The content and structure of reputation domains across human societies: a view from the evolutionary social sciences

Abstract

Reputations are an essential feature of human sociality and the evolution of cooperation and group living. Much scholarship has focused on reputations, yet typically on a narrow range of domains (e.g. prosociality and aggressiveness), usually in isolation. Humans can develop reputations, however, from any collective information. We conducted exploratory analyses on the content, distribution and structure of reputation domain diversity across cultures, using the Human Relations Area Files ethnographic database. After coding ethnographic texts on reputations from 153 cultures, we used hierarchical modelling, cluster analysis and text analysis to provide an empirical view of reputation domains across societies. Findings suggest: (i) reputational domains vary cross-culturally, yet reputations for cultural conformity, prosociality, social status and neural capital are widespread; (ii) reputation domains are more variable for males than females; and (iii) particular reputation domains are interrelated, demonstrating a structure consistent with dimensions of human uniqueness. We label these features: cultural group unity, dominance, neural capital, sexuality, social and material success and supernatural healing. We highlight the need for future research on the evolution of cooperation and human sociality to consider a wider range of reputation domains, as well as their social, ecological and gender-specific variability.

This article is part of the theme issue ‘The language of cooperation: reputation and honest signalling’.

1. Introduction

Reputations are essential for human sociality. Whether used to punish norm violators in small communities or orient behaviour in anonymous online markets, reputations matter [1]. Reputations represent collective beliefs and evaluations a community forms about an individual's behavioural or emotional tendencies [2,3]. They function as currencies in a social marketplace with individuals signalling qualities relative to peers [4,5]. Such signals can reduce transaction costs in the formation, maintenance and termination of relationships by providing information about others without direct experience [6]. Because reputations can facilitate prosocial behaviour and punish deviancy [7], they provide some cognitive scaffolding supporting human sociality, including, the formation of status hierarchies [8,9], social institutions [10] and prosociality [11,12]. Many species rely on reputation-information exchange [13]. Among humans, however, language and gossip create a selective environment whereby reputations have significant social consequences [14–17].

Individual reputations can develop for any domain in which collective information exists on people's behavioural or emotional tendencies [18]. As new formats of social interaction emerge, the human behavioural repertoire becomes unbounded [19], suggesting an unlimited number of potential reputation domains. Nevertheless, evolutionary scholars have typically focused on a narrow range of reputation domains, such as prosociality [15,20–23], competency [3,24,25], aggressiveness [26–28] and sexuality [29,30]. This research has produced valuable insights on the influence of particular reputation domains on facets of social interaction [4], gendered relationships [30] and the evolution of social systems [31].

Research on reputations has remained agnostic, however, about the scope of reputation domains within societies, their frequency across cultures and potential gender biases [32]. Furthermore, research has often occurred in a piece-meal fashion focusing on a single domain, obfuscating the degree to which domains interact and shape behavioural responses as a suite of integrated parts (however, see [3]). Current scholarship lacks a clear understanding of the content, structure and diversity of reputation domains across societies.

We seek to build a foundation for comparative approaches to reputation domain diversity through exploratory analyses of the ethnographic record. We first derive a list of a priori reputation domains (discussed in the electronic supplementary material). We then assess the cross-cultural frequency of evidence for reputation domains and how evidence for gender-specific reputations varies. Lastly, we identify features of reputation domain co-occurrence and the semantic content of ethnography describing reputations. The following aims guide our study:

• — aim 1: to assess the distribution of ethnographic evidence for reputation domains;
• — aim 2: to identify if reputation domains exhibit systematic gender biases and if these domains vary within gender;
• — aim 3: to identify the structure and interrelationships of reputation domains; and
• — aim 4: to reveal the semantic content of reputation domains in the ethnographic record.

Existing work provides a strong rationale for both putative universalities in human reputation domains, as well as variation by social, ecological or gender-specific pressures. Using the ethnographic record in a systematic framework, despite limitations and potential biases (see Materials and Methods, Discussion), is a first step in uncovering patterns across human societies.

2. Material and methods

(a) . Ethnographic sample and coding

To accomplish our aims we relied on the electronic Human Relations Area Files (eHRAF)—an online database of primary ethnographic documents. It should be noted, the ethnographic record is male-biased given the majority of ethnographers have been men and their writings and observations have generally prioritized (deliberately or not) the behaviour and social lives of men [33,34].

The eHRAF includes thousands of documents from over 300 cultures indexed by subject at the paragraph-level [35]. Users can generate a sample of ethnographic texts (i.e. paragraphs) using Boolean searches of subject codes and/or keywords. Our dataset was compiled using a keyword and eHRAF's indexing system, the outline of cultural materials (OCM), which associates each paragraph with any of over 700 subject codes covering a range of topics relevant for the human sciences. We conducted an ‘Advanced Search’ of the keyword ‘reputation’ with any of the OCM subjects: social personality, personality traits, or status, role and prestige. This search aimed to strike a balance between retrieving a generalizable yet manageable sample of the ethnography of reputations. A limitation is that our search may have omitted particular domains of reputations. See the electronic supplementary material for additional details.

We read the resulting 1383 paragraphs for content, excluding those referencing reputations for groups, non-human entities or ethnographers. We applied these inclusion criteria because our goal is to understand individual reputations within a particular culture. We then aggregated paragraphs from the same document. This resulted in a dataset containing 319 documents from 153 diverse cultures with broad geographical coverage (see the electronic supplementary material, figure S1 and table S1). These documents had a mean word count of 140 (s.d. of 160 and range of 14 to 1957). We refer to this as our document dataset, which is publicly available in the reputation diversity database R package [36], including bibliographic information, culture sample and all data.

We derived, a priori, 20 reputation domains from the scientific literature on human sociality. These include aggressiveness, bravery, coercive ability, cooperation, cultural conformity, honesty, industriousness, material capital, medicine, neural capital, oration, parental care, prosociality, sexual fidelity, social capital, social status, sociosexuality, somatic capital, supernatural ability and teaching (see the electronic supplementary material for discussion on operationalization and inclusion). Each domain is operationalized as having both a positive and negative valence. For example, evidence for the reputation domain neural capital—which includes reputations for generalized or specialized intelligence, special knowledge or cognitive abilities—could be based on evidence that a given society values expertise, as well as evidence indicating that a group actively detests mental ineptitude (or vice versa). Using these operationalized reputation domains, we coded the 319 documents in the document dataset for supporting evidence across the 20 reputation domains. Authors decomposed into groups of two were allocated a subset of documents (approximately 106 per pair) to read and code, indicating supporting evidence for each domain and whether the evidence was gendered: male-specific, female-specific or gender neutral. We did not compute inter-coder reliability measures given coders varied in experience reading and coding ethnographic texts and common inter-rater reliability statistics can produce misleadingly low-reliability metrics despite relatively high levels of simple agreement for sparse matrices, such as our data [37,38]. Author-pairs compared coded data to resolve disagreements. For divergent codings, the text and operational definitions were reviewed and consensus reached on the appropriate coding. The aggregated resolved codings constitute our data. See the electronic supplementary material for example text and coding.

(b) . Data analysis

The current study is primarily exploratory. We rely on descriptive and exploratory statistical approaches to accomplish our aims. We assess the cross-cultural support for each reputation domain by estimating the proportion of documents providing supporting evidence, including across our gender coding (represented as a percentage estimate). Because our document dataset is a sample of the ethnographic record and because multiple documents often described the same culture, we incorporated uncertainty accounting for this non-independence and hierarchical structure with generalized linear mixed-effects regression models (GLMM) with random effects for culture using the lme4 package [39]. Some analyses are agnostic to gender-specific codings and any coding (female-specific, male-specific or gender neutral) counts as supporting evidence, while others account for gender-specific codings.

To estimate the frequency of supporting evidence for each reputation domain, we fit intercept-only GLMMs with random intercepts for culture, with each binary-coded reputation domain as outcomes, for all coded data (i.e. evidence for each reputation domain independent of gender-specific codings). We also fitted identical models for female-specific and male-specific evidence. These GLMMs estimate the proportion of documents providing evidence for the reputation domains (i.e. the fixed effect with 95% confidence interval (CI)) adjusting for the non-independence of documents from the same culture (aims 1 and 2). We also compute the percentage of cultures with at least one document providing supporting evidence for each reputation domain (independent of gender coding), with 95% CI estimated using a cluster bootstrap and 1000 samples with replacement (aim 1).

Using all data, where each row represents the gender-specific evidence for each document, we assess gender-biased evidence for each reputation domain by comparing (via information criterion model selection) an intercept-only GLMM (with random intercepts for documents nested within culture and for culture language family) to similar models which include a gender-term covariate (aim 2).

We rely on hierarchical cluster analysis to identify structure (i.e. features) among reputation domains (aim 3). We then use text-analytic methods and a document-term matrix (DTM) of our corpus of ethnography with penalized regression to identify semantic content predictive of evidence for reputation features (aim 4).

We also investigated sources of bias in our coded data owing to features of the ethnographic record. We used the presence of a female coauthor, document publication year and total pages of ethnography per culture in the eHRAF as predictors of our reputation domains (accounting for the hierarchical document-culture structure and culture language family).

All analyses were conducted with R v. 4.0.2 (22 June 2020).

3. Results

Evidence for reputation domains varied across subsistence types with horticulturalists and agriculturalists over-represented relative to pastoralists and hunter–gatherers (electronic supplementary material, table S1). Evidence was also male-biased. Of the 1252 counts of supporting evidence across domains, 695 (56%) were coded as male-specific, 418 (33%) were coded as gender neutral and 139 (11%) were coded as female-specific (electronic supplementary material, table S2).

Bias assessment analyses did not identify strong evidence of bias owing to our meta-ethnographic measures. Consequently, we did not incorporate such measures in analyses. See the electronic supplementary material for results.

(a) . Evidence for reputation domains

For all 20 reputation domains, we report the percentage of cultures that provided at least one count of supporting evidence, independent of gender coding. At the culture-level, the most strongly supported domains, documented in over 50% of cultures included cultural conformity, neural capital, prosociality and social status (figure 1a).

Figure 1.

Evidence for reputation domains. (a) Percentage of cultures providing at least one supporting document (95% CI estimated using a cluster bootstrap). (b) Percentage of documents providing supporting evidence (95% CI computed with intercept-only mixed-effects models). Purple circles: estimates from all data independent of gender coding. Green triangles: estimates from female-specific evidence. Yellow squares: estimates from male-specific evidence.

We report the proportion of documents that provided supporting evidence including gender-specific evidence (figure 1b). At the document-level, the most strongly supported domains, represented in over 30% of documents, included cultural conformity, prosociality, social status, neural capital and industriousness. Evidence for these reputation domains was strongly male-biased, in particular social status and neural capital which were the most supported male-specific domains. The most strongly supported female-specific reputation domains (although male-biased overall) were cultural conformity and industriousness. The between-domain variation among female-specific evidence was minimal compared to the male-specific evidence which was more variable. We emphasize the relatively low levels of female-specific evidence could be a feature of systemic male-bias in the ethnographic record, more so than gendered patterns of social or cultural diversity (see Discussion).

To assess gender biases in the supporting evidence for reputation domains, we fitted two binomial GLMMs of each reputation domain using the entire dataset, where each row represents the gender-specific evidence for each document (i.e. female-specific, male-specific, gender neutral; three rows per document). The first model was an intercept-only GLMM with the binary-coded reputation domains as outcomes and random intercepts for document nested within a culture (to account for the repeated measures of evidence type per document and multiple documents per culture) and a random intercept for culture language family (to partially account for shared ancestry). These intercept-only models were compared to similar models which included gender-evidence type as a covariate. We compared the intercept-only models to their respective gender-term models using Akaike information criterion (AIC) [40]. Gender was deemed to be a predictor of reputation domain evidence when $\mathrm{AIC}\mathrm{\Delta }<-2$ [41]. Results are reported in the electronic supplementary material, table S3 and support patterns in figure 1b. Evidence for all domains was male-biased with the following exceptions: sociosexuality, parental care and teaching did not demonstrate gender biases and sexual fidelity was female-biased. Two reputation domains (bravery and honesty) did not produce female-specific evidence and were not included.

(b) . Structural features of reputation domains

Evidence for different reputation domains may co-occur within documents, putatively suggesting domain interrelatedness and structure. To identify features (i.e. clusters) of domains we used agglomerative hierarchical cluster analysis. See the electronic supplementary material for details.

Figure 2 displays a dendrogram from cluster analysis of the 20 reputation domains, which includes two estimates of significance for how strongly each cluster is supported by the data. We rely on the approximately unbiased (AU) p-values (represented in red at each cluster's ‘edge’), which are computed by multiscale bootstrap resampling and represented as percentages (clusters with AU values greater than 95 are strongly supported; top-level clusters are automatically outlined by red rectangles). This revealed five strongly supported clusters we post hoc identify as sexuality, dominance, supernatural healing, social and material success, and cultural group unity. We used these clusters to compute new variables, henceforth reputation domain features. Although the cluster capturing neural capital and oration was only moderately supported (AU = 76), given neural capital was among the most frequent domains and oratory abilities are a type of neural capital we computed a neural capital feature from this cluster. For each document, these six reputation domain features are coded as 1, when any of the associated domains provided supporting evidence.

Figure 2.

Cluster analysis of reputation domains. Distances were $1-\mathrm{cor}$. Ward agglomeration method. AU p-values (red) computed with 10 000 bootstrap samples using the pvclust package [42]. Edge number in grey.

We estimated the percentage of cultures providing support for each reputation domain feature using the same cluster bootstrap methods used to estimate the culture-level support for domains. Supporting evidence for the cultural group unity and social and material success features was common across cultures, documented in 82% and 64% of cultures, respectively. Evidence for the neural capital feature was documented in 59% of cultures, the dominance feature in 53% of cultures, the supernatural healing feature in 44% of cultures and the sexuality feature in 23% of cultures (figure 3).

Figure 3.

Percentage of cultures providing evidence for reputation features (95% CI estimated via cluster bootstrap).

(c) . The ethnography of reputation domains

We used text analysis to explore the ethnography of reputation domains in reference to our six features. We created a DTM of all ‘informative’ words in our corpus of texts which captures the frequency of each unique term within each document. We fitted an elastic net logistic regression model (with the lasso penalty, $\alpha =1$) of each of the six features as a function of the frequencies of all 8770 unique words (using the glmnet package [43]). Words that were strong positive predictors epitomized the semantic content of documents which provided evidence for that feature. Figure 4 displays non-zero coefficients from elastic net lasso regression models of each reputation domain feature.

Figure 4.

Non-zero coefficients from text analysis elastic net regression models of evidence for reputation features. Coefficients indicate the words in each document which best predicted evidence for the feature. Positive coefficients as purple triangles. Negative coefficients as yellow circles. (a) Cultural group unity, (b) social and material success, (c) neural capital, (d) dominance, (e) supernatural healing and (f) sexuality.

Evidence for the cultural group unity feature was positively predicted by terms related to social relationships and community (e.g. family, person and wife) and negatively predicted by terms related to the supernatural (e.g. spirit and shaman). Evidence for the social and material success feature was positively predicted by wealth, prestige and terms for leadership and status. Evidence for the neural capital feature was positively predicted by skill, leader and village. Evidence for the dominance feature was positively predicted by war, strong, kill and physical implicating reputations for dominance with conflict, physical formidability and aggression. Negative predictors of the dominance feature included status and wealth, suggesting a distinction between dominance and prestige. Evidence for the supernatural healing feature was positively predicted by the terms, shaman, cure, power and medicine; woman was a weak negative predictor. Evidence for the sexuality feature was predicted by girl, woman and sexual.

4. Discussion

The content, structure and diversity of reputation domains across societies are understudied from a holistic perspective. The current study was motivated by a lack of cross-cultural research, despite widespread theorizing in biology, psychology and anthropology regarding the role of reputations for sociality and evolutionary dynamics. Using the eHRAF database, we extracted ethnographic accounts of individual-level reputation domains. Results suggest: (i) there is considerable cross-cultural variability in evidence for reputation domains—some domains are common in the ethnographic record (e.g. cultural conformity, prosociality) while others are relatively rare (e.g. teaching and honesty); (ii) evidence for most reputations are male-biased with male-specific reputation domains more variable than female-specific domains; and (iii) reputation domains cluster within six features: cultural group unity, dominance, neural capital, sexuality, social and material success, and supernatural healing. Below we interpret results from an evolutionary social science perspective.

(a) . Diversity in reputation domains

Most reputation domains (16 of 20) were documented in less than half of sampled cultures (figures 1a and 3). Despite variability, some were more common than others, including reputations for cultural conformity, prosociality, social status, neural capital and industriousness. These results are notable because of what is missing: ‘cooperation’. Evolutionary-oriented scholars have implicated cooperative reputations for explaining human ultrasociality [5,14,15], yet reputations for ‘cooperativeness’ were documented in only 23% of cultures. Reputations indirectly related to cooperation (e.g. conformity, honesty, social relationships and industriousness), however, were common across societies. Reputations for cooperation were also captured by the most common feature: cultural group unity (figure 3). The limited evidence of reputations for cooperativeness could be owing in part to the nature of the ethnographic record (see Limitations) or a product of our operational definition. We follow developmental and neuro-psychologists [25,44,45] by differentiating cooperation—defined as the likelihood an individual intentionally assists another in order to achieve a joint goal—from prosociality—defined as the likelihood one will invest in group welfare or act in group-altruistic ways (see the electronic supplementary material for discussion).

While we wish to avoid sweeping claims and emphasize the exploratory nature of our study, these results signal a need to expand research on reputations beyond cooperativeness, incorporating a variety of domains and examining their effect on sociality, particularly in experimental settings (sensu, [3]). Across cultures, distinct reputations capturing inter-individual variation in personality, experiences, capacities and reliability, probably underpin much of human sociality, including cooperativeness.

(b) . Gender differences in reputation domains

Evidence for most reputation domains was male-biased, and there was greater variance among male than female reputation domains (figure 1b). While this finding is consistent with research demonstrating that, across cultures, male social life is typically more public than female social life [46–48] we cannot disentangle male bias in ethnography from putative male biases in more overt sociality and reputation diversity. This male-biased pattern is consistent, however, with perspectives suggesting societies disproportionately channel opportunities to men to differentiate themselves, at the detriment of women who have fewer avenues to develop social capital [32,47,49,50]. As Rosaldo [48, pp. 393–394] suggests reviewing much ethnography, ‘the vast majority of opportunities for public influence and prestige, the ability to forge relationships, determine enmities, speak up in public, use or forswear the use of force are all recognized as men's privilege and right’.

Competition among women, however, has been suggested to be more indirect and reputation-based, compared to men [27,51], which would predict at least some female-specific reputation domains or limited variance between reputation domains of women and men [52]. Some empirical studies of gender differences in social influence among relatively egalitarian societies have found similarity in the weights of particular status-determining attributes between genders, despite male biases in overall influence [50,53]. Future comparative studies should more comprehensively define female-specific reputation domains and design targeted methods to document supporting evidence [54,55].

The only reputation domain more strongly associated with women than men was sexual fidelity; reputations for sociosexuality did not demonstrate gender bias (see [56] for similar results). These findings support evolutionary psychology models drawing on sexual selection theory which predict gender-specific evaluations related to reproductive strategies [29,51] and widespread male reproductive skew specific to influential men [38,57–59]. Overall, reputations related to sexuality were rare in our data. Sexuality may have been a taboo topic in some ethnographic contexts, but the ethnographic record includes rich descriptions of human sexuality [60,61]. It is possible that our search strategy did not capture much of the ethnography of reputations related to sexuality. Nonetheless, findings do not provide support for a universal psychology dedicated to evaluating female sexuality vis-à-vis males and do support perspectives emphasizing flexibility in reputations, strategies and norms related to sexuality [62].

(c) . Reputation domain structure and evolutionary theories

We find reputation domains are structured along six features which we termed cultural group unity, social and material success, neural capital, dominance, supernatural healing and sexuality. This data-driven, exploratory analysis comports well with theory from evolutionary psychology and the framework of human uniqueness in evolutionary anthropology.

Evolutionary psychologists examining the content of competitor derogation [29] have predicted men will often be evaluated for abilities to control resources necessary for status achievement, attracting mates and reproductive success. We find some support for this claim given the reputation features of dominance and social and material success. Additionally, evidence for the reputation domains social status, material capital and coercive ability were among the most male-biased domains (figure 1b). Status hierarchies shape the priority of access to resources and scholars have suggested they can be navigated through two distinct (though non-mutually exclusive) pathways: dominance and prestige [8,63,64]. These results support a distinction between dominance and social status or prestige [65,66], indicated by the cluster and text analyses (figures 3 and 4c,d).

Reputations for prestige (our social and material success feature) are associated with social networks as well as material resources, more so than reputations for dominance (figures 2 and 4b). These results are consistent with analyses among the Tsimane illustrating inter-relationships between status, social networks and social and material gains from cooperation with high-status individuals [67]. Results also support associations between reputations for dominance and coercion, physical aggression and conflict (figure 4d) [68]. Reputations for bravery were also captured by the dominance feature and cross-cultural research identified bravery as a universal feature of prosocial moral values [69]. Taken together, these results suggest reputations for social status and prestige are often associated with capacities for resource control while reputations for dominance may, in some contexts, be associated with prosocial investments [66,70–72].

The clustering of reputations for cooperation, prosociality, conformity, honesty, teaching and industriousness fits conceptions of the distinct nature of human social cognition, as well as fundamental structures of human groups. For example, scholars suggest human uniqueness relies on an evolved psychology dedicated to reasoning about others having cooperative and prosocial motivations [44,45,73]. These models suggest cultural conformity and learning biases lead to the evolution of well-structured groups and better equip groups to compete with others groups [11,74,75]. Such between-group competitive dynamics can occur through altruistic provisioning of group members or through intergroup violence [76,77] and can in turn, further support within-group cooperation [78,79].

Lastly, the supernatural healing feature is associated with unique features of the human niche (i.e. religion) and fits long-standing anthropological notions about the important role of religious practitioners (e.g. shamans) who manipulate the supernatural to provide benefits for and impose costs on group members [71,80–82].

5. Limitations

Our study has several limitations. First, our data are limited to the content ethnographers recorded and published. Information on reputations that the ethnographer was unaware of, not interested in, nor permitted to research, constrains available data. Therefore, while we can conclude the widespread ethnographic evidence of some reputation domains probably indicates their cross-cultural importance, we cannot conclude reputation domains lacking substantial evidence are indeed rare across cultures. Additionally, the terms an ethnographer uses for reputation domains may reflect their worldview (etic), rather than the worldview of the focus population (emic). We attempted to assess potential biases in our data owing to meta-ethnographic measures (see the electronic supplementary material); however, it is possible other features of ethnography or ethnographers influenced results.

Ethnographic materials related to the social, economic and cultural lives of women are systematically underreported, especially in the early history of the field [32,33,83]. Thus, the extent to which women have fewer avenues for gaining reputations cross-culturally remains unclear and cannot be evaluated via these methods. However, the evidence of gender biases we discovered comport with the common notion that patriarchy is pervasive globally and negatively impacts women's ability to achieve recognition, political power, economic capital and autonomy (see [84]).

We identified the 20 reputation domains a priori, drawing on the literature on human uniqueness and sexual selection theory, which itself is likely to be biased by authors and general biases across the human sciences. While a useful starting point for exploring reputational diversity, we imagine that other domains could exist. Lastly, we constrained our eHRAF search using the keyword ‘reputation’, which could have missed other content on reputations that used adjacent language (e.g. personality, gossip). Recognizing these limitations, these results provide greater cross-cultural validity to existing theories of reputation and can spark future empirical and theoretical work better incorporating the cultural diversity, structure and gendered dimensions of reputation domains.

6. Conclusion

Reputations are a critical component of human social life and have fundamental implications for human evolution. From a socio-structural perspective, reputations are the pathways by which societies evaluate individuals and are the mechanisms through which individuals can distinguish themselves. Despite their centrality to much of human sociality, little systematic cross-cultural research exists on the content and structure of reputation domains. We find that ethnographic evidence for reputations is variable across societies, tends to focus on cultural conformity and prosociality, displays large gender biases with greater variance among males and is structured around themes related to human uniqueness.

Drawing on Chapais’ [85] distinction between context-independent versus context-dependent human universals, we hypothesize reputations for cultural group unity will be a context-independent universal, likely to manifest in all human societies, whereas reputations for social and material success, neural capital, and dominance are more likely to be context-dependent universals, promoted or suppressed by socio-ecological or cultural evolutionary processes [86].

Data accessibility

Original data produced for the current study are available in the archived reputationdiversitydata R package at https://doi.org/10.5281/zenodo.4740791 and at https://github.com/zhgarfield/reputationdiversitydata.

Authors' contributions

S.J.M. conceived of the project. S.J.M. and Z.H.G. designed the coding scheme and methods. S.J.M., Z.H.G. and R.S. wrote the manuscript. E.R.P. and D.I. collected the data. S.J.M., Z.H.G., R.S., E.R.P., D.I. and A.U. processed the data. Z.H.G. performed all statistical analyses.

Competing interests

We declare we have no competing interests

Funding

This work was supported in part by the Global Change and Sustainability Center and the Office of Undergraduate Research at the University of Utah. Z.H.G. acknowledges IAST funding from the French National Research Agency (ANR) under the Investments for the Future (Investissements d'Avenir) program, grant no. ANR-17-EURE-0010.