Handouts without handshakes: Patronizing out-group preferences in humans

Transcription

1 Handouts without handshakes: Patronizing out-group preferences in humans Tomomi Tanaka and Colin F. Camerer* Arizona State University, Tempe, AZ * camerer@hss.caltech.edu, California Institute of Technology, Pasadena, C Abstract Effectiveness of social groups depends on norms of pro-social cooperation with in-group members, which are often shared and policed. 1-5 Typically these norms of in-group favoritism are also associated with out-group prejudice. 6-8 However, low-status groups frequently exhibit out-group favoritism toward higher-status groups We report rare evidence of high-status groups (Vietnamese and Chinese) exhibiting favoritism toward a lower-status outgroup (Khmer) in southern Vietnam. This pattern is evident in high-stakes experimental games that measure altruistic giving and third-party punishment of giving-norm violations. The high-status groups give more to the Khmer, and defend the Khmer more aggressively when they are treated poorly. However, the high-status groups show more typical out-group prejudice against Khmer in trust games involving risky investment, or when forming mutually-beneficial limited coalitions. The high-status groups seem to offer a handout, but not a handshake, a pattern we call patronizing out-group preference. We also report measures of cross-group stereotyping which are consistent with patronizing treatment of out-groups. Patronizing treatment of out-groups is a challenge for evolutionary theories which explain indirect group selection through individuallyselected tendencies toward in-group favoritism. 1,2,12,13

2 Charitable giving, voting for governmental income redistribution, and experimental findings show that people often sacrifice their own material payoffs to improve the well-being of others. Pro-social sacrifices occur for many reasons: reducing inequality, 14,15 increasing the social welfare and helping the poorest, 16,17 reciprocity, 17,18 socially policing norm violators. 19 Human behavior seems to reflect a mixture of both selfishness and pro-sociality toward non-kin to an extraordinary degree, 20 compared to other species. Pro-social preferences lubricate complex economic exchange and are associated with national wealth. 21 Pro-social behaviors are also strongly affected by group membership. In experiments people are sometimes pro-social toward in-group counterparts even when groups are created in simple random ways Social identity theory hypothesizes that ingroup bias arises because groups offer identity and positive self-image. 25 Anthropologists and theoretical biologists have shown that the capacity for in-group favoritism is a key ingredient in gene-culture evolution, which creates indirect group selection and potentially explains the rapid progress of human civilization (and its failings, such as wars and genocides). 1,2,7,12 They hypothesize that groups that demonstrate altruistic behaviors toward ingroup members are more likely to thrive when groups compete. However, some studies find out-group favoritism, but only among low-status groups who favor higher-status group members Field experiments have also shown in-group favoritism in naturally-occurring groups, 6 although few (or no) covariates of group membership are included in these studies. We did detailed experiments in Vietnamese villages to investigate how social group status associated with ethnicity affects preferences for altruistic distribution and for economic cooperativeness. Experiments were conducted in 12 villages with three ethnic groups, Vietnamese (the majority), Chinese (a rich minority) and Khmer (a poor minority), five decisions and games for substantial financial stakes. Three games measure altruism and altruistic punishment. In a dictator game subjects divided a fixed sum with another person. In an envy game a subject received a fixed payment and then decided whether another person got a smaller, equal, or larger payment than their own. In the

3 dictator and envy games (Fig. 1), groups typically exhibit in-group favoritism by giving significantly more to their own group members than to other group members, with one unusual cross-group exception the Vietnamese give more to Khmer in both games and the Chinese give more to Khmer in dictator game. These outgroup favoritism is statistically significant after controlling for individual differences in income, education and other socioeconomic factors (see Supplementary Information). In a third-party punishment game (3PP) 26 two people played the dictator game and the third person decided whether to spend money to punish a dictator who gave too little. Most groups (a) preferentially defend own-group members who receive too little, which is consistent with the findings from Papua New Guinea, 6 and (b) are more reluctant to punish own-group senders who give too little (Fig. 2). There is again an interesting exception favoring out-groups the Vietnamese are quicker to punish a fellow Vietnamese for offering too little, compared to punishment of a Khmer sender making equal offers. The Vietnamese also defend Khmer receivers more aggressively than Vietnamese receivers. This out-group favoritism is statistically insignificant but the tendency to defend another group more strongly than one s own has never been observed in many 3PP game experiments. 6,27 Two games measured prosocial attitudes in stylized versions of business cooperation. In the trust game, player 1 can divide a sum evenly or can invest money which earns a social return (growing by a factor of two). If player 1 invests, then player 2 can reciprocate by splitting the larger sum evenly or can give only 25% of it to player 1. In a coalition game, 28 there are three players. If two players both choose each other as business partners they earn a fixed amount; a player who is excluded (unchosen) get nothing. In both of these games the modest pro-khmer favoritism shown in the three distributional games largely disappears. All groups generally invest more often (Fig. 3), as player 1, if they know that their partner player 2 is in their own group, except for Vietnamese who invest more often to Chinese than to their own group. Player 2 s in all groups reciprocate more often when the player 1 is in their own group. After controlling for demographic variables, Khmer are trusted significantly less by other ethnic groups (See Supplementary Information).

4 In the coalition game, Vietnamese choose Chinese over Khmer 61 % of the time as coalition partners and Chinese exhibit a similar bias, choosing Vietnamese over Khmer 56 % of the time. The likelihood of cross-group coalition formation by Vietnamese and Chinese is highest; as a result, the Khmer earn the least. The expected earnings in the coalition game for Vietnamese, Chinese and Khmer are 8,839, 12,609 and 6,620 dong, respectively. The general pattern of group behavior across all five games shows patronizing treatment of the out-group Khmer: Other groups are willing to give to the Khmer, and protect them when others are not generous, about as strongly as they give to and defend their own groups, but also show the usual pattern of out-group prejudice toward the Khmer in trust and coalition-formation games. The other groups are quick to offer the Khmer a handout, but not a handshake, a form of noblesse oblige, in which high-status groups are expected to help the low-status groups. This pattern is also illuminated by social psychology scales which measure competence, warmth, status, and competitiveness. 29 The Chinese and Vietnamese see the Khmer as warm but incompetent, and the Khmer see the other two groups in the opposite way (Fig. 4). We observe significant relations between these individual-level measures and individual-level behavior in the decisions and games (see Supplementary Information). The amount sent in the envy game and the likelihood of reciprocation by player 2 is higher when outgroups are perceived as warm. The amount sent in the envy game reduces when outgroups are perceived as competitive threat. Importantly, there are many ways in which government policies and social practices in southern Vietnam parallel the experimental behavior. Khmer households are much more likely to receive scholarships, tuition reductions, and exemptions from paying land tax (see SI). However, the Khmer are less likely to own businesses and employ other people. The intermarriage between Vietnamese and Chinese is much more common than intermarriage with Khmer. This evidence of patronizing out-group treatment poses a challenge to theories that explain ingroup favoritism by indirect group selection, and to psychological theories which explain outgroup favoritism. One possibility is that the evolved psychology that generally disfavors outgroups can be displaced by a paternalistic psychology of pseudo-familial protection if

5 outgroup members appear helpless, pose no threat, and are perceived as warm and deserving (consistent with evidence from the stereotyping measures). The other groups treat the Khmer as childlike, helping them with education and tax relief but not taking them seriously as business partners or political leaders. Of course, if the perception of warmth and deservingness is replaced by a perception of being cold, impure or undeserving, the result could be exaggerated out-group disfavoritism. 27,30 A possible clue is the fact that the Khmer in southern Vietnam are aboriginal they were natives before their region became part of Vietnam in the 17th century. One can draw parallels between their history and modern treatment, and history and treatment of other aboriginal peoples in Australia, Canada, and the United States. Modern policies are often generous in helping aboriginal people directly, but those groups have not achieved substantial economic success and are typically not broadly involved in large-scale economic activity (e.g., running large companies or leading politically). Testing whether patronizing out-group favoritism is evident toward other aboriginal groups, and developing theory to account for this exceptional pattern, is a natural next step. Methods We conducted experiments in twelve villages in the Mekong Delta in July and August A total of 326 subjects (141 Vietnamese, 127 Khmer and 58 Chinese) participated in the experiments. A week before the experiments, research coordinators contacted local government officials in each research site and asked them to invite one person from each of the 25 previously surveyed households to the experiments. Before the experiments started, Vietnamese, Chinese and Khmer subjects received white, yellow and red ID tags, respectively. In the instruction we informed the subject the colors of their identification tags were based on ethnicity. We used the strategy method, asking subjects their decisions contingent on the color group of their opponents.

6 In each experiment, a Vietnamese experimenter read a general instruction of the experiment in Vietnamese, and a Khmer experimenter read the general instruction in Khmer. After the general instruction, subjects went out of the room, one by one, and a student assistant read the instruction for each game for each subject, and assisted the subjects to play each game. The average experimental earning for each subject was 97,300 dong (about 6 US Dollars, roughly four days wages for casual unskilled labor). Experiments lasted about 2 hours.

7 References 1 Bowles, S., Choi, J.-K. & Hopfensitz, A. The co-evolution of individual behaviors and social institutions. J. Theor. Biol. 223, (2003). 2 Henrich, J. & Boyd, R. Why people punish defectors: Weak conformist transmission can stabilize costly enforcement of norms in cooperative dilemmas. J. Theor. Biol. 208, (2001). 3 Boyd, R., Gintis, H., Bowles, S. & Richerson, P. J. The evolution of altruistic punishment. Proc. Natl. Acad. Sci. 100, (2003). 4 Fehr, E. & Fischbacher, U. The nature of human altruism. Nature 425, (2003). 5 Fehr, E. & Gächter, S. Altruistic punishment in humans. Nature 415, (2002). 6 Bernhard, H., Fischbacher, U. & Fehr, E. Parochial altruism in humans. Nature 442, (2006). 7 Choi, J.-K. & Bowles, S. The coevolution of parachial altruism and war. Science 318, (2007). 8 Hewstone, M., Rubin, M. & Willis, H. Intergroup bias. Annu. Rev. Psychol. 53, (2002). 9 Boldry, J. G. & Kashy, D. A. Intergroup perception in naturally occuring groups of differential status: A social relations perspective. J. Pers. Soc. Psychol. 77, (1999). 10 Jost, J. T., Banaji, M. R. & Nosek, B. A. A decade of system justification theory: Accumulated evidence of conscious and unconscious bolstering of the status quo. Polit. Psychol. 26, (2004). 11 Haines, E. L. & Jost, J. T. Placating the powerless: Effects of legitimate and illegitimate explanation on affect, memory, and stereotyping. Soc. Just. Res. 13, (2000). 12 Bowles, S. Group competition, reproductive leveling, and the evolution of human altruism. Science 314 (2006). 13 Gintis, H., Bowles, S., Boyd, R. & Fehr, E. Explaining altruistic behavior in humans. Evol. Hum. Behav. 24, (2003). 14 Bolton, G. E. & Ockenfels, A. Erc: A theory of equity, reciprocity, and competition. Am. Econ. Rev. 90, (2000). 15 Fehr, E. & Schmidt, K. M. A theory of fairness, competition, and cooperation. Quart. J. Econ. 114, (1999). 16 Engelmann, D. & Strobel, M. Inequality aversion, efficiency, and maximin preferences in simple distribution experiments. Am. Econ. Rev. 94, (2004). 17 Charness, G. & Rabin, M. Understanding social preferences with simple tests. Quart. J. Econ. 117, (2002). 18 Rabin, M. Incorporating fairness into game-theory and economics. Am. Econ. Rev. 83, (1993). 19 Fehr, E. & Gachter, S. Cooperation and punishment in public goods experiments. Am. Econ. Rev. 90, (2000).

8 20 Nowak, M. A. Five rules for the evolution of cooperation. Science 314, (2006). 21 Knack, S. & Keefer, P. Does social capital have an economic payoff?: A cross-country investigation. Quart. J. Econ. 112, (1997). 22 Charness, G., Rigotti, L. & Rustichini, A. Individual behavior and group membership. Am. Econ. Rev. 97, (2007). 23 Chen, Y. & Li, S. X. Group identity and social preferences. Am. Econ. Rev. 99, (2009). 24 Goette, L., Huffman, D. & Meier, S. The impact of group membership on cooperation and norm enforcement: Evidence using random assignment to real social groups. Am. Econ. Rev. 96, (2006). 25 Tajifel, H. Social psychology of intergroup relations. Annu. Rev. Psychol. 33, 1-39 (1982). 26 Fehr, E. & Fischbacher, U. Third-party punishment and social norms. Evol. Hum. Behav. 25, (2004). 27 Fehr, E., Hoff, K. & Kshetramade, M. Spite and development. Am. Econ. Rev. 98, (2008). 28 Habyarimana, J., Humphreys, M., Posner, D. & Weinstein, J. Social focal points (Georgetown University, 2006). 29 Fiske, S. T., Cuddy, A. J., Glick, P. & Xu, J. A model of (often mixed) stereotype content: Competence and warmth respectively follow from perceived status and competition. J. Pers. Soc. Psychol. 82, (2002). 30 Fershtman, C. & Gneezy, U. Discrimination in a segmented society: An experimental approach. Quart. J. Econ. 116, (2001).

9 Figure 1: The mean amount sent in Envy Game and Dictator Game. Colored bars indicate owngroup. Within-subject Wilcoxon rank-sum test results for own-other differences are shown with asterisks; *p<.01, **p<.05, ***p<.001. Vietnamese give more to Khmer in both games and Chinese give more to Khmer in dictator game. These outgroup favoritism toward Khmer is not statistically significant in within-subject Wilcoxon rank-sum test. However, after controlling for individual differences in income, education and other socioeconomic factors, the outgroup favoritism toward Khmer is significant in both games (see SI).

10 Figure 2: Thresholds of dictator allocations at which punishment occurs in the third party punishment game. A high number indicates a higher willingness to punish. Colored bars indicate own-group. Generally punishment thresholds are lower for own-group senders (i.e., subjects are more reluctant to punish own-group senders) and higher for own-group receivers (i.e., subjects defend own-group members who receive too little). The only exception is Vietnamese punishment involving Khmer: Vietnamese are quicker to punish a fellow Vietnamese for offering too little, compared to punishment of a Khmer sender making equal offers. The Vietnamese also defend Khmer receivers more aggressively than Vietnamese receivers. Statistical tests are within-subjects tests across different ethnic groups of senders or receivers (i.e., between graph columns) *p<.01, **p<.05, ***p<.001.

11 Figure 3: Proportion of Player 1 and Player 2 who trust and reciprocate in trust game. Colored bars indicate own-group. The results of within-subject Wilcoxon rank-sum test for own-other differences are shown with asterisks; *p<.01, **p<.05, ***p<.001.

12 Figure 4: Mean estimated factors of stereotypes (Warmth and Competence) by ethnicity

13 Handouts without handshakes: Patronizing out-group preferences in humans Tomomi Tanaka and Colin F. Camerer * tomomi.tanaka@asu.edu, Arizona State University, Tempe, AZ * camerer@hss.caltech.edu, California Institute of Technology, Pasadena, CA Supplementary Information Subjects Experimental Procedures Games Supporting Analyses Stereotypes Econometric Analysis of Experimental Results Economic Status by Ethnicity in the Mekong Delta Game Instructions

14 Subjects We conducted experiments in twelve villages in the Mekong Delta in July and August The experimental subjects were previously interviewed during a 2002 living standard measurement survey. In the 2002 survey, 142 villages were randomly selected for the survey in the Mekong Delta, and 5813 Vietnamese, 85 Chinese and 388 Khmer households were interviewed. The mean income of Vietnamese, Chinese and Khmer households surveyed in the entire Mekong Delta was 21.6, 40.5, 15.1 million dong (1,350, 2,531, and 944 US dollars), respectively. We selected twelve villages where at least two ethnic groups were interviewed as research sites. In each village, 25 households were interviewed. Some descriptive statistics of the twelve experimental sites are given in Supplementary Table 1. In villages VCK1 and VCK2, Vietnamese, Khmer, and Chinese households were interviewed. In villages indexed by VK1, VK2, --- VK8 (where VK1 indexes the highest mean village income and VK8 indexes the lowest), Vietnamese and Khmer households were interviewed. In villages VC1 and VC2, only Vietnamese and Chinese households were interviewed. The mean income of Vietnamese, Chinese and Khmer households in our research sites was 23.8, 27.3 and 14.0 million dong (1,490, 1,709, and 873 US dollars), respectively. The mean income of the Chinese households in our study villages is lower than the mean income of Chinese households interviewed in the Mekong Delta. Nevertheless, it s not significantly different (unpaired t-test t=0.81, p=0.42). The mean income of Vietnamese, Chinese and Khmer subjects was 25.5, 24.7 and 15.3 million dong (1,565, and 1,541 and 957 US dollars), respectively. Supplementary Table 2 summarizes correlations between key variables. See Supplementary Table 4 for variable definitions. Education correlates negatively with age (-0.41) and being Khmer (- 0.38), and positively with mean village income (0.33) and relative income within the village (0.29). Political power, defined as the share of village official positions normalized by population ratio, is correlated with being Vietnamese (0.32) and negatively correlated with being Chinese (-0.40). Chinese live in high mean income villages, and Khmer live in low mean income villages (correlation efficient are 0.35 and 0.40, respectively). The mean village income and Gini coefficient are highly correlated (0.69). Thus, when conducting econometric analysis, we use only mean village income as

15 an independent variable. Most of the correlations among demographic variables except for mean village income and Gini coefficient are not vary large, so multicollinearity between variables will not do much harm in multiple regressions. Supplementary Table 3 shows the acquaintance ratio, percentage of interethnic marriage and ethnicity of neighbors among experimental subjects. Supplementary Table 1: Descriptive statistics of study villages VKC1 VKC2 VK1 VK2 VK3 VK4 VK5 VK6 VK7 VK8 VC1 VC2 Number of Household By ethnicity of household head (%) Vietnamese Khmer Chinese Percentage of village officials by ethnicity (%) Vietnamese Khmer Chinese Number of Subjects Vietnamese Khmer Chinese Total Mean household income of subjects (million dong) Vietnamese Khmer Chinese Total Age (mean) Gender (male=1) (mean) Education (year) (mean)

16 Supplementary Table 2: Correlations between key variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (1) Age 1 (2) Gender (3) Edu (4) Traditional (5) Trade (6) Business (7) Public (8) Private (9) Casual (10) Power (11) ReIncome (12) MnIncome (13) Gini (14) Vietnamese (15) Chinese (16) Khmer Supplementary Table 3: Acquaintance ratio, interethnic marriage and neighborhood Ethnicity of Spouse Ethnicity of Subject V K C V K C Acquaintance Ratio by opponent s ethnicity Ethnicity of Subject V K C V K C Neighbor's ethnicity Ethnicity of Subject V K C V K C

17 Supplementary Table 4: Variable definitions Variable name Description Age Age of the subject Gender Gender of the subject, 1=male Edu Number of years the subject attended school Traditional =1 if subject's main occupation is farming, fishery, or raising livestock. Trade =1 if subject's main occupation is trading (street vendors). Business =1 if the subject is engaged in household business. Public =1 if the subject works for a public organization. Private =1 if the subject works for a private organization. Casual =1 if the subject is a daily laborer. Power Political power of the subject s ethnic group measured by [(Number of village officials of the subject s ethnicity)/ (Number of households of the subject s ethnicity)] / [(Total number of village officials)/ (Total number of households)] ReIncome Subject s relative wealth within the community, measured by subject's household income divided by the mean income of the village. MnIncome Mean household income of the village (million dong) among 25 households surveyed in 2002 Gini Gini coefficient of the income among 25 households surveyed in 2002 Vietnamese The subject is Vietnamese. Chinese The subject is Chinese. Khmer The subject is Khmer. Acquaintance Ratio of the subjects the subject knows by name in the opponent s ethnic group. Outgroup The other player is a member of outgroup. K-Outgroup The outgroup is Khmer Competence The mean score of 6 stereo-type questions (competent, confident, independent, competitive, intelligent, skillful) regarding the opponent s ethnic group Warmth The mean score of 4 stereo-type questions (tolerant, warm, goodnatured, sincere) regarding the opponent s ethnic group Status The mean score of 3 stereo-type questions (prestigious job, successful, educated) regarding the opponent s ethnic group Competition The mean score of 3 stereo-type questions (special breaks, power, resources) regarding the opponent s ethnic group Spouse The ethnicity of the opponent is the same as spouse s ethnicity. Parent The ethnicity of the opponent is the same as parent s ethnicity.

18 Experimental Procedures A week before the experiments, research coordinators contacted local government officials in each research site and asked them to invite one person from each of the 25 previously surveyed households to the experiments. In Villages, VCK1, VCK2, VC1 and VC2, we invited two persons from each of the 25 households; one for a morning session scheduled from 9 AM and one for an afternoon session scheduled from 12 PM. While the first group was getting paid, we started the second experiment in another room to make sure the subjects of the first and second sessions did not have a chance to talk with each other. In villages VK1, VK2, VK3, VK8, battle of sexes, coalition, envy, and trust games were played in that order. The dictator game was added after we conducted experiments in these four villages. In villages VKC1, VKC2, VK4, VK5, and VK7, battle of sexes, coalition, envy, trust, dictator games were played in that order. We added the third party punishment game when we conducted experiments in villages VC1, VC2 and VK6. Two new games (dictator game and third party punishment game) were played at the end of the experiments (after battle of sexes, coalition game, envy game, and trust game). We alternated the order of dictator game and third party punishment game across subjects. Subjects were randomly assigned the role of Player 1, 2 or 3 in the coalition game. Subjects played both the roles of Player 1 (investor) and Player 2 (trustee) in the trust game, Player 1 (sender) and Player 2 (receiver) in envy game and dictator game. The order of roles was randomized across subjects in the trust and third party punishment games. In the envy and dictator games, subjects played the role of Player 1 (sender) first, then Player 2 (receiver), since Player 2 make no decisions in the envy and dictator games. Before the experiments started, potential subjects were divided into groups, and were given ID tags of different colors upon arrival. In VCK villages, Vietnamese, Chinese and Khmer subjects were divided into white, yellow and red groups, respectively. In the instruction we informed the subject the colors of their identification tags were based on ethnicity. We used the strategy method, asking subjects their decisions contingent on the color group of their opponents. The strategy method has been widely used in experimental economics to efficiently use one subject to generate responses to a variety

19 of contingencies. It is conceivable in this design that subjects respond to a perceived experimental demand to vary their response across groups. However, this perception will not necessarily produce in-group favoritism, nor is it likely to exhibit the pattern of slight favoritism of Khmer in some games and out-group discrimination in others. Furthermore, it is conceivable to use subtler manipulations to convey information about ethnicity (such as names or pictures). However, if we had not marked the ethnic groups clearly by color, we could not be confident about the internal validity of the treatment, which is crucial for interpreting the results. (If we had used last names, for example, then it is possible subjects do not recognize the names as ethnic cues and a finding of no group favoritism would be due to a failure of basic internal validity). In VK1 and VK8 villages, Vietnamese and Khmer subjects were divided into white and red groups, respectively. In villages VK 2, 3, 4, 5, 6, and 7, we divided subjects into four groups by ethnicity and income. High-income Vietnamese, low-income Vietnamese, high-income Khmer, and low-income Khmer were assigned white, black, red and pink ID tags, respectively. The instruction said the colors of their identification tags were based on ethnicity and income. In each game, subjects were first asked their decisions if their opponents were either white or black groups (Vietnamese), and red or pink groups (Khmer). Afterward, they were asked their choices contingent on each of the four color groups. Similarly, in VC villages, subjects were divided into four groups, white, black, red and pink (high-income Vietnamese, low-income Vietnamese, high-income Chinese, and low-income Chinese, respectively). The subjects were first asked their decisions for each game if their opponents were either white or black groups (Vietnamese), then if their opponents were either red or pink groups (Chinese). In this paper, we report the effects of ethnicity on decisions in games and leave the analysis of ethnicity and income effects to another paper. In each experiment, Vietnamese experimenter Pham Thanh Xuan, an undergraduate student at Can Tho University, read a general instruction of the experiment in Vietnamese, and Danh Na Ret, an undergraduate student at Can Tho University, read the general instruction in Khmer. After the general instruction, subjects went out of the room, one by one, and a student assistant read the instruction for each game for each subject, and assisted the subjects to play each game. Before the subjects played games, we

20 conducted a quiz for the first two games (the battle of sexes and coalition games). Twelve subjects either failed the quiz or decided to leave the session. We prepared Vietnamese instruction for Vietnamese and Chinese subjects, and Khmer instruction for Khmer subjects. Vietnamese instruction was provided to Khmer subjects upon request. In total, 141, 127 and 58 Vietnamese, Khmer and Chinese subjects participated in the experiments. The average experimental earning for each subject was 97,300 dong (about 6 US Dollars, roughly four days wages for casual unskilled labor). Experiments lasted about two hours. After subjects played the games, we conducted a questionnaire on stereotyping images of other ethnicities. Subjects rated other ethnic groups on 5-point scales reflecting competence, warmth, status and competition.

21 Vietnamese experimenter Pham Thanh Xuan reads a general instruction in Vietnamese. Khmer experimenter Danh Na Ret reads a general instruction in Khmer.

22 After the general instruction, subjects go out of the room. They receive instruction and record sheets for each game. Subjects are helped by research assistants when making decisions. Games We conducted the envy, dictator, third-party punishment, binary trust and coalition games. We also conduced the battle-of-sexes game. We report the results of this game in a separate paper. Please refer to Supplementary Table 5 for the description of games. Envy Game In the envy game there are two players, player 1 (sender) and player 2 (receiver). Player 1 receives 12,000 dong, and decides how much player 2 should receive from the experimenter. Player 1 has 7 options on how much to give to player 2; 0, 5,000, 10,000, 12,000, 15,000, 30,000 or 60,000 dong. Notice no matter how much player 1 sends to player 2, the sender s own payoff does not change.

23 The inequality aversion hypothesis 1,2 predicts player 1 sends 12,000 dong to player 2. However, if player 1 is concerned about maximizing efficiency as discussed by Englemann and Strobel 3, he/she will send 60,000 dong to player 2. If player 1 envies player 2, he/she will send less than 12,000 dong to player 2. Charness and Grosskopf 4 conducted a similar game where sender s payoff does not depend on how much the sender decides to allocate to the receiver. Using student subjects, they show that senders generally choose to maximize efficiency. Dictator game We use the dictator game to examine social norm concerning distributional fairness. The dictator game is a game played by two players, player 1 (Sender) and player 2 (Receiver). Player 1 is given 10,000 dong, and decides an allocation of the 10,000 dong. Player 1 is given eleven (11) choices, sending either 0, 1,000, 2,000, 3,000, 4,000, 5,000, 6,000, 7,000, 8,000 9,000 or 10,000 dong to player 2. Player 2 then accepts, no matter what player 1 proposed. In a subgame perfect equilibrium, player 1 would offer player 2 nothing. However, experimental evidences indicate that player 1 often offers relatively large shares to player 2 (often 50-50) 5. Pablo Brañas-Garza 6 find that when subjects know that receivers are poor, they send significantly higher amounts. Fershtman and Gneezy 7 conducted dictator game with two major ethnic groups in Israel and found no significant in-group bias. Third party punishment game If the dictator game reflects social norms on distributional fairness among different ethnic groups, the question arises whether violators of such norms will be punished. We conduct third party punishment game 8,9 to study altruistic punishment, i.e., punishment of a violator of social norms at a cost to punisher. Altruistic punishment is considered a key to maintaining social order and cooperation. 9,10 The third party punishment game is similar to dictator game except the third party (punisher) is added to the game. Like the dictator game, player 1 decides what portion of 10,000 dong to send to player 2. Player 3 is endowed with 5,000 dong and has an option of punishing player 1, contingent on the

24 amount of money player 1 sends to player 2. If player 3 decides to punish, 1,000 dong will be subtracted from his/her own endowment of 5,000, and 3,000 dong will be taken away from player 1. Player 1 is aware that player 3 can punish him/her when he/she makes a decision. Bernhard, Fischbacher and Fehr 8 conducted a third party punishment game with members of two indigenous groups in Papua New Guinea. They find that third parties show stronger altruism towards ingroup victims and give ingroup norm violators more lenient judgments. Goette, Huffman and Meier 11 examine the effects of group membership in the Swiss Army. They also find third-party punishment is stronger when a violation affects an ingroup member as opposed to an outgroup member. Trust game We conduct the binary trust game 12 to study trust and reciprocity, which is considered a key element of social capital In a binary trust game, Player 1 (sender) has two choices, A (invest) or B (don t invest). If player 1 chooses A, both players 1 and 2 receive 20,000 dong. If player 1 chooses B, player 2 s decision determines payoffs. If player 2 chooses B1 (don t repay), players 1 and 2 receive 10,000 dong and 50,000 dong, respectively. If player 2 chooses B2 (repay), both players receive 30,000 dong. Player 1 may be worse off if he/she trusts receiver and is not get reciprocated. On the other hand, regardless of the decision he/she makes, player 2 always benefits if player 1 trusts player 2. Nava Ashraf, Iris Bohnet and Nikita Piankov 16, Michael R. Carter and Marco Castillo 17, and Håkan J. Holm and Anders Dalienson 18 demonstrate how trusting behavior can be largely explained by altruism, because trusting investors often do not expect to have much money repaid. Glaeser, Laibson, Scheinkman and Soutter 19 find subjects who are paired with a partner of a different race or nationality send back less money to their partner. Bouckaert and Dhaene 20 use male businessmen of distinct ethnic origins to investigate inter-ethnic trust and reciprocity in Belgium. They find trust and reciprocity on average are independent of the ethnic origin of either participant or the opposite party.

25 Hong and Bohnet 21 conducted modified binary trust games and identified source of distrust among high status and low status individuals. They found individuals with low social status, women, minorities etc, were averse to disadvantageous inequality while individuals with high status, men, Caucasians, etc, dislike being betrayed. We use the decisions made in envy game to control for aversion to disadvantageous relative standing. Coalition game The coalition game was proposed by Habyarimana, Humphreys, Posner and Weinstein 22. There are three players, player 1, player 2 and player 3 in the game. They have three proposals they can vote for simultaneously, proposals 12, 13, and 23. If Proposal 12 receives more than two votes, players 1 and 2 receive 15,000 dong each and player 3 receives nothing. If Proposals 13 receives more than two votes, players 1 and 3 receive 15,000 dong and player 2 receives nothing. Likewise, if Proposal 23 receives more than two votes, players 2 and 3 receive 15,000 dong and player 3 receives nothing. In all games, we use the strategy method and ask subjects their decisions contingent on the group of their opponents. This permits powerful within-subject statistical tests. Supplementary Table 5: Payoff tables (1) Envy game Choice Player 1 12,000 12,000 12,000 12,000 12,000 12,000 12,000 Player 2 0 5,000 10,000 12,000 15,000 30,000 60,000 (2) Trust game Player 2 Player 1 B1 B2 A (20,000, 20,000) (20,000, 20,000) B (30,000, 30,000) (10,000, 50,000) (3) Coalition game Plan 12 Plan 13 Plan 23 Player 1 15,000 15,000 0 Player 2 15, ,000 Player ,000 15,000

26 Supporting Analyses Stereotypes We used questionnaire developed by Fiske et al. 23 to investigate how our subjects view other ethnic groups. The questions are listed in Supplementary Table 6. The factor competence represents images such as competence, confidence, capability, efficiency, intelligence and skillfulness. The factor warmth represents images such as warmth, good nature, sincerity and tolerance. The factor status corresponds images such as prestigious jobs, success, and educational achievement. The factor competition embodies competition for resources and political power. We conducted factor analysis and constructed four factors called competence, warmth, success and competition. Supplementary Table 7 contains factor scores of each question for each factor component. Supplementary Table 8 summarizes the mean of estimated factors. Supplementary Table 8-(1) shows the stereotyping images of Vietnamese by other ethnic groups. Both Chinese and Khmer view Vietnamese as high status but not warm. In addition, Khmer view Vietnamese as competitive threat while Chinese view Vietnamese as incompetent. Under Fiske et al. s model, Khmer s image on Vietnamese matches a high-status competitive outgroup which triggers envy, while Chinese s image of Vietnamese is mixed. Supplementary Table 8-(2) shows the stereotyping images of Khmer by other ethnic groups. Vietnamese and Chinese both think Khmer are not competent and hold low status, but are warm. Under Fiske et al. s model, Khmer would be categorized as low-competence, high-warmth subordinates who are not threat to other groups, and are pitied. Supplementary Table 8-(3) shows the stereotyping images of Chinese by other ethnic groups. Vietnamese and Khmer both think Chinese are highly competent. Vietnamese do not view Chinese as competitive threat while Khmer feel highly competitive toward Chinese. Khmer also think Chinese are not warm. Under Fiske et al. s model, Khmer s image of Chinese matches a high-status competitive outgroup which trigger envy, while Vietnamese hold rather favorable image of Chinese.

27 Supplementary Table 6: Stereotype questions Construct Competence Warmth Status Competition Items How are members of this group? 1: competent, 2: confident, 3: capable, 4: efficient, 5: intelligent, 6:skillful How are members of this group? 7: tolerant, 8: warm, 9: good natured, 10: sincere 11: How prestigious are the jobs typically achieved by members of this group? 12: How economically successful have members of this group been? 13: How well educated are members of this group? 14: If members of this group get special breaks (such as preference in hiring decisions), this is likely to make things more difficult for people like me. 15: The more power members of this group have, the less power people like me are likely to have. 16: Resources that go to members of this group are likely to take away from the resources of people like me. Supplementary Table 7: Factor scores Orthogonally rotated factors 1 (competence) 2 (warmth) 3 (status) 4 (competition) 1: Competent : Confident : Capable : Efficient : Intelligent : Skillful : Tolerant : Warm : Good natured : Sincere : Prestigious jobs : Successful : Educated : Special breaks : Power : Resources

28 Supplementary Table 8: Mean estimated factors of stereotypes by ethnicity (1) Stereotyping image of Vietnamese Mean estimated factors Subjects ethnicity Competence Warmth Status Competition Chinese Khmer (2) Stereotyping image of Khmer Mean estimated factors Subjects ethnicity Competence Warmth Status Competition Vietnamese Chinese (3) Stereotyping image of Chinese Mean estimated factors Subjects ethnicity Competence Warmth Status Competition Vietnamese Khmer

29 Econometric Analysis of Experimental Results We conducted econometric analysis of experimental results to investigate whether we observe significant outgroup bias after controlling for individual differences in income, education and other socioeconomic factors. We used income data from the 2002 Vietnam Living Standard Survey. Supplementary Table 9 contains the regression results of envy game, dictator game, trust game for player 1 and player 2. See Supplementary Table 4 for variable definitions. The regression results show strong evidence of out-group discrimination. Subjects send significantly less to out-groups in envy and dictator games, and trust and reciprocate out-groups significantly less, compared to the members of their own ethnic groups. However, they send significantly more to the Khmer out-group in envy and dictator games, and reciprocate more to the Khmer group compared to their own group. However, Vietnamese and Chinese trust Khmer less than their own ethnic groups, even though it s not significant. The regression results suggest subjects show altruistic behavior to Khmer outgroup but do not necessarily trust Khmer. Supplementary Table 10 shows how stereotyping images affect outgroup discrimination. The ethnic outgroups which are considered warm and less threatening receive significantly more in envy game, and those who are considered as competent receive more in dictator game. The ethnic groups which are considered as high social status are less trusted and those who are considered warm are more reciprocated. Supplementary Table 11 shows whether social distance affects outgroup discrimination. We investigated whether acquaintance rates, having parents or spouses of other ethnic origins affect behaviors observed in games. None of the social distance variables were significant. We also investigated whether political power within village communities affects behaviors in games. Supplementary Table 11 shows political power does not explain outgroup discrimination. We also tested whether being a ethnic minority or relative wealth between ethnic groups affect behavior in games, but none of these variables came out to be significant.

30 In summary, we find that stereotyping images are significant factors affecting outgroup bias but political power and social distance little explain outgroup discrimination.

31 Supplementary Table 9: Regression results I- Envy Game Dictator Game Trust Game (P1) Trust Game (P2) Khmer ** (0.274) (0.249) (0.327) (0.328) Chinese ** (0.279) (0.293) (0.283) (0.298) Age (0.008) (0.007) (0.010) (0.009) Gender * (0.240) (0.213) (0.236) (0.239) Edu (0.034) (0.033) (0.039) (0.037) Traditional *** *** (0.312) (0.250) (0.343) (0.336) Trade (0.378) (0.377) (0.401) (0.356) Business (0.419) (0.407) (0.390) (0.378) Public (0.434) (0.370) (0.549) (0.537) Private * (0.745) (0.487) (0.494) (0.527) Casual ** * (0.329) (0.281) (0.327) (0.335) ReIncome ** (0.105) (0.117) (0.136) (0.136) MnIncome ** *** (17.879) (18.324) (18.136) (16.729) Outgroup *** *** * *** (0.121) (0.096) (0.187) (0.154) K-Outgroup ** * * (0.186) (0.167) (0.281) (0.233) Constant (0.770) (0.760) Pseudo R Observation Note: *** Significant at the 1% level. ** Significant at the 5% level. * Significant at the 10% level. Standard errors are in parentheses. We conducted robust regressions, and adjusted standard errors for correlations within individuals.

32 Supplementary Table 10: Regression results II- Envy Game Dictator Game Trust Game (P1) Trust Game (P2) Khmer * ** (0.270) (0.257) (0.339) (0.330) Chinese ** (0.276) (0.302) (0.295) (0.306) Age (0.008) (0.007) (0.010) (0.009) Gender * (0.239) (0.218) (0.238) (0.238) Edu (0.033) (0.033) (0.039) (0.037) Traditional *** ** (0.303) (0.255) (0.344) (0.337) Trade (0.371) (0.388) (0.395) (0.357) Business (0.412) (0.405) (0.390) (0.382) Public (0.441) (0.377) (0.549) (0.551) Private * (0.797) (0.503) (0.509) (0.528) Casual ** * (0.311) (0.283) (0.328) (0.331) ReIncome ** (0.104) (0.118) (0.137) (0.136) MnIncome *** ** (17.433) (18.745) (18.605) (17.050) Outgroup *** *** ** *** (0.134) (0.117) (0.171) (0.168) K-Outgroup ** ** (0.257) (0.260) (0.281) (0.302) Competence * (0.110) (0.108) (0.136) (0.132) Warmth ** ** (0.107) (0.110) (0.147) (0.126) Status ** (0.112) (0.125) (0.120) (0.145) Competition *** (0.112) (0.110) (0.132) (0.118) Constant (0.786) (0.764) Pseudo R Observation

33 Supplementary Table 11: Regression results III- Envy Game Dictator Game Trust Game (P1) Trust Game (P2) Khmer ** (0.297) (0.253) (0.352) (0.347) Chinese ** (0.276) (0.298) (0.284) (0.295) Age (0.009) (0.007) (0.011) (0.010) Gender * (0.266) (0.215) (0.250) (0.249) Edu (0.037) (0.033) (0.041) (0.038) Traditional *** *** (0.335) (0.256) (0.372) (0.349) Trade * (0.381) (0.399) (0.410) (0.372) Business (0.441) (0.409) (0.394) (0.380) Public (0.447) (0.370) (0.544) (0.533) Private (0.740) (0.493) (0.510) (0.531) Casual ** (0.369) (0.293) (0.352) (0.352) ReIncome ** (0.117) (0.121) (0.143) (0.148) MnIncome *** *** (18.575) (18.872) (19.858) (18.462) Outgroup *** * (0.292) (0.294) (0.420) (0.324) K-Outgroup ** * (0.226) (0.194) (0.324) (0.272) Acquaintance (0.335) (0.321) (0.456) (0.394) Parent (0.434) (0.336) (0.479) (0.429) Spouse (0.326) (0.262) (0.545) (0.420) Neighbor (0.476) (0.371) (0.582) (0.561) Constant (0.813) (0.808) Pseudo R Observation

34 Supplementary Table 12: Regression results IV- Envy Game Dictator Game Trust Game (P1) Trust Game (P2) Khmer ** (0.277) (0.250) (0.328) (0.352) Chinese ** (0.287) (0.298) (0.296) (0.309) Age (0.008) (0.007) (0.010) (0.009) Gender (0.241) (0.213)* (0.237) (0.239) Edu (0.034) (0.033) (0.039) (0.037) Traditional *** *** (0.315) (0.250) (0.339) (0.337) Trade (0.377) (0.377) (0.402) (0.356) Business (0.418) (0.405) (0.383) (0.378) Public (0.433) (0.371) (0.552) (0.536) Private * (0.749) (0.484) (0.488) (0.528) Casual ** * (0.329) (0.285) (0.318) (0.333) ReIncome *** (0.106) (0.117) (0.137) (0.136) MnIncome ** (17.877) (18.342) (18.596) (16.779) Outgroup *** ** ** *** (0.158) (0.151) (0.237) (0.211) K-Outgroup ** * * (0.187) (0.167) (0.285) (0.233) PolPower (0.107) (0.120) (0.137) (0.145) Constant (0.779) (0.759) Pseudo R Observation