EMOTION EXPRESSION AND INDIRECT RECIPROCITY

1. Introduction

Studies have shown that people tend to treat others negatively or positively depending on the previous treatment (negative or positive) they have experienced from others. This is known as indirect reciprocity (Nowak and Sigmund, 2005; Li et al., 2020) or generalized reciprocity (Gray et al., 2014; Strange et al., 2016). The resulting chain effect may have non-negligible welfare implications (Li et al., 2020). For example, if A is unkind to B, and B treats C in the same manner, an innocent person can become the victim of this chain of welfare-reducing interactions. In this study, we examine how emotion expression (EE) may help to disrupt this chain and promote social welfare.

Having the possibility to write messages to express feelings might be an effective way to stop spillover effects and therefore maintain productive relationships. This is important in real life where a substantial part of our days are spent in the company of strangers (Seabright, 2004). A notable example is online shopping, for which we rely on strangers through e-commerce such as Alibaba, Amazon and eBay. Unlike traditional markets where you can give feedback directly to sellers face to face, in online trade, you are in a virtual trading platform, whose setting may cause misunderstandings and prejudices without further communication. In case of being unkindly treated, which commonly happens, the negative feeling of the customers resulting from the unpleasant experience might break the trust in the seller and spill over to other online sellers. In fact, many online markets rely on “feedback” systems to promote trust in transactions (Bolton et al., 2004; Houser and Wooders, 2006).

A number of studies have shown the importance of EE in shaping decisions, for instance, in reducing punishment (Xiao and Houser, 2005, 2009; Bolle et al., 2014; Dickinson and Masclet, 2015). The possibility of verbal expression to an audience is also found to increase cooperation in ultimatum games and hence welfare (Samahita, 2017). Our study is most closely related to Strange et al. (2016) and Li et al. (2020). Both of these studies use two sequential dictator games and focus on the effects of message writing towards a third new person on decisions in the second dictator game. We extend and complement these papers in three ways, which we note here and detail further in the literature review below. First, we clearly identify the dictator’s behavior towards the receiver as neutral, kind, or unkind by setting three conditions from which the dictator can choose in the first interaction. We test, under each condition, how EE impacts receivers’ follow-up decisions towards a new person. Second, we run a message evaluation experiment to classify messages and examine how a message’s content impacts decision-making. Finally, we explore heterogeneity by ethnicity.

We report data from two dictator games by comparing decisions towards another person in the second dictator game with and without EE treatment in the first dictator game. While being otherwise identical to the control experiment, in EE treatment the receivers have an opportunity (but not a requirement) to write a message to their respective dictators in the first dictator game.

Our findings support a clear positive role of EE in generating kind behavior in new interactions. That is, in general, people tend to behave less selfishly as a new dictator towards a new person if they have had a chance to convey emotions in the previous interaction. We find that the treatment effect is driven by both the neutral and unkind conditions, suggesting that recipients take the kind condition as a reference point, and that the reference point varies depending on context (Knetsch, 2010). Further, based on the message experiment, we classify the messages as positive, negative, or neutral. We find no effect of message content on decision-making. However, participants who left messages blank behaved even more generously than those who chose to write messages. This suggests that simply having the chance to convey emotion plays a prominent role in breaking the chain of negative reciprocity. Finally, the exploration of treatment heterogeneity by ethnicity shows that the intervention is less effective for Asians.

The remainder of this paper is organized as follows. Section 2 reviews the related literature. Section 3 introduces the behavioral hypothesis. Section 4 describes the experiment design and procedures. Section 5 reports the results. Section 6 concludes and discusses.

2. Literature Review

Our study is closely related to three branches of literature: behavioral impacts of emotions, emotion venting in reducing punishment and the effects of EE on indirect reciprocity.

2.1. Behavioral impacts of emotions

Emotion contributes to decision-making (Dickinson and Masclet, 2015). Despite the long ignorance of emotions in economic analysis, more and more studies have shown that emotion is an important determinant in many economic decisions. Like conventional preferences, moods and emotions determine the trade-offs that people make between different goods and activities (Loewenstein, 1996, 2000). Many studies have proved the positive role of good mood in promoting pro-social behavior. Using the induced emotion approach, Drouvelis and Grosskopf (2016) find that happy subjects contribute more than angry subjects on average, and happy subjects punish less harshly than angry subjects in one-shot Voluntary Contribution Mechanism (VCM). Many other studies find that anger accompanies the application of costly punishment (Bosman and van Winden, 2002; Ben-Shakhar et al., 2007; Hopfensitz and Ernesto, 2009; Joffily et al., 2014; Dickinson and Masclet, 2015).

The theory of the dual-system frameworks also supports both rational and emotional inputs to a decision (e.g., Camerer et al., 2005; Kahneman, 2011). From a neural science point of view, it has also been shown that when observing opportunistic behavior, anterior insula activation, which is typically associated with aversive stimuli, correlates with subsequent individuals’ decision to punish others (Sanfey et al., 2003). The work of Drouvelis and Grosskopf (2016) used short video clips to induce happiness and anger in a one-shot VCM environment. They found that angry subjects punished more than others, while happy subjects contributed more than angry subjects.

2.2. EE in reducing punishment

An increasing number of studies in experimental economics have supported the positive role of emotion venting in reducing punishment (Xiao and Houser, 2005, 2009; Bolle et al., 2014; Dickinson and Masclet, 2015). Our study is highly inspired by the recent work of Dickinson and Masclet (2015). They use the standard VCM punishment environment by applying several treatments that allow players to vent their emotions prior to making punishment decisions. They administer each additional opportunity to vent emotions: they start with a simple cooling-off period, but then add the opportunity to self-report one’s emotional state as well as assign virtual punishment points. Overall, they find that the venting emotion treatments lead individuals to assign significantly less punishment points to others compared to a treatment without the opportunity to vent emotions.

The current study is also closely connected to the work of Xiao and Houser (2005). By conducting ultimatum games with and without EE treatment, the study of Xiao and Houser (2005) shows that rejection of unfair offers is significantly less frequent in EE treatment when responders can convey their feelings to the proposer concurrently with their decisions. According to their interpretations, responders might feel better after explicitly displaying their emotions to proposers; or, perhaps sending messages of disapproval directly to one’s proposer might be a satisfying alternative form of punishment; or, a responder might believe that accepting a low offer would be interpreted by the proposer as indicating that the responder accepts an inferior position. By expressing anger or disapproval regarding the low offer, responders can deny this interpretation. In all, their study supports the evidence of the human demand to express emotions and its important impacts on decision-making.

2.3. The effects of EE on indirect reciprocity

The studies of Strange et al. (2016) and Li et al. (2020) are closely related to our study. Both studies design two sequential dictator games and focus on the effects of message writing towards a third new person in the second dictator game. Our paper complements theirs in the following ways. Strange et al. (2016)’s study only uses female participants which may limit the validation of the results to a more general group of population. In our study, we use a gender mixed population. Also, in their study, the receivers could only have two choices, either an equal distribution between the dictator and the receiver (12.5 versus 12.5), or an unequal distribution (20 versus 5). In other words, the dictators can offer either 50% or 20% of the total pie. In our setting, we have an initial unequal endowment for both the dictator and the recipient. Based on an initial distribution, the dictator has three options: to keep the initial endowment unchanged, to take some from the recipient, or to give some to the recipient. The initial distribution could be considered as a reference point, while a more equal or a more unequal choice could be considered as two opposite behaviors, either kind or unkind one. Third, in their study, the dictator was not present physically in the lab until the receivers had finished writing the messages. In our setting, the receivers and the dictators are in the same room while making decisions respectively. We believe that the presence of the participants all through the experiment in the lab creates a different setting. One possible difference is that the recipient may feel more connected with the dictator and may lead to higher expectations from the dictator.

The main differences between ours and Li et al. (2020)’s work are as follows. First, though their study allows the dictators to offer any proportion of a 25 yuan pie, ours has an initial endowment for both players and therefore a clear reference point based on an initial distribution, as such, the receivers could identify three types of behaviors: kind, neutral and unkind behavior. Second, our study includes a message evaluation experiment, as such, we can analyze the message contents written by the receivers and relate them to their decision-making. Third, while Li et al. (2020)’s study is conducted in China using Chinese subjects, our experiment is conducted in the US with the participants originating from different countries, allowing us to analyze the behavior under the context of cross-cultural background. Specifically, in the current study, we compare the Asian and non-Asian participants and find their behavioral differences in the decision-making.

3. Behavioral Hypothesis

The hypotheses that we develop are based on evidence from existing studies. In particular, our hypotheses are closely related to consistent findings from several experimental studies suggesting a common human behavior in decision-making: when having the chance to convey feelings, people tend to behave more pro-socially. For example, the study of Xiao and Houser (2005) shows that rejection of unfair offers is significantly less frequent when responders can convey their feelings to the proposer. Dickinson and Masclet (2015) find that venting emotion leads individuals to assign significantly less punishment points compared to cases without the opportunity to vent. Also, Bolle et al. (2014) observe that venting reduces aggression in a vendetta game.

Therefore, we believe that EE plays the role of an external stimulus that shifts one’s “mood”. Specifically, EE could shift the bad mood to good mood (e.g., an angry person becomes a happy person) and turn good mood to better mood (e.g., a happy person becomes a happier person). The “mood shift” role can be understood through the emotion venting procedure. In the framework of the dual system, according to Dickinson and Masclet (2015), emotion venting could reduce the emotional impulse (system 1) before making decisions. This may be accomplished by forcing the decision maker to wait beyond the impulsive time frame when making a punishment decision. In psychology, the Catharsis theory perspective suggests the beneficial effects of expressive writing to release emotions. It is believed that expressing inhibited thoughts and feelings can reduce stress and free the mind of unwanted thoughts (Frattaroli, 2006), therefore improving physical and psychological health outcomes.

In line with the above studies, our first hypothesis is that allowing people to express emotions after an interaction will positively affect their decisions in future interactions.

Nonetheless, we would expect the effects of EE to be different depending on the decisions of the dictators. In our setting, we can classify three clear types of behaviors for the dictator with reference to the initial endowment: a neutral dictator is the one who keeps the initial endowment, while the kind and unkind dictators are those who give or take money from the recipient, respectively. The work of Knetsch (2010) suggests that assessment procedures and circumstances can have a major influence on the reference state on which people’s valuations are based. In line with his work, our particular setting may suggest two possible reference points depending on how people value the “loss”. First, if the recipients believe that kindness should be the common social norm, particularly in a setting where the dictator owns higher payoff than the recipient even in the kind condition, then both the neutral and unkind decisions of the dictator would mean “unfairness” and therefore generate negative feelings for recipients. If the recipients consider the initial endowment as the reference point, then only unkind behavior might generate negative feelings for the recipient, implying we might foresee more effects of EE in this condition. This study thus also provides a way to test how people perceive fairness (Kahneman et al., 1986), that is whether both the neutral and unkind choices or only the unkind choices of the dictators are considered as unfair. As such, we have the following hypothesis.

The contents of a person’s message should generally reflect how they have been treated. For instance, a person who is generously treated is likely to write positive messages, such as a message of gratitude to thank the dictator. However, a person who has been treated negatively may nonetheless choose to write positive messages out of politeness or social consideration, even if they too feel angry. Moreover, some participants may prefer to remain silent even if they have the opportunity to express their feelings. Given the possibilities of various motivations underlying the specific message written, it is possible that the opportunity to write a message is more impactful on behavior than the specific message people write. This suggests the following hypothesis.

Many studies have found behavioral differences relating to cross-cultural and ethnic backgrounds (Cox et al., 1991; Henrich et al., 2005; Gächter et al., 2010). People of different ethnic backgrounds possess different attitudes, values, and norms that reflect their cultural heritages (Cox et al., 1991). Would people of different ethnic backgrounds respond differently to opportunities for EE? We investigate whether EE is more effective for certain ethnic groups.

4. Experimental Design and Procedure

4.1. Dictator games with EE

The experiment is composed of two sequential dictator games with and without EE (EE or NEE). In the first dictator game, randomly assigned Player As and Player Bs receive an initial endowment of $9 and $6. Player As have three options (see Table 1): make no adjustment (Option X), give $1 to Player Bs (Option Y), or take $3 from Player Bs (Option Z). In the second dictator game, Player Bs become Player Cs, i.e., the new dictators. The new initial endowments are $9 for Player Cs and $6 for randomly matched Player Ds. Player Cs can choose to give some of their $9 endowment to Player Ds; to take some amount from Player Ds’ $6 endowment; or to make no adjustment to the endowments.

We follow the design of Xiao and Houser (2005) in applying the EE treatment. It is the same as NEE except that Player B is permitted, but not required, to write a message to Player A with no pecuniary cost. Any message is delivered to Player A. Messages cannot have any strategic implications because (i) Player A has made their decisions before they see Player B’s messages; (ii) all experiments take place anonymously, and (iii) each pair of subjects plays the game only once. Rather, a message provides an opportunity for a receiver to voluntarily display her feelings regarding her matched Dictator’s decision.

4.2. Experiment procedures

The experiment recruited 578 student subjects from the general student population at George Mason University by using standard procedures in place at the Interdisciplinary Center for Economic Science between 2016 and 2017: 146 pairs in EE and 143 pairs in NEE treatment. We ran in total 45 sessions with each session having 12–14 participants.

First, subjects were randomly assigned to Player A and Player B. They were separated by two rows of seats to make sure that they could not see each other. Each subject was randomly assigned a letter as his/her ID in the experiment. After reading the instructions (see Appendix A, we use the instructions of EE treatment for the receivers as an example), each subject was required to successfully complete a quiz to verify comprehension.

Player A (the dictator) chooses his or her option (one of the three) on a decision sheet and puts the decision sheet into an envelope. After all dictators had finished, the experimenter took all the envelopes and randomly matched them with a Player B. The experimenter then distributes the envelopes to the randomly matched Player B. Player B will then open the envelope and read the decisions made by their matched Player A. After they have finished reading, the copy of the decision sheets would be returned to Player A as a record. The first experiment then ended. In the second experiment, Player C (the former Player B), the new dictators needed to make their decisions on a decision sheet whether to adjust the new endowments ($9 and $6) between himself/herself and a new partner Player D. Once they made their decisions, they needed to put the decision sheets into the envelope. After all the dictators had finished, the experimenter took all the envelopes and randomly matched with a new Player D. After Player D read the decisions, the experiment ended. In the EE treatment, following Xiao and Houser (2005), after Player B had read the decisions from Player A, they could choose to write a message on the decision sheets to Player A. After they had finished writing the message, the experimenter collected all the copies of the decision sheets with messages and returned these copies with messages to the corresponding Player A. Player B was asked to avoid indecent language but was otherwise given no guidance regarding what, or whether, to write.

In both treatments, subjects were paid privately with cash at the end of the experiment. Each subject received $5 show-up bonus in addition to money earned in the game. Subjects were in the lab for about 1h and earned about $15 on average. The duration of EE treatment may be slightly longer than NEE treatment due to message writing time. However, this slight time difference would not be a concern for the potential confounding “cooling down” time effect. This is because our experiment is a paper and pencil experiment, and it requires Player B in both EE and NEE treatments to wait for some time before making their decisions in the second experiment. After Player Bs in the first experiment read the decisions from Player As, the experimenter needs to collect all the copies and return them back to the paired Player A. Then, before Player B makes decisions in the second experiment, the experimenter needs to read the instruction for the second experiment first and then distribute the exercise sheet. After all the participants had finished the quiz, the decision sheets were then distributed. That said, both in EE and NEE treatments, Player Bs need to wait for a relatively long time before making decisions. We estimate that this waiting time lasted at least 10min. Also, as the messages are usually quite short, and it took on average less than 2min for participants to finish writing the messages, therefore, relatively, the time duration for writing could be ignored.

4.3. Message evaluation

Testing our hypothesis requires evaluating the content of our responders’ messages. To perform such an evaluation, we follow the work of Xiao and Houser (2005). We run two sessions of message evaluation experiments by recruiting 31 message evaluators from the general undergraduate population at George Mason University (see Appendix B for the instruction for the message evaluation experiment) in 2018. Potential evaluators were excluded if they had previously participated in the former communication experiment. After being seated in the laboratory, each evaluator was given the instruction of the message evaluation task. Besides, to provide the evaluators with the context, we also gave them instruction on the EE treatment of the communication experiment. After completing the instructions, the evaluators were given a randomly ordered listing of all 136 anonymous messages written by the responders in the EE treatment, as the remaining 10 participants in EE treatment did not write any messages. Subjects were asked to classify the messages as showing “positive” or “negative” emotion or as being “neutral” (not positive and not negative). Evaluators were not given any information regarding the situation of the responder who wrote the message, i.e., they did not know the responder’s decision. Subjects were paid $5 for attending and an additional $5 for completing the entire evaluation. To increase the subjects’ attentiveness, they were told that after all evaluations were completed, three messages would be randomly chosen as payoff messages. If the subject’s evaluation matched the most popular evaluation for a message, then they earned an additional $10. Subjects were in the laboratory for about 1h, and median earnings were $30. Messages are classified according to the most popular classification chosen by the evaluators. There was a single most popular classification in 135 of 136 cases.

5. Results

5.1. Taking decisions in EE versus NEE

Table 2 displays the comparisons between EE and NEE treatments for player C’s decisions in a new interaction with a third person. In all cases, Player Cs chose to take less money in EE treatment than in NEE treatment. That said, in general, Player Cs behaved more generously in EE treatment than in NEE treatment. Specifically, on average, player Cs took only $1.59 in EE, but they took $2.36 in NEE. The difference is significant at the 5% level. This is particularly true for the two conditions that Player As behaved neutrally (A takes “0”) or unkindly (A takes “3”) in the first dictator game (also see Figure 1).

We run OLS regressions to control for Player Cs’ personal characteristics as well as their respective conditions in the first dictator game. Table 3 shows the regression results. Consistent with Hypothesis 1, the variable EE treatment in the first column is significantly negative for the “All sample” model, confirming the positive role of EE in generating kinder behavior. In the other columns, we separate the sample according to the specific condition that the recipient was in the previous dictator game. The results show that the treatment effects are mainly driven by the case where recipients were either neutrally or unkindly treated by the dictators (“A takes 0 or 3”). When we separate “A takes 0” and “A takes 3”, both sub-samples are marginally significant, these relatively higher p-values for the sub-samples are likely due to the much smaller sample size for both sub-samples.

In all, our results tend to support Hypothesis 2 that the treatment effect is driven by both the neutral and unkind conditions, suggesting that the recipients are likely to take the option Y (“A gives 1”), the kind behavior, as their reference point of expectation from the dictator. This is possible, as in our setting, even in this kind condition compared to the neutral and unkind condition, the dictator still owns higher payoff than the recipient. As such, the recipients may consider both the neutral and unkind options as unfair decisions.

We find no significant effect of treatment in the case where recipients were kindly treated (“A gives 1”). If people are already happy, writing messages may not significantly increase their happiness. Another possibility is that the (presumably) positive feelings generated by the opportunity to vent one’s emotions are not strong enough to generate significant behavioral changes.

5.2. Messages analysis

We further investigate the contents of the messages in the EE treatment. Figure 2 shows the distribution of messages categorized in the message evaluation experiment. When being kindly treated (option Y, being given $1), around 89.7% of receivers wrote positive messages. A small proportion of responders expressed negative and neutral emotions (1.7% and 5.2%) about this outcome. 3.4% of receivers chose not to leave messages at all. When being neutrally treated (option X, no changes in initial endowment), 47% of responders displayed positive emotions; 32% expressed neutral emotions; and 2.9% expressed negative emotions. Eighteen percent chose not to respond. When being unkindly treated (option Z, being taken $3), 50% of responders wrote negative messages, while 31% wrote neutral messages. Still, a non-negligible proportion of players (around 15%) wrote positive messages to the dictators. The proportion of those choosing not to respond was 4% in this case. The distribution of messages suggests that the contents of messages reflect, but is not dispositive of, how a person was treated in the prior interaction. Also, by comparing the proportion of responders who did not write any messages at all in the three cases, it is worth noting that the tendency to express feelings becomes more obvious when the recipients are either being kindly or unkindly treated rather than being neutrally treated.

We run further regressions to test message effects. We focus only on the EE treatment sample and add dummies of types of messages into the model while keeping all the other control variables consistent as Table 3. Table 4 shows no significant relationship between message content and Player Cs’ decisions. However, it is worth noting that those who chose not to write any messages to the dictators tended to behave more generously than those who wrote messages. All these results seem to support Hypothesis 3, i.e., the message content has little bearing on decision-making, and that having the chance to convey emotions matters.

5.3. Treatment heterogeneity analysis

Do people of different traits all respond in the same way to the intervention? In this section, we further analyze whether and how treatment effects differ across different groups of the population. We divide the sample by ethnicity and run sub-sample regressions using the same set of variables as in the basic regressions. Table 5 shows that the treatment effect disappears for the Asian population, while it holds for the non-Asians. In other words, writing messages seems not to impact Asian behavior.

There are several possible explanations. First, Asians may be less likely to express directly their true feelings. Indeed, our data show that in the case that the recipients are unkindly treated, only 4% of non-Asians write positive messages, 17% of Asians still write positive messages towards the dictators. That said, what is expressed in the message seems largely to deviate from true feelings so that emotion venting could not be fully realized. As a result, for Asians, opportunities for EE may be ineffective in changing behavior. Another possible explanation is that Asians are more likely to maximize earnings than non-Asians. From Table 2, we find that the Asian subjects tend to take significantly more than other participants on average, and even more so when treated unkindly. This may suggest that maximizing earnings may be the dominant consideration in decision-making, leaving little room for EE to play a role in shaping decisions.

6. Conclusion and Discussion

We conducted a two-stage dictator game with and without the opportunity for EE to examine the role of expressing emotions in shaping indirect reciprocity. Consistent with Strange et al. (2016) and Li et al. (2020), our results provide evidence that EE causes more generous behavior in a new interaction. In addition, our study sheds light on the specific conditions under which message writing can help to generate positive indirect reciprocity.

First, our results show that treatment effects are driven by both the neutral and unkind conditions, suggesting that recipients are likely to take kind behavior as their reference point. Alternatively, if the recipient takes the neutral condition as a reference point, i.e., the dictator keeps the initial endowment unchanged, then we would expect to observe treatment effects only in the unkind condition. Therefore, the results suggest that the impact of EE depends on the specific condition the recipients are in. This is in line with the idea of Knetsch and Tang (2006) that the context influences people’s views and different contexts appear to give rise to different valuations. According to our findings, recipients tend to view both the neutral and unkind conditions as unfair, as in fact in our case, even in the kind condition the dictators receive a higher payoff than recipients. Therefore, the recipients are likely to expect that the dictators should choose the kind option as the fair decision.

Second, we found that message content did not affect decision-making in our experiment. Rather, subjects who left messages blank tended to behave even more generously than those who chose to write messages. The results indicate that the opportunity to convey emotion may be more important than the content of the message, perhaps because the chance for communication represents respect.

Third, we test the treatment heterogeneity effects by ethnicity, reaching the conclusion that treatment effects are not homogenous. In fact, we only find effects for non-Asians. It might be that Asians are reluctant to express feelings directly, leaving little room for EE or its opportunity to impact their behavior. Another explanation might be that Asian participants are more likely to be earnings-maximizers than non-Asians.

Overall, our study is a step forward to an improved understanding of the welfare implications of communication and its impact on human behavior. Our findings complement the existing literature by shedding new light on the conditions under which opportunities to express emotions can be effective in breaking cycles of indirect reciprocity. There remains much more to be done in understanding the role of communication in fostering positive social interactions and human flourishing.

Acknowledgments

We are appreciative of the experiment support from Erte Xiao. We thank Marie Claire Villeval and the seminar participants at GATE-LAB (2017) for their helpful comments. We thank Lina Diaz and Yuzhi Wang for research assistance. Support for this research was funded by the National Natural Science Foundation of China (Grant No. 72273014), and the Humanities and Social Science Foundation of the Ministry of Education of China (Grant No. 22YJA790070).