- Email: [email protected]
Selfish punishers
Accepted Manuscript Selfish punishers an experimental investigation of designated punishment behavior in public goods Leonard Hoeft, Wladislaw Mill
PII: DOI: Reference:
S0165-1765(17)30200-8 http://dx.doi.org/10.1016/j.econlet.2017.05.022 ECOLET 7625
To appear in:
Economics Letters
Received date : 27 February 2017 Revised date : 8 May 2017 Accepted date : 19 May 2017 Please cite this article as: Hoeft, L., Mill, W., Selfish punishers an experimental investigation of designated punishment behavior in public goods. Economics Letters (2017), http://dx.doi.org/10.1016/j.econlet.2017.05.022 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
*Highlights (for review)
Highlights: • Designated punishers mitigate free riding while contributing less than non-punishers. • Punishers undercut their own enforced norm.
• The discrepancy between punishers and non-punishers grows over time.
*Manuscript final version Click here to view linked References
Selfish Punishers An experimental investigation of designated punishment behavior in public goods Leonard Hoefta,b , Wladislaw Milla,c,∗ a International
Max Planck Research School on Adapting Behavior in a Fundamentally Uncertain World. Planck Institute for Research on Collective Goods, Kurt-Schumacher-Straße 10, 53113 Bonn Germany. c School of Economics and Business Administration, University of Jena, Bachstraße 18k, 07743 Jena, Germany. b Max
Abstract We show that a second-party punisher forces his peers to contribute to a public good while contributing significantly less himself. This effect increased over time and casts doubt on the prevalent prosocial interpretation of (designated) punishment behavior. Keywords: Punishment; Public Good; Experiment; Selfish JEL: H41, C92, K42 1. Introduction Imagine working with a supervisor on a common project: Everyone benefits from working on it, yet only one enforces contributions. From monarch to managers, many more real-world examples are conceivable. Yet research has concentrated mainly on situations where everyone can enforce contributions. In this situation participants reliably use (peer) punishment to solve social dilemmas (Fehr and G¨ achter, 2002), even though it constitutes a second-order public good in theory. This interpretation has extended to third-party (Baldassarri and Grossman, 2011) and secondparty (O’Gorman et al., 2009) single punishers, who solve the social dilemma efficiently as well. Additional prosocial motivations range from equality concerns (Dawes et al., 2007) to retributive fairness (Falk et al., 2005). However, punishment can also be destructive: Some punishment decisions are motivated by spite and retaliation (Herrmann et al., 2008, Houser and Xiao, 2010, Falk et al., 2005). Traditionally, philosophers have been skeptical of the prosocial use of power and argued for a state monopoly on punishment (Hobbes, 1998). Often, punishment can be self-serving as well as prosocial. Especially a second-party punisher profits from his own use of power. Although he provides a public good among the other players, he could abuse his power by undercutting his enforced contribution norm. Therefore, not only withholding punishment can be selfish (Leibbrandt and Lpez-Prez, 2012), but also punishment itself. The aim of this paper is therefore to investigate the behavior of a second-party single punisher in a public∗ Corresponding author; Tel. +493641 930410; Bachstraße 18k, 07743, Jena Germany. Email addresses: [email protected] (Leonard Hoeft), [email protected] (Wladislaw Mill)
Preprint submitted to Economics Letters
good game. As we are only interested in a comparison of punisher and non-punisher behavior, we used a single treatment with no feedback on individual contribution and costless punishment to have the cleanest angle at selfish and prosocial behavior, excluding inequality, reputational, and leadership concerns, and to give selfishness its best shot. To ensure our results were robust and due to punishment rather than informational asymmetries, we additionally ran a full information treatment. We hypothesize that: H1 Punishers contribute less than non-punishers. H2 This effect increases over time. H3 Punishers enforce higher contributions than their own contribution. Consistent with previous studies in which the punisher changed every round (O’Gorman et al., 2009), we find that punishers stabilize cooperation by enforcing a high contribution level. However, they enforce a double standard by failing to contribute accordingly. They contribute less than their peers and additionally reduce their relative contribution over time, even though this behavior is overwhelmingly condemned (Cubitt et al., 2011, Reuben and Riedl, 2013). This highlights the importance of instrumental motivations for punishment and casts doubt that various prosocial motivations (Leibbrandt and Lpez-Prez, 2012), reputation, leadership (O’Gorman et al., 2009), equality concerns (Johnson et al., 2009, Dawes et al., 2007), or retributive fairness (Falk et al., 2005) are the main drivers in repeated second-party punishment scenarios. Instead, a punisher might simply mitigate the social dilemma for his own benefit. May 7, 2017
2. Materials and Methods
Only one of the thirty rounds was payoff-relevant in case the public good was drawn to be payoff-relevant for the respective subject.
2.1. Measurements 2.1.1. Public goods game task All participants were randomly assigned a role (punisher, non-punisher) and to a group of four in which they remained for the duration of the public goods game. Each session consisted of thirty rounds. Participants were instructed that each round would consist of three stages. In the first stage, participants were asked to allocate 20 tokens to a private and public account (1 token = 25 euro cents). Tokens allocated to the private account were theirs to keep. The tokens that were allocated to the public account (ci ) had a marginal per-capita return (MPCR) of 0.5, so that each group member would receive 0.5 times the total contribution to the public goods game. The payoff πi of the participant i can therefore be formalized in the following way: X cj (1) πi = 5 − ci + 0.5 ·
2.1.2. Additional measurements We also collected data on spite (Marcus et al., 2014), social dominance orientation (SDO) (von Collani, 2002), rivalry & narcissism (Back et al., 2013), and social value orientation (SVO) to increase the robustness of our results. To measure SVO, we used the 6-items primary ring matching version of the Slider Measure (Murphy et al., 2011). At the end of the experiment, only one of the 6 items was randomly chosen to become payoff-relevant in case this task was paid. Either the slider-measure or the public goods game task was chosen with equal probability to be payoff-relevant, while the three questionnaires (Spite, SDO, rivalry & narcissism) were not incentivized. 2.2. Participants and Design 96 participants (47 % female) were recruited with the online registration software Hroot (Bock et al., 2014). The experiment was conducted at the BonnEconLab and consisted of 4 sessions, each with 24 participants. The participants’ age ranged from 16 to 57 years (Mdn = 22). Most students were bachelor students (Semester Mdn=5). The average earning was 14.58 e (including a 4 Euro show-up fee) and the experiment lasted 1.5 hours (including setting, video instructions, payoff etc.). All measurements were computerized with the experimental software z-Tree (Fischbacher, 2007). Participants were randomly assigned to computer cubicles. They received video instructions separately and the opportunity to ask questions for each task in the experiment.3 First, they were asked to complete SVO measurements. Then, they participated in a public goods game for 30 rounds. After that, they completed questionnaires and filled in socio-demographics. At last, they were presented with their payoff information and received their payoff privately.
j∈{1,n}
In the second stage, only the punisher (who was referred to as “D”) was informed about the contributions of all group members in the first stage. The participants were shown in random order to the punisher each round anew to rule out reputation effects from previous rounds. D (the punisher) was now asked to indicate how much she would punish subject i (ςi , i 6= D)1 . For this purpose she was equipped with 30 tokens. Each token could be used by the punisher to deduct one token of the payoff of a targeted subject. Unused tokens were not added to the payoff of D to rule out equality concerns2 , so the contributions of the punisher could be compared to the contributions of others directly. The other three group members were just shown a blank screen asking them to wait for the decision of the punisher. The payoff πi of the participant i 6= D can therefore be formalized in the following way (the payoff of the punisher is described by equation (1)): X πi = 5 − ci + 0.5 · cj − ς i (2)
3. Result
j∈{1,n}
As each observation is nested within a group, we study only the means of each group over all rounds as the test statistics. To test for the first hypothesis (Hyp 1), namely that punishers will use their position to contribute less, we use a simple t-test. As hypothesized, we find a massive discrepancy between the average contribution behavior over all rounds of punishers (M= 8.29, SD4 = 4.39) and non punishers (M= 15.27, SD= 4.98); this is also highly significant: (t(46)= −4.09, p ≤0.01).
In the third stage (feedback stage), participants were informed about their own contribution to the private and group account, the overall group contribution, their own punishment (reduction), and their payoff. Participants were informed neither of the contributions of other group members nor of punishment meted out to others - this was made public in the instructions to avoid leadership and reputational concerns. 1 To avoid framing and demand effects, we referred to the act as “reducing the payoff”. 2 In case of payoff-relevant equipment, the punisher could contribute more in stage one, anticipating extra gains in the second stage. If there was no extra equipment, the punisher could contribute less in stage one, compensating his extra expenditure in stage two.
3 The
video instructions with English subtitles and an English version of the handout can be found in the supplementary materials. 4 Average of the standard deviations over all rounds.
2
Result 1 Punishers contribute significantly less than non-punishers Contribution to the Public Good
20
To strengthen our findings, we additionally estimate contribution behavior in a mixed effects linear model with random effects on groups νj and individuals νj,i :5 ci,t =β0 + β1 t + β2 P unisher + β3 1t=30 + β4 P unisher × t + νj + νi,j + i,j,t
(3)
where t = 1, ..., 30 is the period, 1t=30 is the indicator function to control for last round effects, and punisher is a dummy variable with 1 if we estimate the contribution of a punisher and 0 for non-punishers. Groups are j = 1, ..., 24 and individuals are i = 1, ..., 96. νj and νi,j are the level 1 and level 2 random effects of groups and individuals. The second model additionally includes control variables.6 The results of Table 1 indicate that punishers contribute significantly less to the public good. This effect is additionally increasing over time, which is driven by the increase of non-punisher’s contribution.
Punisher No Yes
15
10
5
0 0
10
Round
20
30
Figure 1: Mean contributions to the public good
Contribution to the public good Constant 13.99∗∗∗ (0.95) 18.13∗∗∗ (3.92) Punisher −5.55∗∗∗ (1.08) −5.81∗∗∗ (1.09) Period 0.09∗∗∗ (0.01) 0.09∗∗∗ (0.01) Punisher:Period −0.09∗∗∗ (0.02) −0.09∗∗∗ (0.02) LastRound −3.05∗∗∗ (0.51) −3.05∗∗∗ (0.51) Controls × X Observations 2,880 2,880 Log Likelihood −8,699.16 −8,701.40 Akaike Inf. Crit. 17,414.31 17,432.80 Bayesian Inf. Crit. 17,462.04 17,522.28
Result 2 The difference between punisher and nonpunisher contributions increased significantly over time. To check if the differences in contribution levels were due to the use of punishment, we defined a measure which represents the contribution norm and checked if the punisher undercut his own norm. The imposed norm is measured by the highest contribution punished.7 The average imposed norm over all rounds is MM axContrP un = 14.05, SDM axContrP un = 2.11. Comparing this to the average contribution of the punisher (M= 8.29, SD= 4.39), it can be seen that punishers contributed significantly less than the established norm (t(46)= −3.377, p ≤0.01). Therefore, average punishment is illegitimate (contributions higher than your own are punished) according to Faillo et al. (2013).
Notes: p : ∗ ∗ ∗ < .001 ∗ ∗ < .01∗ < .05 Standard errors are in parentheses Table 1: Linear mixed effects model of the contribution
4. Robustness Check To ensure that the effect is due to the possibility to punish, and not due to the additional knowledge of the punisher, we conducted a second experiment. This experiment is identical to the first; however, this time the contributions of every player were public knowledge. Thus, the only difference between a punisher and a non-punisher is the punishment option. 96 participants (56 % female) were recruited for the second study. The participants’ age ranged from 18 to 53 years (Mdn = 23). The average earning was 14.94 e. As can be seen in Figure 2, the behavior in Study Two resembles Study One to a surprising degree (see Figure 1). Using the same approach as in Study One we find again a large discrepancy, between the average contribution behavior over all rounds of punishers (M= 9.48, SD= 3.82) and non punishers (M= 16.62, SD= 3.99); This is also highly significant: (t(46)= −4.08, p ≤0.01). Using the same econometric approach as in study one to estimate Model 3, we obtain similar results, as can be seen in Table 2.
Result 3 Punishers contribute significantly below the norm they impose.
5 Note
that a random slope on groups and individuals over time yield qualitatively the same results. Furthermore, we controlled for the robustness of the results by using a pooled OLS with group-level clustering. The results (significance and direction of the estimators) do persist. 6 None of the control variables improved the model significantly. However, to show the robustness of our results, we still included all individual level controls, which were SVO, Spite, SDO, narcism, rivalry, age and gender. 7 In order to account for cases in which the punisher established a sufficient threat level which led to full (35 %) or high contributions persisting without further punishment, we estimate for each group the highest contribution punished so far. We found little evidence of punishers abandoning a once-established punishment norm.
3
Contribution to the Public Good
20
might not be driven by prosocial motivations to create and enforce a beneficial social norm in the group. Solving the dilemma can be in the punisher’s own best interest, and in a setting that gives selfish motivations their best shot we see that they dominate contribution levels over time. Noticeably, transparency does not mitigate the problem. Institutional design should therefore be especially sensitive to situations of designated punishment. The outlook on designated punishers might have been too optimistic. The experiment also indicates that the motivations for punishment might be context-dependent and do not necessarily signal the willingness of the punisher to engage in overall cooperative behavior. On the contrary, this study provides the first evidence of the dangers of power concentration in punishment settings.
Punisher No Yes
15
10
5
0 0
10
Round
20
30
Acknowledgement
Figure 2: Mean contributions to the public good in Study Two
We thank Christoph Engel, Oliver Kirchkamp, Urs Fischbacher, Nikos Nikiforakis, Martin Kocher, Ulrike Malmendier, Daniel Houser, Werner G¨ uth, and an anonymous referee for helpful comments. We appreciate comments from the participants of the Economic Science Association World Conference 2016, the Jena Econ-Seminar and the Bonn Econ-Seminar. We gratefully acknowledge funding from the Max Planck Society and the IMPRS Uncertainty.
If we compare the average imposed norm over all rounds in Study Two MM axContrP un = 13.32 (SDM axContrP un = 2.32) with the average contribution of punishers in Study Two M= 9.48 (SD= 3.82), punishers - again - contributed significantly less than the norm they established (t(46)= −1.993, p =0.05). Result 4 The findings of Study One are qualitatively replicated in a full information setting.8
References Back, M. D., Kuefner, A. C. P., Dufner, M., Gerlach, T. M., Rauthmann, J. F., Denissen, J. J. A., 2013. Narcissistic admiration and rivalry: Disentangling the bright and dark sides of narcissism. Journal of Personality and Social Psychology 105 (10), 1013–1037.
Contribution to the public good Constant 13.97∗∗∗ (0.81) 12.79∗∗∗ (3.06) Punisher −4.38∗∗∗ (0.93) −4.04∗∗∗ (1.04) Period 0.18∗∗∗ (0.01) 0.18∗∗∗ (0.01) Punisher:Period −0.18∗∗∗ (0.02) −0.18∗∗∗ (0.02) LastRound −2.14∗∗∗ (0.44) −2.14∗∗∗ (0.44) Controls × X Observations 2,880 2,880 Log Likelihood −8,302.87 −8,310.71 Akaike Inf. Crit. 16,621.74 16,651.43 Bayesian Inf. Crit. 16,669.46 16,740.91
Baldassarri, D., Grossman, G., 2011. Centralized sanctioning and legitimate authority promote cooperation in humans. Proceedings of the National Academy of Sciences 108, 11023–11027. Bock, O., Baetge, I., Nicklisch, A., 2014. hroot: Hamburg registration and organization online tool. European Economic Review 71 (C), 117–120. Cubitt, R. P., Drouvelis, M., Gchter, S., Kabalin, R., 2011. Moral judgments in social dilemmas: How bad is free riding? Journal of Public Economics 95 (34), 253–264.
Notes: p : ∗ ∗ ∗ < .001 ∗ ∗ < .01∗ < .05 Standard errors are in parentheses
Dawes, C. T., Fowler, J. H., Johnson, T., McElreath, R., Smirnov, O., 2007. Egalitarian motives in humans. Nature 446 (7137), 794– 796.
Table 2: Linear mixed effects model of the contribution in Study Two
Faillo, M., Grieco, D., Zarri, L., 2013. Legitimate punishment, feedback, and the enforcement of cooperation. Games and Economic Behavior 77 (1), 271–283.
5. Conclusion
Falk, A., Fehr, E., Fischbacher, U., 2005. Driving Forces Behind Informal Sanctions. Econometrica 73 (6), 2017–2030.
Single punishers can indeed solve the social dilemma. However, punishers do not necessarily adhere to the contribution norm they enforce themselves. Contrary to interpretations of existing literature, designated punishment
Fehr, E., G¨ achter, S., 2002. Altruistic punishment in humans. Nature 415 (6868), 137–140. Fischbacher, U., 2007. z-tree: Zurich toolbox for ready-made economic experiments. Experimental Economics 10 (2), 171–178. Herrmann, B., Th¨ oni, C., G¨ achter, S., 2008. Antisocial punishment across societies. Science 319 (5868), 1362–1367.
8 Transparency
did not affect the discrepancy of contributions, which is the main point of this paper. More subtle differences will be discussed as part of a larger investigation into power abuse in a future study.
Hobbes, T., 1998. Leviathan. Oxford University Press, Oxford. Houser, D., Xiao, E., 2010. Inequality-seeking punishment. Economics Letters 109 (1), 20–23.
4
Johnson, T., Dawes, C. T., Fowler, J. H., McElreath, R., Smirnov, O., 2009. The role of egalitarian motives in altruistic punishment. Economics Letters 102 (3), 192–194. Leibbrandt, A., Lpez-Prez, R., 2012. An exploration of third and second party punishment in ten simple games. Journal of Economic Behavior & Organization 84 (3), 753–766. Marcus, D. K., Zeigler-Hill, V., Mercer, S. H., Norris, A. L., 2014. The psychology of spite and the measurement of spitefulness. Psychological Assessment 26 (2), 563–574. Murphy, R. O., Ackerman, K. A., Handgraaf, M. J. J., 2011. Measuring social value orientation. Judgment and Decision Making 6 (8), 771–781. O’Gorman, R., Henrich, J., Van Vugt, M., 2009. Constraining free riding in public goods games: designated solitary punishers can sustain human cooperation. Proceedings of the Royal Society of London B: Biological Sciences 276 (1655), 323–329. Reuben, E., Riedl, A., 2013. Enforcement of contribution norms in public good games with heterogeneous populations. Games and Economic Behavior 77 (1), 122–137. von Collani, G., 2002. Das konstrukt der sozialen dominanzorientierung als generalisierte einstellung: Eine replikation. Zeitschrift fuer Politische Psychologie 10 (1), 263–282.
5
PII: DOI: Reference:
S0165-1765(17)30200-8 http://dx.doi.org/10.1016/j.econlet.2017.05.022 ECOLET 7625
To appear in:
Economics Letters
Received date : 27 February 2017 Revised date : 8 May 2017 Accepted date : 19 May 2017 Please cite this article as: Hoeft, L., Mill, W., Selfish punishers an experimental investigation of designated punishment behavior in public goods. Economics Letters (2017), http://dx.doi.org/10.1016/j.econlet.2017.05.022 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
*Highlights (for review)
Highlights: • Designated punishers mitigate free riding while contributing less than non-punishers. • Punishers undercut their own enforced norm.
• The discrepancy between punishers and non-punishers grows over time.
*Manuscript final version Click here to view linked References
Selfish Punishers An experimental investigation of designated punishment behavior in public goods Leonard Hoefta,b , Wladislaw Milla,c,∗ a International
Max Planck Research School on Adapting Behavior in a Fundamentally Uncertain World. Planck Institute for Research on Collective Goods, Kurt-Schumacher-Straße 10, 53113 Bonn Germany. c School of Economics and Business Administration, University of Jena, Bachstraße 18k, 07743 Jena, Germany. b Max
Abstract We show that a second-party punisher forces his peers to contribute to a public good while contributing significantly less himself. This effect increased over time and casts doubt on the prevalent prosocial interpretation of (designated) punishment behavior. Keywords: Punishment; Public Good; Experiment; Selfish JEL: H41, C92, K42 1. Introduction Imagine working with a supervisor on a common project: Everyone benefits from working on it, yet only one enforces contributions. From monarch to managers, many more real-world examples are conceivable. Yet research has concentrated mainly on situations where everyone can enforce contributions. In this situation participants reliably use (peer) punishment to solve social dilemmas (Fehr and G¨ achter, 2002), even though it constitutes a second-order public good in theory. This interpretation has extended to third-party (Baldassarri and Grossman, 2011) and secondparty (O’Gorman et al., 2009) single punishers, who solve the social dilemma efficiently as well. Additional prosocial motivations range from equality concerns (Dawes et al., 2007) to retributive fairness (Falk et al., 2005). However, punishment can also be destructive: Some punishment decisions are motivated by spite and retaliation (Herrmann et al., 2008, Houser and Xiao, 2010, Falk et al., 2005). Traditionally, philosophers have been skeptical of the prosocial use of power and argued for a state monopoly on punishment (Hobbes, 1998). Often, punishment can be self-serving as well as prosocial. Especially a second-party punisher profits from his own use of power. Although he provides a public good among the other players, he could abuse his power by undercutting his enforced contribution norm. Therefore, not only withholding punishment can be selfish (Leibbrandt and Lpez-Prez, 2012), but also punishment itself. The aim of this paper is therefore to investigate the behavior of a second-party single punisher in a public∗ Corresponding author; Tel. +493641 930410; Bachstraße 18k, 07743, Jena Germany. Email addresses: [email protected] (Leonard Hoeft), [email protected] (Wladislaw Mill)
Preprint submitted to Economics Letters
good game. As we are only interested in a comparison of punisher and non-punisher behavior, we used a single treatment with no feedback on individual contribution and costless punishment to have the cleanest angle at selfish and prosocial behavior, excluding inequality, reputational, and leadership concerns, and to give selfishness its best shot. To ensure our results were robust and due to punishment rather than informational asymmetries, we additionally ran a full information treatment. We hypothesize that: H1 Punishers contribute less than non-punishers. H2 This effect increases over time. H3 Punishers enforce higher contributions than their own contribution. Consistent with previous studies in which the punisher changed every round (O’Gorman et al., 2009), we find that punishers stabilize cooperation by enforcing a high contribution level. However, they enforce a double standard by failing to contribute accordingly. They contribute less than their peers and additionally reduce their relative contribution over time, even though this behavior is overwhelmingly condemned (Cubitt et al., 2011, Reuben and Riedl, 2013). This highlights the importance of instrumental motivations for punishment and casts doubt that various prosocial motivations (Leibbrandt and Lpez-Prez, 2012), reputation, leadership (O’Gorman et al., 2009), equality concerns (Johnson et al., 2009, Dawes et al., 2007), or retributive fairness (Falk et al., 2005) are the main drivers in repeated second-party punishment scenarios. Instead, a punisher might simply mitigate the social dilemma for his own benefit. May 7, 2017
2. Materials and Methods
Only one of the thirty rounds was payoff-relevant in case the public good was drawn to be payoff-relevant for the respective subject.
2.1. Measurements 2.1.1. Public goods game task All participants were randomly assigned a role (punisher, non-punisher) and to a group of four in which they remained for the duration of the public goods game. Each session consisted of thirty rounds. Participants were instructed that each round would consist of three stages. In the first stage, participants were asked to allocate 20 tokens to a private and public account (1 token = 25 euro cents). Tokens allocated to the private account were theirs to keep. The tokens that were allocated to the public account (ci ) had a marginal per-capita return (MPCR) of 0.5, so that each group member would receive 0.5 times the total contribution to the public goods game. The payoff πi of the participant i can therefore be formalized in the following way: X cj (1) πi = 5 − ci + 0.5 ·
2.1.2. Additional measurements We also collected data on spite (Marcus et al., 2014), social dominance orientation (SDO) (von Collani, 2002), rivalry & narcissism (Back et al., 2013), and social value orientation (SVO) to increase the robustness of our results. To measure SVO, we used the 6-items primary ring matching version of the Slider Measure (Murphy et al., 2011). At the end of the experiment, only one of the 6 items was randomly chosen to become payoff-relevant in case this task was paid. Either the slider-measure or the public goods game task was chosen with equal probability to be payoff-relevant, while the three questionnaires (Spite, SDO, rivalry & narcissism) were not incentivized. 2.2. Participants and Design 96 participants (47 % female) were recruited with the online registration software Hroot (Bock et al., 2014). The experiment was conducted at the BonnEconLab and consisted of 4 sessions, each with 24 participants. The participants’ age ranged from 16 to 57 years (Mdn = 22). Most students were bachelor students (Semester Mdn=5). The average earning was 14.58 e (including a 4 Euro show-up fee) and the experiment lasted 1.5 hours (including setting, video instructions, payoff etc.). All measurements were computerized with the experimental software z-Tree (Fischbacher, 2007). Participants were randomly assigned to computer cubicles. They received video instructions separately and the opportunity to ask questions for each task in the experiment.3 First, they were asked to complete SVO measurements. Then, they participated in a public goods game for 30 rounds. After that, they completed questionnaires and filled in socio-demographics. At last, they were presented with their payoff information and received their payoff privately.
j∈{1,n}
In the second stage, only the punisher (who was referred to as “D”) was informed about the contributions of all group members in the first stage. The participants were shown in random order to the punisher each round anew to rule out reputation effects from previous rounds. D (the punisher) was now asked to indicate how much she would punish subject i (ςi , i 6= D)1 . For this purpose she was equipped with 30 tokens. Each token could be used by the punisher to deduct one token of the payoff of a targeted subject. Unused tokens were not added to the payoff of D to rule out equality concerns2 , so the contributions of the punisher could be compared to the contributions of others directly. The other three group members were just shown a blank screen asking them to wait for the decision of the punisher. The payoff πi of the participant i 6= D can therefore be formalized in the following way (the payoff of the punisher is described by equation (1)): X πi = 5 − ci + 0.5 · cj − ς i (2)
3. Result
j∈{1,n}
As each observation is nested within a group, we study only the means of each group over all rounds as the test statistics. To test for the first hypothesis (Hyp 1), namely that punishers will use their position to contribute less, we use a simple t-test. As hypothesized, we find a massive discrepancy between the average contribution behavior over all rounds of punishers (M= 8.29, SD4 = 4.39) and non punishers (M= 15.27, SD= 4.98); this is also highly significant: (t(46)= −4.09, p ≤0.01).
In the third stage (feedback stage), participants were informed about their own contribution to the private and group account, the overall group contribution, their own punishment (reduction), and their payoff. Participants were informed neither of the contributions of other group members nor of punishment meted out to others - this was made public in the instructions to avoid leadership and reputational concerns. 1 To avoid framing and demand effects, we referred to the act as “reducing the payoff”. 2 In case of payoff-relevant equipment, the punisher could contribute more in stage one, anticipating extra gains in the second stage. If there was no extra equipment, the punisher could contribute less in stage one, compensating his extra expenditure in stage two.
3 The
video instructions with English subtitles and an English version of the handout can be found in the supplementary materials. 4 Average of the standard deviations over all rounds.
2
Result 1 Punishers contribute significantly less than non-punishers Contribution to the Public Good
20
To strengthen our findings, we additionally estimate contribution behavior in a mixed effects linear model with random effects on groups νj and individuals νj,i :5 ci,t =β0 + β1 t + β2 P unisher + β3 1t=30 + β4 P unisher × t + νj + νi,j + i,j,t
(3)
where t = 1, ..., 30 is the period, 1t=30 is the indicator function to control for last round effects, and punisher is a dummy variable with 1 if we estimate the contribution of a punisher and 0 for non-punishers. Groups are j = 1, ..., 24 and individuals are i = 1, ..., 96. νj and νi,j are the level 1 and level 2 random effects of groups and individuals. The second model additionally includes control variables.6 The results of Table 1 indicate that punishers contribute significantly less to the public good. This effect is additionally increasing over time, which is driven by the increase of non-punisher’s contribution.
Punisher No Yes
15
10
5
0 0
10
Round
20
30
Figure 1: Mean contributions to the public good
Contribution to the public good Constant 13.99∗∗∗ (0.95) 18.13∗∗∗ (3.92) Punisher −5.55∗∗∗ (1.08) −5.81∗∗∗ (1.09) Period 0.09∗∗∗ (0.01) 0.09∗∗∗ (0.01) Punisher:Period −0.09∗∗∗ (0.02) −0.09∗∗∗ (0.02) LastRound −3.05∗∗∗ (0.51) −3.05∗∗∗ (0.51) Controls × X Observations 2,880 2,880 Log Likelihood −8,699.16 −8,701.40 Akaike Inf. Crit. 17,414.31 17,432.80 Bayesian Inf. Crit. 17,462.04 17,522.28
Result 2 The difference between punisher and nonpunisher contributions increased significantly over time. To check if the differences in contribution levels were due to the use of punishment, we defined a measure which represents the contribution norm and checked if the punisher undercut his own norm. The imposed norm is measured by the highest contribution punished.7 The average imposed norm over all rounds is MM axContrP un = 14.05, SDM axContrP un = 2.11. Comparing this to the average contribution of the punisher (M= 8.29, SD= 4.39), it can be seen that punishers contributed significantly less than the established norm (t(46)= −3.377, p ≤0.01). Therefore, average punishment is illegitimate (contributions higher than your own are punished) according to Faillo et al. (2013).
Notes: p : ∗ ∗ ∗ < .001 ∗ ∗ < .01∗ < .05 Standard errors are in parentheses Table 1: Linear mixed effects model of the contribution
4. Robustness Check To ensure that the effect is due to the possibility to punish, and not due to the additional knowledge of the punisher, we conducted a second experiment. This experiment is identical to the first; however, this time the contributions of every player were public knowledge. Thus, the only difference between a punisher and a non-punisher is the punishment option. 96 participants (56 % female) were recruited for the second study. The participants’ age ranged from 18 to 53 years (Mdn = 23). The average earning was 14.94 e. As can be seen in Figure 2, the behavior in Study Two resembles Study One to a surprising degree (see Figure 1). Using the same approach as in Study One we find again a large discrepancy, between the average contribution behavior over all rounds of punishers (M= 9.48, SD= 3.82) and non punishers (M= 16.62, SD= 3.99); This is also highly significant: (t(46)= −4.08, p ≤0.01). Using the same econometric approach as in study one to estimate Model 3, we obtain similar results, as can be seen in Table 2.
Result 3 Punishers contribute significantly below the norm they impose.
5 Note
that a random slope on groups and individuals over time yield qualitatively the same results. Furthermore, we controlled for the robustness of the results by using a pooled OLS with group-level clustering. The results (significance and direction of the estimators) do persist. 6 None of the control variables improved the model significantly. However, to show the robustness of our results, we still included all individual level controls, which were SVO, Spite, SDO, narcism, rivalry, age and gender. 7 In order to account for cases in which the punisher established a sufficient threat level which led to full (35 %) or high contributions persisting without further punishment, we estimate for each group the highest contribution punished so far. We found little evidence of punishers abandoning a once-established punishment norm.
3
Contribution to the Public Good
20
might not be driven by prosocial motivations to create and enforce a beneficial social norm in the group. Solving the dilemma can be in the punisher’s own best interest, and in a setting that gives selfish motivations their best shot we see that they dominate contribution levels over time. Noticeably, transparency does not mitigate the problem. Institutional design should therefore be especially sensitive to situations of designated punishment. The outlook on designated punishers might have been too optimistic. The experiment also indicates that the motivations for punishment might be context-dependent and do not necessarily signal the willingness of the punisher to engage in overall cooperative behavior. On the contrary, this study provides the first evidence of the dangers of power concentration in punishment settings.
Punisher No Yes
15
10
5
0 0
10
Round
20
30
Acknowledgement
Figure 2: Mean contributions to the public good in Study Two
We thank Christoph Engel, Oliver Kirchkamp, Urs Fischbacher, Nikos Nikiforakis, Martin Kocher, Ulrike Malmendier, Daniel Houser, Werner G¨ uth, and an anonymous referee for helpful comments. We appreciate comments from the participants of the Economic Science Association World Conference 2016, the Jena Econ-Seminar and the Bonn Econ-Seminar. We gratefully acknowledge funding from the Max Planck Society and the IMPRS Uncertainty.
If we compare the average imposed norm over all rounds in Study Two MM axContrP un = 13.32 (SDM axContrP un = 2.32) with the average contribution of punishers in Study Two M= 9.48 (SD= 3.82), punishers - again - contributed significantly less than the norm they established (t(46)= −1.993, p =0.05). Result 4 The findings of Study One are qualitatively replicated in a full information setting.8
References Back, M. D., Kuefner, A. C. P., Dufner, M., Gerlach, T. M., Rauthmann, J. F., Denissen, J. J. A., 2013. Narcissistic admiration and rivalry: Disentangling the bright and dark sides of narcissism. Journal of Personality and Social Psychology 105 (10), 1013–1037.
Contribution to the public good Constant 13.97∗∗∗ (0.81) 12.79∗∗∗ (3.06) Punisher −4.38∗∗∗ (0.93) −4.04∗∗∗ (1.04) Period 0.18∗∗∗ (0.01) 0.18∗∗∗ (0.01) Punisher:Period −0.18∗∗∗ (0.02) −0.18∗∗∗ (0.02) LastRound −2.14∗∗∗ (0.44) −2.14∗∗∗ (0.44) Controls × X Observations 2,880 2,880 Log Likelihood −8,302.87 −8,310.71 Akaike Inf. Crit. 16,621.74 16,651.43 Bayesian Inf. Crit. 16,669.46 16,740.91
Baldassarri, D., Grossman, G., 2011. Centralized sanctioning and legitimate authority promote cooperation in humans. Proceedings of the National Academy of Sciences 108, 11023–11027. Bock, O., Baetge, I., Nicklisch, A., 2014. hroot: Hamburg registration and organization online tool. European Economic Review 71 (C), 117–120. Cubitt, R. P., Drouvelis, M., Gchter, S., Kabalin, R., 2011. Moral judgments in social dilemmas: How bad is free riding? Journal of Public Economics 95 (34), 253–264.
Notes: p : ∗ ∗ ∗ < .001 ∗ ∗ < .01∗ < .05 Standard errors are in parentheses
Dawes, C. T., Fowler, J. H., Johnson, T., McElreath, R., Smirnov, O., 2007. Egalitarian motives in humans. Nature 446 (7137), 794– 796.
Table 2: Linear mixed effects model of the contribution in Study Two
Faillo, M., Grieco, D., Zarri, L., 2013. Legitimate punishment, feedback, and the enforcement of cooperation. Games and Economic Behavior 77 (1), 271–283.
5. Conclusion
Falk, A., Fehr, E., Fischbacher, U., 2005. Driving Forces Behind Informal Sanctions. Econometrica 73 (6), 2017–2030.
Single punishers can indeed solve the social dilemma. However, punishers do not necessarily adhere to the contribution norm they enforce themselves. Contrary to interpretations of existing literature, designated punishment
Fehr, E., G¨ achter, S., 2002. Altruistic punishment in humans. Nature 415 (6868), 137–140. Fischbacher, U., 2007. z-tree: Zurich toolbox for ready-made economic experiments. Experimental Economics 10 (2), 171–178. Herrmann, B., Th¨ oni, C., G¨ achter, S., 2008. Antisocial punishment across societies. Science 319 (5868), 1362–1367.
8 Transparency
did not affect the discrepancy of contributions, which is the main point of this paper. More subtle differences will be discussed as part of a larger investigation into power abuse in a future study.
Hobbes, T., 1998. Leviathan. Oxford University Press, Oxford. Houser, D., Xiao, E., 2010. Inequality-seeking punishment. Economics Letters 109 (1), 20–23.
4
Johnson, T., Dawes, C. T., Fowler, J. H., McElreath, R., Smirnov, O., 2009. The role of egalitarian motives in altruistic punishment. Economics Letters 102 (3), 192–194. Leibbrandt, A., Lpez-Prez, R., 2012. An exploration of third and second party punishment in ten simple games. Journal of Economic Behavior & Organization 84 (3), 753–766. Marcus, D. K., Zeigler-Hill, V., Mercer, S. H., Norris, A. L., 2014. The psychology of spite and the measurement of spitefulness. Psychological Assessment 26 (2), 563–574. Murphy, R. O., Ackerman, K. A., Handgraaf, M. J. J., 2011. Measuring social value orientation. Judgment and Decision Making 6 (8), 771–781. O’Gorman, R., Henrich, J., Van Vugt, M., 2009. Constraining free riding in public goods games: designated solitary punishers can sustain human cooperation. Proceedings of the Royal Society of London B: Biological Sciences 276 (1655), 323–329. Reuben, E., Riedl, A., 2013. Enforcement of contribution norms in public good games with heterogeneous populations. Games and Economic Behavior 77 (1), 122–137. von Collani, G., 2002. Das konstrukt der sozialen dominanzorientierung als generalisierte einstellung: Eine replikation. Zeitschrift fuer Politische Psychologie 10 (1), 263–282.
5