Anger Management: Aggression and Punishment in the Provision of Public Goods

Abstract

The ability to punish free-riders can increase the provision of public goods. However, sometimes, the benefit of increased public good provision is outweighed by the costs of punishments. One reason a group may punish to the point that net welfare is reduced is that punishment can express anger about free-riding. If this is the case, then tools that regulate emotions could decrease the use of punishments while keeping welfare high, possibly depending on pre-existing levels of aggression. In this lab experiment, we find that adopting an objective attitude (objective), through a form of emotion regulation called cognitive reappraisal, decreases the use of punishments and makes a statistically insignificant improvement to both net earnings and self-reported emotions compared to a control condition (natural). Although the interaction between the emotion regulation treatment and level of aggression is not significant, only low aggression types reduce their punishments; the results are of the same direction, but statistically insignificant for high aggression types. Overall, our findings suggest that pairing emotion regulation with punishments can decrease the use of punishments without harming monetary and mental welfare.

1. Introduction

There are many examples of small groups trying to provide a public good: the football team keeping their locker room clean, MBA students doing a group project together or faculty making sure all departmental needs are met. Additionally, something that can increase the provision of these public goods is allowing peers to punish each other. For example, football teammates may not guard an inconsiderate quarterback very well; a student can give a bad rating to a lazy group member; or a faculty member can speak negatively about a bad citizen at a faculty meeting. Such punishments are a common mechanism for increasing the provision of public goods in real life and in lab experiments. However, punishments do not always increase net welfare because often, the costs of punishment outweigh any gains in public good provision. For example, the football team may have a spotless locker room, but no one to enjoy it due to excessive injuries.

There are many reasons a group may punish to the point that net welfare is reduced. One such reason is that punishment can express anger about free-riding. If this is the case, then tools that manage that anger could decrease the use of punishments while keeping welfare high. Additionally, it seems likely that such emotion regulation may be particularly beneficial to individuals and groups who are more aggressive. One can imagine that the highly aggressive football team may benefit from tools that reduce anger more than a group of Quakers raising a barn.

In the present lab experiment, we examine the impact of both emotion regulation and levels of aggression on monetary and mental welfare. Subjects play a linear public goods game with punishment in groups. We randomly assign groups to receive one of two types of instructions. In sessions with objective instructions, subjects are told to “please adopt a neutral, analytical and objective attitude in order to stay calm,” while in those with natural instructions, they are told to “please respond naturally with whatever thoughts and emotions occur to you spontaneously.” In addition, we use a pre-screen survey to measure subjects’ baseline aggression levels to randomly assign subjects to homogeneous groups of all high or low aggression types.

We find that objective instructions decrease the use of punishments. However, the interaction of aggression type and objective instructions is not significant. Only the low aggression types have a statistically significant reduction in punishments, while the results are in the same direction but statistically insignificant for high aggression types. In addition, objective instructions improve both net earnings and self-reported negative emotion when compared to natural instructions. However, these improvements are not statistically significant for low or high aggression types.

Our findings suggest that using emotion regulation instructions decreases punishments and does not harm either monetary or emotional welfare. Our emotion regulation is a very light touch. As such, it is easy to imagine how a message like “adopt a neutral, analytical and objective attitude in order to stay calm” could be paired with opportunities to punish a free-riding peer in the real world (e.g., when being asked to rate a colleague or group member).

In the next section, we briefly describe the related experimental literature. We detail the experimental design in Section 3, report results in Section 4 and offer conclusions in Section 5.

2. Related Experiments

Peer punishment can raise contributions to the public good in lab experiments, but the effects on monetary welfare as measured by net earnings (group earnings minus punishment costs) are mixed. The majority of the previous experiments find that peer punishments decrease net earnings 1, with only a handful of studies finding that peer punishments increase monetary welfare [6,7,8].

Previous work has shown that there are a number of ways to increase the effectiveness of punishment by increasing the severity of punishment [8], coordinating punishment through a centralized mechanism [9,10,11,12], eliminating counter-punishment [13,14,15], eliminating antisocial punishment [16] or by allowing a longer timeline for learning how to use punishment [4]. In this paper, we propose using emotion regulation to increase the effectiveness of punishment.

There is evidence that people use punishments to express their anger about free-riding. The work in [6] shows that subjects will use non-monetary disapproval messages to punish free-riders, and they suggest that the use of such messages is a means of expressing anger with free-riders. The work in [17] finds that subjects will pay for the right to assign punishment to free-riders personally rather than let someone else do the punishing. Additionally, [14] find that punishment of a non-cooperator by another subject is not a substitute for one’s own punishment of that non-cooperator. The fact that subjects have a preference to assign punishments themselves suggests that they may be using punishments to express anger about free-riding. The work in [18] finds that observing free-riding triggers negative emotions and that punishing free-riders is associated with higher self-reported positive feelings. However, they also find that those who are punished experience negative emotions. Overall, the effect of punishments on group emotional welfare is ambiguous.

To our knowledge, only one paper tries to improve group monetary welfare by controlling subjects’ emotions. The work in [19] designs three treatments to help subjects release their negative emotions before making punishment decisions. The first treatment adds a cooling off period. The second treatment adds a self-reporting of emotional states to the cooling off period. The third treatment adds virtual punishment points to the cooling off period and the self-reporting. Results show that subjects have the highest net earnings in the cooling off-only treatment. Excessive emotion venting in the second and third treatments reduces contribution levels and decreases the punishment’s effectiveness.

Dickinson and Masclet [19] nicely illustrate the potential for interventions that alter emotions to have an impact on behavior in public goods’ contexts. It remains unclear, though, what it is about the cooling off period that has the effects they observed. What were participants doing in that 5-min period? One possibility is that participants were thinking about things unrelated to the game (“can’t forget to buy eggs on the way home”); or, maybe they were attempting to think differently about the game itself (“it’s just a game; keep a level head”). Both of these options represent effective ways to alter emotions, but they represent different forms of emotion regulation [20], with the former focused on altering attention to the situation and the latter focused on altering appraisal of the situation. Our study makes a novel contribution by explicitly manipulating just one form of emotion regulation, cognitive reappraisal.

Cognitive reappraisal refers to the process of reinterpreting emotion-triggering situations in order to alter one’s emotional response. For example, faced with criticism from a friend, a person might interpret the criticism as an opportunity to learn or consider the situation from the friend’s perspective. These alternative interpretations may mitigate the anger one might otherwise feel. Indeed, a recent meta-analysis summarizing the results of previous experiments indicates that cognitive reappraisal is an effective way of regulating emotions as reflected in self-report and behavioral measures [20]. In this work, we apply this strategy to the context of a public goods scenario. Specifically, we instruct participants to adopt a neutral, analytical and objective attitude in order to remain calm while making decisions. It was our hope that cognitive reappraisal would reduce the use of punishments while still improving net earnings and emotional welfare.

Previous work has found that emotions can affect actions in a variety of settings and specifically in the use of punishment2. Among all of the negative emotions that might increase the amount of punishment assigned, anger’s role is especially emphasized. The work in [15,28] shows that higher anger intensity is positively related to higher punishment. The work in [29] uses a short video to induce subjects’ happiness or anger before they play the public goods game and find that angry subjects tend to contribute less and punish more than happy subjects. The work in [29] shows that induced anger increases the use of punishments, so a next step is to test if those with a pre-existing tendency to anger/aggression punish more than those with lower pre-existing aggression levels.

Finally, the design of this experiment allowed us to determine whether pre-existing aggression levels might moderate the effect of cognitive reappraisal on punishments and/or monetary or emotional welfare. Previous studies have shown these kinds of individual-difference moderation effects in other domains. For example, [30] found that those with a pre-existing tendency to ruminate on negative aspects of the self and life in general exhibited different patterns of change in brain activation when using cognitive reappraisal. In this work, we examine whether those with a pre-existing tendency to be aggressive exhibit different levels of punishments, earnings and/or emotional welfare when using cognitive reappraisal in a public goods scenario. To our knowledge, no existing studies point to the specific interaction pattern we might expect in this context. Intuitively, lower aggression people may be less likely to be overly emotional in their decision making and so might be less affected by emotion regulation. We acknowledge, however, that other patterns are possible.

3. Experimental Design

3.1. Pre-Screen and Recruitment

In October 2014, we began collecting responses to an online pre-screen survey using Qualtrics from students at Tufts University. The median time to complete the pre-screen survey was 2 min. Respondents answered 29 Buss–Perry Aggression Scale questions [31] and reported their gender and age (see the Appendix for full pre-screen survey). The Buss–Perry score, which exhibited acceptable internal consistency reliability in this sample (Cronbach’s alpha = 0.89), is highly associated with aggressive behaviors in economic decision making and video games [32,33]. Respondents with a total aggression score above 2.58 were defined as a high aggressive type3. We used 2.58 as the cut-off value to define two aggression groups because this was the median value of 223 responses collected before our experiment started in October 2015 4. Subjects were never informed of their aggression score or that of any other player.

In total, we had 487 respondents to our pre-screen survey between October 2014 and January 2016. We invited these respondents to participate in our 21 experimental sessions, which ran between October 2015 and March 20165.

3.2. Session Procedures

Subjects played ten periods of a linear public goods (LPG) game in fixed groups of four with random identifiers that changed each period to mitigate reputations and revenge motives. Each session had at least eight participants to ensure subjects did not know who they were specifically interacting with6.

3.2.1. Group Assignment

At the beginning of each session, subjects checked in with an experimenter and were assigned to a computer. Using data from the pre-screen survey, subjects were assigned to play in a four-person group that was either all high aggression individuals or all low aggression individuals. Subjects were not made aware of the group’s aggression makeup during the session, and groups of both types were allowed to be in the same session.

We use groups with all of the same aggression types because there is a well-documented tendency for people’s social networks to be made up of those who are similar to themselves [35]. Furthermore, if we had groups with heterogeneous aggression types, it would be difficult to disentangle the effect of emotion regulation from the effect of group makeup. Last, we expect that if there are differing treatment effects by aggression type, they will be most pronounced in homogeneous groups, so all high or low aggression types seemed a natural starting point7.

Table 1. Buss–Perry aggression score by group type and regulation condition.

	Mean	SD	Min	Max	Subjects
High aggression, natural	3.47	0.42	2.83	4.41	32
High aggression, objective	3.29	0.54	2.61	4.72	32
Low aggression, natural	2.04	0.32	1.41	2.54	32
Low aggression, objective	2.03	0.39	1.07	2.55	36

shows the individual aggression scores by aggression type (low or high) and whether subjects received objective emotion regulation or natural instructions. Using G*Power 3.1.9.2, we determined that, with a minimum of 32 subjects in each group, we had sufficient power to detect medium effects (Cohen’s d = 0.50) with power (1 − β) = 0.80 for the objective versus natural emotion regulation instruction comparison within each aggression type, aggregated across periods.

3.2.2. Overview of Experiment

Subjects were seated at a computer and made choices using ztree software [36]. First, all subjects completed a mood rating survey8. Then, the experimenter read instructions for the first five periods, which consisted of a linear public goods game with punishment and instructions on “how to approach this task.” Although every session followed the same general experimental procedures, sessions with the natural versus objective treatments differed in “approach” instructions, which we will detail in a moment. Subjects had to correctly complete a comprehension quiz before playing the first five periods. Subjects were told there would be a total of ten periods, and they would be paid for one randomly-chosen period9.

Upon completion of the fifth period, instructions for Periods 6 to 10 were read aloud. In these last five periods, subjects played the same game with the same instructions on “how to approach this task.” The only difference was that at the end of each period, after all actions had taken place and results had been reported, subjects rated their positive and negative emotions using the well-validated evaluative space grid [38], as shown in Figure 1. We chose to collect self-reported emotions because we wanted to measure both emotional and monetary welfare. However, self-reporting emotions is itself a form of emotional venting [19], so we only did this in Periods 6 to 10.

We used the evaluative grid space because it allows subjects to efficiently report negative and positive emotions, both of which might be particularly important in our study. Previous work shows the co-existence of both positive and negative emotions in other economic games [27]. For example, how might a subject feel who earned a low payoff this period, but who punished the free-riding group member? This subject may feel negative about the low payoff, but positive about punishing the free-rider. The grid rating procedure yielded two scores that ranged from zero to four, one for negative affect and one for positive affect, for Periods 6 to 10.

After all ten periods had been played, participants completed the same mood rating survey they did at the beginning of the session.

3.2.3. Linear Public Goods Game with Punishment

Each period, subjects were given a five-token endowment. In Stage 1, subjects can keep tokens in a private good that pays $1.50 per token to only them or they can contribute tokens to a public good that pays $1.00 per token to all four members of their group10. Because the individual return to keeping a token ($1.50) is greater than the individual return to contributing to the public good ($1.00), the Nash equilibrium of the one shot game is zero public contributions11. Stage 1 payoffs are:

$${\pi }_i^1=1.5(5-g_i)+\sum _{j=1}^4{g}_j$$

(1)

In Stage 2, subjects are shown the Stage 1 contributions and earnings of their group members with randomly-created identifiers that change each period. Subjects can then assign punishment points to any other person in their group12. A punishment point costs $1 to assign and reduces the income of the subject receiving the punishment by $313. Thus, Stage 2 payoffs are Stage 1 payoffs minus the cost of punishment assigned by i to others and minus three-times the punishment points assigned to i:

$${\pi }_i^2={\pi }_i^1-\sum _{j\ne i}{p}_{i,j}-3\sum _{k\ne i}{p}_{k,i}$$

(2)

In the final period of play, there is no reason to pay the costs to assign punishment points as they cannot deter future free-riding behavior, so we expect $p_{i,j}=0$ in the final period. As such, if no punishment points are assigned in Stage 2, an own profit-maximizing player will set $g_i=0$. The game is played a finite number of times, so using backward induction, we expect that own profit-maximizing players will choose $g_i=0$ for all ten periods, and if all players choose this amount, they will each earn $7.50 from the game. However, the group welfare-maximizing contribution is $g_i=5$. If all players choose this amount, they will each earn $20.00 from the game.

3.2.4. Emotion Regulation: How to Approach the Task

We ran sessions with either natural instructions or objective emotion regulation instructions. The difference between these two types of sessions was reflected in how subjects were instructed to approach the task, as follows:

• Natural: “For this experiment, we’d like you to please respond naturally with whatever thoughts and emotions occur to you spontaneously. Thus, while remaining focused on the task, please allow yourself to respond naturally to the intentions of other group members and the investment and deduction decisions they are making.”
• Objective: “For this experiment, we’d like you to please think about your decisions in this task in a way that helps you stay calm. Thus, while remaining focused on the task, please do your best to adopt a neutral, analytical, and objective attitude about the intentions of other group members and the investment and deduction decisions they are making.”

Subjects were also reminded of how to approach the task each period with a screen prior to making punishment decisions that displayed the following messages:

• Natural: “Remember: As you consider your decisions, please respond naturally with whatever thoughts and emotions occur to you spontaneously.”
• Objective: “Remember: As you consider your decisions, please adopt a neutral, analytical, and objective attitude in order to stay calm.”

Last, these same messages appeared on the screen used to make punishment decisions as shown in Figure 2.

The objective instructions were intended to encourage subjects to use a specific strategy, cognitive reappraisal, to regulate their emotions in order to stay calm. The natural instructions, roughly matched in terms of length and focus on other group members, were intended to serve as an active control condition as a basis of comparison. Subjects were not encouraged to regulate their emotions. Instead, they were encouraged to respond naturally. We felt it was important to compare the effect of the objective instructions to a treatment with other written instructions because the mere mention of how to approach the decision might have an effect, and we wanted to keep this constant across our treatments.

It is important to note that most if not all subjects who received objective instructions probably intuited that the instructions were meant to impact their behavior and emotions in the game (i.e., reduce punishments, increase contributions and/or improve emotional state). Much like how a patient undergoing anger management training knows that the goal of their counselor is to reduce the patient’s anger, this experimenter demand effect is part of the treatment of interest, not a limitation in this study context.

4. Results

We sought to determine whether emotion regulation via cognitive reappraisal might reduce the use of punishment while improving monetary and emotional welfare. We further sought to determine whether pre-existing tendencies toward anger/aggression might have an opposing effect on punishments and, possibly, moderate the effect of the cognitive reappraisal treatment on punishments and welfare. Importantly, subjects who were randomly assigned to objective emotion regulation instruction sessions did not differ from those assigned to natural instruction sessions on age, gender, average aggression score or starting mood14.

Table 2. Summary statistics by aggression type and instruction.

	Contribution	Punishment Assigned	Net Earnings	Positive Affect	Negative Affect
High aggression, natural	4.04	0.58	15.30	2.58	0.91
High aggression, objective	4.07	0.30	16.46	2.63	0.69
Low aggression, natural	3.52	0.58	13.97	2.08	1.14
Low aggression, objective	3.32	0.20	15.00	1.97	0.79
Obs.	1320	1320	1320	660	660
Min	0	0	−21.5	0	0
Max	5	13	21	4	4

Note: objective means the subjects received objective emotion regulation instructions, while natural means subjects received natural instructions. There were a total of 132 subjects. Contribution, punishment and net earnings were calculated for each subject across all ten periods of play; positive and negative affect scores from the evaluative space grid were collected and thus calculated in Periods 6 to 10 only. If a subject made negative profits in the period that was chosen for payment, they received the $5 show-up fee only.

presents summary statistics by aggression type and instructions.

4.1. Public Contributions

We did not have expectations before the experiment about the level of contribution by aggression type because we were most interested in measuring the use of punishments and monetary/emotional welfare. However, it is interesting to note that contributions to the public good are higher for the high aggression types than the low aggression types and that these differences become larger over time as illustrated by Figure 315. As such, we include group public contributions as a control in our analysis of punishment behavior to account for the different contributions by low and high types.

4.2. Punishment

Figure 4 shows that, in all ten periods, subjects of either aggression type punish less when they read the objective emotion regulation instructions. We did not have expectations about the use of punishment over time before the experiment. However, Figure 4 shows that the difference in the use of punishments by type of instructions falls over time for both types. This may be partially driven by changes in public contributions over time as illustrated in Figure 316.

Figure 5 shows the effect of the objective instructions on the number of punishment points assigned from a random effects regression that controls for group level public good provision. We use a random effects model to account for each subject playing 10 periods, and we use standard errors clustered at the group level to account for groups interacting over 10 periods17. Using this preferred specification, we find that for low aggression types, the objective emotion regulation instructions result in a statistically-significant decrease of 0.43 punishment points, about a 74% decline from the mean for low aggression types using natural instructions ($\frac{-0.43}{0.58}=-0.74$). This represents a roughly medium effect size (Cohen’s d = 0.56, 95% CI (0.08, 1.05))18. The objective emotion regulation instructions also decrease punishment for high aggression types, in this case by 0.27 points, a 47% decline, $\frac{-0.27}{0.58}=-0.47$, from the mean for high aggression types using natural instructions. This difference is not statistically significant, but represents an effect that is small to medium in size (Cohen’s d = 0.36, 95% CI (−0.14, 0.85))19. Overall, this pattern suggests that objective emotion regulation instructions reduce the use of punishments. The objective emotion regulation instructions reduce the use of punishments more for low than high aggression types, although this difference is not statistically significant.

4.2.1. Antisocial Punishment

One particular type of punishment that has been shown to be harmful to net earnings is antisocial punishment [16,44,45,46]. Antisocial punishment is punishment assigned to contributors rather than free-riders. The work in [16] defines weak antisocial punishment as occurring when a subject i assigns punishment points to subject j, but subject j’s public contribution was greater than or equal to subject i’s contribution. Using this measure, we found that under the natural instructions, 34% of punishments by high aggression types are antisocial, but only 19% are antisocial for low aggression types. The use of the objective instructions lowers the amount of antisocial punishment to 27% for high types, but had no effect on the level for low types. The use of the objective instructions does not result in a statistically-significant difference in the number of antisocial punishment points assigned for either low or high types. We take this as suggestive, but inconclusive evidence that the objective instructions decrease the amount of antisocial punishments for high aggression types20.

We have shown that the objective emotion regulation instructions decrease the use of punishment points and the amount of antisocial punishment, at least for high aggression types. In the next section, we determine whether emotion regulation instructions and levels of aggression improve monetary and emotional welfare.

4.3. Monetary and Emotional Welfare

Figure 6 shows that, within each aggression group, subjects have higher net earnings when there is emotion regulation (objective) compared to when there is not (natural) in eight of ten periods21. However, the difference is not statistically significant, as shown in Figure 7a. Figure 7 shows the effect of the objective instructions on our three welfare measures: net earnings, positive affect and negative affect. These estimates come from random effects regressions (Appendix

Table A2. Dependent variable: welfare measures.

	Net Earnings	Positive Affect	Negative Affect
	1	2	3
High Aggression	1.33	0.50	−0.22
	(0.81)	(0.27) ${}^{+}$	(0.23)
	[1.66]	[0.36]	[0.37]
objective	1.04	−0.11	−0.34
	(0.79)	(0.26)	(0.22)
	[1.61]	[0.26]	[0.32]
High Aggression × Objective	0.13	0.16	0.12
	(1.13)	(0.38)	(0.32)
	[2.15]	[0.48]	[0.46]
Period	0.26	0.00	0.05
	(0.03) ***	(0.02)	(0.02) *
	[0.09] **	[0.04]	[0.03]
Constant	12.53	2.06	0.73
	(0.60) ***	(0.27) ***	(0.24) **
	[1.38] ***	[0.39] ***	[0.33] *
N	1320	660	660
Test of high Agg natural = high Agg objective with unclustered SE
$P>\left|t\right|$ or $P>\left|z\right|$	0.15	0.85	0.34
Test of high Agg natural = high Agg objective with clustered SE
$P>\left|t\right|$ or $P>\left|z\right|$	0.41	0.90	0.51

There was a total of 132 subjects. The random effects regressions account for each subject playing ten periods. Clustered standard errors (SE) at the group level are reported in brackets using 33 clusters. Unclustered SE are reported in parentheses. In the final two rows, we report the results of a test that the objective instructions had a statistically-significant effect for high aggression types by testing the significance of combined coefficients objective + high aggression × objective. *** p < 0.001, ** p < 0.01, * p < 0.05, ${}^{+}$ p < 0.1.

)22.

Figure 7a shows that net earnings increase with the objective emotion regulation instructions for high aggression types by $1.17 a 8% rise from the mean for high types with natural instructions ($\frac{1.17}{15.30}=0.08$; Cohen’s d = 0.38, 95% CI (−0.12, 0.87)) and for low aggression types by $1.03 a 7% rise for low types with natural instructions ($\frac{1.03}{13.97}=0.07$; Cohen’s d = 0.31, 95% CI (−0.17, 0.78)), but neither of these differences is statistically significant.

Figure 7b shows that positive affect is similar in Periods 6 to 10 for subjects with the natural instructions or with the objective emotion regulation instructions. With the objective emotion regulation instructions, high aggression types report an increase of 0.05 points in positive affect, a 2% rise from the mean for high types with natural instructions ($\frac{0.5}{2.58}=0.02$; Cohen’s d = 0.04, 95% CI (−0.45, 0.53)). By contrast, low aggression types report a decrease of 0.11 points, a 5% decline from the mean for low types with natural instructions ($\frac{-0.11}{2.58}=-0.05$; Cohen’s d = 0.11, 95% CI (−0.37, 0.59)). Neither of these differences are statistically significant.

In parallel, with the objective emotion regulation instructions, high aggression types report a decrease of 0.22 points in negative affect, a 24% decrease from the mean for high types with natural instructions ($\frac{-0.22}{0.69}=-0.24$; Cohen’s d = 0.24, 95% CI (−0.25, 0.73)). Similarly, low aggression types report a decrease of a 0.34 points in negative affect, a 43% decrease from low types with natural instructions ($\frac{-.34}{0.79}=-0.43$; Cohen’s d = 0.37, 95% CI (−0.11, 0.85)). Again, neither of these differences are statistically significant, as shown in Figure 7c.23

We also explore whether there are any changes in the 11 mood ratings taken at the beginning and end of the experiment. We find that the objective emotion regulation instructions reduce feeling rejected and embarrassed for low aggression types, while they reduce feeling lonely and active for high aggression types24.

In short, although the point estimates for most of the measures of monetary and emotional welfare improve, the differences are not statistically significant.

5. Conclusions

Much of the previous work has found that peer punishments increase public good provision, but they can decrease group earnings net of the costs of punishments. In this paper, we show that using objective emotion regulation instructions instead of natural instructions decreased the use of punishments and caused a statistically insignificant increase in net earnings and emotional well-being. The punishment benefit accrued most clearly to our low aggression groups in that, although emotion regulation instructions prompted our high aggression groups to decrease punishment, the decrease was not statistically significant. Overall, it appears that objective emotion regulation instructions improve outcomes, but effects are sometimes small and perhaps noisy. The promise represented in these results suggests that more research into how various types of emotion regulation can affect behavior is needed.

Those in our study who receive the objective emotion regulation instructions see the message “please adopt a neutral, analytical and objective attitude in order to stay calm”, while those who see the natural instructions are told “please respond naturally with whatever thoughts and emotions occur to you spontaneously”. This type of emotion regulation is a very light touch. One can easily imagine displaying such messages in environments where peer punishment might be exacted on group members (e.g., when being asked to rate a colleague or teammate). In fact, the emotion regulation manipulation may have been too subtle as evidenced by the non-significant decrease in negative affect. A less subtle manipulation may have more powerfully reduced anger and, thus, produced fewer null effects.

Although the changes are not always statistically significant, both high and low aggression types use less punishment, have higher net earnings and report lower negative affect when receiving the objective emotion regulation instructions. This pattern indicates that emotion regulation might improve outcomes. The policy implication of our experiment might be that messages encouraging emotion regulation should be implemented alongside the availability of punishment.