensions 8

B. Social value orientation 10

III. Experimental design and methods 11

IV. Descriptive Statistics and manipulation check 16

A. Descriptive Statistics 16

B. Manipulation Check 17

V. Experimental Results 18

A. Hypotheses I- Dependency of emotions on degree of fairness 18 B. Hypotheses II- influence of social value orientation 23 C. Exploratory findings and additional analyses 24

VI. Limitation, Discussion and Conclusion 33

A. Limitation 33

B. Discussion and Conclusion 33

VII. Appendix 34

VIII. References 39

(4)

I. Introduction

Why do humans act prosocially? Are altruism and cooperation learned skills or are these patterns of behavior deeply rooted in our genetic make-up? Taking a look at the world of (other) animals also observe cooperation and helping behavior, but mostly if the behavior and outcome happens to serve the animal’s self-interest.

Turning to humans, prosocial behavior in strategic decision making may be explained by two different hypotheses:

The first focuses on an evolutionary set of criteria: (a) the altruism has to benefit the recipient more than it costs the altruist, (b) an expectation of a repetition of interaction should be present and (c) humans should be able to detect non-reciprocators (Axelrod & Hamilton, 1981). Nevertheless, prosocial behavior often occurs in economic games, although the listed criteria do not apply. Apparently humans developed prosocial patterns of behavior which are not based on any expectation of reciprocity, relatedness or other factors serving their own interests. Reducing your own payoff, in order to cooperate with another person you have never seen and you won't ever see again does not seem to be intuitive and rationally right. Why then do still act prosocially in such situations? What leads us to do so?

Addressing this, a second hypothesis has been proposed. This one assumes humans to have cognitive limitations, such that they are very used to cooperating and acting in a prosocial manner and are not able to inhibit this default behavior in the lab (Rand, Greene & Nowak, 2012; Jordan, Peysakhovich & Rand, 2014). Accordingly, the explanation for non rational behavior would simply be that humans are not able to change common behavior.

Although these hypotheses might help to understand prosocial behavior in particular circumstances, they do not completely explain behavior deviating from standard economic theory. Other factors that might work as moderators of decision making and by this help in explaining are more or less neglected. Taking these moderating factors into account, the concept of emotions, their influence on decision making and their explanatory role in this

(5)

context have attracted more and more interest and are suggested to have a striking influence on decision making in economical contexts (e.g. Elster, 1998).

Acting on this suggestion, propose further (behavioral) hypothesis on how decision making is determined by factors not as yet included in standard economic models. first suggest

emotions to be one of the strongest drivers of decision making of any kind, and especially

social decision making, i.e., assume them to show a moderating effect. Second, when it comes to social decision making, the social value orientation of humans may work as a moderator of emotions. In this paper we will focus on the influence of emotions on social decision making and will integrate into this analysis the impact of social value orientation on and via emotions. All of the games assume that subjects only interact in this particular situation (one shot game) and do not know each other.

The following analysis proceeds as follows: section 2 offers a literature review which is the basis for the derivation of the hypotheses. Section 3 introduces the experimental design and methods. The results of the experiment are presented in section 4, and section 5 contains an interpretation of those results. A summary of the analysis, the identification of some limitations as well as further research desiderata are presented in the final section 6. The appendix contains some additional results, charts and tables.

II. Literature review

One of the perhaps most important features of economic theory and game theory is the conception of homo oeconomicus, a perfectly self-rational decision-making human. Based on this assumption, models predict humans to behave according to a so-called Nash equilibrium in strategic and economic games. In a Nash equilibrium, none of the interacting players can do better (increase the outcome) by deviating from the chosen behavior if the partner does not also change. In a significant number of experiments, though, this Nash equilibrium is not reached. The following will summarize findings of four famous economic games and how human behavior deviated from prediction. These games can be divided into three different classes: (a) single player games, where only person makes a decision; (b) sequential games, where the decision of the first mover is

(6)

followed by a decision of the second mover; and (c) simultaneous games, where both players decide at the same time.

It is important to note that each of the games taken into account in this thesis is referred to as a one-shot game no matter the class of which it is part. “One shot” thereby describes the property of the games such that players only choose one single action per game. They will not make an additional choice or interact with their partner within a game again. In the following, each game, its theoretical implications and main findings will be explained.

Dictator Game

In a one-shot Dictator Game (DG), a proposer is endowed with a certain amount of money. Now he has to decide whether he wants to yield a part of the endowment to an unknown partner (called the responder). The responder has no power of decision and can thus only accept the offer. Accordingly, the DG is classified as a single player game. Because the proposer’s decision is not dependent on anyone but himself, pure altruism is measured in a DG. Since individuals only interact once in this particular situation, economic theory suggests the (rational) proposer to keep all the endowment in order to maximize the individual outcome (and behave according to the Nash equilibrium). Looking at experimental results, surprisingly, in almost 50% of the cases, the subdivision turns out to be about an equal split (Engel, 2011). On average, people seem to be truly altruistic, since by sending even a very small amount to the responder, the proposer actively turns down monetary reward.

Ultimatum Game

Similarly surprising results can be found in sequential games. Whereas the DG has only one active decision maker, the Ultimatum Game (UG) adds a decision that is made by the responder as a second step (sequential decisions). The responder can now either accept or reject the proposed amount. If the offer is rejected, both players will be unable to keep any money. If it is accepted, though, both keep the agreed sums. Whereas economic theory expects the proposer to offer the smallest amount possible and the responder to accept this (since otherwise he will end up with nothing), we observe significantly anomalous behavior. In line with the findings in the DG, proposers in the UG offer “too much” on average. Something really interesting also happens on the side of the responder - a significant number of actually profitable proposals get rejected. Responders rather decide

(7)

to end up empty handed than to agree on a perceived “too small” offer. Thus they seem actively to turn down monetary reward if they find the offer to be unfair (Oosterbeek et al., 2004).

Trust Game

In another sequential game, the Trust Game (TG), reciprocal exchange is measured. Both players are endowed with the same amount of money. The first mover (the investor) can decide how much of this endowment to send to the second player (the trustee). The decided amount gets multiplied by a certain factor and transferred to the trustee. Following that, the trustee can decide how much money to return. It is important to note that the trustee is not obligated to send anything back. If the trustee returns the trust and sends back money, both can end up with a higher amount than the endowment. If the trustee does not return the trust, though, the investor takes a loss. Since the players only interact in this particular situation, the Nash equilibrium of this game would be for the investor to not send and to keep all the endowment. Despite this prediction, the majority of investors sends some amount of money and the trust is usually returned (Sanfey, 2007; Johnson et al., 2011)

Prisoner's Dilemma

Another game used to measure reciprocal exchange is the Prisoner's Dilemma (PD). Different from the games previously explained, players make simultaneous rather than sequential decisions, either choosing to cooperate with or to defect from the respective other at the same time. The highest payoff for a player would occur when he chooses to defect and the partner chooses to cooperate, which is the worst outcome when actions are reversed. Both choosing cooperation leads to a moderate and equal outcome for both players. Both choosing defection instead leads to a still equal but much worse output than mutual cooperation. As subjects interact solely in this situation and thus trust is not expected to be returned, theory predicts the Nash equilibrium -mutual defection- to occur. Despite this prediction, mutual cooperation happens in almost 50% of the cases (Sanfey, 2007; Sally, 1995)

Anticipation of reciprocity, a reunion in the future or some kind of relatedness would explain the found behavior. But what leads individuals to behave in this manner in an anonymous one-shot game? Trying to answer this question, research in the field of behavioral economics started to include other factors and developed models aiming to

(8)

predict and to explain human behavior correctly. Still it seems as though there is no conclusion yet. In order to get even more insights, it is important to find out which and how different factors actually influence decision making and then, in turn, lead to non rational behavior. In this present study, we focus on the indisputable importance of emotions in this context.

A. Emotions - behavioral dimensions

As suggested in numerous studies, emotions exert significant influences on cognitive processes and thus on decision making (Engelmann & Hare, 2008; see also Pessoa, 2008; Phelps, 2006). This assumption has not only been confirmed for individual decision making, but also for social decision making. Findings from neuroeconomics analyzing activated brain structures during social decision making processes find emotions to play a crucial role (Sanfey, 2007). Accordingly, behavior in social and economic decision making situations is predicted by a model that does not take into account any kind of emotional sensation, though we have evidence that people actually experience emotions during the decision making process. The important component of emotions actually exerting influence on the behavior of interest is obviously not taken into account. Still, we wonder why people deviate from the explained theoretical predictions. Is it not tempting to suggest that these very emotions that are experienced during the decision making process and are not included in traditional economic models could serve as an explanation for the deviation? If we do so, two kinds of emotions have to be considered (Rick et al. 2008):

1) Incidental emotions: those that are already present before the decision is made and that are triggered by events unrelated to the decision.

and

2) anticipatory emotions induced by the outcome.

Mood/emotional state

Addressing the first point, several studies provide some evidence for the significant influence of moods and emotional states on decision making. To demonstrate these influences of emotions on prosocial behavior, a field study was conducted by Isen et al. (1976). Results show that after a positive mood manipulation (free gift at the door),

(9)

subjects were more willing to help at a telephone survey than were members of a control group. These findings suggest that the feeling of happiness leads to increased helping behavior. Closely related to this, several experiments have been conducted showing the influence of emotions on decision making in a social setting. Sanfey et al. (2007), for example, let people play a UG after inducing positive or negative emotions. Their results show a significant difference on the responder side. Indeed, the acceptance rate of responders is dependent on the emotion induced. Positive emotion induction leads to a higher, negative emotion induction to a lower acceptance rate. In line with these results, Andrade and Ariely (2009) find that inducing anger increases the rejection rate of responders significantly. The mentioned studies strongly show the influence of emotions on decision making in a social context.

Emotion related to outcome

Addressing the second point, it seems not only to be interesting how the current state of mind affects the decision, but also which decisions and outcomes trigger which kind of emotions and whether the anticipation of these emotions can possibly explain non-rationality. In this connection, Pillutla and Murnighan (1996) asked subjects how they felt after deciding in a UG and which emotions led them to this decision. Their results confirm that unfair offers lead to higher levels of anger and higher rejection rates by the responder. As they did not control for the emotional state of the subjects, these findings should be interpreted as more correlational than causal. Another study not manipulating but rather monitoring developed emotions was conducted by Fehr et al. (2002). There, subjects played a one shot public goods game where altruistic punishment of free riders was possible but costly. A significant number of subjects made use of this costly punishment. Whereas punishment would rationally seem to make sense in an iterated game, it does not in a one shot game and only leads to a reduction of individual payoff. Nevertheless, subjects decided to do so. The authors suggest emotions serve as a proximate mechanism. To investigate this hypothesis, subjects read scenarios of hypothetical outcomes of the game and rated their emotions. The results show that free riding leads to strongly negative emotions which in turn trigger the decision for punishment. These findings suggest that humans anticipate the specific emotions that occur during or with the outcome and select their mode of behavior accordingly. (Dependent on whether this emotion is desirable or not, people decide.) If it is actually the case that outcome related emotion has a strong influence on decision making, it could be reasoned that certain patterns of behavior

(10)

cause strong negative/positive feelings and the anticipation of experiencing those in turn generates non-rational decisions (Fehr et al., 2002).

To be able to make any assumption about the possible anticipation of certain emotions in an economic situation, first it is essential to know if and what kind of emotions are experienced in particular situations. To my knowledge there is as yet no such systematical investigation about emotions in specific economical situations. Gaining those very insights is the aim of this research. We decided to distinguish between prosocial and rather selfish situations and simply ask subjects what kind of emotions they sense. Behaving fairly and being the recipient of fair behavior should lead to highly positive emotions, whereas unfair behavior should result in highly negative emotions. A factor that might affects these emotional reactions is the social value orientation.

B. Social value orientation

Social value orientation describes “stable preferences for certain patterns of outcomes for oneself and others” (Van Lange et al. 1997). Van Lange et al. (1997) investigated this dimension intensively, developing a questionnaire - the SVO - distinguishing between three different social value orientations: a) prosocial, b) individualistic and c) competitive. Prosocial individuals want to maximize their own, as well as other’s output. Individualistic individuals more or less only care about themselves and their own output. Competitive individuals by contrast tend to maximize their output relative to others, thereby intending to end up with an advantage.

Depending on the distinctness of this orientation, people will decide differently in a social decision making situation. This allows the development of hypotheses about emotional reactions in different economic games.

Summarizing the last two passages, we propose the following hypotheses:

1) Positive emotions are connected to prosocial outcomes and the opposite for negative emotions.

2) Prosocial classified individuals react with significantly stronger negative emotions when being treated selfishly.

(11)

Moreover, we expect these subjects to experience more negative emotions when they treat others unfairly, which may be one mechanism that prevents them from such anti-social (but here also rational and theory congruent) behavior. Less prosocial oriented subjects instead should not react with equally strong emotions when treated unfairly or when behaving unfairly towards others. Including the SVO into the present research will help in achieving greater insight about the role of emotions on decision making.

III. Experimental design and methods

The hypotheses derived will be tested by using data from an online questionnaire. This questionnaire was created using Qualtrics software, Version (07/2016) of Qualtrics (Qualtrics, Provo, UT). In the following, the experimental design, the specific scenarios, their classification as fair vs. unfair and the selection of important scenarios are introduced and explained.

Overall design

All participants were exposed to the four economic games (UG, DG, PD, TG) in random sequence to avoid order effects (within subject design). Within each game subjects were randomly assigned to a particular situation. Depending on the type of game, the design was set up as follows: for sequential games (UG and TG), a 2 (proposer vs. responder) x 2 (behave prosocially vs. selfishly) vs. (get treated prosocially vs. selfishly) design was used since it might be important if you are the initiator or the victim of unfair behavior. For simultaneous (PD) and one-player games (DG), we used a 2 (behave fair vs. unfair) x 2 (get treated fair vs. unfair) design. See Table 1 for an overview.

(12)

Table 1: Summary of all possible scenarios

Notes: This table presents every possible scenario we

wanted to analyze. Within the next passages we will explain which of the cases had to be excluded due to certain explanations.

Scenarios

Each scenario was designed in the following way:

The particular game was explained and was thereby framed as realistically as possible (not as an economic game but rather as a daily situation) to avoid framing effects. Playing one game consisted of two steps. First, subjects were assigned the role (proposer vs. responder for sequential games) and the mode (prosocial vs. selfish) for each game and had to decide how they would behave in this particular situation. Second, they were put in a hypothetical

situation, for which their decision and the response of their partner was already

determined. Accordingly, subjects’ task was to first imagine and afterwards put themselves in the described scenario as well as possible.

Classification as fair vs. unfair scenario

The distinction between fair and unfair situations was implemented in a specific way, and so “fair” and “unfair” were interpreted as follows: “Fair” is always connected to prosocial or trusting behavior, while “unfair” relates to selfish or rather distrustful behavior.

In the following, the classification will be explained for each game. within

subject between subject

GAME ROLE MODE

single player game Dictator Game proposer prosocial selfish sequential games Ultimatum Game proposer prosocial offer Selfish offer responder prosocial offer selfish offer Trust Game investor trustful distrustful

trustee returns trust

selfish simultaneous games Prisoners Dilemma no role mutual cooperation one-sided cooperation

(13)

Fair situations consisted of two fair actions, and so both players chose fair behavior. An unfair situation, by contrast, involved at least one unfair action, and so either one player or both players chose to behave unfairly. Since the DG contains only one action, a situation in the DG was fair when the proposer acted fairly and unfair when he chose to behave unfairly.

For the DG and UG, sending half (5€) of the endowment was used to show fair behavior by the proposer, whereas sending 0€ in the DG and 1€ in the UG, respectively, represented unfair behavior by the proposer. This classification was based on findings by Camerer, (2003) and Andrade & Ariely, where subjects rejected offers at around 20% - 30% of the endowment due to the violation of their sense of fairness. Categorization of fair vs. unfair behavior on the responder’s side is not possible in the UG. Whether a rejection of an unfair offer is fair or unfair behavior is in the eye of the beholder and we were not comfortable with making generalized assumptions.

In the TG, sending the whole endowment (10€, which gets tripled and transferred) represented prosocial (trusting) behavior by the investor. Since selfish behavior by the investor (sending 0€) would end the game right away, in all possible trustee-situations, trustees held 40€, whereas the investor held 0€. Sending back 20€ (trustworthy) therefore exemplified fair behavior, while sending back 0€ represented unfair (selfish) behavior. This classification is not based on any empirical findings and is thus an individual interpretation. Certainly scenarios can be classified differently.

In the PD, situations where both players chose to “accept” were characterized as cooperative (prosocial) scenarios. If one of the players instead chose to “defect,” a selfish situation was described. Again, these categorizations are just my individual point of view and can certainly be seen differently.

Selection of important scenarios

Due to the inconclusiveness of some scenarios, we decided only to report upon the most interesting combinations of actions for each game. Some possible scenarios (combinations of role and mode) were excluded before the investigation:

(14)

a) Dictator game. No scenario was excluded.

b) Ultimatum game. Scenarios where the responder rejected an unfair offer were

excluded. First, as mentioned above, the evaluation of (un)fair behavior in the case of a responder is difficult to make. Second, if an unfair offer gets rejected, one could refer to it as “poetic justice” and emotions might not appear noticeably. Thus, in all scenarios, responders accepted the offer no matter if they were fair or unfair.

c) Trust Game. Scenarios where the investor decides to keep the whole endowment

were excluded since the game would end in this situation without the response of the trustee, with both players holding their initial endowments. Thus, in every scenario, the trustee decided to send the whole endowment.

d) Prisoner’s Dilemma. Scenarios where both players decide to “reject” were

excluded because, in these situations, both players should not feel as though they were treated unfairly and emotional responses should not be considered as significant. Again, one could refer to as “poetic justice”.

Experimental procedure

After a short welcome, subjects were presented with four scenarios. After each scenario, subjects had to rate their experienced emotions in the particular situation on a 7-point Likert scale. The listed emotion included in our view relevant basic emotions as well as social emotions. See Table 2 for an overview.

Table 2: Summary of emotions

Note: All interrogated positive and negative emotions

classified into basic and social emotions

This sequence was repeated in an equivalent way for each of the four games. (Please see

Figure 1 for the exact sequence of either sequential or simultaneous and one- player

games.)

positive negative

Basic emotions happiness anger

sadness

Social emotions self-contented

satisfied camaraderie relieved betrayed ashamed disappointed

(15)

Figure 1: Sequence for sequential and simultaneous games and one-player games

Note: The figure presents the sequence of sequential games (UG&TG; left) and the simultaneous (PD; right) and

one-player game (DG, right).

After finishing the games part of the survey, participants had to fill in three different standardized questionnaires - a questionnaire testing the social value orientation (SVO, Van Langen et al, 1997), the Emotion Regulation Questionnaire (ERQ; Gross & John, 2003) and the Positive And Negative Affect Schedule (PANAS, Watson et al., 1988). The SVO measures the individual social value orientation of each subject. It was included to test the second hypotheses. The ERQ assesses emotion regulation strategies of individuals, and the PANAS positive and negative affect in the previous week. The latter two questionnaires where included to capture the influence of outcome related emotions by controlling for incidental emotions and emotional states.

Scales for positive and negative emotions

In order to compare experienced positive and negative emotions in different situations, we combined all measured positive emotions and all negative emotions into separate variables. The variable for “positive emotions” thus includes the average score of all positive

rating of emotions hypothetical scenario decision in this particular situation explanation of the game random assignment to mode rating of emotions hypothetical scenario decision in this particular situation explanation of the game random assignment to

(16)

emotions in the particular situation, whereas the variable for “negative emotions” includes the average score for of all negative emotions.

SVO scale

Because the standardized classification of individuals as “prosocial” vs. “individualistic” vs. “competitive” only applies when an individual chooses a predetermined combination of answers and many participants could not be categorized in that way, we calculated an additional “prosocial score”. Depending on how many prosocial answers the subjects chose, a certain score was assigned. Individuals who chose one out of nine as prosocial categorized answers were assigned a score of 1, individuals who choose 4 out of 9 prosocial options received a score of 4, and so forth.

As a direct correlate for the SVO, we decided to choose the amount transferred by first movers in the dictator game since this is a direct measure of altruism (Sanfey, 2007) and thus also of prosociality.

IV. Descriptive Statistics and manipulation check

A. Descriptive Statistics

A total of 142 subjects (67 males and 75 females) took part in the online survey. Participants were mostly students (111 subjects) and employed workers (23 subjects), whereas the rest of the participants was either not employed (four subjects) or retired (one subject).

88 Participants were classified as “prosocial”, 31 as “individualistic, four as “competitive” and 19 could not be classified via the SVO. As the small sub-sample of competitive individuals does not have explanatory power (Simmons et al., 2011), we decided to exclude them from further analysis. Instead we focused on differences between the “not classified,” “individualistic” and “prosocial” subjects. After excluding the four competitive subjects, 138 remaining subjects were analyzed. Table A1 (appendix) summarizes the descriptive statistics of the independent variables used in the analysis of hypotheses 1 and 2. We controlled for multicollinearity by correlating all independent variables. See Table

(17)

B. Manipulation Check

As a manipulation check, we tested the assumption that participants classified as prosocial in the SVO also transferred a significantly higher amount to their partner in the DG compared to individualistic ones and non-classified subjects. Figure 2 presents the average amount sent by the proposer subjects to the SVO classification. We found a significant effect, supporting the assumption (prosocial vs. individualistic: t(135)=3.51, p=0.0006; prosocial vs. non-classified: t(135)=2.48, p=0.0114). Correlation of the transferred amount with the prosocial scale reveals a significant effect, r(138)=0.32, p<0.001.

Figure 2: Average amount transferred in the DG subject to SVO classification Note: Prosocial individuals transfer significantly more of their endowment compared to individualistic

and not classified subjects. 0

1 2 3 4

noclass individualistic prosocial

classification svo average transf erred amount DG

(18)

V. Experimental Results All analyses in this paper were done in R 3.3.1.

For all regressions we used the package for “Regression Modeling Strategies” (rms) by Harrell Jr (2016).

A. Hypotheses 1 - Dependency of emotions on degree of fairness

Overall games

Figure 3 represents the average positive and

negative emotions for each game indicated by the subjects as a function of the degree of fairness within a scenario (collapsed across roles). Panel A shows average positive emotions in fair and unfair situations. Panel B shows average negative emotions in fair and unfair scenarios. In order to underpin the treatments effect (fair vs. unfair scenarios) statistically, we conducted a regression analysis, controlling for positive and negative affect as well as for emotions regulation, as follows:

(1) !_"#= %_&+ %₍)*+,_" + %_./0102_"+ %₃456_" + %₇8*9: _" + %_;8) + <_"#

The dependent variable !_"# represents the average positive (negative) emotions a subject i indicated in a scenario s. fairi is a dummy prosocial (fair)

behavior, game is a dummy for the particular game (DG, UG, TG, PD). PANAS and ERQ are sets of two control variables each for subject’s

DV pos neg 6.05* 11.47*** (2.18) (4.31) 11.71*** -6.47*** (11.00) (-7.95) 0.23*** -0.96 (4.36) (-0.39) 0.00 0.15** (-0.05) (3.15) -0.03 -0.06 (-0.53) (-1.22) 0.03 0.03 (0.52) (0.52) 2.33* 0.96 (2.49) (-1.10) -0.02 4.89*** (-0.02) (5.21) 0.177 1.59 (0.20) (1.66) -2.97* 1.84 (-2.04) (1.77) 1.67 -5.03*** (1.31) (-4.65) 0.53 -1.68 (0.34) (1.50) R2 0.47 0.37 adjusted R2 0.49 0.36 observations 552 552 PD fairness*UG fairness*TG fairness*PD pa na reap sup UG TG constant fairness

Note. This table reports the OLS (with robust

covariance’s) coefficient estimates (t-values in parentheses). The dependent variables are positive (pos) and negative (neg) emotions. Fairness is a dummy for prosocial behavior. Pa and na are control variables included in the PANAS, reap and sup are control variables included in ERQ. The games are represented by UG, TG and PD and are compared to the DG.

Table 3: Estimation results of emotions

(19)

incidental emotions and emotion regulation. The set of interactions between game and fair is represented by the term gf, where interactions between UG, TG and PD with degree of fairness are calculated, and DG is used as the reference game. <_"# is the idiosyncratic error term. We test this model using OLS with robust covariance, since all subjects participated in each game.

Table 3 presents the estimation results. Here we find a main effect for the degree of

fairness. Subjects indicated significantly higher average positive emotions in fair situations, t(540)=11.00, p<0.001. In line with this, significantly lower average negative emotions in fair situations, t(540)=7.95 p<0.001, were indicated. We also find a significant main effect for the control variables within the PANAS. Positive emotions are significantly higher when scores in positive affect are high (t(540)=4.36, p<0.0001),while negative emotions are higher when scores in negative affect are high (t(540)=3.15, p=0.002). Interactions between games and level of fairness reveal an effect in the TG, where negative emotions are significantly lower in fair situations, t(540)=4.65, p<0.0001. Interestingly, positive emotions are significantly lower in fair situations in the UG, t(540)=2.04, p=0.04. Additionally, we controlled for the role (proposer vs. responder) in the UG and TG, using a similar regression model:

(2) !"# = %&+ %()*+," +%7/0102" + %;456"+ %.8*9: + %3,=>: +

%_?8),_" + <_"# ,

where all variables are interpreted as explained above, role is a dummy for being a proposer in the specific situation or not and gfr is the set of interactions between game,

fairness and role. Again, an OLS with robust covariance was used.

The main effect for fairness stays highly significant for positive as well as for negative emotions. No significant effect on emotions was found between the games. More specifically, proposers indicated significantly higher average negative emotions compared to responders. This effect increases in a UG such that fair (prosocial) proposers indicated even higher

negative emotions, t(276)= 2.53, p=0.018. No significant differences were found in the

connection with positive emotions.

where is the table?

this shows that incidental emotions can influence emotional reactions

(20)

Panel A. positive emotions Panel B. negative emotions

Figure 3: Average emotions in fair and unfair scenarios for each game

Note: The figure presents the average positive (Panel A) and average negative emotions (Panel B) indicated

by the subjects in prosocial (fair) and selfish (unfair) scenarios for each game.

Separate games

In order to get more insights about the particular games, the behavior in the specific situation and the emotional responses of subjects within the games, we analyzed each game separately. First, we wanted to investigate whether subjects indeed behaved anomalously from theoretical prediction and thus replicate the findings explained in the literature review. Although we have to be careful to compare hypothetical outcomes to actual experiments, we found some evidence that no significant difference occurs between hypothetical situations and the “real” world (Kühberger et al. 2002; Camerer et al. 2003, Kang et al. 2011). Second, we used adjusted versions of the above mentioned regression model (without dummies for game type and their interactions with fairness) to analyze indicated positive and negative emotions within each game.

For the DG and PD we used the following model:

(3) !_"#= %_&+ %₍)*+,_"+%_./0102_"+ %₃456_" + <_"# . In the UG and TG we additionally controlled for the role:

(4) !_"#= %_&+ %₍)*+,_" +%_./0102_"+ %₃456_" + %₇,=>: + %_;), <_"# . 0 10 20 DG PD TG UG games

p

o

sti

ve

emo

ti

o

n

s

fairness unfair fair 0 5 10 15 20 DG PD TG UG games

n

eg

ati

ve

emo

ti

o

n

s

fairness unfair fair

(21)

Interactions between fairness and role are represented by the term fr. Table A3 (appendix) presents the estimations results for each game. In the following, the results will be reported game by game.

• Dictator game

Replication of findings. Findings by Engel (2011) were supported. We used a one sample

t-test to show that individuals deviated from the Nash equilibrium by transferring an amount significantly greater than zero to their partner (M=3.82, SD=2.01, p<0.001), even in the current experiment where choice scenarios were hypothetical.

Hypothesis 1. We used regression model (3) to compare the experienced emotions in fair vs.

unfair situations. We found a significant difference in emotional experiences as participants indicated significantly more positive emotions when they (hypothetically) acted fairly (transfer > 0 Euros) than when they acted unfairly, t(132)=10.86, p<0.001. Moreover, in situations, in which the dictator acted unfairly (transfer = 0 Euro), they indicated significantly higher negative emotions, t(132)=7.89, p<0.001, compared to fair transfers. These findings suggest that being prosocial and giving something to others leads to positive emotions.

• Ultimatum Game

Replication of findings. Our results support the findings by Oosterbeek et al. (2004) showing that proposers in the Ultimatum game send significantly higher amounts (more than 1€) to the responder than predicted by theory (M=4.86, SD=1.90, p<0.001, (t-test)). Additionally, the responder behavior deviates from theoretical predictions, as 45% (17 out of 38) would- even in hypothetical situations- reject 1€-offers, thereby turning down monetary outcomes and rather ending up with nothing than 1€.

Hypothesis 1. Using regression model (4) we analyzed differences in emotional experience

between proposer and responder in fair and unfair scenarios. Collapsed over roles, participants indicated significantly higher positive emotions in a fair scenario compared to an unfair scenario, t(136)=6.4, p<0.001. Average negative emotions were shown to be significantly higher in unfair situations, t(136)=5.2, p<0.001.

Examining the differences between proposer and responder, we found an interesting effect. Fair as well as unfair behavior by the proposer does not lead to any significant difference in

not clear enough - do you mean for both sides, the proposer and responder?

(22)

only 1 Euro), responders and proposers have similar emotional experiences. The same is found for prosocial behavior by the proposer (transferring 5€). These findings suggest that being selfish (proposer) leads to negative emotions similarly to being treated unfairly (responder). Acting prosocially instead leads to a similar level of positive emotions as receiving a fair treatment. The latter suggestion mirrors findings from the DG where prosocial behavior leads to positive emotions as well.

• Trust Game

Replication of findings. Analysis of the Trust Game revealed results in line with findings by

Johnson (2011). Investors send significantly more of their initial endowment to the other person than predicted by theory (M=5.71, SD=3.24, p<0.001 (t-test)). Although theory predicts the game to end at the first step, because the investors’ rational behavior is to keep the whole endowment, we tried to analyze the trustee’s behavior as well. What happens if the investor sends 10€ (acted fairly)? Do trustees live up to expectations and return the trust or keep the money to themselves (which would maximize their own profit)? Prior research (Johnson et al. 2011) found positive amounts and thus also trust returned by the trustees. In our study, on average, 17.04 € (from the 30 they received) were returned (and thereby the outcome is not maximized but trust is returned).

Hypothesis 1. We used regression model (4) to analyze differences in emotional response

between investor and trustee in fair and unfair scenarios. Collapsed over roles, we found a significant difference for average positive emotions, where individuals experienced significantly higher average positive emotions in a fair scenario, t(130)=7.97 ,p<0.0001. In line with this, subjects indicated significantly lower scores in negative emotions in a fair situation, t(130)=6.96, p<0.0001. No significant differences were found between investor and trustee in fair and unfair situations. These findings lead to further interesting suggestions. Acting prosocially (return the trust as a trustee and send back 20 Euros to the investor) leads to a similar level of positive and negative emotions as receiving a prosocial treatment (retrieving the trust as an investor and receiving 20 Euro from the trustee). Acting selfishly (return 0 Euro as a trustee) instead leads to a comparable level of positive and negative emotions as receiving this selfish treatment (neither retrieving the trust nor any money as an investor).

(23)

• Prisoners Dilemma

Replication of findings. In 47,1% (65 out of 138) of cases subjects chose “cooperate” in the

Prisoner’s Dilemma and thus behaved in violation of the Nash equilibrium (choosing to defect). This confirms results found by Sanfey (2007).

Hypothesis 1. Using regression model (3) we analyzed the differences in emotional

experience in cooperative and non-cooperative situations. Subjects indicated significantly higher positive average emotions in situations of mutual cooperation, t(140)=10.86, p<0.001. In situations of single cooperation (one player chose “defect”), significantly higher scores in negative emotions where shown, t(140)=8.53,p<0.001. This leads to the assumption that subjects simply prefer cooperation and choose their behavior accordingly. In order to get more insights into emotional experience, we analyzed the difference between being the cooperator and inciting defection (C/D) and being the defector while the other is willing to cooperate (D/C). We tested this by comparing means of positive and negative emotions in subsets of C/D and D/C situations. Significantly higher positive emotions were found in D/C situations compared to the reverse scenario (p=0.0005, t-test). Significantly higher negative emotions were indicated in C/D situations compared to D/C scenarios (p=0.0002, t-test).

B. Hypothesis 2 - influence of social value orientation

In order to identify the effect of social value orientation on the emotional experience, we included the SVO-classification into the regression models (1) - (4) (see Table A4 (appendix) for a summary of estimation results). First, it has to be generally noted, that in each game the main effect for fairness stays highly significant (higher positive and lower negative emotions in fair situations). Secondly, for the SVO classification we find only significant effects in the UG and TG. These findings will be discussed in the following.

Ultimatum game. In the UG we find a significant 3-way interaction between fairness, role and

the group “not classified”. Since the two out of three main effects for those variables are not significant (only fairness), we should be really careful with interpreting this interaction.

Trust game. In the TG subjects, classified as individualistic, indicated significantly higher

(24)

subjects experience lower positive emotions in a fair scenario (t(122)=2.05, p=0.04). Interactions between role and the group of individualistic subjects as well as interaction between those variables and fairness are difficult to interpret, since there is no main effect found for role.

C. Exploratory findings and additional analyses

After reviewing all separate basic and social emotions, it seems interesting that average positive and negative emotions (each as an aggregate of 5 specific emotions) differ and also that there are specific emotions that are outstanding in different situations and between different roles. Therefore, we compared mean scores of all requested single emotions indicated in the specific scenario. Since the indication of the single emotions is not independent, we were not able to use an ANOVA to test for significant changes. Accordingly, we had to go back to visual inspection to determine whether single emotions were dominant in specific scenarios. Afterwards we compared means of these dominant emotions in the most interesting situations using a two-sample t-test. Since this paper aims to find answers for the question why individuals act prosocially even if this reduces their outcome, we found it especially interesting to look at emotional differences between (a) actively choosing prosocial and selfish behavior, (b) actively choosing and receiving a prosocial treatment and (c) actively choosing or receiving a selfish treatment1.

By doing so we have to distinguish between sequential and simultaneous games. On this view, the expectation and experience of emotions in sequential games are different from those in simultaneous games. We will start with sequential games, where the anticipated emotions are not directly influenced by the other’s action. Anticipated emotions are not dependent on expectations about the other’s action but rather on own decision. Afterwards we will continue with the simultaneous game.

(25)

Sequential games

(a) Acting prosocial vs. acting selfish. Figure 4 gives on overview of the indicated

separate emotions in all scenarios where the subject acted prosocially (Panel A) compared to scenarios where the subject acted selfishly (Panel B). Looking at all games in Panel A, it appears that happiness, camaraderie and satisfaction are the dominant emotions; to filter out one dominant emotion is hard to accomplish, though. A clearer picture presents itself when looking at all the games in Panel B, where shame seems to be the dominant emotion experienced when acting selfishly.

After filtering out those dominant emotions, we compared the mean scores of those dominant emotions between scenarios where subjects acted prosocially compared to situations where subjects acted selfishly. Since we compared 2 different situations and thus two different groups, we were able to use a two-sample t-test for the analysis (see Table 4 for detailed results.) we found that prosocial behavior (compared to selfish behavior) leads to significantly higher scores of happiness, satisfaction and camaraderie in each of the games. Shame scores significantly higher in situations of selfish behavior. This is quite interesting, since acting prosocially also means lowering one´s output. Thus the maximization of one´s output (acting selfish) leads to shame, whereas sharing the output leads to feelings of happiness. These findings support the assumption that subjects actively chose prosocial over selfish behavior in order to promote feelings of happiness and to avoid shame.

(26)

Panel A Panel B

Figure 4: Average emotions in prosocial (Panel A) and selfish (Panel B) scenarios Note: Compared are the average single emotions in situations in which individuals hypothetically acted

prosocially (Panel A) to scenarios where individuals acted selfishly (Panel B). Subjects indicated higher levels

of happiness, camaraderie and satisfaction in prosocial situations. In selfish situations shame is the predominant

0 1 2 3 4 5 happ y relie ved camar ader y satisfied _selfcontent sad _angr y

disappointed ashamed betr

a yed emotions a vera g e score prosocial offer in DG 0 1 2 3 4 happ y relie ved camar ader y satisfied _selfcontent sad _angr y

a yed emotions a vera g e score selfish offer in DG 0 1 2 3 4 5 happ y relie ved camar ader y satisfied _selfcontent sad _angr y

a yed emotions a vera g e score

prosocial offer in UG − proposers view

0 1 2 3 happ y relie ved camar ader y satisfied _selfcontent sad _angr y

selfish offer in UG − proposers view

0 2 4 happ y relie ved camar ader y satisfied _selfcontent sad _angr y

prosocial trustee in TG − trustees view

0 1 2 3 4 5 happ y relie ved camar ader y satisfied _selfcontent sad _angr y

selfish trustee in TG − trustees view

incentivize emotions to test subjects awareness

(27)

Table 4: Comparison of emotions in situations of prosocial behavior to situations of selfish behavior

Notes: This table presents the average indicated (mean) emotions in situations where subjects decided to act prosocially

compared to situations where subjects decided to act selfishly. To test differences, we used a two-sample t-test, where tdiff gives the t-value. We find all differences to be significant, using the following classification:

* p<0.05 ** p<0.01 *** p<0.001

(b) Acting prosocially vs. receiving prosocial treatment. Treating someone fairly and

receiving a fair treatment leads to positive emotions on both sides, whereas negative emotions score rather low. Filtering out a dominant positive or negative emotion turns out to be difficult. Figure 5 provides some exciting insights. Interestingly, individuals who actively chose be prosocial and in turn lower their reward (transfer fair amount in UG; return money in TG) seem to experience similar emotions as responders.

Comparing means of single emotions did not lead to significant differences in happiness or satisfaction between subjects who chose to share and receivers (Table 5), although the former actively chose to decrease the output. Additionally, there are no significant differences in any negative emotion.

GAME mode happiness satisfaction camaraderie shame sadness mean tdiff mean tdiff mean tdiff mean tdiff mean tdiff

DG prosocial 5.09 -8.18*** 4.98 -8.96*** 4.91 -10.38*** 1.26 12.78*** 1.28 7.12*** selfish 2.78 2.48 2.07 4.60 2.96 UG prosocial 4.77 -2.59* 4.87 -3.88** 5.26 -6.06*** 1.58 -5.64*** 1.65 3.06** selfish 3.62 3.29 2.78 3.75 2.76 TG prosocial 5.32 -7.32*** 5.38 -7.06*** 4.92 -5.76*** 1.35 9.79*** 1.27 6.63*** selfish 2.44 2.47 2.24 5.21 3.50

(28)

Panel A Panel B

Figure 5: Average emotions in situation of prosocial behavior (Panel A) and prosocial treatment (Panel B) Note: This figure compares the average single emotions in prosocial situations. Panel A shows emotions

indicated by subjects who (hypothetically) chose the prosocial behavior. This is compared to Panel B, presenting the emotions indicated by the subjects who (hypothetically) received the prosocial treatment. Positive and negative single emotions are indicated on a comparable level. No difference between the situations is found between happiness, camaraderie or satisfaction.

Table 5: Comparison of means of emotions between acting prosocial and receiving a prosocial

treatment

Notes: This table presents the average indicated (mean) single emotions in situations where subjects acted prosocially

(proposer/trustee) compared to situations where subjects received a prosocial treatment (responder/investor). Means were compared by using a two-sample t-test, where tdiff presents the t-value of this test. No significant differences were found

between any of the listed single emotions.

0 1 2 3 4 5 happ y relie ved camar ader y satisfied selfcontent sad _angr y

prosocial offer in UG − proposers view

prosocial offer in UG − responders view

0 2 4 happ y relie ved camar ader y satisfied selfcontent sad _angr y

prosocial trustee in TG − trustees view

0 2 4 happ y relie ved camar ader y satisfied selfcontent sad _angr y

prosocial investor & trustee in TG − investors view

GAME role happiness satisfaction camaraderie sadness anger disappointment shame betrayal mean tdiff mean tdiff mean tdiff mean tdiff mean tdiff mean tdiff mean tdiff mean tdiff UG proposer 4.77 1.08 4.87 -0.42 5.26 -1.04 1.65 -0.82 1.65 -1.4 1.74 -0.65 1.58 -1.07 1.65 -1.07

responder 5.19 4.69 4.81 1.44 1.25 1.56 1.34 1.36

TG trustee 5.32 -0.87 5.38 -0.33 4.92 -1.63 1.27 -0.35 1.11 -1.21 1.32 -1.07 1.35 0.32 1.14 -1.21

(29)

(c) Acting selfishly vs. receiving selfish treatment. Comparing scenarios where the

subjects actively chose to behave selfishly to scenarios where the subjects (passively) received a selfish treatment gives some interesting insights (Figure 6). Sorting out dominant positive emotions turns out to be difficult. For negative emotions, though, shame on the proposers and disappointment and anger on the receiver’s sight seem to be dominant.

Comparing means of these dominant emotions between the scenarios within each game reveals significant differences, which are presented in Table 6. Shame is significantly higher when acting selfishly compared to receiving a selfish treatment. Disappointment and anger are significantly higher in the opposite situations. Interestingly, sadness does not differ between the two scenarios. This leads to the assumption, that a selfish treatment leads to comparable levels of general sadness on the active (acting selfish) and the passive (receiver of selfish treatment) side. Additionally, shame or disappointment is experienced.

(30)

Panel A Panel B

Figure 6: Average emotions in situation of selfish behavior (Panel A) and selfish treatment (Panel B) Note: Compared are the average single emotions in prosocial situations. Panel A show emotions indicated by

subjects who (hypothetically) chose the selfish behavior. This is compared to Panel B, presenting the emotions indicated by the subjects who (hypothetically) received the selfish treatment. Actively choosing selfish behavior leads to shame, whereas receiving it leads to disappointment. Positive emotions are not experienced at high levels.

Table 6: Comparison of means of emotions between acting selfish and receiving a selfish treatment

Note. This table presents the average indicated (mean) single emotions in situations where subjects acted selfishly

(proposer/trustee) compared to situations where subjects received a selfish treatment (responder/investor). Means

were compared by using a two-sample t-test, where tdiff presents the t-value of this test. We used the following

classification of significance levels: * p<0.05 ** p<0.01 *** p<0.001 0 1 2 3 happ y relie ved camar ader y satisfied selfcontent sad _angr y

selfish offer in UG − proposers view

0 1 2 3 happ y relie ved camar ader y satisfied selfcontent sad _angr y

selfish offer in UG − responders view

selfish trustee in TG − trustees view

prosocial investor & selfish trustee in TG − investors view

GAME role happiness satisfaction camaraderie sadness anger disappointment shame mean tdiff mean tdiff mean tdiff mean tdiff mean tdiff mean tdiff mean tdiff

UG proposer 3.62 -2.18* 3.30 -1.4 2.78 -1.53 2.76 -0.23 1.38 4.80*** 1.84 3.70*** 3.76 -4.93***

responder 2.74 2.74 2.21 2.66 2.82 3.26 1.79

TG trustee 2.44 0.30 2.47 0.03 2.23 0.42 3.50 -1.36 2.18 -4.64*** 3.09 -3-39** 5.21 8.02***

(31)

Simultaneous games

A slightly different situation occurs when actions are chosen simultaneously (in the PD). Here the anticipation of emotions is directly related to the behavior of the other. Thus the anticipated emotions are dependent on the expectation about the behavior of the other. Here we found it especially interesting to analyze differences between situations where subjects cooperated and the other defected to situations where subjects defected while their partner cooperated (see Figure 7 for an overview.):

In situations where the individual cooperated and defection occurred (C/D), disappointment and anger were indicated, which is comprehensible and in line with the hypothesis. Comparing means of these emotions leads to significant differences between the two scenarios (Table 7). If the subject chose defection while the other cooperated (D/C), subjects experienced significantly higher emotions of happiness and satisfaction (Table 7), which could be explained by receiving a high payoff. Though, subjects were happier and more satisfied they experienced high levels of shame, a finding that is in line with previous findings. Acting non-cooperatively (in this case, being more or less antisocial) made subjects feel ashamed. This leads to the suggestion that subjects want to avoid the possible experience of shame and in turn choose cooperation (which would explain the high percentage of cooperation in our study as well as in previous findings explained in the literature review).

(32)

Panel A Panel B

Figure 7. Average emotions in situations where subjects cooperated solely (Panel A) and defected solely (Panel B)

Note: This figure shows the average indicated single emotion in situations where subjects cooperated while their

partner defected (Panel A) and situations where subjects defected while their partner cooperated (Panel B).

Emotions of happiness and satisfaction score higher in situations where the subject defected and their partner cooperated (Panel B). Though, also shame is indicated at high levels in those situations. In situations where the subject cooperated and the partner defected, disappointment was indicated at high levels.

Table 7. Comparison of means of emotions between two different situations of one sided cooperation

Note. This table presents the average indicated single emotions in situation where the subjects cooperated and the

partner defected (C/D) compared to the reverse situation (D/C). Means were compared by using a two-sample t-test, where tdiffpresents the t-value of this test. We used the following classification of significance levels:

* p<0.05 ** p<0.01 *** p<0.001 0 1 2 3 4 happ y relie ved camar ader y satisfied _selfcontent sad _angr y

cooperation while other defects

0 1 2 3 happ y relie ved camar ader y satisfied _selfcontent sad _angr y

defection while other cooperates

GAME mode happiness satisfaction anger disappointment shame mean tdiff mean tdiff mean tdiff mean tdiff mean tdiff

PD C/D 1.79 -4.07*** 1.76 -3.88*** 4.00 6.07*** 4.51 5.32*** 2.07 -4.60***

(33)

VI. Limitation, Discussion and Conclusion

A. Limitations

Interpreting these results, certain limitations should be considered. First of all, in these kinds of experiments demand effects are possible. Even though the survey was conducted online, subjects might try to answer the questions as they anticipate the experimenter would prefer. Additionally, 95 of the participants indicated having an economics background, which can also lead to biased data. Another possible limitation is the usage of hypothetical scenarios, which might not picture the reality appropriately. However, several studies (Kühberger et al. 2002; Kang et al. 2011) show that there are non significant differences in decisions and brain activity between hypothetical and real situations. In connection with the limitation due to hypothetical scenarios, one other aspect should be mentioned. Since subjects are put in situations rather than actually decide on their own, the hypothetical behavior could be more salient and thus lead to different levels of emotions. Salient thereby means that subjects might react in a different way when the hypothetical situation is in line with the behavior they would actually choose compared to scenarios where they are put in situations that deviate from their self chosen behavior. This aspect should be investigated in following studies.

An obvious further limitation in this study is using visual inspection in the additional analyses. No statistical consolidated results but rather trends are found that can possibly be used as pilot data for future research in this field.

B. Discussion and Conclusion

Summarizing the results, we find hypothesis 1 supported. Positive emotions are indeed significantly higher in fair situations, whereas negative emotions are increased in unfair situations. This is not only shown for each game separately but also over all games. Treating someone prosocially and receiving a prosocial treatment does lead to highly positive emotions, whereas treating someone unfairly and/or receiving and unfair treatment leads to higher negative emotions.

Controlling for the social value orientation (hypothesis 2) does not lead to any significant changes. There is neither a significant main effect for prosociality nor any interaction effect

(34)

that the degree of fairness in an economic situation determines the experienced emotions, no matter what social value orientation the individual shows.

Additionally, it may be noted that individuals experience significantly higher levels of happiness, camaraderie and satisfaction when acting prosocially, whereas shame is predominant when acting in a selfish way. This is not only found in sequential but also in simultaneous games. These interesting findings support the assumption that individuals - in order avoid feelings of shame – choose prosocial behavior over selfish behavior, even it that includes a lower outcome. Additionally, one could suggest that subjects seek the experience of positive emotions such as happiness, satisfaction and also camaraderie and therefore act prosocially.

The findings lead to the assumption that individuals might profit from being prosocial and thus develop some kind of personal emotional utility from sharing (money) with others. Aiming to increase or even maximize this emotional utility could provide an interesting explanation for the prosocial behavior which deviates from theoretical prediction so often observed.

Further research should focus on analyzing emotional responses in real and not hypothetical situations. This could be done in a similar way, where subjects would have to explain why they acted in a certain way. Although this might be a quite obvious way to analyze behavior, the findings here could be useful as a means of comparison. Establishing even more detailed connections between certain situations and experienced as well as expected emotions might make it possible to explain human behavior that deviates from economic standard theory.

Outlook

Establishing systematic connections between emotions and specific social and economic decisions could be helpful in the field of neuroeconomics. As stated above, neuroeconomics already sheds some light on the role of emotions in decision making situations. Several studies have been conducted in order to find evidence for the importance of emotions in social decision making (for a detailed review see Sanfey, 2007). For instance, Rilling et. al (2008) let subjects play an iterated prisoner's dilemma while analyzing neural correlates using fMRI. After every possible outcome, subjects had to rate the strength of several emotions. Subjects sensed significantly more happiness when both