1 1 Inter-ethnic Interaction, Strategic Bargaining Power, and the Dynamics of 2 Cultural Norms: A Field Study in an Amazonian Population 3 4 John Andrew Bunce1,2,3 and Richard McElreath1,2 5 6 1Department of Human Behavior, Ecology, and Culture, Max Planck Institute for Evolutionary 7 , , Germany 8 2Department of Anthropology, University of California, Davis, CA USA 9 3Department of Anthropology, Indiana University, Bloomington, IN USA 10 11 Corresponding author: John Andrew Bunce (tel: +49 341 3550 347, email: 12 [email protected]) 13 14 ORCID 15 John Andrew Bunce: 0000-0003-4092-485X 16 Richard McElreath: 0000-0002-0387-5377 17 18 In press in Human

2 19 ABSTRACT 20 Ethnic groups are universal and unique to human societies. Such groups sometimes have 21 norms of behavior that are adaptively linked to their social and ecological circumstances, and 22 ethnic boundaries may function to protect that variation from erosion by inter-ethnic interaction. 23 However, such interaction is often frequent and voluntary, suggesting that individuals may be 24 able to strategically reduce its costs, allowing adaptive cultural variation to persist in spite of 25 interaction with out-groups with different norms. We examine five mechanisms influencing the 26 dynamics of ethnically-distinct cultural norms, each focused on strategic individual-level choices 27 in inter-ethnic interaction: bargaining, interaction frequency-biased norm adoption, assortment 28 on norms, success-biased inter-ethnic social learning, and childhood socialization. We use 29 Bayesian item response models to analyze patterns of norm variation and inter-ethnic interaction 30 in an ethnically-structured Amazonian population. We show that, among indigenous Matsigenka, 31 inter-ethnic education with colonial Mestizos is more strongly associated with Mestizo-typical 32 norms than is even extensive inter-ethnic experience in commerce and wage labor. Using 33 ethnographic observations, we show that all five of the proposed mechanisms of norm adoption 34 may contribute to this effect. However, of these mechanisms, we argue that changes in relative 35 bargaining power are particularly important for ethnic minorities wishing to preserve distinctive 36 norms while engaging in inter-ethnic interaction in domains such as education. If this mechanism 37 proves applicable in a range of other ethnographic contexts, it would constitute one cogent 38 explanation for when and why ethnically-structured cultural variation can either persist or erode 39 given frequent, and often mutually beneficial, inter-ethnic interaction. 40 41 Keywords: Ethnic boundaries, Cultural evolution, Education, Norms, Indigenous peoples, 42 Amazonia 43 44 1. INTRODUCTION 45 The extent of human cultural variation is unprecedented in nature, and its evolutionary 46 origins remain mysterious. Much human cultural variation is structured in symbolically marked 47 groups, i.e., ethnic groups, which are often associated with different suites of beliefs about what 48 constitutes appropriate behavior in a given context, i.e., cultural norms (Barth 1998[1969]; see 49 also Bicchieri 2006). Life in such groups may have characterized our species for 80,000 years or 50 more (d'Errico et al. 2009; Foley and Lahr 2011). The fact that ethnic groups are nearly universal 51 in human populations at all but the smallest geographic scales, are absent in all other primates, 52 and are highly dynamic presents important puzzles to evolutionary social scientists: What are the 53 psychological and cultural processes that make and remake ethnic groups? What role did such 54 processes play in the adaptive history of humankind? 55 56 1.1. Evolutionary Models of the Norms and Behaviors of Symbolically Marked Groups 57 One idea to explain the maintenance of ethnic groups is that assortment on symbolic 58 markers helps to preserve locally-adaptive norms, knowledge, and beliefs from erosion by 59 admixture (Richerson and Boyd 2005; Henrich and McElreath 2003). For example, McElreath et 60 al. (2003) demonstrate that symbolically marked groups with differences in norms can arise 61 spontaneously under a minimal set of assumptions: benefits to inter-individual interaction 62 depend on norm coordination, interaction partners are chosen on the basis of markers, and norms 63 and markers tend to be acquired from locally successful individuals. Once such marked groups 64 evolve, intra-group coordination interactions yield greater benefits than interaction between

3 65 groups. A similar model by Boyd and Richerson (1987) focuses instead on ecological 66 adaptations rather than norms, but also produces symbolically marked groups that protect 67 adaptive, culturally transmitted behavioral variation from erosion by mixing between residential 68 groups. Once evolved, ethnic groups in these models have salient, and relatively fixed, 69 boundaries (see also Pagel and Mace 2004), and isolating mechanisms attenuate costly inter- 70 group interaction (analogous to reproductive isolating mechanisms in biological species, e.g., 71 Price 2008). 72 A shortcoming of these models is that interaction between groups is exogenously 73 imposed, rather than a result of the strategic decisions of individuals. In these models, there are 74 no benefits to inter-group over intra-group interaction. In contrast, in nearly all ethnographic and 75 historically-described contexts, people voluntarily engage in at least some inter-ethnic interaction 76 (e.g., Wolf 1982). Such interaction sometimes entails the adoption of out-group cultural norms 77 (Nagel 1996; Kopenawa and Albert 2013; Behrens 1992) and is not necessarily viewed by 78 participants as detrimental (Rosengren 2004; Baer 2004). There may be adaptive reasons to 79 interact with other ethnic groups, and even to adopt culturally transmitted behavior from them. 80 Thus, models that exclude such benefits miss a potentially important factor affecting cultural 81 dynamics. 82 One potential resolution to this conflict between models and empirical observation is that 83 individuals strategically adjust how, and with whom, they interact, in order to either reduce costs 84 of norm mis-coordination, or, rather, to preferentially adopt adaptive behavior from members of 85 other groups. The strategic nature of between-group interaction could help explain how, in some 86 contexts, ethnic-typical norms can be maintained for generations despite frequent inter-ethnic 87 interaction (e.g., Fur and Baggara: Haaland 1998 (1969)), while, in other cases, the norms of one 88 ethnic group are rapidly replaced by those of another (e.g., assimilation of some immigrants in 89 the : Gans 1979). The lack of mechanistic models of these dynamics represents an 90 important challenge for our understanding of the evolution of cultural norms in ethnically- 91 structured populations. Ultimately we seek a body of theory that can explain the uniqueness of 92 human ethnic variation among primates, contribute to an understanding of the dynamics of 93 historical and contemporary ethnic groups, and enhance the ability of ethnic minority members 94 to reverse or slow the loss of valued cultural norms, should they wish to. 95 96 1.2. Mechanisms of Norm Adoption at Ethnic Boundaries 97 Previous research on ethnic variation has emphasized links between the adoption of out- 98 group cultural norms (e.g., assimilation and marginalization), participation in inter-ethnic 99 interaction (e.g., through social networks), individual personality traits (e.g., attitude toward the 100 out group, bicultural efficacy), and the strategic maintenance or change of ethnic identity 101 (Wimmer 2013; Berry 1997; LaFromboise et al. 1993). However, the mechanistic links between 102 inter-ethnic interaction and norm adoption are often not specified in sufficient detail to facilitate 103 empirical testing. For instance, the pioneering study of Graves (1967) among minority Spanish- 104 speaking Americans and Native Americans in a majority Anglo-American town in the 105 Southwestern U.S., found a strong relationship between inter-ethnic interaction (e.g., in domains 106 such as education, military service, out-group friendships, employment, etc.) and the probability 107 that minority individuals hold majority-typical norms. From this, it is argued that exposure to 108 outgroup norms, identification with the outgroup, and access to resources controlled by the 109 outgroup all play roles in the adoption of outgroup norms. However, the mechanisms underlying 110 such relationships are not well understood, making it difficult to distinguish among them with

4 111 the available data. As an initial step forward, we focus on five theoretically-grounded and 112 previously-published mechanisms whose predictions can be compared using quantitative and 113 ethnographic data at hand. An individual’s adoption of outgroup norms may depend on: 1) 114 differential bargaining power during inter-ethnic interactions, 2) the relative importance (e.g., 115 relative frequency) of inter-ethnic coordination interactions; 3) inter-ethnic assortment on 116 coordination norms, coupled with success-biased social learning among co-ethnics; 4) success- 117 biased inter-ethnic social learning; and 5) favorable exposure to outgroup norms during 118 childhood socialization. These five mechanisms are neither exhaustive nor mutually exclusive. 119 However, they make distinct predictions that we can evaluate using empirical data in order to 120 better understand their roles in the norm dynamics of a particular ethnically-structured 121 population. In the remainder of this section, we sketch each and provide citations to further 122 discussion. 123 124 1.2.1. Bargaining 125 Norms play an important role in inter-ethnic interactions because many such interactions 126 have the form of coordination games. In a coordination game, all participants receive a higher 127 (though not necessarily equal) payoff if they act in concordant rather than discordant manners 128 (Bicchieri 2006). Consequently, players should seek to interact with others holding similar norms 129 in the context of interaction (McElreath et al. 2003). Many domains of social life, such as 130 commerce, healthcare, education, and marriage (Nave 2000) can be modeled as coordination 131 games. Inter-ethnic interactions can be especially challenging because, initially, distributions of 132 norms often differ among ethnic groups (Barth 1998[1969]), potentially frustrating attempts at 133 coordination (e.g., 18th century Chinese-British commerce: Sahlins 1994[1988]). 134 When two individuals with different norms desire to coordinate, they must negotiate 135 about whose norm they will use for the interaction. In many contexts of social coordination, the 136 relevant norms may be deeply held and psychologically costly to change, e.g., norms of fairness, 137 child-rearing, education. Individuals may be able to coordinate using norms different from those 138 they personally identify with (e.g., cross-cultural or bicultural competence: Johnson et al. 2006; 139 LaFromboise et al. 1993), but the resulting psychological costs (e.g., cognitive dissonance: 140 Festinger 1962), among other potential costs, should motivate them to try to impose their 141 preferred norm on the coordination interaction. Following Ensminger and Knight (1997; Knight 142 1992), we define an individual’s ability to do this as bargaining power. Bargaining power is 143 contingent on, among other things, control over the resources (material or immaterial) that one 144 brings to the coordination interaction, as well as control over choice of interaction partner. An 145 individual’s ability to impose coordination norms is reduced if she has less control over (or less 146 valuable) coordination resources and if she has less choice about who she coordinates with 147 (Knight 1992; Ensminger and Knight 1997). 148 When there are benefits to inter-ethnic coordination but limited numbers of out-group 149 coordination partners, members of one ethnic group (e.g., residents) may compete among 150 themselves to interact with members of the other ethnic group (e.g., visitors). One way that 151 residents can compete for visitors is to offer to coordinate using visitor-typical, rather than 152 resident-typical, norms. Thus, in a given coordination interaction, a visitor has greater bargaining 153 power than a resident, all else being equal, as she can choose from among residents the one most 154 willing to coordinate using her preferred norm. In the terminology of economics, the visitor has 155 short-side power in a contested exchange (Bowles and Gintis 1993). In this way, where there is 156 inter-ethnic asymmetry in resource control, intra-ethnic competition among members of the

5 157 resource-favored group tends to decrease any inter-ethnic asymmetry in bargaining power 158 (Ensminger and Knight 1997). The effect of differential bargaining power on norm adoption is 159 mediated by the frequency and importance of inter-ethnic interactions. The more an individual 160 engages in subjectively important interactions using particular norms, the more likely she may be 161 to adopt and internalize those norms, changing her cognition to align with her behavior 162 (Festinger 1962; Barth 1966). Thus, we predict that the probability that an individual adopts out- 163 group norms, as opposed to maintaining her original norms, is a decreasing function of the 164 relative amount of bargaining power she has in frequent and important inter-ethnic coordination 165 interactions. 166 167 1.2.2. Interaction Frequency-biased Norm Adoption 168 Regardless of relative bargaining power, the norms of frequently-interacting ethnic 169 groups would be expected to evolve toward similarity in any domain of inter-ethnic interaction 170 (Barth 1998[1969]), as both parties then receive the benefits of coordination (e.g., market-based 171 interactions: Henrich et al. 2010). We predict that, in a given interaction context, individuals on 172 one side of the interaction should directly adopt the norm of their most frequent or most 173 important coordination partners, all else being equal. For instance, given a degree of asymmetry 174 in bargaining power across a range of interaction contexts, norm adoption by low-power 175 individuals will be most likely in those domains where interaction is most frequent or most 176 important to them. If inter-ethnic interactions are important, this mechanism will produce a 177 positive association between out-group norms and inter-ethnic interaction experience on one side 178 of the ethnic boundary. 179 180 1.2.3. Inter-ethnic Assortment on Norms 181 If there is pre-existing intra-group variation in norms, individuals may preferentially 182 assort such that those with the most out-group-typical norms engage in the most inter-ethnic 183 coordination. If there is an added benefit to inter-ethnic over intra-ethnic coordination, 184 individuals who have out-group-typical norms in the context of interaction may attain high 185 prestige. Out-group norms can then spread within a group as people copy prestigious or 186 successful co-ethnics (Barth 1966; Henrich and Gil-White 2001), with the possible result that the 187 norms of one group eventually replace those of the other. This process can result in a positive 188 association between the amount of inter-ethnic interaction and the probability of holding out- 189 group-typical norms, an empirical pattern potentially indistinguishable from that generated by 190 the previous mechanism of interaction frequency-biased norm adoption. However, if inter-ethnic 191 assortment on norms is operating, we expect the probability that an individual initially engages 192 in inter-ethnic interaction to be higher if she already holds out-group-typical norms. Without 193 assortment, initial inter-ethnic interaction is expected to be random with respect to norms. 194 195 1.2.4. Success-biased Inter-ethnic Social learning 196 An individual’s decision to adopt the norms of another ethnic group may be based on the 197 relative perceived success or prestige of individuals in the other ethnic group relative to that of 198 those in her own group. When there is a causal link between a socially-learned norm and group- 199 level success (e.g., when group benefits depend, in a non-additive manner, on the frequency of 200 the norm among its constituents), the selective imitation of successful groups can be 201 operationalized as cultural group selection (Richerson et al. 2016; Henrich 2004). This may 202 occur with or without direct interaction between individuals in the two groups, as long as the

6 203 norms of the successful or prestigious group are known to individuals in the other. Thus, 204 success-biased inter-ethnic social learning of norms need not depend on the benefits or costs to 205 individuals of inter-ethnic coordination interactions. Boyd and Richerson (2002) showed that, 206 under a range of conditions, this mechanism can result in the spread of individually-costly group- 207 beneficial norms among ethnic groups, and may explain the spread of early Christian norms of 208 mutual aid (reviewed in Richerson and Boyd 2005). However, such success-biased social 209 learning is easily generalized to any norm that people wish to copy, whether or not the norm 210 itself entails group benefits. One type of evidence for this mechanism would come from an 211 observation that the norms typical of a successful or prestigious ethnic group are directly copied 212 by individuals in other ethnic groups, regardless of those individuals’ personal inter-ethnic 213 interaction experience in the social contexts in which such norms apply. Note that the principal 214 difference between this mechanism and the previous mechanism of inter-ethnic assortment on 215 norms is that, here, the targets of success-biased social learning tend to be individuals in the out- 216 group rather than co-ethnics with out-group-typical norms. 217 218 1.2.5. Childhood Socialization 219 Socialization is the process by which children are taught the norms of the society or 220 ethnic group in which they live. Although some norms may correspond with a near-universal 221 genetic predisposition (e.g., mating aversion for very close kin: van den Berghe 1983), it is likely 222 that, on average, children are not born with a strong predisposition one way or the other toward 223 most norms responsible for ethnic variation in our species. The cultural pluripotency of young 224 children is most apparent in situations of cross-cultural fostering, where children readily adopt 225 norms very different from those of their biological parents (reviewed in Richerson and Boyd 226 2005), and in societies undergoing rapid social and economic change, i.e., where norms covary 227 with birth cohort (reviewed in Chen and French 2008). Children have evolved high receptivity to 228 learning whichever norms are used to socialize them (Legare and Nielsen 2015), as this 229 facilitates access to the benefits associated with integration into the society where they presume 230 to grow up (Chen and French 2008; Ochs and Izquierdo 2009). 231 The purpose of norms, by definition, is to restrict the range of available behavioral 232 choices in a given decision context, at times eliminating choices that might otherwise seem 233 desirable (e.g., a norm against cheating). In order to function, a norm must be reasonably 234 inflexible given contextual variation in the behavior to which it applies (e.g., cheating on an 235 exam versus cheating in sports). However, this same quality of inflexibility may make norms, 236 once learned, relatively resistant to change later in life, even when there may be benefits to doing 237 so (e.g., norms of circumcision learned prior to immigration into a non-circumcising host 238 society: Morison et al. 2004). For this reason, a child socialized using an in-group norm may 239 become progressively more resistant to subsequent adoption of a conflicting out-group norm. It 240 may also be the case that children pass through a developmental window (perhaps prior to 241 adolescence) in which they are cognitively more receptive to the adoption of cultural norms 242 (Cheung et al. 2011; but see Chudek et al. 2015). If either norm inflexibility or a developmental 243 norm-adoption window occur (or if both occur), we predict that, on average over a broad range 244 of ages, the earlier in the socialization process (e.g., the younger) an individual is favorably 245 exposed to out-group norms through inter-ethnic interaction, the more likely she is to adopt 246 norms of the other ethnic group. 247 248 1.3. Overview of the Present Study

7 249 Incorporating the mechanisms described above, we develop and evaluate hypotheses for 250 why inter-ethnic interaction variably results in persistence or erosion of cultural norm 251 differences. We do so using ethnographic data on inter-ethnic interaction and norm adoption 252 decisions in a relatively remote population in Amazonian Peru, where a salient ethnic boundary 253 exists between indigenous Matsigenka and neighboring colonist Mestizos. Matsigenka live in a 254 community legally protected from incursion by Mestizos, and with very little exposure to mass 255 media. Matsigenka-Mestizo interaction is almost entirely constrained to the domains of 256 commerce, wage labor, and education, and such inter-ethnic interaction nearly always requires 257 Matsigenka to leave their community and travel to Mestizo towns. Hence, the circumstances of 258 inter-ethnic interaction are highly asymmetric and limited to specific memorable contexts. This 259 allows us to abstract away from the full complexity of inter-ethnic frontiers and collect 260 individual-level data on both norms and histories of inter-ethnic interaction that address 261 hypothetical mechanisms for the persistence or erosion of norm differences. 262 We find that Matsigenka who interact voluntarily, and even frequently, with Mestizos in 263 the contexts of labor and commerce show little evidence of change towards Mestizo-typical 264 norms, while those who attended Mestizo schools are more likely to hold Mestizo-typical norms 265 as adults. A number of mechanisms likely contribute to this empirical pattern. However, using 266 ethnographic data we argue that, in particular, the differential bargaining power of each ethnic 267 group in each domain of interaction provides both a theoretically cogent and empirically accurate 268 explanation that extends existing evolutionary models of the cultural dynamics of ethnic groups 269 in human societies. Furthermore, we suggest that changes in bargaining power in the context of 270 inter-cultural education may contribute to the maintenance of cultural norms valued by ethnic 271 minorities. 272 This paper employs multi-level Bayesian item-response models to analyze patterns of 273 individual-level variation in cultural norms. This framework is consistent with, but meaningful 274 advances, quantitative anthropological approaches to norm variation (e.g., cultural consensus and 275 consonance: Romney et al. 1986; Dressler et al. 2005; Oravecz et al. 2014; Oravecz et al. 2015). 276 These models treat norm variation as directly unobservable, but nevertheless inferable from 277 patterns of behavior (e.g., responses to interview questions), and make no assumptions about the 278 correctness or incorrectness of any norm or behavior. We provide a script sufficient to repeat and 279 extend our analysis, so that other researchers can engage with these models and adapt them to 280 other contexts. 281 282 2. METHODS 283 2.1. Study Population 284 The study was conducted among residents of the Matsigenka Native Community of 285 Tayakome (adult population: 79), located inside Manu National Park, in the Department of 286 Madre de Dios, in the lowland Amazonian region of Southeastern Peru, and in the Mestizo towns 287 of Boca Manu (adult population: ~80) and Atalaya (adult population: ~65), located just outside 288 the boundary of the park, in the Departments of Madre de Dios and Cusco, respectively (Figure 289 1). 290

8

291 292 Figure 1 Map of the Matsigenka study community of Tayakome, and the Mestizo study 293 communities of Boca Manu and Atalaya, as well as the locations of other nearby settlements in 294 and around Manu National Park (in grey), Peru 295 296 Tayakome residents practice swidden horticulture, fishing, and hunting. Most Matsigenka 297 understand basic Spanish but prefer to interact in their own language. Some travel outside the 298 park and live among Mestizos for weeks to months working as wage-laborers (32%), or to attend 299 Mestizo-run schools (20%), e.g., boarding secondary schools. Shorter excursions are 300 occasionally made to purchase items (e.g., pots, flashlights) in Mestizo general stores. There is a 301 government health post in Tayakome, though most adult Matsigenka interact little with the 302 Mestizo technician. The primary school in Tayakome is staffed by externally-trained Matsigenka 303 teachers. Nearly all inter-ethnic interaction with Mestizos occurs in the three domains of 304 commerce, wage labor, and education, and takes place in Mestizo towns. There is very little 305 Matsigenka-Mestizo inter-marriage or domestic interaction. Most Mestizos are colonists from 306 the Andean highlands who usually interact with each other in Spanish. They are not permitted to 307 enter the national park without a government permit. Residents of Boca Manu and Atalaya tend 308 small general stores and restaurants, build wooden boats, grow plantains for market, log the 309 surrounding forest, and/or work in tourism for the park. Further details of the study communities 310 can be found in Online Resource Appendix A.1., as well as in Shepard et al. (2010) and Llosa 311 Isenrich and Nieto Degregori (2003). 312 313 2.2. Data Collection

9 314 JAB lived in Boca Manu for approximately three months (September and November 315 2012, January 2014), Atalaya for two months (December 2012 and February 2014), and 316 Tayakome for 13 months (January-December 2013, March 2014). Over several rounds of semi- 317 structured interviews, he recorded interviewees’ life history and recollections of personal intra- 318 and inter-ethnic interaction experience, emphasizing the domains of commerce, wage-labor, and 319 education. He then designed a set of fourteen vignette questions (Table 1) for the purpose of 320 learning about specific norms in eight contexts of social coordination (commerce, wage labor, 321 education, spousal relations, parent-offspring relations, inheritance, healthcare, and religion), and 322 administered these questions privately to 74 (94%) residents of Tayakome (including the Mestizo 323 health technician), 45 (56%) residents of Boca Manu, and 42 (65%) residents of Atalaya, all of 324 whom had been previously interviewed regarding life history and inter-ethnic interaction 325 experience. Further description of data collection methodology, as well as translations of vignette 326 questions are provided in Online Resource Appendix A.2. 327 328 Table 1 Vignette questions administered in this study, and their respective social contexts. The 329 response corresponding to a meta-norm for practical inter-dependence is indicated in column 330 four. An alternative response corresponds to a meta-norm for respectful autonomy. The number 331 of Matsigenka and Mestizo interviewees answering each question is indicated in column five. A 332 parenthetical “A” indicates that the question was only asked of Mestizos in Atalaya, not Boca 333 Manu. Further explanation and translations of these questions are provided in Online Resource 334 Appendix A.2. Number Social Question Practical inter- N (Matsi, Mest) Context dependence 1 Spousal There is a married couple with no children. Okay 71, 80 Relations The woman hunts and fishes (Mestizo: has a job and makes more money). The man stays home and cooks, weaves (Mestizo: cleans), and washes clothes. Is this okay or not okay? 2 Parent- After school, a ten-year-old daughter Okay 77, 78 Offspring cannot go to friend's house to play because Relations she has to care for her two-year-old brother until their parents come home at night. Is this okay? 3 Inheritance A man always wears his favorite hat. After Not okay 74, 79 he dies, his son takes the hat and wears it. When he wears it he remembers his father. Is this okay? 4 Education A teacher hits students when they don't Okay 75, 74 learn. Is this okay? 5 Education A student pays attention to the teacher and Okay 79, 81 never asks any questions. Is this okay? 6 Healthcare If you get a respiratory illness (influenza), Health post 79, 81 do you first go to the health post, first use home remedies, or first go to a shaman or curandero? 7 Inheritance An old woman has two new pots and two Both pots to the 76, 82 adult daughters. One daughter has her own daughter who has two pots, but wants her mother’s pots. The none other daughter has no pots, and also wants

10 her mother’s pots. When the mother dies, who should inherit the pots? Illustrated with a diagram. Options: one pot to each daughter, both pots to the daughter who has none. 8 Religion A good person does not want to be Hell or 64, 60 baptized. Where does his or her soul go somewhere else when they die? Options: up (heaven), down (hell), somewhere else. 9 Religion A bad person is baptized. Where does her Heaven 64, 29 (A) or his soul go when they die? Options: up (heaven), down (hell), somewhere else. 10 Wage Labor A man is hired to prepare an agricultural Okay 75, 77 field. He stops work at noon in order to go visit a friend. He returns the next day to finish the job. Is this okay? 11 Commerce There are two stores. One is cheap with a Cheap store with 63, 77 mean owner. The other is expensive with a mean owner nice owner. Where would you buy? 12 Spousal A man wants to marry. His mother is the Okay 70, 30 (A) Relations sister of the woman's father. Is it okay for him to marry this woman? Illustrated with a diagram. For the Matsigenka, examples were provided of potential marriages between people known to the interviewee. 13 Parent- Parents want their daughter to marry a Obey parents 63, 81 Offspring certain man that she does not like. She Relations wants to marry someone else. Should she obey her parents and marry him anyway or not? 14 Wage Labor A man is hired to work two days: Monday Okay to go and 67, 80 and Tuesday. Monday night there is a become drunka party (Matsigenka: hosted by a Matsigenka). Should he go and get drunk? (Matsigenka: He goes and gets so drunk that he can’t work on Tuesday. Is this ok?) 335 aSee Online Resource Appendix A.2. for an explanation of the cultural context of drunkenness in 336 Matsigenka society, which will differ from that of most readers. 337 338 2.3. Statistical Analysis 339 The norms measured by the vignette questions may co-vary, such that knowing how an 340 individual answered one question gives you information about how she answered another 341 question, and, in the ideal case, about how she answered all of the other questions. If true, then 342 people’s responses to the fourteen vignettes can be represented by a smaller number of latent 343 dimensions, and, ideally, by a single latent dimension. We use Item Response Theory (IRT) 344 models (Schacht and Grote 2015; Bafumi et al. 2005; Jackman 2001; Fox 2010) in a Bayesian 345 framework (McElreath 2016) to show that, for this study, the vignette responses of each 346 interviewee are well represented by a single dimension. This latent dimension constitutes a 347 convenient way to compare individuals on the basis of all fourteen measured norms 348 simultaneously. It does not necessarily represent a unitary over-arching belief held by actual 349 people (e.g., a meta-norm). For instance, it may be that the fourteen measured norms are

11 350 functionally independent, but happen to covary within this sample of people. However, for ease 351 of exposition below, we will refer to the latent dimension as representing a meta-norm. We 352 interpret the negative pole of the dimensional axis as a meta-norm prioritizing respectful 353 autonomy, while the positive pole represents a meta-norm prioritizing practical inter- 354 dependence. See Online Resource Appendix A.2. for the relation between individual vignette 355 questions and interpretation of the latent dimension. Note, however, that the conclusions below 356 do not depend on interpretation of this constructed dimension. The location on the latent axis of 357 individual (and thus a continuous measure of ’s meta-norm) is represented by the parameter 358 . We model as a linear function of a population-mean intercept (), an individual-level 359 random effect ([], zero-centered offset from for each individual ), and various 360 combinations of hypothesized predictors, including ethnicity, and inter-ethnic commerce, wage 361 labor, and education experience: 362 363 = + [] + [] … , for =1,…,

364 where J is the number of interviewees. An example fixed effect predictor, [], is the product 365 of the coefficient for ethnicity and the binary ethnicity indicator for individual . 366 To construct an IRT model, we follow Bafumi et al. (2005) by embedding the linear 367 function within a logistic function. This allows us to simultaneously evaluate properties of 368 each individual and each vignette question with respect to the latent dimension. The probability 369 that the response y of a particular interviewee j to a particular vignette question k is the practical 370 inter-dependence response (column four of Table 1), Pr = , is given by a 371 logistic function (inverse logit) ranging between zero and one: 372 373 Pr =1 = logit [( − )] 374 where practical inter-dependence and respectful autonomy responses are represented by 1 and 0, 375 respectively. The domain (x axis) of this logistic function is the latent dimension. The slope at 376 the function’s inflection point, , is the degree to which an affirmative versus negative response 377 to question k discriminates among individuals holding the meta-norm for respectful autonomy 378 versus practical inter-dependence. The location of the inflection point on the latent axis, , is 379 the degree to which a person must hold the meta-norm for practical inter-dependence in order for 380 the model to predict that she give the inter-dependence-associated response to question k. See 381 Online Resource Appendix Figure B1 for illustrations. 382 To check the robustness of results to the effects of different predictors, we fit a series of 383 19 models varying in the fixed effect predictors included in the linear function for . Parameter 384 estimation for each model was accomplished with RStan (Stan Development Team 2016), 385 running four Hamiltonian Monte Carlo chains in parallel until convergence was suggested by a 386 high effective number of samples (> 500) and estimates of 1.00 (McElreath 2016). This 387 entailed 4000 samples per chain, half of which were warm-up. We compared model fit with 388 WAIC (McElreath 2016). All data and statistical analysis scripts in (R Core Team 2014) 389 implementing RStan are available from Github (https://github.com/jabunce/bunce-mcelreath- 390 HN-2016-matsigenka-norms). Further explanation and priors for model parameters are provided 391 in Online Resource Appendix A.3. 392

12 393 3. RESULTS 394 3.1. Associations between Norms and Inter-ethnic Experience 395 Figure 2 shows that, for all fourteen vignette questions, a larger proportion of Matsigenka 396 than Mestizos gave responses corresponding to practical inter-dependence. These raw 397 proportions suggest an overall ethnic difference in the distributions of the norms applied by the 398 interviewees to answer the questions, and demonstrate the utility of the interview instrument to 399 distinguish between these two ethnic groups. See Online Resource Appendix C.1. for further 400 discussion of this result. In all nineteen IRT models, posterior estimates of the discrimination 401 parameters () for all vignette questions are non-zero (Online Resource Appendix Figures B1 402 and B2). This indicates that all questions can reasonably distinguish among individuals along a 403 single latent axis (Jackman 2001) which, as described above, we interpret as representing an 404 individual’s meta-norm for practical inter-dependence versus respectful autonomy. 405

406 407 Figure 2 Proportions of Matsigenka (n=79) and Mestizo (n=82) interviewees giving the practical 408 inter-dependence response to the fourteen vignette questions in Table 1. The diagonal is the line 409 of equal proportions between Matsigenka and Mestizos. The vertical (or horizontal) distance 410 from a point to the diagonal is the difference in proportion between ethnic groups. Example 411 differences for questions 5 and 11 are given. Note that, for all questions, a larger proportion of 412 Matsigenka than Mestizos gave inter-dependence responses, i.e., all points fall above the 413 diagonal 414 415 Next we turn to individual norm differences among Matsigenka, and examine the extent 416 to which such differences are predicted by inter-ethnic interaction experience. Figure 3 plots the 417 locations of all 161 interviewees on the latent (meta-norm) axis. As expected, there is marked 418 separation by ethnicity, with Matsigenka tending toward the positive pole for practical inter-

13 419 dependence and Mestizos tending toward the negative pole for respectful autonomy. However, it 420 is apparent that a number of Matsigenka individuals have axis locations very similar to those of 421 average Mestizos, suggesting that they hold the Mestizo-typical meta-norm for respectful 422 autonomy. The right three columns of Figure 3 demonstrate that these Matsigenka with the 423 Mestizo-typical norm tend to have inter-ethnic commerce, wage labor, and education experience. 424 In contrast, Matsigenka with the more Matsigenka-typical norm for practical inter-dependence 425 may have inter-ethnic commerce and wage labor experience, but almost none have education 426 experience with Mestizos. 427

428 429 Figure 3 Mean posterior probability estimates for the locations () of all 161 interviewees on 430 the latent axis, as predicted by an IRT model with a random effect for individual and no 431 predictors (model m1 in Online Resource Appendix Table B1). Plotted in columns from left to 432 right are: all Mestizos, all Matsigenka, only Matsigenka who have commerce experience with 433 Mestizos (w/ Com), only Matsigenka who have wage labor experience with Mestizos (w/ Lab), 434 and only Matsigenka who have education experience with Mestizos (w/ Edu). Horizontal lines 435 are drawn at the mean of each column. Within columns, points are jittered on the x-axis for 436 clarity 437 438 The same effect is seen in Figure 4, plotting counterfactual contrasts of posterior 439 predictions from the best-fitting model (m19 in Online Resource Appendix Table B1, with 33% 440 of model weight). The model predicts that Matsigenka who have never engaged in commerce, 441 wage labor, or education interactions with Mestizos have the Matsigenka-typical meta-norm of 442 practical inter-dependence, which differs from the Mestizo-typical meta-norm of respectful 443 autonomy (Figure 4A). Similarly, Matsigenka with commerce and wage labor experience tend to 444 hold the norm of practical inter-dependence like those of other Matsigenka (Figure 4B), but 445 unlike the norm of most Mestizos (Figure 4C). In contrast, Matsigenka who went to school with 446 Mestizos tend to hold the norm of respectful autonomy like most Mestizos (Figure 4C), and

14 447 unlike the norm typical of their fellow Matsigenka (Figure 4B) (further details of analysis in 448 Online Resource Appendix B). 449 An exploratory model including sex and categorical age predictors achieved 7% of model 450 weight, although the coefficients of each of these predictors could not be distinguished from zero 451 (m12 in Online Resource Appendix Table B1). Furthermore, coefficient estimates for ethnicity 452 and the inter-ethnic experience predictors were not distinguishable from those of the best-fitting 453 model, above. This gives us confidence that inclusion of sex and age in the model has little effect 454 on the results. Coefficient estimates for all models are provided in Online Resource Appendix 455 Table B1. 456

457 458 459 Figure 4 Predicted contrasts on the latent axis for groups defined by ethnicity and interaction 460 experiences with Mestizos, based on the best-fitting IRT model (m19 in Online Resource 461 Appendix Table B1). 90% highest probability density intervals (HPDI) are shown in grey. Panel 462 A: Predicted difference in location between an average Matsigenka without any inter-ethnic 463 interaction experience (Matsi w/o exp) and an average Mestizo with only commerce experience 464 with other Mestizos (Mest w/ com) (posterior mean contrast 2.52, 90% HPDI: [1.09, 3.98]), an 465 average Matsigenka without inter-ethnic experience and an average Mestizo with only labor 466 experience with other Mestizos (Mest w/ lab) (2.34, [0.93, 3.65]), and an average Matsigenka 467 without inter-ethnic experience and an average Mestizo with only education experience with 468 other Mestizos (Mest w/ edu) (3.66, [2.22, 5.14]). Here and below, probability distributions that

15 469 do not strongly overlap zero indicate a detectable difference in location, i.e., in meta-norm. Panel 470 B: Predicted difference in location between an average Matsigenka without and with only inter- 471 ethnic commerce (0.74, [-0.05, 1.46]), or wage labor (0.36, [-0.36, 1.09]), or education (3.04, 472 [2.1, 4.06]) experience. Panel C: Predicted difference in location between an average Matsigenka 473 with only inter-ethnic commerce, wage labor, or education experience, and an average Mestizo 474 with each type of experience only among other Mestizos (respectively, 1.78, [0.49, 3.08]; 1.98, 475 [0.61, 3.33]; 0.62, [-0.73, 1.99]) 476 477 478 3.2. Bargaining 479 There is ethnographic evidence that Matsigenka have less bargaining power in the 480 domain of inter-ethnic education than they do in inter-ethnic wage labor and commerce. This 481 coincides with the result above that Matsigenka with inter-ethnic education experience are more 482 likely to hold Mestizo-typical norms than are Matsigenka with inter-ethnic labor and commerce 483 experience. In the context of a strict boarding school environment, Matsigenka children usually 484 do not control their presence or absence in classrooms, nor the time they must allocate to 485 studying, which constitute their contributions to coordination in the domain of education. 486 Additionally, students usually cannot choose their coordination partners, i.e., the Mestizo 487 teachers and care-takers with whom they interact. By leveraging their physical strength and 488 institutional authority to adjust punishments and incentives, teachers can thus compel students to 489 coordinate using Mestizo norms, which many students likely adopt and internalize. Mestizo-run 490 secondary schools are attractive to many Matsigenka parents and children, and generally 491 function near maximum capacity. Thus, competition among Mestizo-run schools for Matsigenka 492 students is largely absent. 493 In contrast, for both wage labor and commerce, Matsigenka control valued resources for 494 coordination, i.e., their labor and their money, respectively. Because laborers and customers are 495 usually in demand, Mestizos compete among themselves to interact with Matsigenka, who can 496 often choose to work for or buy from those Mestizos most willing to coordinate using 497 Matsigenka-typical norms. For instance, Matsigenka wage laborers usually have the option of 498 changing employers if Mestizos try to impose Mestizo-typical norms either inside or outside of 499 the context of labor. As a Mestizo from Atalaya who routinely contracted Matsigenka field hands 500 stated, 501 “[Matsigenka] work well, but I let them work in their own way, because their world is 502 different. A type of person like us [Mestizos] already knows how this kind of work is. 503 [We] work until late, [and are] more demanding. But a Machi, when you bring him [to 504 the agricultural field], you let him work in his own way. If he wants to leave, then we 505 leave. If he wants to go for a little while because he is tired, I let him . . . If I say 506 something to them, like demanding that they do something, they get angry and they leave 507 you, just like that. They’re not even interested in the money.” 508 509 Thus, compared to education, there may be less incentive for Matsigenka to adopt Mestizo- 510 typical norms during interaction in domains such as wage labor and commerce, where 511 Matsigenka have greater bargaining power (additional examples in Online Resource Appendix 512 C.2.). 513 514 3.3. Interaction Frequency-biased Norm Adoption

16 515 It seems likely that Matsigenka directly adopt Mestizo-typical norms as a result of inter- 516 ethnic coordination interactions. For instance, describing Matsigenka boarding school students in 517 Boca Manu, who also engage extensively in commerce and wage labor on weekends, one 518 Mestizo resident stated, 519 “When they first come, they have many problems, and they are used to not working. That 520 is one of their customs. And when they come here they have to change their way of 521 thinking. They have to see how to get ahead. It’s a struggle for them. He who does the 522 most, is the most ambitious, obtains the best things. They see that among themselves, and 523 they change.” 524 There is evidence of a positive relationship between the frequency of inter-ethnic 525 interaction and the probability of holding out-group norms. Compared to Matsigenka with inter- 526 ethnic wage labor and commerce experience, Matsigenka who went to school with Mestizos tend 527 to have the most Mestizo-like norms and also the most lifetime inter-ethnic interaction 528 experience. For instance, of the 16 Matsigenka who attended Mestizo-run schools, 14 also have 529 inter-ethnic wage labor experience, which is often quite extensive. Thus, Matsigenka with inter- 530 ethnic education experience tend to have more years of interaction experience with Mestizos in 531 the combined domains of education and labor (~12 years, on average) than do Matsigenka wage 532 laborers who did not attend a Mestizo-run school (~ 4 years, on average). It is thus likely that the 533 total amount of interaction experience with Mestizos, regardless of domain, is positively 534 associated with Mestizo-like norms among Matsigenka. 535 However, it is also likely that education has a unique effect on out-group norm adoption, 536 independent of the frequency or amount of time spent in interaction. Two Matsigenka 537 interviewees did not go to Mestizo-run schools but have approximately 15 years of wage labor 538 experience, living and working for more than half of each year with Mestizos outside of 539 Tayakome. Despite this extensive inter-ethnic experience, these individuals tend to have more 540 Matsigenka-typical norms (larger ) than do the nine Matsigenka who each have combined 541 education and labor experience with Mestizos totaling approximately 10 to 15 years (Online 542 Resource Appendix Figure C1). This anecdotal evidence suggests that, on average, accounting 543 for the total amount of inter-ethnic experience, inter-ethnic education is more strongly associated 544 with Mestizo-like norms than is wage labor alone. 545 546 3.4. Inter-ethnic Assortment on Norms 547 There is ethnographic evidence for a moderate degree of assortment on norms in inter- 548 ethnic interaction. Matsigenka children with the most Mestizo-like norms may be preferentially 549 selected for, and retained in, Mestizo boarding schools. For instance, two of the three Mestizo 550 boarding secondary-schools currently attended by Tayakome children require matriculation 551 interviews, including a cursory evaluation of norms, e.g., readiness to shake hands and return a 552 greeting in Spanish. Additionally, several Matsigenka students have been expelled from these 553 two schools for failure to conform to Mestizo norms of acceptable students, e.g., obedient and 554 nulliparous. In contrast, one of the three boarding schools attended by Tayakome students does 555 not require matriculation interviews and has not, to our knowledge, expelled students. Thus, at 556 this school, initial assortment on norms is not enforced from the Mestizo side of the inter-ethnic 557 interaction. Similarly, in informal conversations in Tayakome, several Matsigenka with very 558 Matsigenka-typical norms (high ) expressed a strong desire, and detailed future plans, to 559 engage in wage labor interactions with Mestizos (Online Resource Appendix Figure C1). Thus,

17 560 assortment on norms may occur in some contexts of inter-ethnic education, but it may be weak 561 or absent for many other Matsigenka-Mestizo interactions. 562 In order for inter-ethnic assortment on norms to affect norm dynamics, the norms of 563 Matsigenka with Mestizo-typical norms should be preferentially copied by their fellow 564 Matsigenka (i.e., success- or prestige-biased intra-ethnic social learning). Although we did not 565 explicitly investigate such a preference in the present study, we note its plausibility. For 566 example, the mother of a Mestizo-educated Matsigenka man in Tayakome told JAB that she 567 stopped wearing a traditional Matsigenka nose ring (koriki) at the insistence of her son, whom 568 she esteems, and who had apparently adopted a Mestizo norm for the inappropriateness of such 569 ornaments. 570 571 3.5. Success-biased Inter-ethnic Social Learning 572 As shown above, the probability of holding Mestizo-typical norms is contingent on the 573 type of inter-ethnic interaction experience of Matsigenka individuals (i.e., commerce or labor 574 versus education). This is contrary to the expectation under success-biased social learning, 575 assuming Matsigenka regard Mestizos as successful or prestigious, and assuming all Matsigenka 576 regardless of inter-ethnic experience are aware of Mestizo-typical norms. Although Mestizos do 577 control access to desired manufactured goods, JAB’s ethnographic observations suggests that 578 most Matsigenka may not generally regard the Mestizos of Boca Manu and Atalaya as more 579 prestigious than they are. For instance, a Matsigenka with extensive experience working on a 580 Mestizo-owned tourist boat in Atalaya viewed Mestizos as inferior workers, stating, 581 “Mestizos are a little lazy. Matsigenka surpass them in work. As crew members, 582 sometimes [the boss] tells us to wash the boat. The Mestizo doesn’t wash; he just stays 583 watching. The Matsigenka washes everything.” 584 If inter-ethnic differences in prestige or perceived success are small, success-biased social 585 learning will be weak. On the other hand, it may also be the case that many of the norms 586 represented by the vignette questions are difficult for Matsigenka to observe, or hear about, 587 without extensive experience living among Mestizos. However, once a Matsigenka learns of 588 such norms, she may adopt them via success-biased social learning even if she never employs 589 such norms in her personal interactions with Mestizos. There is some limited evidence for this 590 type of learning. Among the twenty Matsigenka with four or more years of combined inter- 591 ethnic education and wage labor experience, approximately half held a Mestizo-typical 592 (respectful autonomy) norm concerning inheritance (Question 7, 11/20 interviewees), compared 593 to only 15% (9/59) among Matsigenka with less extensive inter-ethnic experience. This occurred 594 despite the fact that only four Matsigenka interviewees have the opportunity to coordinate with 595 Mestizos in the context of inheritance, i.e., two have children with a Mestizo, and two have one 596 Mestizo parent. Thus, success-biased social learning may occur to a limited extent among 597 Matsigenka with extensive inter-ethnic interaction experience. 598 599 3.6. Childhood Socialization 600 There is an association between early age of inter-ethnic socialization and the probability 601 of holding out-group typical norms. Matsigenka interviewees who engaged in inter-ethnic 602 interaction at the earliest ages were those who attended Mestizo-run schools. These included five 603 students who attended just inter-ethnic primary school (when aged approximately 6-12), five 604 who just attended inter-ethnic secondary school (when aged approximately 13-17) or beyond, 605 and six who attended both inter-ethnic primary and secondary schools. As shown above, for

18 606 Matsigenka, inter-ethnic education has the strongest association with Mestizo-typical norms. 607 Socialization of children to a broad range of Mestizo norms, including those outside of 608 academics, appears to be one goal of Mestizo-run education. For instance, a Mestizo staff 609 member at one boarding school attended by Tayakome children explained: 610 “In the boarding school there are rules … [What] we’re doing here is simply the way to 611 put away dishes, [to] put things in order, [to go about] cleaning, [to] have a clean room, 612 [to] have everything in order, [to] live with beauty and honor.” 613 No Matsigenka engaged in inter-ethnic wage labor or commerce as a young child without 614 also attending Mestizo schools. Thus, in this sample, the effect of early age of out-group 615 socialization on norm adoption cannot be distinguished from that of other aspects of inter-ethnic 616 education experience. 617 618 4. DISCUSSION 619 Our objective was to investigate the mechanisms by which individuals adopt out-group 620 norms, potentially leading to cultural dynamics in ethnically-structured populations. We have 621 shown quantitatively that indigenous Matsigenka who attended Mestizo-run schools tend to hold 622 more Mestizo-typical norms across a range of social contexts, compared to Matsigenka with only 623 wage labor and commerce experience with Mestizos. Cross-sectional data of this type cannot be 624 used directly as evidence that inter-ethnic interaction, such as education, causes cultural change 625 (see also Graves 1967). However, we have leveraged our ethnographic observations to show that 626 the observed pattern is likely due to all five of the mechanisms we identified as potentially 627 influencing individual adoption of out-group cultural norms: bargaining, interaction frequency- 628 biased norm adoption, inter-ethnic assortment on norms, success-biased inter-ethnic social 629 learning, and childhood socialization. Below we argue that, of these mechanisms, bargaining is 630 particularly important in this ethnographic context, and may additionally explain why inter- 631 ethnic education can be such a potent driver of cultural dynamics in other ethnic groups engaging 632 with colonial powers. However, first we address limitations and strengths of our methodology 633 and analysis. 634 635 4.1. Limitations and Strengths of the Study 636 This study systematically examined norms, but not the behaviors to which they apply. 637 Norms and behavior may correspond (e.g., Atran et al. 2002), though they need not. Of particular 638 importance are cases of bicultural competence (LaFromboise et al. 1993), where an individual 639 may modify her behavior according to the norms of the ethnic group in which she finds herself. 640 In this study we assume that each individual holds a single norm with respect to each vignette 641 question (or a single meta-norm applicable to each question), and that this norm is reflected in 642 each response. Under this assumption, if the behavior of a biculturally-competent individual 643 changes with ethnic context, it may therefore periodically deviate from the norms that she 644 personally holds. It is possible that, rather than reflecting individually-held norms, vignette 645 responses of a biculturally-competent individual reflect behavioral decisions corresponding to 646 the ethnic context that she imagines while listening to each vignette (which are mostly designed 647 to be neutral with regard to Matsigenka and Mestizo ethnic context). This study is limited by the 648 assumption that any such imagined context is one in which the subjectively appropriate behavior 649 corresponds to the norms that the interviewee personally holds. 650 We employed Bayesian item-response models to analyze individual-level responses to a 651 range of questions about norms. Although similar in underlying mathematical form to the

19 652 widely-used cultural consensus model framework (Romney et al. 1986), IRT models are more 653 flexible, and, most importantly, facilitate direct estimation of the effects of predictors on latent 654 outcome variables in hierarchically-structured data (Schacht and Grote 2015; Oravecz et al. 655 2014). An important goal of this study is to assess between-group variation and explain within- 656 group variation in a variety of norms, apparent in Figure 2. The IRT models allow us to represent 657 this norm variation in a lower-dimensional space, where effects of the hypothesized predictor 658 variables can be meaningfully estimated and understood. We expect this analysis strategy to 659 prove particularly useful for anthropological interview data in any context where the researcher 660 employs a battery of questions to investigate and explain individual-level variation in cognitive 661 states (e.g., norms, preferences, beliefs) that cannot be directly observed. We encourage other 662 researchers to explore our analysis scripts (https://github.com/jabunce/bunce-mcelreath-HN- 663 2016-matsigenka-norms) and modify them to suit their needs. 664 665 4.2. Bargaining Power Interacts with Other Mechanisms 666 The low bargaining power of Matsigenka children in Mestizo-run schools interacts with 667 other identified mechanisms of norm adoption. For instance, the fact that Mestizo-educated 668 Matsigenka have more lifetime inter-ethnic interactions than Matsigenka without such 669 experience, may contribute to their adoption of Mestizo-typical norms over the course of their 670 lives. However, it also seems likely that Mestizo-typical norms learned in school facilitate 671 subsequent inter-ethnic coordination in domains such as commerce and wage labor. After 672 graduating, Mestizo-educated Matsigenka may self-assort into commerce and wage labor 673 coordination interactions with Mestizos, domains where subsequent norm adoption may occur, 674 though at comparatively lower rates due to the increased bargaining power of Matsigenka. Thus, 675 adoption of Mestizo-typical norms by Matsigenka students with low bargaining power may 676 facilitate subsequent interaction frequency-biased norm adoption. 677 The low bargaining power of Matsigenka students may cause feelings of inferiority. In a 678 boarding school environment, Matsigenka children may view Mestizo teachers and peers as 679 more prestigious and successful, thereby facilitating success-biased social learning of Mestizo 680 norms unrelated to those that Matsigenka actually use to interact with Mestizos. In contrast, in 681 domains like wage labor, the higher bargaining power of Matsigenka may reduce feelings of 682 inferiority and thereby reduce the success-biased adoption of such norms. Thus, bargaining 683 power may interact with success-biased inter-ethnic social learning. 684 Finally, the young age of Matsigenka students may contribute to their susceptibility to 685 adoption of Mestizo-typical norms. However, consistent with JAB’s observations, Ochs and 686 Izquierdo (2009) argue that Matsigenka children as young as six years old have already learned 687 important Matsigenka-typical norms, such as sharing, hard work, and intra-group harmony. 688 Simple exposure to alternative Mestizo-typical norms would not necessarily induce these 689 children to change the norms they have already learned. Thus, in addition to early exposure, the 690 potent effect of inter-ethnic education on norms is likely also due to the fact that children are 691 easier to physically manage and strategically isolate from elder co-ethnics (e.g., in a boarding 692 school), such that their control over coordination resources and choice of adult coordination 693 partners is constrained, thereby reducing bargaining power. In summary, the low bargaining 694 power of Matsigenka students likely plays an important role in norm adoption, despite the fact 695 that other identified mechanisms may operate concurrently. 696 697 4.3. Education, Bargaining Power, and Cultural Change

20 698 The potency of education as a driver of cultural change has long been recognized and 699 exploited by colonial powers attempting to eradicate the cultural norms of (i.e., “civilize”, 700 “develop”) indigenous peoples, e.g., Native South Americans (Aikman 2003), Native North 701 Americans (Parliament of Canada 1985; Trennert 1979), Indians under British colonial rule (Seth 702 2007). The contemporary movement for inter-cultural bilingual education attempts to combat 703 this legacy. One strategy is to introduce non-Western students to desired Western academic 704 knowledge using the pedagogical norms of the students’ own society, and require local 705 community oversight of, and participation in, students’ education (Aikman 2003; Trapnell 2003). 706 Only local communities themselves can judge the appropriateness of such an educational 707 program. However, we note that several elements of this approach may, in effect, increase the 708 relative bargaining power of indigenous students and parents in educational coordination with 709 out-group teachers, e.g., community choice of teachers (hiring and firing power), students learn 710 from both co-ethnics and out-group members, attendance is voluntary. Thus, if such a program is 711 effective in transcending the historical paradigm of colonial education by reducing the loss of 712 valued ethnic-typical cultural norms, an increase in relative bargaining power may be an 713 important part of the explanation. 714 We argue that differential bargaining power is likely to be one powerful driver of cultural 715 dynamics in contexts of inter-ethnic interaction, and is at least partially responsible for the 716 pattern among Matsigenka that out-group cultural norms strongly co-vary with inter-ethnic 717 education experience. Bargaining power can vary by domain of inter-ethnic interaction (e.g., 718 education versus wage labor), affecting the costs and benefits accruing to individuals of such 719 interactions. Consequently, even when norm distributions differ across an ethnic boundary, as 720 they do here, inter-ethnic interaction does not invariably entail mis-coordination costs for all 721 participants (contra McElreath et al. 2003), and does not always lead to the erosion of ethnic- 722 typical cultural variation. Similarly, although ethnic communities may have compelling reasons 723 to curtail it, the adoption of out-group norms, when it does occur, should not be viewed as 724 invariably costly. Rather, to advance theory of cultural change in ethnically-structured 725 populations, inter-ethnic norm adoption decisions can be more productively investigated as 726 resulting from the interaction of several mechanisms, including strategic choice of benefit- 727 seeking individuals in contexts of low bargaining power. Future work in this and other 728 ethnographic settings should incorporate longitudinal data on individual norm development, and 729 look for situations affording natural experimental control over inter-ethnic interaction 730 experience, in order to more effectively distinguish among the predictions of the five 731 mechanisms of norm adoption identified here. 732 733 ACKNOWLEDGMENTS 734 We thank the residents of Tayakome, Boca Manu, and Atalaya for their participation. For 735 help with fieldwork and data analysis we thank Caissa Revilla, Cecilio Huamantupa and family, 736 Rosa Maria Chigueti, Napoleon and Lila Oyeyoyeyo, Berta Chura, Mircia Metaki, Veronica 737 Chavez, Cesar Flores, Fortunato Rayan, and staff of the Cocha Cashu Biological Station, Patricia 738 Alvarez, Cecilia Carrasco, Miriam Minaya, Oscar and Natalia Revilla, Gonzalo Lugon, 739 Rainforest Flow, SERNANP Cusco, Oscar Espinosa and the Pontificia Universidad Católica del 740 Perú, Mark Grote, and Lynne Isbell. Ronny Barr assisted in the design of Figure 1. Anne Pisor, 741 Cody Ross, Bret Beheim, Cristina Moya, and staff in the Department of Human Behavior, 742 Ecology, and Culture at the Max Planck Institute for Evolutionary Anthropology, as well as 743 Doug Yu and two anonymous reviewers, provided valuable comments and insightful criticism on

21 744 earlier versions of this manuscript. Funding was provided by the U.S. National Science 745 Foundation (BCS 1227152) to JAB. All work was conducted under University of California, 746 Davis IRB 226284-2. 747 748 Conflict of Interest: John Bunce and Richard McElreath declare that they have no conflict of 749 interest. 750 Ethical approval: All procedures performed in studies involving human participants were in 751 accordance with the ethical standards of the institutional and/or national research committee and 752 with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. 753 Informed consent: Informed consent was obtained from all individual participants included in 754 the study. 755 756 REFERENCES 757 Aikman, S. (2003). La Educación Indígena en Sudamérica: Interculturalidad y Bilingüismo en 758 Madre de Dios, Perú (M. I. Griffiths, Trans.). Lima: IEP Ediciones.

759 Atran, S., Medin, D., Ross, N., Lynch, E., Vapnarsky, V., Ucan Ek’, E., et al. (2002). 760 Folkecology, cultural epidemiology, and the spirit of the commons: a garden experiment 761 in the Maya lowlands, 1991–2001. Current Anthropology, 43(3), 421-450.

762 Baer, G. (2004). ¿Comes carne de armadillo? acerca del problema de la identidad de los 763 Matsiguenka del Bajo Urubamba en el Oriente Peruano. Boletín de Antropología 764 (Universidad de Antioquia), 18(35), 317-336.

765 Bafumi, J., Gelman, A., Park, D. K., & Kaplan, N. (2005). Practical issues in implementing and 766 understanding Bayesian ideal point estimation. Political Analysis, 13(2), 171-187.

767 Barth, F. (1966). Models of social organization. Royal Anthropological Institute of Great Britain 768 and Ireland, Occasional Papers, No. 23.

769 Barth, F. (1998[1969]). Introduction. In F. Barth (Ed.), Ethnic Groups and Boundaries: The 770 Social Organization of Culture Difference (pp. 9-38). Long Grove: Waveland Press, Inc.

771 Behrens, C. A. (1992). Labor specialization and the formation of markets for food in a Shipibo 772 subsistence economy. Human Ecology, 20(4), 435-462.

773 Berry, J. W. (1997). Immigration, acculturation, and adaptation. Applied Psychology, 46(1), 5- 774 68.

775 Bicchieri, C. (2006). The Grammar of Society: The Nature and Dynamics of Social Norms. 776 Cambridge: Cambridge University Press.

777 Bowles, S., & Gintis, H. (1993). The revenge of Homo economicus: contested exchange and the 778 revival of political economy. Journal of Economic Perspectives, 7(1), 83-102.

22 779 Boyd, R., & Richerson, P. J. (1987). The evolution of ethnic markers. Cultural Anthropology, 780 2(1), 65-79.

781 Boyd, R., & Richerson, P. J. (2002). Group beneficial norms can spread rapidly in a structured 782 population. Journal of Theoretical Biology, 215, 287-296.

783 Chen, X., & French, D. C. (2008). Children’s social competence in cultural context. Annual 784 Review of Psychology, 59, 591-616.

785 Cheung, B. Y., Chudek, M., & Heine, S. J. (2011). Evidence for a sensitive period for 786 acculturation: younger immigrants report acculturating at a faster rate. Psychological 787 Science, 22(2), 147-152.

788 Chudek, M., Cheung, B. Y., & Heine, S. J. (2015). US immigrants’ patterns of acculturation are 789 sensitive to their age, language, and cultural contact but show no evidence of a sensitive 790 window for acculturation. Journal of Cognition and Culture, 15, 174-190.

791 d'Errico, F., Vanhaeren, M., Barton, N., Bouzouggar, A., Mienis, H., Richter, D., et al. (2009). 792 Additional evidence on the use of personal ornaments in the Middle Paleolithic of North 793 Africa. Proceedings of the National Academy of Sciences of the USA, 106(38), 16051- 794 16056.

795 Dressler, W. W., Borges, C. D., Balieiro, M. C., & Dos Santos, J. E. (2005). Measuring cultural 796 consonance: examples with special reference to measurement theory in anthropology. 797 Field Methods, 17(4), 331-355.

798 Ensminger, J., & Knight, J. (1997). Changing social norms: common property, bridewealth, and 799 clan exogamy. Current Anthropology, 38(1), 1-24.

800 Festinger, L. (1962). A Theory of Cognitive Dissonance. Stanford: Stanford University Press.

801 Foley, R. A., & Lahr, M. M. (2011). The evolution of the diversity of cultures. Philosophical 802 Transactions of the Royal Society of London B, 366, 1080-1089.

803 Fox, J.-P. (2010). Bayesian Item Response Modeling: Theory and Applications (Statistics for 804 Social and Behavioral Sciences). New York: Springer.

805 Gans, H. J. (1979). Symbolic ethnicity: the future of ethnic groups and cultures in America. 806 Ethnic and Racial Studies, 2(1), 1-20.

807 Graves, T. D. (1967). Psychological acculturation in a tri-ethnic community. Southwestern 808 Journal of Anthropology, 23(4), 337-350.

809 Haaland, G. (1998 (1969)). Economic determinants in ethnic processes In F. Barth (Ed.), Ethnic 810 Groups and Boundaries: The Social Organization of Culture Difference (pp. 58-73). 811 Long Grove: Waveland Press, Inc.

23 812 Henrich, J. (2004). Cultural group selection, coevolutionary processes and large-scale 813 cooperation. Journal of Economic Behavior & Organization, 53(1), 3-35.

814 Henrich, J., Ensminger, J., McElreath, R., Barr, A., Barrett, C., Bolyanatz, A., et al. (2010). 815 Markets, religion, community size, and the evolution of fairness and punishment. 816 Science, 327, 1480-1484.

817 Henrich, J., & Gil-White, F. J. (2001). The evolution of prestige: freely conferred deference as a 818 mechanism for enhancing the benefits of cultural transmission. Evolution and Human 819 Behavior, 22, 165-196.

820 Henrich, J., & McElreath, R. (2003). The evolution of cultural evolution. Evolutionary 821 Anthropology, 12, 123-135.

822 Jackman, S. (2001). Multidimensional analysis of roll call data via Bayesian simulation: 823 Identification, estimation, inference, and model checking. Political Analysis, 9(3), 227- 824 241.

825 Johnson, J. P., Lenartowicz, T., & Apud, S. (2006). Cross-cultural competence in international 826 business: toward a definition and a model. Journal of International Business Studies, 827 37(4), 525-543.

828 Knight, J. (1992). Institutions and Social Conflict. Cambridge: Cambridge University Press.

829 Kopenawa, D., & Albert, B. (2013). The Falling Sky: Words of a Yanomami Shaman (N. Elliott, 830 & A. Dundy, Trans.). Cambridge: The Belknap Press of Harvard University Press.

831 LaFromboise, T., Coleman, H. L. K., & Gerton, J. (1993). Psychological impact of biculturalism: 832 evidence and theory. Psychological Bulletin, 114(3), 395-412.

833 Legare, C. H., & Nielsen, M. (2015). Imitation and innovation: the dual engines of cultural 834 learning. Trends in Cognitive Sciences, 19(11), 688-699.

835 Llosa Isenrich, E., & Nieto Degregori, L. (2003). El Manu a través de la historia. Lima: PRO- 836 MANU.

837 McElreath, R. (2016). Statistical Rethinking: A Bayesian Course with Examples in R and Stan 838 (Texts in Statistical Science). Boca Raton: CRC Press.

839 McElreath, R., Boyd, R., & Richerson, P. J. (2003). Shared norms and the evolution of ethnic 840 markers. Current Anthropology, 44(1), 122-129.

841 Morison, L. A., Dirir, A., Elmi, S., Warsame, J., & Dirir, S. (2004). How experiences and 842 attitudes relating to female circumcision vary according to age on arrival in Britain: a 843 study among young Somalis in London. Ethnicity and Health, 9(1), 75-100.

844 Nagel, J. (1996). American Indian Ethnic Renewal: Red Power and the Resurgence of Identity 845 and Culture. New York: Oxford University Press.

24 846 Nave, A. (2000). Marriage and the maintenance of ethnic group boundaries: the case of 847 Mauritius. Ethnic and Racial Studies, 23(2), 329-352.

848 Ochs, E., & Izquierdo, C. (2009). Responsibility in childhood: three developmental trajectories. 849 Ethos, 37(4), 391-413.

850 Oravecz, Z., Anders, R., & Batchelder, W. H. (2015). Hierarchical Bayesian modeling for test 851 theory without an answer key. Psychometrika, 80(2), 341-364.

852 Oravecz, Z., Vandekerckhove, J., & Batchelder, W. H. (2014). Bayesian cultural consensus 853 theory. Field Methods, 1525822X13520280.

854 Pagel, M., & Mace, R. (2004). The cultural wealth of nations. Nature, 428, 275-278.

855 Parliament of Canada (1985). Indian Act, R.S., c. I-6, s. 1. (http://laws- 856 lois.justice.gc.ca/eng/acts/I%2D5/index.html).

857 Price, T. (2008). Speciation in Birds. Greenwood Village: Roberts and Company.

858 R Core Team (2014). R: a language and environment for statistical computing. R Foundation for 859 Statistical Computing, Vienna, Austria. URL http://www.R-project.org/.

860 Richerson, P. J., Baldini, R., Bell, A. V., Demps, K., Frost, K., Hillis, V., et al. (2016). Cultural 861 group selection plays an essential role in explaining human cooperation: a sketch of the 862 evidence. Behavioral and Brain Sciences, 39, e30, doi:doi: 863 10.1017/S0140525X1400106X.

864 Richerson, P. J., & Boyd, R. (2005). Not by Genes Alone: How Culture Transformed Human 865 Evolution. Chicago: University of Chicago Press.

866 Romney, A. K., Weller, S. C., & Batchelder, W. H. (1986). Culture as consensus: a theory of 867 culture and informant accuracy. American Anthropologist, 88(2), 313-338.

868 Rosengren, D. (2004). Los Matsigenka. In F. Santos, & F. Barclay (Eds.), Guía Etnográfica de la 869 Alta Amazonía (Vol. IV, pp. 1-157, Travaux de l'Institut Français d'Études Andines, Vol. 870 181). Balboa: Smithsonian Tropical Research Institute, Instituto Francés de Estudios 871 Andinos.

872 Sahlins, M. (1994[1988]). Cosmologies of capitalism: the trans-Pacific sector of "The World 873 System". In N. B. Dirks, G. Eley, & S. B. Ortner (Eds.), Culture/Power/History: A 874 Reader in Contemporary Social Theory (pp. 412-455). Princeton: Princeton University 875 Press.

876 Schacht, R., & Grote, M. (2015). Partner choice decision making and the integration of multiple 877 cues. Evolution and Human Behavior, 36, 456-466.

878 Seth, S. (2007). Subject Lessons: The Western Education of Colonial India (Politics, History, and 879 Culture). Durham: Duke University Press.

25 880 Shepard, G. H., Rummenhoeller, K., Ohl, J., & Yu, D. W. (2010). Trouble in paradise: 881 indigenous populations, anthropological policies, and biodiversity conservation in Manu 882 National Park, Peru. Journal of Sustainable Forestry, 29, 252-301.

883 Stan Development Team (2016). RStan: the R interface to Stan, Version 2.14.1. http://mc- 884 stan.org.

885 Trapnell, L. A. (2003). Some key issues in intercultural bilingual education teacher training 886 programmes - as seen from a teacher training programme in the Peruvian Amazon basin. 887 Comparative Education, 39(2), 165-183.

888 Trennert, R. A. (1979). Peaceably if they will, forcibly if they must: the Phoenix Indian School, 889 1890-1901. The Journal of Arizona History, 20(3), 297-322.

890 van den Berghe, P. L. (1983). Human inbreeding avoidance: culture in nature. Behavioral and 891 Brain Sciences, 6, 91-123.

892 Wimmer, A. (2013). Ethnic Boundary Making: Institutions, Power, Networks. Oxford: Oxford 893 University Press.

894 Wolf, E. R. (1982). Europe and the People Without History. Berkeley: University of California 895 Press. 896

897 Biographic Sketches 898 John Bunce is a postdoctoral fellow in the Department of Human Behavior, Ecology, and Culture 899 at the Max Planck Institute for Evolutionary Anthropology. He studies cultural change at ethnic 900 boundaries using both theoretical mathematical models and ethnographic fieldwork in 901 Amazonian Peru. For his Ph.D. at the University of California, Davis, in 2009 he studied the 902 ecology and genetics of color vision in Neotropical primates. 903 904 Richard McElreath (BS, MA, PhD) is director at the Max Planck Institute for Evolutionary 905 Anthropology in Leipzig and professor in the Department of Anthropology at the University of 906 California, Davis. He studies the role of culture in human evolution. 907 1

ONLINE RESOURCE APPENDIX

Inter-ethnic Interaction, Strategic Bargaining Power, and the Dynamics of Cultural Norms: A Field Study in an Amazonian Population (Human Nature)

John Andrew Bunce1,2,3 and Richard McElreath1,2

1Department of Human Behavior, Ecology, and Culture, Max Planck Institute for Evolutionary Anthropology, Leipzig, Germany 2Department of Anthropology, University of California, Davis, CA USA 3Department of Anthropology, Indiana University, Bloomington, IN USA

Corresponding author: John Andrew Bunce ([email protected])

A. Supplemental Methods

A.1. Study Population Matsigenka Study Population This study was conducted among residents of the Matsigenka Native Community of Tayakome, located inside Manu National Park, in the Department of Madre de Dios, in the lowland Amazonian region of Southeastern Peru. The Matsigenka (also spelled Machiguenga or Matsiguenka) are indigenous Amazonians who speak a language in the Arawakan family. Most currently live in the watersheds of the Urubamba and Manu Rivers in the Departments of Cusco and Madre de Dios (respectively) in lower-elevation tropical forests. The current population is estimated at 12,000 across both watersheds (Shepard 2002). Dialects, subsistence strategies, market integration, and certain other aspects of culture differ to varying degrees between Urubamba and Manu, with Manu populations (the focus of this study) generally less-well integrated into the local market economy. Tayakome, with a population of 79 adults (plus one health care technician in the health post) in 41 nuclear families, in 12 household clusters (or clans) at the time of this study, is the smaller of two officially recognized Matsigenka Native Communities inside Manu National Park, a UNESCO Biosphere Reserve and World Heritage Site. The community was established in the late 1960s by missionaries of the Summer Institute of Linguistics, which were evicted from Tayakome after creation of the national park in 1972 (Shepard et al. 2010). As a consequence of living inside a national Protected Natural Area, residents are required to maintain “traditional” subsistence practices of swidden horticulture (mostly manioc), fishing, and bow hunting, which are believed to be less detrimental to the extraordinarily high biodiversity of the park (Levi et al. 2009; Ohl-Schacherer et al. 2007). Domesticated animals (excepting dogs and chickens), firearms, and the extraction and sale of forest products are prohibited by the park administration. These restrictions have been, and continue to be, a source of tension between the Matsigenka and park authorities (Shepard et al. 2010). Tayakome can only be reached by a one- to two-day boat trip up the Manu River from the nearest Mestizo town of Boca Manu, located just outside the park boundary. Non-Matsigenka may only enter the park and visit Tayakome while conducting government-sanctioned activities (e.g., medical services, Catholic religious 2

services, research and infrastructure projects, monitoring by park guards). There is a government health post in Tayakome staffed by one or two Mestizo technicians, as well as a primary school staffed by two externally-trained Matsigenka teachers. Nearly all Matsigenka in Tayakome prefer to communicate exclusively in their native language. However, most adults can speak and understand basic Spanish, and a few are fluently bilingual. At the time of this study, there were three battery-powered short-wave radios in Tayakome, and a satellite pay telephone at the health post (seldom used by residents). There were no functioning televisions. However three households had devices capable of playing dvd movies. Some households had battery-powered stereos to play recorded music. An NGO had recently installed an ingenious potable running water system (http://www.rainforestflow.org/index.html), and the regional government had recently provided each household with a solar panel, battery, and two light-bulbs. All cooking was done over fire. Most inter-ethnic interaction between Matsigenka and majority-culture Mestizos occurs outside the park in the domains of boarding school education, wage labor, and commerce. Upon completion of primary school (usually 12 to 15 years old), 15 Tayakome children have been sent to four-year boarding high schools in the Mestizo towns of Boca Manu, Shintuya, and Salvación, all within a one- to two-day boat trip from Tayakome. At the time of this study, 4 of these young adults had successfully completed high school. Additionally, 5 current adult residents of Tayakome grew up outside of the park (e.g., in Urubamba) and attended primary and/or secondary schools in other Mestizo towns. Approximately half of Tayakome adults have varying degrees of experience working as wage-laborers among Mestizos outside of the park. This experience can include harvesting Mestizo banana fields over the course of a few weeks, or working as a crewmember in a Mestizo logging outfit or on a Mestizo-owned tourist boat for months at a time. Most adult Tayakome residents have experience buying goods in Mestizo general stores in towns like Boca Manu, using money earned through wage labor. Commonly purchased items include clothes, flashlights, batteries, soap, machetes, pots, crackers, and carbonated drinks. It is also common for Matsigenka arriving without money in a Mestizo town to initially exchange labor for food and lodging. A Catholic priest visits Tayakome twice per year to hold mass, baptize children, and distribute gifts in well-attended ceremonies.

Mestizo Study Population This study was also conducted among residents of the Mestizo towns of Boca Manu and Atalaya, located just outside the boundary of Manu National Park, in the Departments of Madre de Dios and Cusco, respectively. Travel down the Alto Madre de Dios River from Atalaya to Boca Manu takes approximately one day. Most Mestizos living in the towns bordering the park are colonists (or their descendants) originating from highland regions of Puno, Cusco, and Apurimac. Many are native Quechua speakers, all are fluent in Spanish, and nearly all prefer to interact among themselves exclusively in Spanish. As a convenience, we use the term Mestizo to refer to the non-Matsigenka residents of these towns. However, it must be noted that few of these individuals actually use this term to refer to themselves. When asked to distinguish themselves from the Matsigenka, colonists from the highlands often referred to themselves as “Gente Blanca” (White People), a label with complex meanings that do not, in this case, correspond objectively with skin color. Additionally, there are residents of Mestizo towns in the region who identify as members of other local indigenous groups, such as Yine and Amarakaeri, or as having mixed indigenous-colonist ethnicity. These individuals make up a small minority in the present 3 study, and, as they are all well-integrated into the market economy, are grouped in the Mestizo category. Boca Manu is a Mestizo town with a population of approximately 80 adults in approximately 55 family units at the time of this study. The population of this town fluctuates, as many families spend months away from town working in other areas. There are also several residents who are unmarried, or whose spouses live elsewhere. It is located at the entrance to Manu National Park, at the confluence of the Manu and Alto Madre de Dios Rivers, and can only be reached by boat. It is the seat of government for the District of Fitzcarrald, and the Matsigenka communities inside Manu National Park compose an important constituency for the mayor in Boca Manu. Most residents of Boca Manu are Mestizos, although there are three Matsigenka families that live in the town. Residents tend small general stores (five in town), own small restaurants (four in town), make traditional wooden boats (mostly to sell downriver to small-scale gold miners), buy permits to log on nearby indigenous territory, and/or capture and sell valuable tree-trunks that float down the Manu River and out of the park in the rainy season (October – March). Few residents have agricultural fields, and most buy at least some of their food at the general stores, whose merchandise is periodically restocked from Cusco. There is a kindergarten, primary, and secondary school in Boca Manu, including a boarding house close to the school for Matsigenka secondary-school students from the Manu communities. There is a government health post with a resident doctor and technician in town. A generator provides central electricity for three hours each night. Most families have a television, and most people have cellular telephones. A public computer with internet access in the municipal government building is used infrequently. Most Boca Manu residents have interacted at some point with adult Matsigenka from Tayakome, either selling them food or merchandise, or hiring them as part of logging crews. A few men from Boca Manu have visited Tayakome, either as part of political campaigns during mayoral elections, or as boat crews on government-sanctioned business. Nearly all residents are familiar with (know the names of), have sold to, served meals to, or hired as day-laborers, the Matsigenka students in the boarding school. Most residents of Boca Manu identify as belonging to either Catholic, Evangelical, or Adventist congregations in town. Atalaya is a Mestizo town with a population of approximately 65 adults in approximately 30 households at the time of this study. The population of this town fluctuates due to seasonal workers. It is located on the Alto Madre de Dios River and is the first river terminus of the only road (un-paved) connecting this region to the highland city of Cusco. It is the primary river embarkation point for tourists traveling down from Cusco who have booked boat tours of Manu National Park. Residents tend small general stores (three in town), own small restaurants (six in town), and/or own or work as crewmembers on boats taking tourists for several-day cruises into Manu National Park. Several residents have agricultural fields in which they grow plantains for market in Cusco, a day’s truck ride away. Almost no one in Atalaya grows food for auto- consumption. The larger town of Pilcopata is one hour by truck or motorcycle from Atalaya, and most residents buy food in the market there, though smaller items may be purchased in the local general stores. There is a kindergarten and primary school in Atalaya, and secondary (high) schools in the nearby towns of Pilcopata and Salvación. The nearest boarding school housing Matsigenka students is in Salvación (an hour away by truck), and most Atalaya residents have little contact with Matsigenka children. While there is no health post in Atalaya, there are small hospitals nearby in Pilcopata and Salvación. However, many residents prefer to go to Cusco for moderate to serious healthcare needs. Central electricity is available continuously. Most families have a television, and most people have cellular telephones. Several fee-for-use computers with 4

internet access are used occasionally. At the time of this study, there was one resident Matsigenka family in Atalaya, and two unmarried Matsigenka young adults living with, and working for, a Mestizo family. Several Matsigenka men from Tayakome are well-known to most residents as they spend up to seven months each year living in Atalaya and working as crewmembers on tourist boats. When they don’t have tours scheduled, these men are often hired out as day-laborers in Mestizo plantain fields. A few Mestizo men from Atalaya have visited Tayakome as boat crews on government-sanctioned business. Most residents identify as Catholic, Evangelical, or Adventist, but no religious services are conducted in town.

A.2. Data Collection In Tayakome and Atalaya, JAB lived with a host family, routinely participating in household and community activities and sharing meals. In Boca Manu, he lived in a rented room, ate at local restaurants, and participated in activities such as boat building and wood extraction at the invitation of residents. In all three communities JAB conducted several rounds of individual semi-structured and structured interviews with adult residents. All interviews were conducted by him in the interviewee’s preferred language (Matsigenka or Spanish) while out of earshot of other community members. He conducted at least one interview with all but two residents of Tayakome, and with approximately two thirds of the residents of Boca Manu and Atalaya. Included in this study are interviews with members of two of the three Matsigenka families living in Boca Manu, two Matsigenka residents of Atalaya, as well as one Mestizo technician working in the health post in Tayakome. The first rounds of interviews entailed soliciting a life- history narrative. From these interviews, JAB identified education, labor and commerce as three domains of common Matsigenka-Mestizo inter-ethnic interaction. In follow-up interviews he recorded each individual’s recollection of personal intra- and inter-ethnic experience in these three domains. Based on his own experience living in these communities, he then designed a set of fourteen vignette questions (Bernard 2006), presented below. From the perspective of this study, none of the vignette questions has a correct or incorrect answer, so no value is assigned to affirmative versus negative responses to any question. In cases of individuals who are cross- culturally competent, we assume that these vignette questions measure an individual’s favored norms. Thus, during some inter-ethnic interactions, such an individual may actually coordinate using non-favored norms which do not coincide with her answers to the vignette questions (e.g., Mestizos hiring Matsigenka wage laborers). Below are the fourteen vignette questions employed in this study, across eight social coordination contexts. JAB developed questions in Spanish with the help of an English-Spanish bilingual Peruvian collaborator. He translated the questions into Matsigenka, and then had them back-translated from Matsigenka to Spanish with the help of a Matsigenka-Spanish bilingual Matsigenka collaborator in Tayakome. Due to organizational difficulties, questions were presented in the same order to interviewees from Tayakome and Boca Manu, but the order of questions 13 and 14 changed for interviewees in Atalaya. Interviewees responded affirmatively or negatively after JAB initially presented each vignette question. For a subset of interviewees, he then asked them why they responded the way they did. Upon hearing his request for an explanation, occasionally they then changed their answer from negative to affirmative, or vice versa. If the interviewee did not wish to explain why they changed their initial answer, we did not include their response to that particular question in the present analysis. We exclude these responses in case JAB’s request to explain their initial answer led them to believe that he was not in agreement with their answer, which 5

they then changed in order to answer in the way they believed he wanted them to. This occurred despite assurances that their answers were confidential and that JAB believed there were no right or wrong answers to the vignette questions. These cases of response changes may be due to the fact that interviewees were unaccustomed to answering such unusual questions in this unfamiliar format. On other occasions, some interviewees explained that their response (affirmative or negative) to a vignette question was contingent on additional information that JAB had not provided in the vignette itself. For example, for Question 1 (below), an informant responded that it is okay for a wife to go hunting sometimes while her husband stays home, but it is not okay if she does it often. Such responses resulted from interviewees thinking more deeply about a vignette than JAB had when he designed it. We excluded these contingent responses from the present analysis, because JAB had not provided the requested details (e.g., frequency of the behavior in question) to the rest of the interviewees. For all those interviewees who answered the question without requesting additional details, we assume that they interpreted the question in a similar way (e.g., the wife always goes hunting, not just sometimes). As a result of these two classes of interviewer-interviewee misunderstanding, we excluded a total 73 out of 2094 responses (3.5%). To check the effect of excluding these data, we included people’s initial responses, and included contingent responses after providing the additional requested information such that these interviewees interpreted the questions in the way we assumed all other interviewees had. Including or excluding these data has no qualitative effect on the conclusions. The latent axis constructed by the domain-general IRT analysis from these fourteen vignette questions is interpreted according to how individuals with different positions on the latent axis answered each question. The negative pole of the axis corresponds to a meta-norm prioritizing respectful autonomy in interaction behavior. “Respectful” here refers to formality, honor, and conforming to the stereotype of a “civilized” member of Peruvian national society. Autonomy is here defined as individualism and self-reliance. The positive pole of the latent axis corresponds to a meta-norm emphasizing practical inter-dependence. “Practical” here stands for prioritizing immediate material benefits over conformity to custom or societal expectations, e.g., valuing money over honor. Inter-dependence here stands for altruistic behavior (e.g., obedience) benefiting the social group, which, in this case, is usually the family unit or, at its largest, the clan (of which there are 12 in Tayakome). Explanations of how answers to each vignette question contribute to interpretation of the latent axis are provided below each translation. Note that there are multiple ways of interpreting the answers to each question. We provide a set of interpretations that emphasize one common theme uniting most of the questions. This theme constitutes one meaning of the latent axis constructed through the IRT analysis. There are undoubtedly other themes uniting the questions, which would lead to alternative, and equally correct, interpretations of the latent axis. Note, however, that interpretation of the meaning of the latent axis has no effect on the conclusions of this study. The latent axis is important only insofar as it distinguishes Matsigenka and Mestizos on the basis of norms, whatever those norms may be. The contribution of each question to the construction of the latent axis through the IRT models is shown in Figure B1, below. The social context of each vignette question (education, wage labor, commerce, spousal relations, parent-offspring relations, inheritance, healthcare, and religion) is included in parentheses.

Question 1, wife hunts/works (Spousal Relations) 6

There is a married couple with no children. The woman hunts and fishes (Mestizo: has job and makes more money). The man stays home and cooks, weaves (Mestizo: cleans), and washes clothes. Is this okay or not okay?

Hay una pareja sin hijos. La mujer trabaja y gana plata. El hombre se queda en la casa, cocina, limpia, y lava ropa. ¿Está bien?

Ainho ogari tsinane, mameri otomi, mameri oshinto. Iroro ogotake omatsigatake ochacopi cameti. Onti kovinsari. Iroro oati onkogira ivatsa inkenishkue. Yogari ojina ainho iripitake ivanko, yonkotia sekatsi, intiri ijamatia, ikiviasarati igamisate. Cameti o tera cameti?

Interpretation for latent axis: An affirmative response suggests a more flexible and practical subsistence strategy, allowing the family unit to adapt to a wider range of ecological, economic, and social conditions, thus prioritizing well-being of the family group over individual conformity to gender roles, e.g., those stereotypical of “civilized” Peruvian society. Differences between questions asked of Matsigenka and Mestizo interviewees reflect the different tasks stereotypical of each gender in the two ethnic groups.

Question 2, daughter babysits (Parent-Offspring Relations)

After school, a ten-year-old daughter cannot go to friend's house to play because she has to care for her two-year-old brother until their parents come home at night. Is this okay?

Después de salir de la escuela en la tarde, una hija de diez años no puede ir a jugar con sus amigas porque tiene que volver a la casa y cuidar a su hermano menor de dos años hasta que lleguen sus papás en la noche. ¿Está bien?

Oga oshinto oshiriagakota diez años. Hatanai chavini, okonteiganae escuela. Hanki agavea omagempitira. Ogari oatae ovankokue ogierira iariri itiomiani ishiriagakota dos años, iripigaira iriegi tsitenigeti. Cameti o tera cameti?

Interpretation for latent axis: An affirmative response suggests prioritizing a child’s responsibility to the family group over her individual desire to play.

Question 3, not wear dead hat (Inheritance)

A man always wears his favorite hat. After he dies, his son takes the hat and wears it. When he wears it he remembers his father. Is this okay?

Un hombre siempre lleva un gorro favorito. Se muere. El hijo coge el gorro y se lo pone. Cuando se lo pone, piensa en su papá. ¿Está bien?

7

Yogari sidadi kantani yamea igorate. Imbogini ikamake. Itomi yagakero igorate irashi iriri ikamake. Itomi kantani igoraterora, kantani ikenkirira iriri. Cameti o tera cameti?

Interpretation for latent axis: A negative response corresponds with a common Matsigenka belief that the soul of a dead person will return to haunt anyone who appropriates his or her old possessions, making them sick (J. Bunce, pers. obs., see also Johnson 2003). To prevent this, used (but not new unused) belongings of a dead person should be burned or buried. Because of the physical proximity and spiritual bonds between the bodies of Matsigenka family members, sickness poses a threat to the whole family (Izquierdo et al. 2008). Thus a negative response to this vignette suggests prioritizing wellbeing of the family-group over an individual desire to remember a loved one. This cultural practice also, incidentally, curtails most inter-generational transmission of what little material wealth individual Matsigenka accumulate in their lifetimes.

Question 4, hit students (Education)

A teacher hits students when they don't learn. Is this okay?

Un profesor golpea a sus alumnos cuando no aprenden. ¿Está bien?

Yogotagantasirira ipasatakeri estudiante tera irogoigia sankevantera. Cameti o tera cameti?

Interpretation for latent axis: An affirmative response suggests prioritizing obedience and extrinsic motivation for learning over strategies to stimulate individual student interest and to accommodate variation in individual learning style.

Question 5, no questions (Education)

A student pays attention to the teacher and never asks any questions. Is this okay?

Un estudiante siempre escucha y hace caso al profesor. Nunca hace ninguna pregunta. ¿Está bien?

Yogari estudiante kantani ikemisantake yogatagantasirira. Tenka inkogakote. Cameti o tera cameti?

Interpretation for latent axis: An affirmative response suggests prioritizing conformity and faithful copying of information over curiosity and exploration.

Question 6, post flu (Healthcare)

If you get a respiratory illness (influenza), do you first go to the health post, first use home remedies, or first go to a shaman or curandero? (Affirmative: go first to the health post. Negative: anything else) 8

¿Si tiene gripe, Ud. va primero a la posta, primero usa remedios caseros, o primero va al curandero?

Agakempi merentsi. Okityo piatake postakue? Okityo povetsike inchashi pankotsikue? Okityo piatake irishipokempira? Okityo piatake iritasonkempira?

Interpretation for latent axis: An affirmative response for the Matsigenka may suggest that, as respiratory infections originate almost exclusively from outside the community and are viewed as non-Matsigenka diseases, they must be treated with non-Matsigenka remedies at the health post. Additionally, as respiratory infections are particularly virulent among the Matsigenka and usually spread rapidly through the community, prompt individual treatment benefits the community. A negative response for Mestizos (nearly always indicating home remedies) suggests that they recognize the inefficacy of most medicines (e.g., antibiotics) for treatment of minor respiratory infections, which are usually viral. It may also suggest a norm for self-reliance, rather than dependence on doctors.

Question 7, pot to needy (Inheritance)

An old woman has two new pots and two adult daughters. One daughter has her own two pots, but wants her mother’s pots. The other daughter has no pots, and also wants her mother’s pots. When the mother dies, who should inherit the pots? (Illustrated with a diagram. Affirmative = one pot to each daughter. Negative = both pots to the daughter who has none)

Hay una mujer vieja con dos ollas nuevas. Tiene dos hijas adultas. Una hija tiene sus propias dos ollas, pero quiere las ollas de su mama. La otra hija no tiene ollas. También quiere las ollas de su mama. ¿Cuando la mama se muere, a quién debería heredar las dos ollas?

Ogari tsinane okamake. Aityo pitieti ojiromanga otierira. Ainho piteni oshinto antaroni. Paniro oshinto aityo pitieti ohiromangane. Okogake oka otierira hiromanga. Ogari apiteni oshinto mameri ohiromangane. Ariompa okogake oka otierira hiromanga. Tyani gakerone otierira hiromanga ashi iniro?

Interpretation for latent axis: An affirmative response suggests prioritizing individual entitlement over the needs of others in the kin group.

Question 8, good nonbaptized no heaven (Religion)

A good person does not want to be baptized. Where does his or her soul go when they die? Options: up (heaven), down (hell), somewhere else. (Affirmative: heaven. Negative: hell or somewhere else)

Una persona es buena, pero no quiere bautizarse. ¿Cuando se muere, adónde va su alma?

9

Yogari sidadi inti cameti. Tenka ikogira ibautizatakeri Padre. Ikamake. Tyara iriatae isire? Enokue? Saviakue? Pashini?

Interpretation for latent axis: An affirmative response suggests a belief that individual autonomous actions are more important in determining supernatural benefits than is formal acceptance by a representative of an organized religious community. For some Matsigenka, there is a distinction between a soul going into the ground (kipatsikue) and going to a place with flames below the world (morekakue) (see also Rosengren 2004). Both of these places can be indicated by the preposition “down” (saviakue), but they are not necessarily equivalent in terms of favorable or unfavorable outcomes after death. JAB realized this after most of the interviews were completed. Therefore, a negative response to this question by Matsigenka may be interpreted as “anywhere other than a heaven-like place above the world”.

Question 9, bad baptized heaven (Religion)

A bad person is baptized. Where does her or his soul go when they die? Options: up (heaven), down (hell), somewhere else. (Affirmative: heaven. Negative: hell or somewhere else)

Una persona es mala gente. Es bautizado. ¿Cuando se muere, adónde va su alma?

Yogari sidadi inti tera cameti. Padre ibautizatakeri. Ikamake. Tyara iriatae isire?

Interpretation for latent axis: An affirmative response suggests a belief that once a person is socially accepted as a Christian, he or she will receive the supernatural reward due Christians, regardless of autonomous individual behavior. Negative responses by Matsigenka should be interpreted as they are for Question 11, above.

Question 10, stop work to visit (Wage Labor)

A man is hired to prepare an agricultural field. He stops work at noon in order to go visit a friend. He returns the next day to finish the job. Is this okay?

Un hombre está contratado limpiar una chacra. Deja de trabajar a mediodía porque quiere visitar a su amigo. El día siguiente termina de limpiar la chacra. ¿Está bien?

Kamatitya, ipuinatakeri koriki yogari sidadi itsamaitera imagashipogote irashi Viracocha. Katinga poriatsiri, yoga sidadi yapakuianae itsamaitera, iroventi ikogake ikamosote ishaninka. Paita onkuita yagatake itsamaitera, itsongatero imagashipogote irashi Viracocha. Cameti o tera cameti?

Interpretation for latent axis: An affirmative response suggests prioritizing the maintenance of social bonds over dedication to contractual labor.

10

Question 11, cheap mean store (Commerce)

There are two stores. One is cheap with a mean owner. The other is expensive with a nice owner. Where would you buy? (Affirmative: expensive store)

Hay dos tiendas. Una tienda es barata pero el dueño es malo. La otra tienda es cara pero el dueño es bueno. ¿Dónde iría Ud. para comprar?

Aityo pitieti otiendate Viracocha. Patiro otiendate tenka opuinatasanotempa. Yogari shintarorira inti kisantari. Apiteni otiendate onti puinatasanori. Yoga shintarorira tenka irikisante. Tyaka picompratake? Hanta puinatasanori irashi cameti shintarorira o apiteni tenka opuinatasanotempa irashi kisantari?

Interpretation for latent axis: An affirmative response suggests prioritizing respectful behavior over the practicality of saving money.

Question 12, xcousin marriage (Spousal Relations)

A man wants to marry. His mother is the sister of the woman's father. Is it okay for him to marry this woman? (Illustrated with a diagram. For the Matsigenka, examples were provided of potential marriages between people known to the interviewee.)

Un hombre quiere casarse con una mujer. La madre de él es la hermana del padre de la mujer. ¿Está bien?

Yogari sidadi ikogake ijinantempera. Iriniro onti iritsiro iri oga tsinane. Cameti o tera cameti?

Interpretation for latent axis: An affirmative response indicates acceptance of cross-cousin marriage, which may serve to strengthen social ties within clans. This suggests prioritizing benefits to the clan (e.g., cohesion) over any perceived or un-perceived individual-level benefits associated with avoidance of consanguineous mating. Note that an affirmative response does not necessarily indicate a preference for cross-cousin marriage over other forms of marriage. Rather it indicates a lack of opposition to cross-cousin marriage.

Question 13, arranged marriage (Parent-Offspring Relations)

Parents want their daughter to marry a certain man that she does not like. She wants to marry someone else. Should she obey her parents and marry him anyway or not? (Affirmative = she marries who she wants and does not obey her parents)

Los papás quieren que su hija se case con un chico. Pero a la hija no le gusta el chico. Ella quiere otro. ¿Debería hacerles caso a sus papás y casarse con ese chico, o no?

11

Iriegi ikogake ojinantempera irishinto antaroni. Ikantakero "Noshinto, gaeri yogari sidadi, inti cameti.” Kantangicha, ishinto tenka okogiri yoga sidadi, okogakeri pashini. Iroventi, iroro, irishinto, tenka okematsataeri iri. Tenka ojinantaigiri yoga sidadi. Agakeri pashini. Cameti o tera cameti?

Interpretation for latent axis: An affirmative response suggests prioritizing individual choice over obedience to the family group.

Question 14, laborer party (Wage Labor)

A man is hired to work two days: Monday and Tuesday. Monday night there is a party (Matsigenka: hosted by a Matsigenka). Should he go and get drunk? (Matsigenka: He goes and gets so drunk that he can’t work on Tuesday. Is this ok?) (Affirmative: it is okay to go and get drunk. Negative: it is not okay to go and/or not okay get drunk)

Un hombre está contratado trabajar dos días: lunes y martes. Lunes en la noche hay una fiesta. ¿El hombre debería ir y emborracharse?

Itinkame ipuinatakeri koriki yogari sidadi intsamaitera pitieti kuitagita: Lunes y Martes. Lunes tsitenigeti aityo owidoki irashi ishaninka. Iriro iati, para ishinkitaka. Tenka iragavea intsamaitira Martes. Cameti o tera cameti?

Interpretation for latent axis: A positive response suggests prioritizing the maintenance of social (i.e., friendship and kin) bonds over honoring an individual labor contract. The differences between questions asked of Matsigenka and Mestizo interviewees result from the fact that contractual labor occurs exclusively in Mestizo towns. A party there would nearly always be hosted by a resident Mestizo, and a Matsigenka might not be invited or feel comfortable going. Thus, for Matsigenka interviewees JAB specified that the party was hosted by a Matsigenka. He also specified a more stringent condition for Matsigenka interviewees, namely, that the laborer is so drunk that he cannot work the next day. Even with this difference in stringency, the only interviewees who answered positively were Matsigenka. It should be noted that inebriation is an important part of both Mestizo and Matsigenka society. Among Mestizos it usually involves mass-produced beer, is generally viewed as a vice, and is both cause and consequence of much domestic discord. In contrast, among Matsigenka, women control production and serving of a fermented manioc drink called owidoki (Spanish: masato). Owidoki parties usually take place at a couple’s home and are important social events attended by many community members. Both men and women drink, and often become drunk. Gossip, myths, songs, and (often ingeniously irreverent) jokes are shared, as well as grievances with other community members. Such parties are one of the important contexts in which it is socially acceptable to air (and often resolve) such grievances (Shepard 2002), as well as transmit important cultural knowledge such as myths and songs. The importance of these parties to Matsigenka culture may be the reason why only Matsigenka believed it acceptable to attend such a party at the expense of a contractual labor obligation.

12

A.3. Statistical Analysis

A.3.1. General Description Based on the life history interviews, JAB developed hypotheses (a priori relative to model construction) that four binary predictors would be important in determining an individual’s norms, and hence her location on the latent axis: Ethnicity (1=Matsigenka, 0=Mestizo), Education (1=school with Mestizos, 0=no such experience), Labor (1=wage labor with Mestizos, 0=no such experience), and Commerce (1=commerce or exchange with Mestizos, 0=no such experience). For further description of these predictors, see Appendix section A.3.3., below. Note that Matsigenka individuals may take on values of 0 or 1 for Education, Labor, and Commerce predictors, while all Mestizo participants in this sample have values of 1 for these three predictors. We did not have a priori hypotheses about the effects of sex or age on norms, but we included these in the models as exploratory predictors. We also included the following two-way interactions: Ethnicity and Education, Ethnicity and Labor, Ethnicity and Commerce, and Ethnicity and Sex. Explanations of how these predictors were coded is provided below. Further information and examples of IRT models can be found in Schacht and Grote (2015), Jackman (2001), and Fox(2010). Because these data were collected across three communities (Tayakome, Boca Manu, and Atalaya), it is reasonable to include a random effect for community in the IRT models. However, community and ethnicity are highly correlated in this population, as very few Matsigenka live in Atalaya and Boca Manu, and no Mestizos live in Tayakome (with the exception of the health post technician). For most questions, mean responses from Mestizo residents of Boca Manu and Atalaya were very similar. Therefore, the analysis presented here does not include a random effect for community. Including a community-level random effect in the best-fitting model does not qualitatively change the interpretations presented below. As unidimensional IRT models provide a reasonably good fit to these data (indicated by all discrimination parameters estimated as non-zero: Figure B1, below), and as we had no a priori hypotheses of multiple simultaneous meta-norms, here we do not pursue higher- dimensional models. Following Jackman (2001), the justification for using discrimination parameters to decide on the number of dimensions to model in an exploratory analysis is as follows. If, in a one-dimensional model, a given question’s discrimination parameter is zero, then an individual’s position on the latent axis provides no information about how that individual answered that particular question. If several questions are found to have discrimination parameters of zero, then perhaps the responses of these questions are strongly correlated among themselves, but not detectably correlated with responses to the rest of the questions. This would justify fitting a two-dimensional model, in the hope that the discrimination parameters of these questions will be non-zero in the second dimension. Jackman (2001) provides an example of such data exploration. However, for a model of given dimension, if the discrimination parameters of all questions are estimated to be non-zero, then an individual’s position in the modeled latent space is informative with regard to how she answered all questions. This is the case for the one-dimensional models used here. A series of 19 models (Appendix Table B1, below) with different combinations of predictors for ethnicity, inter-ethnic education, wage labor, and commerce experience with Mestizos, as well as sex, age, and interactions was fit to the responses of the fourteen vignette questions (see Appendix A.2., above). Models have the following generic form, but vary in the fixed-effect predictors (e.g., for Ethnicity) included in the linear function for : 13

Pr 1logit

… , for 1, … ,

∼N0, 1,

∼N0, 1, for 1, … ,

∼N0, 1, ∼0, , for 1, … ,

∼0, , for 1, … , ∼ Exponential1 ∼ Exponential1, where J is the number of interviewees and K is the number of vignette questions. See Bafumi et al. (2005) for notational changes if not all interviewees answered all questions. Informative N0, 1 priors for the intercept (), random effect for individual (), and additional predictor coefficients (e.g., ) identify the model by constraining the position and scale of the ’s to fall within a reasonable distance (e.g., usually within two standard deviations) on either side of zero, resolving additive and multiplicative aliasing (Bafumi et al. 2005). Any choice of position and scale in IRT models is arbitrary and results in equivalent inference. Reflection (or rotational) invariance (Jackman 2001; Bafumi et al. 2005) is not a concern as mean discriminations of all questions are positive, effectively polarizing the latent axis. Variances for and are given exponential hyperpriors to control ceiling effects common to logistic models (McElreath 2016, pg 378). The complete series of 19 models is shown in Table B1, below. Parameter estimation for each model was accomplished with RStan (Stan Development Team 2016), running four Hamiltonian Monte Carlo chains in parallel until convergence was indicated by a high effective number of samples (> 500) and estimates of 1.00 (McElreath 2016). This usually entailed 4000 samples per chain, half of which were warm-up. Data and analysis script are available from Github (https://github.com/jabunce/bunce-mcelreath-HN-2016-matsigenka- norms).

A.3.2. Coding of Fixed-Effect Predictors

Ethnicity (1=Matsigenka) Interviewees were coded as Matsigenka if they self-identified as Matsigenka. Two siblings had a Matsigenka mother and a Mestizo father. Both of them had lived for several years in Matsigenka communities in Manu, self-identified as Matsigenka, and were coded as such. All non-Matsigenka in this study were classified as Mestizos. A description of these interviewees is provided above in Appendix A.1.

14

Sex (1=Male) Sex was coded according to self-identified gender.

Age Category (Adolescent, Adult, Elder) Many Matsigenka, especially older adults, do not know their exact age in years. For this reason we use age categories as predictors, estimating ages where necessary. Adolescents were younger than 20, adults were 20 to 50, and elders were older than 50.

Education Experience (1=attended school with Mestizos) All Mestizos attended primary and/or secondary school with other Mestizos, so all were coded as 1. Several Matsigenka interviewees grew up outside of Tayakome and went to either a boarding- or non-boarding primary school with Mestizos. These individuals were coded as 1. Most Matsigenka in Tayakome attended primary school in Tayakome, with Matsigenka teachers and all Matsigenka students. If this was an interviewee’s only education experience, she or he was coded as 0. There is no secondary school in Tayakome. A few Matsigenka from Tayakome attended boarding secondary schools with Mestizos outside of Tayakome for at least four of the requisite five years, and some had additional educational training after high school (e.g., for tour guide certification). These boarding school attendees were coded as 1. Two Matsigenka interviewees attended a boarding secondary school for a few months before either being expelled or leaving because they did not like it. These interviewees were coded as 0. The average amount of inter-ethnic education experience among Matsigenka scored as 1 was approximately 7 years.

Wage Labor Experience (1=wage labor experience with Mestizos) All Mestizos were coded as having wage labor experience with other Mestizos. The vast majority of these interviewees, both men and women, had earned money themselves, helped their spouse or relative earn money, and/or hired people to work for them. In contrast to the wage labor experience of most Matsigenka, most Mestizos were self-employed (e.g., banana farmers, small restaurant owners, boat builders) and did not have a “boss”. Matsigenka were coded as 1, i.e., having wage labor experience with Mestizos, if they spent an approximate total of at least 12 months living with and working under Mestizos in their lifetime. This experience occurred in Mestizo towns, Mestizo-run tourist lodges or work camps (e.g., for logging or oil companies), or Mestizo-staffed park guard posts. Several Matsigenka interviewees attended boarding secondary school in Mestizo towns for several years and were hired as wage laborers on most weekends. These individuals were also scored as 1. The average amount of wage labor experience among Matsigenka scored as 1 was approximately 5.5 years.

Commerce Experience (1=commerce experience with Mestizos) All Mestizos were coded as 1, i.e., having commerce experience with other Mestizos, as all had purchased items in stores. Matsigenka were coded as 1 if they had ever directly bought or sold/traded items to Mestizos. Most Matsigenka had bought from Mestizo stores at least a few times in their life, although quantification based on memory was difficult.

A.3.3. Model Comparison and Inference Models with different combinations of predictors were compared using the Widely Applicable Information Criterion (WAIC) (McElreath 2016, pg 205). Inference was based on the 15

posterior predictive distributions of the best fitting model (Appendix Table B1, below). We then use these posterior predictions to calculate counterfactual contrasts for the position on the latent axis of each combination of ethnicity and interaction experience in order to directly address the hypotheses under investigation (Schenker and Gentleman 2001). The general form of these contrasts is as follows: 1) Matsigenka without experience (i.e., education, labor, or commerce) – Mestizo (all of whom have each type of interaction experience only with other Mestizos); 2) Matsigenka without interaction experience – Matsigenka with each type of inter-ethnic experience; and 3) Matsigenka with inter-ethnic experience – Mestizo with experience only with other Mestizos. Data and analysis scripts used to fit all models using the open-source RStan package (Stan Development Team 2016) in R (R Core Team 2014) are available from Github (https://github.com/jabunce/bunce-mcelreath-HN-2016-matsigenka-norms).

B. Supplemental Results

16

Figure B1. Logistic functions generated from the mean posterior location (, x-value of vertical grey line) and discrimination (, slope of curve at intersection with vertical line) of each of the 14 vignette questions, derived from an IRT model with a random effect for individual and no fixed effect predictors (m1 in Table B1 below). Questions with higher absolute value discrimination contribute more to the construction (and hence, interpretation) of the latent axis. The discriminations of questions 8 and 11 are marginally distinguishable from zero (Figure B2, below). The fact that discriminations are distinguishable from zero suggests that questions map well onto a single dimension (Jackman 2001). The posterior mean location of each individual on the latent axis is plotted in red. A person’s y-value of 0 or 1 corresponds with a respectful autonomy or practical inter-dependence response, respectively, to the given question. For a given question’s location and positive discrimination , increasing an individual’s location increases the probability that she gives the practical inter-dependence response. For a given and positive , increasing decreases the probability of a practical inter-dependence response. For a given , increasing a positive increases the probability of a practical inter- dependence response, while simultaneously decreasing the probability of such a response for .

17

Figure B2. Posterior probability density estimates for discrimination parameters () in the best- fitting IRT model (m19 in Table B1) for the fourteen vignette questions. 90% HPDI are shown in grey. Note that all discrimination estimates are markedly non-zero, with the exception of those for questions 8 and 11, which are marginally non-zero. This suggests that nearly all questions contribute substantially to the construction of the latent axis, and supports the decision to limit analysis to a single latent dimension (Jackman 2001).

The best-fitting IRT model (m19 in Table B1, below) contained predictors for ethnicity, education with Mestizos, wage-labor, and commerce experience (and their interactions), but did not include predictors for sex and age category. Strong associations are detected between individuals’ norms and both ethnicity and education experience, while more modest associations are evident for wage labor and commerce experience, and interactions between ethnicity and experience (Figure B3, below). Sex and age are not present in the best-fitting model, and show no detectable associations with norms represented by the latent axis. 18

The difference in WAIC between models m19 and m17 is slight, and some sampling iterations give m17 the highest WAIC weight. This suggests that m19 is slightly over-fit, in the sense that, if one knows a person’s ethnicity, commerce experience, and education experience, little additional predictive value is gained by also knowing that person’s wage labor experience. However, because the objective of this study is to compare the associations between norms and inter-ethnic experience in these three domains, model m19 is used for calculation of contrasts in Figure 4.

19

20

Figure B3. Posterior probability density estimates for parameters in the best-fitting IRT model (m19 in Table B1) fit to all fourteen vignette questions. 90% HPDI are shown in grey. Effect size is on the latent axis, where more-positive values correspond to a domain-general norm for practical inter-dependence, and more-negative values correspond to a norm for respectful autonomy. Ethnicity (Matsi) and inter-ethnic education experience (Education) have strong associations with individuals’ positions on the latent axis, i.e., very little posterior density overlapping zero. Wage labor, commerce, and [ethnicity by experience] interactions have more modest associations.

C. Supplemental Discussion

C.1. Ethnic Differences in Norms

The raw proportions of Matsigenka and Mestizo responses to the vignette questions (Figure 2), as well as IRT analyses (Figures 3 and 4), suggest marked ethnic differences in the 21 distributions of norms across eight contexts of social coordination: commerce, wage labor, education, spousal relations, parent-offspring relations, inheritance, healthcare, and religion. Matsigenka-typical norms, represented by a meta-norm for practical inter-dependence, coincide with observations of other ethnographers regarding the importance of harmony and cohesion within the Matsigenka extended family (Johnson 2003; Izquierdo et al. 2008; Izquierdo 2005; Ochs and Izquierdo 2009), and that family members, to some extent, share a collective body (Shepard 2002). Interestingly, norms for such social inter-dependence may not apply to Matsigenka who are not close kin. This coincides with the subsistence autonomy of Matsigenka family units (Johnson 2003), and could explain the individualistic behavior observed by Henrich (2000; 2004) in experimental games that deliberately excluded framing cues of kinship (Hagen and Hammerstein 2006). Mestizo-typical norms, represented by a meta-norm for respectful autonomy, seem apt for pioneers who left kin groups in the highlands to colonize an unfamiliar tropical environment, and who now depend on integration into a competitive local economy. This norm also coincides with JAB’s experience living in Mestizo towns, where emphasis is placed on individual self-sufficiency (even among spouses), formality (e.g., using the formal Usted conjugation of Spanish verbs), and adherence to the rituals of “civilized” society (e.g., stylized political speeches at town meetings). Some of these characteristics may stem from processes of ethnic transformation documented among both highland Quechua-speakers (de la Cadena 1992) and lowland mixed-indigenous populations (Gow 1993).

C.2. Matsigenka control of labor resources

Mestizos who attempt to impose Mestizo-typical labor norms on Matsigenka, risk losing their labor. While living in Atalaya, JAB witnessed a tour-boat owner warn one of his Matsigenka bow-men that he wouldn’t pay him if he caught him drinking at a local bar. Later, he attempted to convince two of his Matsigenka boat drivers to make a season-long commitment to him and become his permanent employees with healthcare and retirement benefits, while giving him more control over their professional lives. Within a few months, these Matsigenka men had switched to working as temporary employees of other Mestizo boat owners. In Boca Manu, a Mestizo who routinely contracted Matsigenka boarding school boys to help him haul wood on weekends, told me that he had to pay the Matsigenka a day early to be sure that no one else in town would hire them away at the last minute. Thus, in both Atalaya and Boca Manu, Matsigenka workers appear to be in demand, and Mestizos compete for their labor. They nearly always have the option of switching employers if they are dissatisfied with their current employer.

22

C.3. Amount vs. Type of Inter-Ethnic Interaction Experience

Figure C1. Mean posterior locations of all 161 interviewees on the latent axis, from an IRT model with a random effect for individual and no fixed-effect predictors (m1 in Table B1, above). Mestizos are open circles and Matsigenka are solid circles. The grey horizontal line is the overall population posterior mean location. Purple: two Matsigenka with approximately 15 years of inter-ethnic wage labor experience, and no inter-ethnic education experience. Orange: Matsigenka with a range of approximately 10 to 15 years of combined inter-ethnic education and wage labor experience. Notice that the two purple individuals, with extensive inter-ethnic experience but none in the domain of education, have more-positive values on the latent axis than all orange individuals, who have comparable amounts of inter-ethnic experience including education. This suggests that, for the Matsigenka, after controlling for the total amount of inter- ethnic interaction experience, the domain of inter-ethnic education has a stronger association with the Mestizo-typical meta-norm (more negative values on the latent axis) than does the domain of inter-ethnic wage labor. The blue point in the lower left is an atypical Matsigenka, with a very Mestizo-like meta-norm, who has four years of inter-ethnic wage labor experience and no inter-ethnic education experience. Interestingly, this individual is also unusual for being marginalized by many residents of Tayakome (for reasons withheld to preserve anonymity). Green: two Matsigenka individuals with very Matsigenka-typical norms (high values on the latent axis) who expressed desire and future plans to work as wage laborers with Mestizos. This suggests that these individuals are not concerned (or do not know) that their norms are very different from those of the average Mestizo.

REFERENCES

Bafumi, J., Gelman, A., Park, D. K., & Kaplan, N. (2005). Practical issues in implementing and understanding Bayesian ideal point estimation. Political Analysis, 13(2), 171-187.

Bernard, H. R. (2006). Research Methods in Anthropology: Qualitative and Quantitative Approaches (4ed.). Lanham: Altamira. de la Cadena, M. (1992). Las mujeres son más indias: etnicidad y género en una comunidad del Cuzco. Isis Internacional, Ediciones de las Mujeres (Santiago de Chile), 16, 25-71. 23

Fox, J.-P. (2010). Bayesian Item Response Modeling: Theory and Applications (Statistics for Social and Behavioral Sciences). New York: Springer.

Gow, P. (1993). Gringos and wild indians: images of history in western Amazonian cultures. L'Homme, 33(126), 327-347.

Hagen, E. H., & Hammerstein, P. (2006). Game theory and human evolution: a critique of some recent interpretations of experimental games. Theoretical Population Biology, 69, 339– 348.

Henrich, J. (2000). Does culture matter in economic behavior? Ultimatum Game bargaining among the Machiguenga of the Peruvian Amazon. American Economic Review, 90(4), 973-979.

Henrich, J., & Smith, N. (2004). Comparative experimental evidence from Machiguenga, Mapuche, Huinca, and American populations. In J. Henrich, R. Boyd, S. Bowles, C. Camerer, E. Fehr, & H. Gintis (Eds.), Foundations of Human Sociality: Economic Experiments and Ethnographic Evidence from Fifteen Small-Scale Societies (pp. 125- 167). Oxford: Oxford University Press.

Izquierdo, C. (2005). When ‘‘health’’ is not enough: societal, individual and biomedical assessments of well-being among the Matsigenka of the Peruvian Amazon. Social Science and Medicine, 61, 767-783.

Izquierdo, C., Johnson, A., & Shepard, G. H., Jr. (2008). Revenge, envy, and cultural change in an Amazonian society. In S. Beckerman, & P. Valentine (Eds.), Revenge in the Cultures of Lowland South America (pp. 162-186). Gainesville: University of Florida Press.

Jackman, S. (2001). Multidimensional analysis of roll call data via Bayesian simulation: Identification, estimation, inference, and model checking. Political Analysis, 9(3), 227- 241.

Johnson, A. (2003). Families of the Forest: The Matsigenka Indians of the Peruvian Amazon. Berkeley: University of California Press.

Levi, T., Shepard, G. H., Jr., Ohl-Schacherer, J., Peres, C. A., & Yu, D. W. (2009). Modelling the long-term sustainability of indigenous hunting in Manu National Park, Peru: landscape-scale management implications for Amazonia. Journal of Applied Ecology, 46, 804–814.

McElreath, R. (2016). Statistical Rethinking: A Bayesian Course with Examples in R and Stan (Texts in Statistical Science). Boca Raton: CRC Press.

Ochs, E., & Izquierdo, C. (2009). Responsibility in childhood: three developmental trajectories. Ethos, 37(4), 391-413. 24

Ohl-Schacherer, J., Shepard, G. H., Jr., Kaplan, H., Peres, C. A., Levi, T., & Yu, D. W. (2007). The sustainability of subsistence hunting by Matsigenka native communities in Manu National Park, Peru. Conservation Biology, 21(5), 1174-1185.

R Core Team (2014). R: a language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL http://www.R-project.org/.

Rosengren, D. (2004). Los Matsigenka. In F. Santos, & F. Barclay (Eds.), Guía Etnográfica de la Alta Amazonía (Vol. IV, pp. 1-157, Travaux de l'Institut Français d'Études Andines, Vol. 181). Balboa: Smithsonian Tropical Research Institute, Instituto Francés de Estudios Andinos.

Schacht, R., & Grote, M. (2015). Partner choice decision making and the integration of multiple cues. Evolution and Human Behavior, 36, 456-466.

Schenker, N., & Gentleman, J. F. (2001). On judging the significance of differences by examining the overlap between confidence intervals. The American Statistician, 55(3), 182-186.

Shepard, G. H. (2002). Three days for weeping: dreams, emotions, and death in the Peruvian Amazon. Medical Anthropology Quarterly, 16(2), 200-229.

Shepard, G. H., Rummenhoeller, K., Ohl, J., & Yu, D. W. (2010). Trouble in paradise: indigenous populations, anthropological policies, and biodiversity conservation in Manu National Park, Peru. Journal of Sustainable Forestry, 29, 252-301.

Stan Development Team (2016). RStan: the R interface to Stan, Version 2.14.1. http://mc- stan.org.