Monopolizing sanctioning power under noise eliminates perverse punishment but does not increase cooperation

ORIGINAL RESEARCH published: 29 September 2016 doi: 10.3389/fnbeh.2016.00180 Monopolizing Sanctioning Power under Noise Eliminates Perverse Punishment But Does Not Increase Cooperation Sven Fischer 1*, Kristoffel Grechenig 2 and Nicolas Meier 3 1 Newcastle University, Newcastle University Business School, Newcastle upon Tyne, UK, 2 Max Planck Institute for Research on Collective Goods, Bonn, Germany, 3 School of Business and Economics, RWTH Aachen University, Aachen, Germany We run several experiments which allow us to compare cooperation under perfect and imperfect information in a centralized and decentralized punishment regime. Under perfect and extremely noisy information, aggregate behavior does not differ between institutions. Under intermediate noise, punishment escalates in the decentralized peer-to-peer punishment regime which badly affects efficiency while sustaining cooperation for longer. Only decentralized punishment is often directed at cooperators (perverse punishment). We report several, sometimes subtle, differences in punishment behavior, and how contributions react. Keywords: cooperation, public good, centralized punishment, imperfect information, anti-social punishment, perverse punishment 1. INTRODUCTION Edited by: Benedikt Herrmann, University of Nottingham, UK Reviewed by: Mark Bernard, Goethe University Frankfurt, Germany Daniele Nosenzo, University of Nottingham, UK *Correspondence: Sven Fischer Received: 01 December 2014 Accepted: 09 September 2016 Published: 29 September 2016 Citation: Fischer S, Grechenig K and Meier N (2016) Monopolizing Sanctioning Power under Noise Eliminates Perverse Punishment But Does Not Increase Cooperation. Front. Behav. Neurosci. 10:180. doi: 10.3389/fnbeh.2016.00180 Modern societies have centralized sanctioning power as a means to enforce norms (Weber, 1919). This monopoly has often been justified on the premise that private, decentralized enforcement has (higher) negative externalities (Clotfelter, 1978; Polinsky, 1980). But, is centralization necessarily better? Experiments on voluntary cooperation repeatedly demonstrate that, compared to an environment without punishment, decentralized, informal, peer-to-peer punishment increases cooperation under perfect information (Yamagishi, 1986; Ostrom et al., 1992; Fehr and Gächter, 2000, 2002), and welfare in the long run (Gächter et al., 2008). Various studies, however, challenge the idea that peer-to-peer punishment generally enhances cooperation (for an overview, see Nikiforakis, 2014): for example on the basis of punishment which is targeted at cooperators, referred to as anti-social (Herrmann et al., 2008) or perverse punishment (Cinyabuguma et al., 2006), or on the basis of counter-punishment (targeted either at the group or the punisher directly as in Nikiforakis, 2008; Nikiforakis et al., 2012)1 . If punishment is centralized, however, there is no opportunity for counter punishment and such sources of inefficiency are less likely. Furthermore, with only one punisher, there are no coordination problems and no problems resulting from possibly conflicting contribution norms. Some experiments test the effectiveness of formal, centralized enforcement mechanisms compared to informal, decentralized regimes, while capturing important aspects of institutions. This literature characterizes centralization as a mechanism that allows to commit to a sanctioning scheme, such that punishment is automatically carried out and/or determined according to some exogenous voting rule (Kosfeld and Riedl, 2004; Tyran and Feld, 2006; Guillen et al., 2007; Kube and Traxler, 2011; Putterman et al., 2011; Andreoni and Gee, 2012; Ambrus and Greiner, 2015). Centralization is viewed as a commitment mechanism Frontiers in Behavioral Neuroscience | www.frontiersin.org 1 September 2016 | Volume 10 | Article 180 Fischer et al. Monopolizing Punishment institutional factors, we return to the origins of formal punishment as a centralization of informal sanctioning regimes (Turnbull, 1962; Guala, 2012). Furthermore, by allowing participants to interact over 30 periods, we obtain enough observations for a detailed analysis of individual behavior and group dynamics. To the best of our knowledge, we are the first to analyze the effects of centralization of punishment per se under imperfect information. and these papers do not explore the effect of centralization per se, i.e., of merely concentrating all power into one hand. In an environment with a centralized punisher rather than an automated mechanism, an important question is how the authorization to punish is granted. Baldassarri and Grossman (2011) compare two centralized punishment regimes. In one, authority is granted by chance, in the other by electing the punisher from within the group after two trial rounds. While punishment behavior is very similar, in the elected authority treatment participants cooperate more. This suggests that in order to test whether centralization per se matters, one needs to abstract from any mechanism that grants legitimacy, and restrict the analysis to a fair random selection. There are a few studies which allow to compare a centralized with a decentralized punishment regime in this way, and the resulting evidence is mixed. In Carpenter et al. (2012), where the role of the punisher is randomly allocated but fixed and punishment is cheap, contributions and overall efficiency are larger in the decentralized regime. Similarly, in Nosenzo and Sefton (2014) contributions are substantially larger in a mutual punishment regime than in a centralized one. In O’Gorman et al. (2009), on the other hand, where authority is also granted at random but changes every round, and where punishment is expensive, there are no significant differences in cooperation. However, due to significantly more punishment, the decentralized regime is less efficient. One conclusion from these studies is that the effectiveness of a centralized regime stands and falls with the ability and willingness of just one person to invest resources for punishment, which results in considerably more variability in performance between groups. Decentralized regimes, on the other hand, suffer from the problems already mentioned. Such negative effects are likely to be aggravated by noisy information, which so far was only tested in decentralized regimes. Several studies show that peer to peer punishment is not able to sustain cooperation under imperfect information (Grechenig et al., 2010; Ambrus and Greiner, 2012; Grechenig et al., 2015) and it remains unclear whether this equally holds for centralized punishment. For example, imperfect information may result in some to stop punishing altogether, even irrespective of total group contributions. While in a decentralized regime this could partly be compensated by more punishment of others, if punishment is centralized, this is not possible. In order to test the effect of centralization per se under differe (...truncated)