Relative wealth and inequality associate with health in a small-scale subsistence society

In high-income countries, relative wealth and inequality affect health by causing psychosocial stress. We test this hypothesis in a small-scale subsistence society, the Tsimane. We associated relative household wealth (n=1003) and community-level wealth inequality (n=35, Gini = 0.15 - 0.43) with a range of psychosocial and health outcomes (depressive symptoms [n=663], social conflicts [n=393], non-social problems [n=390], social support [n=392], cortisol [n=828], BMI [n=9378], blood pressure [n=1614]), self-rated health [n=809], morbidities [n=3140]) controlling for absolute wealth, age, sex, community size, distance to town and relevant random effects. Relative wealth and inequality were associated with self-rated health and morbidity, especially respiratory disease, the leading cause of mortality in the Tsimane. Inequality was also associated with higher blood pressure. However, psychosocial stress did not mediate these associations, suggesting other mechanisms. These findings are consistent with socio-economic hierarchies affecting some health outcomes in any society, while others might be exacerbated in high-income countries.


Abstract 21
In high-income countries, relative wealth and inequality may affect health by causing psychosocial 22 stress. We test this hypothesis in a small-scale subsistence society, the Tsimane. We associated relative 23 household wealth (n=1003) and community-level wealth inequality (n=35, Gini = 0.15 -0.43) with a 24 range of psychosocial and health outcomes ( community size, distance to town and relevant random effects. Relative wealth and inequality were 28 associated with self-rated health and morbidity, especially respiratory disease, the leading cause of 29 mortality in the Tsimane. Inequality was also associated with higher blood pressure. However, 30 psychosocial stress did not mediate these associations, suggesting other mechanisms. These findings are 31 consistent with socio-economic hierarchies affecting some health outcomes in any society, while others 32 might be exacerbated in high-income countries. 33

Introduction 34
It is relatively uncontroversial that people with greater access to resources -usually operationalized as 35 income, wealth, or broader indicators of socio-economic position, rank or status 1 -should be in better 36 health, as resources can be converted into better nutritional status, access to health care, or insulation 37 against health risks. Such benefits of absolute rank are also commonly found in non-human primates 38 Furthermore, studies have found that the steepness of socio-economic hierarchies (i.e. income 47 or wealth inequality) is associated with both physical and mental health outcomes -including self-rated 48 health, all-cause mortality, heart disease, respiratory disease, obesity, or homicide -independent of the 49 effects of absolute wealth (Nowatzki, 2012; Pickett and Wilkinson, 2015; Wilkinson and Pickett, 2006). , a formal meta-analysis did find significant 53 associations between inequality and mortality or self-rated health in high-income countries (Kondo et 54 and not just for low-rankers: depending on how rank is achieved and maintained, high-or low-ranking 79 individuals may be more stressed (Abbott et al., 2003;Sapolsky, 2005). Crucial to who is stressed is the 80 availability of social support, which can be as or even more important for health and fitness as rank per 81 se (Sapolsky, 2005; Snyder-Mackler et al., 2020). Other factors primarily impact low-ranking individuals: 82 in many primate (and some human) societies, subordinates are regularly subjected to aggression and 83 intimidation by higher-ranking individuals (Silk, 2003), resulting in the lack of control and learned 84 helplessness that often cause depression (Sapolsky, 2005(Sapolsky, , 2004. Greater inequality, i.e. steeper 85 hierarchies entail more skewed payoff distributions, favoring more intense competition and risk-taking, 86 especially among low-ranking individuals; this is argued to explain the persistent association between 87 income inequality and homicide rates, as most homicides result from escalated contests over status 88 (Daly andWilson, 1997, 1988  negative health consequences can result from perpetual status-striving, the distribution of social 97 support, from lack of control and learned helplessness, from intensified competition especially among 98 low-ranking individuals, and perhaps from generally "faster" life histories or developmental constraints. 99 In summary, the argument for why and how being low in a hierarchy negatively influence health 100 is that humans, much like other primates, are sensitive to their relative rank and the distribution of 101 fitness outcomes, and that we adjust our behaviors and physiological responses accordingly. Several 102 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 12, 2020. . https://doi.org/10.1101/2020.06.11.20121889 doi: medRxiv preprint open questions remain, however. First, the inequality hypothesis remains hotly debated, and testing it 103 requires careful statistical methods. Second, it remains unclear to what extent the observed health 104 consequences of relative status and possibly inequality in high-income countries represent (i) tradeoffs 105 of potentially adaptive responses to lower relative rank and/or to inequality, or (ii) are caused by 106 evolutionary mismatch, i.e. novel conditions that cause maladaptive outcomes. If the health 107 consequences stem from tradeoffs to adaptive responses, e.g. people taking more risks and making 108 Further, many traditional societies have immune systems that are well calibrated by frequent exposure 120 to pathogens and microbiota, and predominantly experience acute responses to infections (Blackwell et 121 al., 2016a;McDade, 2005), unlike the chronic low-grade inflammation that links stress to hypertension, 122 cardiovascular disease, and depression in high-income countries . Lastly, 123 competition (for mates, resources, etc.) in such societies is usually fairly local, meaning that the scale at 124 which relative rank and inequality should be measured is more obvious than in large-scale modern 125 societies with television and social media, where people are simultaneously part of many hierarchies. 126 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 12, 2020. the population studied here, traditional forms of status generally supported a status-health gradient, 136 but studies on income or wealth showed mixed results. In a sample of four communities, politically 137 more influential men had lower cortisol and a lower incidence of respiratory infection, though there 138 were also many null results, and higher income was associated with higher cortisol (von Rueden et al., 139 2014). Across 13 Tsimane villages, relative wealth was associated with better self-reported health 140 (Undurraga et al., 2010); however, average self-reported health was lower in wealthier villages. In a 141 larger sample of Tsimane villages, relative income associated with lower (higher) BMI among individuals 142 with smaller (larger) support networks (Brabec et al., 2007). While results are mixed, there is some 143 converging evidence that suggests market integration generates psychosocial stress in traditional 144

societies. 145
In terms of the relationship between inequality and health, studies among Tsimane have also 146 shown mixed results. One study found no association between income inequality and body fat (Godoy 147 et al., 2005), but income inequality was associated with more negative emotions (Godoy et al., 2006). 148 Greater village wealth inequality did not associate with self-reported health in one study (Undurraga et 149 al., 2010) but associated with better self-reported health and lower self-reported stress in another, 150 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 12, 2020. . https://doi.org/10.1101/2020.06.11.20121889 doi: medRxiv preprint controlling for individual and village wealth level (Undurraga et al., 2016). Overall, these results provide 151 mixed evidence for associations between inequality and health. 152 Here we test for links between hierarchy and health among the Tsimane, expanding upon 153 previous studies in several ways. First, we simultaneously assess the effects of relative resource access 154 and resource inequality, controlling for absolute resource access (measured as relative wealth, wealth 155 inequality, and absolute wealth, respectively), thus addressing major methodological critiques of the 156 If, on the other hand, hierarchy-health associations in high-income countries are predominantly 180 caused by evolutionary mismatch stemming from the combination of a) rigid socio-economic 181 hierarchies, and b) conditions favoring chronic disease (e.g. obesogenic diets, sedentary lifestyle, chronic 182 inflammation) and psychosocial stress (e.g. residential isolation from kin), we predict that among the 183 Tsimane: 184 P4a: Associations between relative wealth or inequality and health will be absent or inconsistent 185 and not necessarily mediated by psychosocial stress 186 P4b: Associations between relative wealth or inequality and health will only be found for 187 outcomes associated with chronic disease risk (e.g. BMI, blood pressure), but not necessarily with 188 acute illness such as infectious disease 189 These predictions are based on the assumptions that, while material wealth is a novel source of status 190 among the Tsimane, it has not yet led to the kind of rigid socio-economic hierarchies typical of 191 industrialized populations; and that the relationship between hierarchy and BMI or blood pressure is 192 relatively linear, i.e. that these measures will respond to hierarchy even at levels that are not yet 193 associated with chronic disease. Table 1 gives an overview of all variables and their predicted 194 associations with wealth and inequality. In addition to the predictions listed in the table, testing P2b 195 involves exploring interactions between inequality and relative wealth, and testing P3 involves adding 196 psychosocial variables as covariates in models of health outcomes (see Methods). 197 198 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 12, 2020. . https://doi.org/10.1101/2020.06.11.20121889 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 12, 2020.

Results 227
Wealth varied considerably by age (Fig. 1A), hence we used an age-corrected measure of relative wealth 228 (see Methods) that reflects one's trajectory along this age gradient. At the high end of the wealth 229 distribution (Fig. 1B), much of the variation was driven by livestock, especially cattle, which were 230 introduced to the region by ranchers but are only owned by a small minority of Tsimane. There was 231 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 12, 2020. . https://doi.org/10.1101/2020.06.11.20121889 doi: medRxiv preprint substantial variation in mean wealth and wealth inequality among the study communities ( Fig. 1C-D). 232 Mean wealth was generally lowest in communities located in the interior forest ( Fig 1C, bottom right), 233 which is remote and inaccessible by road for much of the year (due to washed out bridges), and those 234 downriver from San Borja (Fig 1C, top), which experience frequent flooding and are within or adjacent to 235 a protected bioreserve that limits resource extraction to residents. We operationalized inequality by 236 calculating a village-level Gini coefficient for wealth (see Methods). Wealth inequality was generally 237 lower in communities farther from the market towns of San Borja and Yucumo, where Tsimane can sell 238 produce and purchase market goods, though some villages near towns also show low inequality ( Fig 1D). 239 240 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 12, 2020. . https://doi.org/10.1101/2020.06.11.20121889 doi: medRxiv preprint uncertain. There was no support for an association with social conflicts. In terms of health, relative 266 wealth was associated with lower total morbidity (β=-0.03, P<0=0.92), and lower odds of respiratory 267 illness (OR=0.88, P<1=0.98). The association with self-rated health (β=0.03, P>0=0.78) was highly 268 uncertain but in the expected direction. The only detrimental association was with gastrointestinal 269 illness, which was more likely in relatively wealthier people (OR=1.07, P>1=0.93). The were no or very 270 weak associations between relative wealth and BMI, systolic and diastolic blood pressure, and other 271 infections. However, absolute wealth was strongly associated with lower systolic (β=-0.32, P<0=0.99) and 272 diastolic (β=-0.38, P<0=0.98) blood pressure ( Figure S1, Tables S7 & S8); people with higher absolute 273 wealth also had more labor partners (β=0.33, P>0=0.88). Other than blood pressure and labor partners, 274 absolute wealth was not associated with any outcomes, reaffirming the notion that relative rank matters 275 more than absolute resource access. 276 To examine whether psychosocial variables mediated associations between relative wealth and 277 health (P3), we included psychosocial variables (depression, non-social problems, and cortisol levels) as 278 covariates in models of blood pressure, self-rated health and morbidities (Tables S8-S14). This did not 279 change the strength or certainty of wealth-health associations in any meaningful way, despite several 280 psychosocial variables being themselves strongly associated with the health outcomes and sometimes 281 substantially improving goodness of fit of these models. Specifically, depression was strongly associated 282 with worse self-rated health (β=-0.24, P<0=1.00), more total morbidities (β=0.12, P>0=1.00), and a greater 283 risk of respiratory (OR=1.21, P>1=0.85) and gastrointestinal illness (OR=1.26, P>1=0.92). Having fewer 284 non-social problems was associated with reduced total morbidity (β=-0.10, P<0=1.00) and a reduced risk 285 of gastrointestinal illness (OR=0.70, P<1=1.00); each SD increase in cortisol associated with a greatly 286 increased risk of respiratory illness (OR=2.61, P>1=1.00). This indicates that psychosocial variables are 287 indeed associated with health outcomes (Stieglitz et al., 2015), but independently of wealth. 288 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 12, 2020. . https://doi.org/10.1101/2020.06.11.20121889 doi: medRxiv preprint distribution that supports an association between wealth and the outcome (P>0 or P<0). All predictions 293 control for age, sex, inequality, distance to market town, community size, and mean community wealth, 294 holding all other variables at the mean, with sex=female. For the first two rows, the outcomes are 295 measured as Z scores, the bottom row as probabilities. Rough categories of dependent variables 296 (psychosocial, continuous health outcomes, and binary health outcomes) are distinguished by rows and 297 colors. 298 299 For wealth inequality (Fig 3, Tables S2-S14), associations with psychosocial variables were mostly 300 negligible, with the strongest being fewer non-social problems in more unequal communities (β=-0.15, 301 P<0=0.92). However, wealth inequality was associated with some detrimental health outcomes, 302 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 12, 2020. . https://doi.org/10.1101/2020.06.11.20121889 doi: medRxiv preprint association between inequality and the outcome. All predictions control for age, sex, household wealth, 316 distance to market town, community size, and mean community wealth, holding all other variables at 317 the mean, with sex=female. For the first two rows, the outcomes are measured as Z scores, the bottom 318 row as probabilities. Rough categories of dependent variables (psychosocial, continuous health 319 outcomes, and binary health outcomes) are distinguished by rows and colors. 320 321 Finally, we conducted several post-hoc tests to examine whether wealth-health associations 322 were contingent on sex, or whether relative wealth effects were contingent on levels of inequality and 323 vice versa. For example, inequality could trigger increased stress and competitiveness only in men given 324 a history of higher reproductive skew in males (Daly and Wilson, 1997Wilson, , 1988) and inequality might affect 325 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 12, 2020. . https://doi.org/10.1101/2020.06.11.20121889 doi: medRxiv preprint the wealthier and poorer differently (P2b), i.e. poorer individuals may fare even worse in more unequal 326 contexts. For this we included wealth x inequality, wealth x sex, or inequality x sex interactions. Several 327 associations were contingent (Table S15). In particular, wealth or gini associations with depression, 328 cortisol, diastolic blood pressure, infections, respiratory and gastrointestinal illness varied by sex, albeit 329 not consistently. While only the labor partners and diastolic blood pressure models favored a gini x 330 wealth interaction. Thus, inequality seems to generally affect the wealthier and poorer equally, contra 331 P2b, while sometimes men and sometimes women are more affected by wealth or inequality. 332

Discussion 333
We tested whether relative wealth and wealth inequality were associated with a broad range of 334 psychosocial and health outcomes, independently of absolute wealth, in a small-scale subsistence 335 society. We found that relative wealth within a community and community-level wealth inequality were 336 generally more strongly associated with outcomes than absolute wealth, which clearly indicates relative 337 position in a hierarchy matters above and beyond absolute access to resources, and reaffirms our 338 decision to treat the community as the relevant scale of analysis. Indeed, absolute wealth was hardly 339 associated with any outcomes ( Figure S1), likely because the most common health risks affect everyone 340 similarly within a community. 341 Consistent with the hypothesis that higher relative position in a socio-economic hierarchy 342 improves outcomes and the steepness of that hierarchy worsens them (P1 and P2a), wealthier people 343 had better psychosocial outcomes, fewer morbidities and a reduced risk of respiratory illness, while 344 people in more unequal communities suffered from worse self-rated health and an increased risk of 345 respiratory illness. Similarly, von Rueden et al (2014) found lower risk of respiratory infection among 346 influential men. Thus, converging results implicate status in respiratory morbidity among the Tsimane. 347 Given that respiratory illness is the leading cause of mortality at all ages in this population (Gurven et al.,348 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 12, 2020. Despite some clear hierarchy-health associations, we found limited support for the hypothesis 361 that these associations were mediated by psychosocial stress (P3). Associations between wealth and 362 psychosocial outcomes were generally weaker and, importantly, none of the wealth-health associations 363 were altered when including psychosocial measures, including cortisol levels, as covariates (even though 364 cortisol was itself associated with some health outcomes). In contrast, a study of four Tsimane 365 communities found that influential men with greater social support had lower cortisol (von Rueden et 366 al., 2014), and higher cash income associated with higher cortisol in two traditional societies (Konečná 367 and Urlacher, 2017; von Rueden et al., 2014). In another study of the Tsimane, higher incomes predicted 368 lower BMIs, unless individuals had relatively more social support (Brabec et al., 2007). For traditional 369 societies experiencing market integration, whether relative status increases, decreases, or has no effect 370 on stress and health may depend on the status measure and its association with social support. It 371 therefore remains unclear what mechanisms were responsible for the wealth-health associations found 372 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 12, 2020. . https://doi.org/10.1101/2020.06.11.20121889 doi: medRxiv preprint here, though hierarchy is known to affect immune function, and thereby infectious disease morbidity 373 somewhat independently of HPA activity (Aiello et al., 2018;Snyder-Mackler et al., 2020. 374 Arguably, our results also lend some support to the hypothesis (P4) that hierarchy-health 375 associations in high-income countries are due to, or at least exacerbated by evolutionary mismatch 376 (Sapolsky, 2004). Specifically, wealth-health associations were somewhat inconsistent (P4a), with some 377 associations being negligible or counter to predictions. One of the strongest associations of inequality 378 was with blood pressure, a major contributor to chronic disease (P4b). While the vast majority of  , 2012)). Thus, assuming that the associations between hierarchy and blood pressure persist 394 across a wider range, we found support for the prediction (P4b) that hierarchy contributes to chronic 395 disease risk. 396 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 12, 2020. . https://doi.org/10.1101/2020.06.11.20121889 doi: medRxiv preprint In sum, we present the most comprehensive test of hierarchy-health associations in a traditional 397 society to date. Thanks to more detailed data and better methods than those used in many studies in 398 high-income countries, we also provided a stronger test of the relative wealth and inequality hypotheses 399 from local organic materials (e.g. canoes, bows and arrows), market goods, i.e. industrially produced 419 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 12, 2020. . https://doi.org/10.1101/2020.06.11.20121889 doi: medRxiv preprint items obtained through trade or purchase (e.g. bicycles, motorbikes), and livestock (e.g. pigs, cows), 420 which were subsequently converted into their local market value in Bolivianos and summed (Fig. 1). 421 Objective household wealth arguably provides only an indirect measure of people's subjective 422 wealth and status (Norton, 2013) but these data were most widely available for this study. unclear whether that will be the case given low residential mobility and concentration of work and 440 socializing within communities. However, Tsimane will occasionally visit families in other communities 441 and sporadically engage in market-based interaction with non-Tsimane, and comparisons with wealther 442 neighbors can contribute to Tsimane status aspirations (Schultz, 2019). Nevertheless, as mentioned 443 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 12, 2020. . https://doi.org/10.1101/2020.06.11.20121889 doi: medRxiv preprint above (Study population), we consider the community to be the most relevant arena for status 444 competition among Tsimane. Note that most studies on health effects of inequality use income 445 inequality (but see Nowatzki, 2012), which is less unequally distributed than wealth. Cash income among 446 the Tsimane during this study period was sporadic and many households may have no income in a given 447 sampling period, which leads to overestimated Ginis. We therefore preferred wealth and wealth 448 inequality as a more reliable measure of households' long-term access to resources and its distribution. 449 Psychosocial and health variables. Depressive symptoms were measured using an adapted 18-item questionnaire (Stieglitz et al.,459 2014), the responses to which were summed to yield an overall depression score. The same interview 460 also asked whether participants experienced conflicts with several kinds of social partners as well as 461 non-social problems, such as food insecurity, illness, or debt; affirmative answers were summed to yield 462 a composite measure of social conflicts and non-social problems, respectively. A household's 463 cooperation network was measured as the number of people from different households who helped in 464 that household's fields in a given year. 465 Morbidity was assessed during regular medical check-ups using the ICD-10 classification 466 (International Classification of Disease, 10 th edition) and then grouped into 18 clinically meaningful 467 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 12, 2020. . https://doi.org/10.1101/2020.06.11.20121889 doi: medRxiv preprint categories following the Clinical Classifications System (https://www.hcup-468 us.ahrq.gov/toolssoftware/ccs/ccsfactsheet.jsp); morbidities in any of these categories were summed to 469 give a total morbidity score potentially ranging from 0 (no morbidities) to 18 (at least one morbidity in 470 each category). In addition, we also examined the presence/absence of infectious and parasitic diseases 471 (CCS 1, hereafter "infections"), diseases of the respiratory system (CCS 8, "respiratory illness") and 472 diseases of the digestive system (CCS 9, "gastrointestinal illness"), which represent the most common 473 causes of morbidity and mortality in this population . See Table S1 for examples of 474 the six most common diagnoses in these three categories. Self-rated general health was measured at 475 each check-up using a five-point scale from ("very bad" [1] to "excellent" [5]). Distance to the town of 476 San Borja was measured as nearest route (whether by river or road) from the center of the community 477 and provides a proxy for access to modern amenities. 478

Data analysis 479
Prior to analysis, all variables were transformed into z-scores (in case of cortisol levels after log 480 transformation due to their skewed distribution) such that coefficients were standardized across 481 outcomes for comparability. All outcomes were modeled as Gaussian, except the presence/absence of 482 specific morbidities (Bernoulli). Each analysis modeled an individual-level outcome as a function of 483 individual-, household-, and community-level characteristics (Table 1). Thus, we fit the following base 484 model for each outcome: 485 Outcomeijkl ~ β0 + (β1 * Sexj) + (β2 * Agej)+ (β3 * relative household wealthk) + (β4 * Community-level 486 Ginil) + (β5 * Community-level mean wealthl) + (β6 * Community Sizel) + (β7 * Distance of community to 487 market townl) + uj + uk + ul + eijkl 488 wherein the subscripts denote measurement i, individual j, household k, and community l, respectively. 489 β0 is the intercept, all other β's are slopes, u's are random intercepts, and e is the residual error (not 490 available for Bernoulli responses). Variance inflation factors indicated virtually no collinearity among 491 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 12, 2020. . https://doi.org/10.1101/2020.06.11.20121889 doi: medRxiv preprint predictors (all VIFs <3). After fitting the models, the β for absolute wealth was calculated as the sum of 492 the posterior samples for β3 and β5. 493 In order to test whether potential wealth-health associations were mediated by psychosocial 494 stress we re-ran all health models (blood pressure, self-rated health, total morbidity, infections, 495 respiratory and gastrointestinal illness) with pertinent psychosocial variables as covariates and checked 496 whether this changed the associations with wealth and inequality. In particular, we included depression, 497 non-social problems and cortisol levels as covariates as they had the greatest overlap with the health 498 samples. Furthermore, PCAs revealed that depression was capturing up to 90% of the variation in 499 psychosocial stress. Because of missing values in the psychosocial covariates, we used Bayesian 500 imputation while fitting these models, which makes no additional assumptions on missingness 501 compared to complete-case analysis, preserves uncertainty and leads to less biased estimates 502 (McElreath, 2016). In addition, we also ran a series of exploratory analyses in which we added 503 interaction terms. We reported interactions if they improved the expected log pointwise predictive 504 density [ELPD] in a stepwise forward selection procedure. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 12, 2020. . measure are given in the supplementary materials (Tables S2-S14). All data and R code will be made 516 available at https://github.com/adrianjaeggi/tsimanewealthandhealth. 517

Acknowledgements 518
We thank the Tsimane for their generous participation and years of collaboration, and THLHP personnel 519 for their herculean efforts and dedication in data collection. We also thank the Santa Fe Institute's 520 working group on wealth inequality in small-scale societies, led by Monique Borgerhoff Mulder and Sam 521 Bowles, for stimulating discussions. 522 Bürkner P-C. 2017. brms: An R package for Bayesian multilevel models using Stan. J Stat Softw 80. 560 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 12, 2020. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 12, 2020. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 12, 2020.  Figure S1: Predicted associations between absolute wealth and all outcomes 806 Lines are posterior means and shaded areas are 95% credible intervals. Numbers in each panel 807 represent the posterior probability, i.e. the proportion of the posterior that supports an association 808 between inequality and the outcome. All predictions hold all other variables at the mean, with 809 sex=female. For the first two rows, the outcomes are measured as Z scores, the bottom row as 810 probabilities. Rough categories of dependent variables (psychosocial, continuous health outcomes, and 811 binary health outcomes) are distinguished by rows and colors. 812 813 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 12, 2020. . https://doi.org/10.1101/2020.06.11.20121889 doi: medRxiv preprint Tables   814   Table S1: Overview of most common morbidities 815 Three of the most common Clinical Classification Systems categories (CCS, number in parentheses) and 816 the 6 most prevalent diagnoses within each category (in decreasing order down rows, ICD-10 codes in 817 parentheses). Musculoskeletal conditions (CCS 13) were also common but not analyzed independently 818   is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 12, 2020.    is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 12, 2020.   is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 12, 2020.    is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 12, 2020. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 12, 2020. . https://doi.org/10.1101/2020.06.11.20121889 doi: medRxiv preprint  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 12, 2020. . https://doi.org/10.1101/2020.06.11.20121889 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 12, 2020. . https://doi.org/10.1101/2020.06.11.20121889 doi: medRxiv preprint Gini-BP association much stronger in richer people Sex * Gini Gini-BP association only seen in women Self-rated health -Total morbidity -Infections Sex * Gini Women slightly higher than men in more equal places Respiratory illness Sex * Wealth Wealth-health association only seen in women Gastrointestinal illness Sex * Gini Gini effect slightly stronger for men, weak support a Note that stepwise model selection was not computationally feasible for BMI because robust 886 comparisons took days to weeks to run. A comparison of just the base model and a model with all 887 possible interactions favored the base model 888 889 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 12, 2020. . https://doi.org/10.1101/2020.06.11.20121889 doi: medRxiv preprint