Identity, Group Conflict, and Social Preferences

This paper presents a novel experiment on group conflict. Subjects are divided into groups according to preferences on paintings, and subjects are divided into groups according to self-declared political affiliations and leanings. Using a unique within subject design, we find twenty percent of subjects destroy social welfare – at personal cost – when facing a subject outside their group. This effect relates to individual identities. In the political treatment, Democrats and Republicans, in contrast to Independents, behave more selfishly and competitively towards out-group members. The results show social preferences for fairness and social welfare maximization are not universal and depend on the social context. * The author order combines alphabetical order (the convention in economics) and lab director last (the convention in psychology & neuroscience norm). We are grateful to Jeff Butler, Pedro Rey-Biel and participants at Conference on the Economics of Interactions & Culture (EIEF), Ecole Polytechnique, Pompeu Fabra, Institute for Economic Analysis & Universitat Autònama de Barcelona, and Paris School of Economics for their comments. We thank the Social Science Research Institute at Duke for sponsoring our faculty fellows program in 2010-2011, “From Brain to Society (and Back),” and we are grateful to the Transdisciplinary Prevention Research Center (TPRC) at Duke for funding this project. I. Introduction This paper presents a novel experiment on identity, group conflict, and social preferences. There is now a storied academic literature that argues human beings are not purely selfish. Rather, people are concerned about the well being of others when making decisions and allocating income. A series of economic experiments has demonstrated that subjects will give up own income in order to achieve higher social welfare and allocations that are more equitable. Yet this picture of preferences and allocation decisions does not jibe with much of human history. While many cultural and religious traditions involve charity and help to those less fortunate and redistribution is a feature of modern societies and democratic governments, throughout time people have been unfair and cruel to others. Human history is full of prolonged inter-group conflict, forced extraction of goods and labor, and genocide. Empirical research in economics has demonstrated that ethnic divisions are related to lower levels of public goods, dysfunctional institutions, and reduced growth. This paper delves into this apparent contradiction. We conduct a novel experiment to ask when people behave selfishly, when they maximize social welfare, and when they destroy the payoffs of others. The group division in our experiment is necessarily mild compared to the historical conflicts recalled above. Yet, even in a congenial university environment, we uncover a significant amount of “statusseeking” or “competitive behavior.” Using a unique within subject design, we find that twenty percent of participants are concerned with relative payoffs to the extent that they destroy social welfare – at personal cost – when facing a subject outside their group. This behavior is not punishment or retaliation for non-cooperative behavior or “negative reciprocity.” Subjects in our experiment are not responding to 1 North, Wallace & Weingast (2009) are among those bold enough to tackle the sweep of human history. 2 Prominent studies include Easterly & Levine (1997), Alesina, Baqir, & Easterly (1999), Alesina & La Ferrara (2005), Miguel & Gugerty (2005). 3 With the recent exception of Iriberri & Rey-Biel (2011), experiments and proposed formulations of utility have largely ignored such behavior. The model proposed by Andreoni & Miller (2002) does not allow for status seeking behavior, and Bolton & Ockenfels (2000, p. 172, Assumption 3) rule out such behavior by assumption on the shape of their proposed utility function. Fehr & Schmidt’s (1999) utility function allows for the possibility of such behavior, but they do not include it in their analysis; they argue would not change equilibrium behavior in the games they consider (p. 824). 4 See, e.g., Fehr & Schmidt (1999), Fehr & Gächter (2000), Charness & Rabin (2002). the choices of others; they are simply choosing allocations. Thus we find that there are a variety of social preferences and these preferences depend critically on the social context. This experiment thus advances the quest for uncovering the distribution of social preferences (Fehr & Schmidt (2009)) and finds support for hypotheses concerning identity and economic outcomes (Akerlof & Kranton (2000, 2010)). In this experiment, subjects allocate money to themselves and to others in three conditions: an asocial control, a minimal group treatment, and a political group treatment. In the minimal group treatment, following the classic method in social psychology, subjects are divided into two groups according to their preferences over images and lines of poetry. In the political group treatment, subjects are divided into two groups according to their self-declared political affiliations and leanings. The asocial treatment serves as a control for both group treatments. The minimal group treatment serves as a control for the political group treatment. Following the work of Akerlof & Kranton (2000, 2010), we test whether a subject’s behavior depends on his or her identity. Identity, here, as in social psychology, indicates an individual’s (self)assignment to a social category, or group. Examples of broad social categories in the real world are gender, race, ethnicity, nationality, political party, etc. Experiments can draw on such existing identities, or experiments can create social categories inside the laboratory, as in the minimal group treatment. The premise of the latter is that studying subjects with temporary identities created in the laboratory serves as a window on behavior outside the laboratory, where identities are longer lasting and more deeply held. Our experiment combines these methods and tests two hypotheses. First, we test whether subjects’ identities affect behavior. In particular, we test if subjects are less willing to allocate money to subjects outside their group. Second, we test whether this effect depends on individual identities and subjects’ affinities for their assigned group. We infer this affinity from subjects’ self-reported political affiliations as Democrat, Republican, Independent, or None. An Independent assigned to the Democrat 5 Iriberri & Rey-Biel (2011) find that about 10% of subjects are “competitive” in a setting like our asocial condition. We find that about 5% are “competitive” in the asocial condition and 20% are competitive in group treatments when allocating income to subjects in the other group. group would likely have less affinity for their assigned group than a Democrat assigned to the Democrat group. We find support for both hypotheses. In the minimal group treatment, subjects are more competitive and more selfish when allocating money to out-group members. But in the political group treatment, only Democrats and Republicans exhibit this behavior. Independents have significantly different behavior, treating out-group subjects similarly to in-group subjects. This study builds on two streams of experimental literature in economics: on social preferences and on social identity. The work on social preferences often pits the theory of a “selfish economic man” against a theory where people also have preferences over the payoffs of others. Charness & Rabin (2002) introduce a series of games and method to estimate social preferences and conclude that subjects are not purely selfish and exhibit preferences for social welfare maximization rather than aversion to inequity. Fehr & Schmidt (1999) suggest that there might not be one way to describe people, as selfish or not, or inequity averse or not, but rather there is a distribution of individual social preferences. Andreoni & Miller (2002) find that indeed different individuals follow consistently different rules for the allocation of payoffs. We find that, in the asocial condition, about 25% of subjects are “selfish,” they put almost no weight on anyone’s payoffs but their own. About 37% of subjects have preferences to maximize social welfare and 33% aim for fair allocations. The remaining 5% are “competitive;” they are willing to reduce their own absolute payoffs in order to increase the difference between their payoffs and the other person’s payoffs. These distributions change in the group treatments, indicating that social preferences are not constant but depend on the social context. In particular, there is a significant increase in selfish behavior and competitive behavior when allocating income to out-group subjects. In the minimal group treatment, 35% of subjects are selfish and 21% are competitive, with only 13% maximizing social welfare and 31% striving for fair allocations. Thus well more than half of the subjects are neither fair nor social welfare maximizing when facing out-group subjects. In the area of social identity, several early experiments showed that the race or ethnicity of subjects changes play in dictator and ultimatum games (e.g., Fershtman & Gneezy (2000), Glaeser, Laibson, Scheinkman, and Souter (2000)). A recent set of experiments has studied social groups created in the laboratory. Our paper is closest to Chen and Li (2009) who use a minimal group paradigm and find that, on average, subjects are more likely to be social welfare maximizing towards in-group members. Our paper is also close in spirit to, and supports the results of, Klor & Shayo (2010) who divide subjects into two groups according to their university fields of study. Subjects are assigned gross incomes and asked to vote over alternative redistributive tax schemes. They find that subjects vote more often for the tax rate that favors in-group members. Our experiment finds a strong effect of the group treatment: effect: on average, subjects are less fair when allocating to in-group members than out-group members, which is similar to Chen & Li‘s (2009) results. This average hides the range of subject behavior. It hides the prevalence of purely selfish behavior and the destructive behavior of subjects that emerges strongly in the group context. To uncover individual preferences, we use a finite mixing model, which is relatively new to experimental economics. 8 We use the utility function proposed by Fehr & Schmidt (1999) and Charness & Rabin (2002) and estimate parameters using a discrete choice maximum likelihood function and a finite mixing model. The mixing model estimates “types” of subjects, where the parameters characterizing each “type” are not assumed but are those that maximize the likelihood function. We can then interpret these “types” according to the utility function: we find subjects are distinctly either “selfish,” “weak social welfare maximizing,” “strong social welfare maximizing,” or “competitive.” Iriberri and Rey-Biel’s (2011) recent contribution also studies the possibility that subjects adopt significantly different and 6 See Chen & Li (1999) and Akerlof & Kranton (2010) for extensive reviews of the experimental literature in economics and social psychology. 7 Klor & Shayo (2010) find further that subjects’ behavior in the experiment relates to answers to questions concerning redistribution in a post-experiment survey using an adaptation of questions from the World Values Survey. 8 To the best of our knowledge, Stahl and Wilson (1994) was the first use of finite mixture modeling in behavioral experiments. They and followers such as Bosch-Domenech et. al (2010) consider Beauty Contest games, estimating the proportion of subjects who reason at different levels. Harrison and Rutstrom (2009) and Conte et. al allow for a mixture of expected utility and prospect theory. Andersen et. al. (2011) allow for part of the population to behave according to traditional exponential discounting and the remainder to behave according to hyperbolic discounting. distinct behavior in dictator games. They estimate four types using the Fehr & Schmidt (1999) and Charness & Rabin (2002) linear utility that we also adopt. 9 We take a further step and classify individual subjects into types in a way consistent with the mixing distribution (Nagin (2005)). We construct a posterior probability that an individual subject is of certain type and assign individuals to the type with the greatest posterior probability. To our knowledge the present study is the only one in behavioral economics that takes this next step and combines this classification with demographics and other subject-specific data to study the sources of individual variation. We use this classification to test the identity hypotheses discussed above. We also study how political ideology and demographic characteristics relate to individual behavior in different treatments. We uncover, in particular, a correlation between social preferences and political ideology; subjects who support “small government” are significantly more selfish than other subjects. This paper is organized as follows. Section II describes our experiment in detail. Section III provides the theoretical framework and empirical strategy for analyzing the data. We report the behavioral results in Section IV. Section V concludes. II. The Experiment The experiment was conducted in the Duke Center for Cognitive Neuroscience, which follows the same protocols as laboratories in experimental economics, in particular the protocol of no deception. The experiment involved 141 subjects drawn from the Duke University community. Summary demographic characteristics of the subjects are presented in the Appendix. 9 Econometrically the present paper differs from Iriberri & Rey-Biel (2011) in that we use the mixture model to calculate the posterior probability that an individual is a particular type and use these posterior probabilities to assign each individual to a type. We then determine the demographic and other factors that are associated with each type. Substantively, the goals of the papers are also different. Iriberri & Rey-Biel study how revealing the distribution of play changes future play, and they take great care to minimize any interpersonal influences that could stimulate other-regarding behavior. The purpose of our experiment, in contrast, is to test how different social contexts affect other-regarding behavior. 10 Klor & Shayo (2010) classify subjects into types according to the individual utility parameter estimates, as in Andreoni & Miller (2002). They then relate this type-classification to individual attributes and answers to survey questions. Sessions were held at various times of day and were spread across January, February, and March 2011. Figure 1. Timeline of Experiment For all subjects, experimental sessions proceeded as illustrated in Figure 1. First, subjects received instructions on the decisions they would be asked to make and practiced using the predefined computer keys that would indicate their choices. All sessions began with the asocial condition. Then each subject made decisions in the minimal group treatment and the political group treatment. The order of the group treatments was randomized across subjects. In the asocial condition, subjects were asked to allocate money to themselves and other participants. There were two kinds of pairings, which occurred randomly. Subjects allocated money between themselves and other subjects, called YOU-OTHER matches. Subjects also allocated money between two random other subjects, called OTHER-OTHER matches. These 11 The Appendix provides a transcript of subject instructions and other details of the experiment, including sample screen shots from the minimal group and political group treatments. Instructions 3-5 minutes Asocial Control 12 minutes 2 Survey 2-5 minutes 78 Choices 17 minutes Minimal Group Treatment 52 Choices