How much cultural variation around the globe is explained by ecology?

Abstract

How much cultural variation is explained by the physical and social ecologies people inhabit? Here, we provide an answer using nine ecological variables and 66 cultural variables (including personality traits, values and norms) drawn from the EcoCultural Dataset. We generate a range of estimates by using different statistical metrics (e.g. current levels, average levels across time, unpredictability across time) of each of the ecological variables. Our results suggest that, on average, ecology explains a substantial amount of human cultural variation above and beyond spatial and cultural autocorrelation. The amount of variation explained depended on the metrics used, with current levels and average levels of ecological conditions explaining the greatest amounts of variance in human culture on average (16% and 20%, respectively).

1. Introduction

In some parts of the world, premarital sex is common, but in others it is rare and a criminal offence [1]. The Reddy caste in India forbids eating beef, yet, within the same state, the Didayi tribe openly eats beef at festivals [2]. Romanians prefer to stand over a metre away from strangers, but right across the border, Bulgarians prefer to stand at nearly half that distance [3]. Americans tend to view the self as independent from others, whereas the self in Japan is viewed as interconnected and overlapping with close friends and family [4]. Around the world, people change their behaviour to fit in with the group, but such conformity effects are weaker in North America and Western Europe than in other world regions [5]. Indeed, societies and individuals vary in many ways including their rules for behaviour [6–10], behavioural tendencies [11,12], preferences [3,13–15], values [16–19], motivations [20,21] and political institutions [22,23].

Where does such human cultural variation come from? One account holds that cultural variation is driven, at least in part, by the ecologies populations of individuals inhabit [24–27]. In this view, ‘ecology’ is defined broadly to include physical geography, climate (e.g. temperature, rainfall), resource availability, infectious diseases, and a range of social ecological dimensions such as the ratio of males to females, resource inequality, rates of interpersonal and intergroup violence and population density. Scholars dating back to at least Steward [28] have argued that much of cultural variation is the product of such ecological conditions, and that populations of individuals inhabiting similar ecologies should show similar cultural patterns [29–32].

Compatible with this account, past research has linked a number of specific ecological dimensions to cultural variation. For example, the prevalence of infectious disease has been linked to societal variation in conformity [33–35], individualism [36,37], openness to experience [38], violence [39] and in-group favouring preferences [40–42]. Other work has linked variation in population density to fertility [26,43] and the strength of social norms [7]. Yet other work has documented correlations between countries' levels of resources and numerous cultural variables, ranging from individualism [44,45], to violence [46–48], to subjective well-being [49,50]. Links have also been found between levels of rainfall and religious beliefs [51]. Yet other research has shown relationships between markers that may serve as cues of relatedness within populations, such as ancestral diversity or the prevalence of cousin marriage, are linked to cultural differences in emotional expression and regulation [52] and willingness to risk one's life for the group [53].

How might ecology exert such influences on human culture? Although not the focus of the present project, here we briefly describe three influential accounts of how such effects may occur: evoked culture, cultural evolutionary theory and socioecological theory. First, some recent accounts have argued that such variation might emerge from the aggregate responses of individuals adapting to the local environment they inhabit [24,26,54]. In this view, culture may be ‘evoked’ by specific environmental affordances that promote particular behavioural and cognitive adaptations [24,55]. From this perspective, humans can be viewed as analogous to jukeboxes, placed in different locales. Each contains the same set of songs so to speak, but placed in different environments, they will be asked to play different songs; placed in similar environments, they will probably be asked to play the same songs. Unlike song choice on a jukebox, these evoked psychological and behavioural responses are generally adaptive for humans. Thus, if a group of humans share an environment, which evokes similar responses within them, then one might expect groups of humans to exhibit systematic variation in aggregate in terms of psychological and behavioural tendencies.

Another, non-mutually exclusive account for how ecology shapes cultural variation is rooted in the framework of cultural evolution. Indeed, many cultural evolutionary accounts hold that cultures evolve, at least in part, in response to ecological pressures [56–58]. This view emphasizes how collective responses to environmental threats and affordances help societies to meet these challenges through the creation of norms, institutions and practices. For example, societies which historically faced high levels of threat in terms of war with other groups, high levels of infectious disease, and higher frequency of natural disasters tend to have tighter social norms, which are thought to enhance group cohesion and survival [7,59]. Similarly, societies which experience more extreme temperatures tend to have lower levels of political freedom [60] but demonstrate greater creativity [61], in part to adapt to the challenges of these extreme environments. It seems plausible that many of ecology's effects on culture may reflect some combination of evoked individual-level responses, as well as these types of more collective cultural responses which involve learning, norms and social influence [62].

Yet another influential account, the socioecological framework, emphasizes the importance of concrete, objective features of the environment in influencing the shared systems of meaning, mental phenomena, and behaviour that make up culture [25,63]. Although this view emphasizes ecological conditions primarily as a cause, and culture and individual psychology as outcomes, it also posits that such relationships may be bidirectional including through processes akin to niche construction in non-human animals [63]. The socioecological perspective is largely compatible with theories emphasizing both evoked and transmitted processes which link ecological conditions to contemporary patterns of variation across human groups.

Thus, there are numerous, often non-mutually exclusive frameworks that have attempted to explain how ecological conditions are linked to human cultural variation. It is noteworthy that even theories which emphasize other types of factors, such as the role of economic and political systems [18], religious institutions [58], historical modes of subsistence [64] or philosophical traditions [65] to explain contemporary patterns of human cultural variation, often also posit a central role for ecology as part of a causal chain.

(a) . The present study

The present study extends past work on ecology and culture in three ways. First, although prior work has documented numerous links between specific ecological variables and cultural outcomes [7,36,38,42,66–70], to date there has been no broader and systematic test of the overall effect of ecology on human cultural variation. The primary goal of the present work is to answer this question: how much of human cultural variation can be explained by ecology? Second, given that multiple ecological variables are often related to the same cultural outcome (e.g. resource abundance, rates of infectious disease and population density are all associated with cultural variations in individualism [36,44,71]), assessing the contribution of ecology to such variations should also entail analyses that consider the contributions of multiple ecological variables simultaneously.

Finally, prior work has also typically operationalized ecology using single timepoint estimates such as current levels of an ecological condition or its level at some discrete point in the past. Yet ecologies are not necessarily stable over time. The extent to which features of the environment are variable or predictable, and their historic as well as contemporary levels may all exert influences on human culture. For example, at the individual level, unpredictability in resource availability is linked to moral thinking, working memory and impulsivity [72–74]. Given that group differences may represent the aggregate of individual differences, cross-cultural researchers interested in ecology might do well to assess linkages between ecological unpredictability and culture as well. In a similar vein, one might expect different adaptations for variation within ecology (i.e. a culture that experiences seasonal variation in population density due to tourism), the frequency of outlier events (i.e. a particularly rainy year) and the average ecological conditions over time (i.e. a consistently dense population) [66,75,76]. Thus, in the present work we also include metrics designed to capture not only contemporary levels of ecological conditions, but also their temporal properties.

Here, we test links between nine ecological variables (rainfall, temperature, gross domestic product per capita (GDP), the Gini index of resource inequality, extrinsic mortality, life expectancy, disease threat, unemployment and population density) and 66 cultural variables (including personality traits, values, social motivations, subjective well-being and government functioning) in 201 countries. Each ecological variable is broken down into 11 statistical metrics (e.g. current levels, average levels across time, temporal variability and temporal unpredictability). As a primary goal, we attempt to estimate the average amount of cultural variation explained by the nine ecological variables after accounting for spatial and cultural autocorrelation, focusing on each statistical metric in turn. In exploratory analyses, we also assess which ecological variables (and statistical metrics of those variables) are most strongly related to overall cultural variation, and which cultural variables are most strongly related to ecology writ large. The study procedures and analysis plan were pre-registered on the Open Science Framework (https://osf.io/45am7/) prior to data collection and analysis (see electronic supplementary material for details including departures from pre-registration). A pre-print of this manuscript was posted on PsyArXiv [77]; all data are publicly available on the Open Science Framework at https://osf.io/45am7/ and Dryad at https://doi.org/10.5061/dryad.f7m0cfz1x [78].

2. Methods

A more-detailed methods and results section is available in the electronic supplementary material. A full description of the EcoCultural Dataset is available in Wormley et al., 2022 (Scientific Data) [77].

(a) . Cultural variables

We gathered data on 66 previously published cultural variables including personality traits, values and social motivations (electronic supplementary material, table S3).¹ To be included in our analyses, these variables had to be composite scores—not single items—and data from at least 20 countries had to be available.

(b) . Ecological variables

We gathered data on nine distinct ecological variables using archival data: rainfall, temperature, GDP, resource inequality, extrinsic mortality, life expectancy, disease threat, unemployment and population density (electronic supplementary material, table S1).² These variables are ones which have been previously linked to human and animal behaviour (for references, see electronic supplementary material, table S1) and have been defined as ecological variables in Sng et al., [26] or Van de Vliert [60,79]. To be included, the data required 20 years of time series data.

(c) . Data aggregation and metrics

For each model, we truncated the ecological time series data so that the last datapoint matched the year before the cultural variable was published, which we took as a proxy for the year of its collection. For example, if we were interested in how affective autonomy, using Schwartz [19], correlated with a particular ecological operationalization, we subset our dataset so the last year of data was 2006. We operationalized the ecological variables in 11 ways:

(i) . Current levels

We operationalized this metric as the last available datapoint within a subset dataset (Current). Contemporary ecological conditions feature heavily in the notion of evoked culture. Thus, to the extent that some of human cultural variation is the aggregate of individual evoked responses, then one would expect that ecological conditions should be linked to cultural variation across groups.

(ii) . Mean levels

We computed mean levels of each ecological variable over time (Mean). Given that culture may involve collective responses that take time to create or change (i.e. institutions, rules, practices, etc.), one might expect culture to be responsive to average ecological conditions as well (i.e. generally resource-rich or generally low disease threat environments).

(iii) . Variability

We computed two indicators of temporal variability for each ecological variable: standard deviation (s.d.), and range (Range). Ecology is not constant over time (e.g. seasons [75]) and such variation may necessitate cultural adaptations such as greater behavioural flexibility and innovation [80,81]. Thus, it may be that variability in the environment might be linked to a range of cultural outcomes.

(iv) . Extreme perturbations

We computed the maximum (Max) and the minimum (Min). We also calculated the percentage of outliers within the dataset, defined as datapoints lying at or beyond 2.5 standard deviations from the mean as a marker of sudden shocks within the environment (Outliers). Organisms should also be expected to adapt to extreme events within their environments and how frequently such extremes occur (e.g. natural disasters, rising global temperatures), which may manifest as cultural variation.

(v) . Unpredictability

We computed several indicators meant to assess the predictability of ecological conditions over time. We did so by computing first-order autocorrelation within each ecological time series (First-order autocorrelation). We assessed trends within the data by fitting a linear model predicting the ecological variable from the year. From this, we took the standardized linear regression coefficient as an indicator of the strength and direction of linear trends within the ecological data (β). We also operationalized predictability using the auto.arima function from the forecast package in R [82]. In the first step, we used the algorithm to fit a variety of ARIMA models to the first 80% of our time series data and select the best one. Next, we evaluated that model's accuracy in predicting the remaining 20% of the data. We extracted indicators of model fit including the mean absolute per cent error (MAPE) and the mean absolute standard error (MASE); greater error scores indicate greater unpredictability in the data with the idea that if such algorithms cannot predict future datapoints, the ecology can probably be described as relatively unpredictable by humans as well.

According to prior work, living within an unpredictable environment may necessitate even greater behavioural flexibility, as well as the presentation of less extreme phenotypes [76,83,84]. Although much prior work on unpredictability has focused on individual differences, it is plausible that ecological unpredictability might also lead to comparable differences between human groups.

3. Results

(a) . Analytic strategy

Our primary analytic strategy was designed to assess how much of the variance on average across our 66 cultural variables was explained by physical and social ecology. Given that spatial autocorrelation (which was present in our data, see electronic supplementary material for details) and cultural autocorrelation may lead to spurious patterns of association when assessing relationships across countries, and given potential issues of multi-collinearity among our ecological predictors, we adopted a multi-step LASSO regression design to quantify how much variance in culture was due to ecology. In our final step of analyses, this resulted in 666 individual models—one for each pairing of cultural variables and ecological operationalizations.

In the first step, we ran a series of LASSO regressions predicting each of the 66 cultural variables from the average latitude and longitude of countries as a means of assessing how much variance in culture was explained by spatial autocorrelation alone. Within LASSO regression, predictors are penalized and reduced to zero if they do not predict a significant unique portion of the outcome variable's variance [85,86]; the coefficients were reduced to zero in 29 cases for latitude and 37 cases for longitude. On average, latitude and longitude—our proxy for spatial similarity—explained around 10% of the variance in cultural variables (electronic supplementary material, table S3; figure 1).

Figure 1.

R² from models predicting cultural variables from latitude and longitude (proxy for spatial autocorrelation).

In the second step, we added Muthukrishna et al.'s estimates of countries' overall cultural distance from the United States as a proxy for cultural autocorrelation [87] to the LASSO regression model, along with latitude and longitude. The average explained variance increased to 25.6% in this step across 66 models (figure 2).

Figure 2.

R² from models predicting cultural variables from latitude, longitude and cultural distance.

In the third step, we added markers of the physical ecology (rainfall, temperature and disease) to our LASSO regression models in addition to latitude, longitude and cultural distance. We did this for each metric type (i.e. mean, outliers, current levels) for a total of 666 models (average N ∼ 47 countries). Again, to reduce the effects of multicollinearity, coefficients were reduced to zero if they did not explain a sufficient amount of unique variance. The average variance explained by the combined predictors in these models ranged from 25.7% (for first-order autocorrelation as an operationalization of ecological conditions) to 31.8% (for maximum levels as an operationalization of ecological conditions), but explained as little as 0% or up to 61% of the variance, depending on the ecological operationalization and cultural variable in the model (figure 3).

Figure 3.

Average R² across 11 models predicting cultural variables from latitude, longitude, cultural distance and physical ecology variables.

In the fourth and final step, we created models in the same fashion as in step 3, but with social ecological variables included as predictors (life expectancy, GDP, extrinsic mortality, population density, income inequality and unemployment as additional predictors; average N ∼ 31 countries; see electronic supplementary material, table S5 for details). Across 666 models, R² ranged from 0.00 to 0.99, depending on the operationalization of ecology used and the cultural variable. The average explained total variance increased to range between 26.5% (for models including the percentage of outliers as an operationalization of ecological conditions; range of R² = [0.00, 0.94]) and 45.2% (for models including mean as an operationalization of ecological conditions; range of R² = [0.00, 0.94]) (figure 4).

Figure 4.

Average R² across 11 models predicting cultural variables from latitude, longitude, cultural distance, physical ecology variables and social ecology variables.

How much cultural variation is due to ecology above and beyond spatial and cultural autocorrelation? Comparisons of the models generated by the final step suggest that ecological metrics of sudden environmental shocks or environmental unpredictability explain a relatively trivial amount of added variance in culture beyond spatial and cultural autocorrelation, whereas current levels (average added variance explained = 16%; range = [0%, 68%]) and mean levels of ecological conditions (average added variance explained = 20%; range = [0%, 68%]) as well as some ecological metrics of environmental variability (i.e. standard deviations of ecological conditions; average added variance explained = 15%; range = [0%, 64%]) explain a more substantial amount of cultural variation (table 1). In Cohen's classification [88], average R²s for most metrics of ecological unpredictability (i.e. MASE, first-order autocorrelation, β) were small effect sizes, as were average R²s for most metrics of ecological variability (i.e. minimum, maximum, range). Other metrics of unpredictability had, on average, trivial effect sizes (i.e. outliers and MAPE). The average R²s for standard deviation, historical means and current levels of ecological conditions were medium effect sizes.

Table 1.

Average R² across 66 cultural variables in a LASSO regression model explained by controls for spatial and cultural autocorrelation and by ecology.

	Model 1:	Model 2:	Model 3:	Model 4:	added variance explained by physical & social ecology
predictors in the model	latitude + longitude	latitude + longitude + cultural distance	latitude + longitude + cultural distance + physical ecology	latitude + longitude + cultural distance + physical ecology + social ecology
	0.100
		0.256
outliers			0.260	0.265	0.009
MAPE			0.275	0.268	0.012
first-order autocorr.			0.257	0.304	0.048
MASE			0.260	0.306	0.050
minimum			0.322	0.315	0.059
linear trend (β)			0.269	0.328	0.072
maximum			0.311	0.371	0.115
range			0.294	0.376	0.120
standard deviation			0.298	0.403	0.147
current			0.321	0.413	0.157
mean			0.306	0.452	0.196

(b) . Additional analyses

The models generated in step 4 also lend insights as to which ecological variables are consistent, independent predictors of culture. No matter the operationalization, the average coefficient for temperature, population density, income inequality and disease threat were never reduced to zero. Models with the average coefficients of these ecological dimensions are presented in the electronic supplementary material (electronic supplementary material, table S6) but should be interpreted with caution as the coefficients are unstandardized.

Details regarding the average amount of added variance explained by the combination of physical and ecological features for each cultural variable in the models in step 4 are also presented in the electronic supplementary material, providing some insight into the relative degree to which different cultural variables are linked to ecology writ large. The R² for well-being (0.70) and innovation (0.69) were the highest in the final models, while break-up concern and national pride were not explained by any predictors in the model (R² = 0.00; electronic supplementary material, tables S3–S4).

Zero-order correlations between each metric of each individual ecological dimension and each cultural outcome are presented in electronic supplementary material, tables S8–S18. Standardized, partial β coefficients for the relationship between these variables with controls for spatial and cultural autocorrelation are presented in electronic supplementary material, tables S19–S29.

4. Discussion

How much of cultural variation across human groups is explained by ecology? Here, we provide a large-scale, systematic test of the overall relationship between ecological conditions and human cultural variation. When accounting for spatial and cultural autocorrelation, it appears that ecological conditions on average account for somewhere between 1 and 20% of the variance in human culture, with historical averages (20%) and current levels (16%) of ecological conditions explaining the most added variance. However, depending on the ecological operationalization and the cultural dependent variable, ecology explained up to an additional 68.4% of the variance in some cases. Taken together, our results confirm past theoretical assertions and empirical findings suggesting that ecology is a source of human cultural variation, and for the first time, to our knowledge, provide a range of estimates for quantifying this relationship writ large.

This study was also designed to capture features of the ecology beyond single timepoint estimates which may have interesting theoretical implications. For example, our results suggest that historical averages in ecological conditions appear to have the strongest linkage to culture, whereas markers of ecological variability had weaker links, and markers of ecological unpredictability appear to have the weakest links to culture. This suggests that, in general, unpredictability and variability in the ecology have relatively modest impacts on patterns of human cultural variation. Further, historical levels of ecological conditions explained more of the variance in culture than did current levels, suggesting perhaps that culture is somewhat more calibrated to historical levels versus momentary, present ecological conditions (although both linkages were of fairly comparable size on average). These comparisons should be interpreted with some caution, as formal model comparison across metrics is not performed in the present work, as models including different metrics of the same variables would severely violate assumptions of independence. Nevertheless, these findings may serve as a jumping-off point for more nuanced theorizing regarding how exactly ecology comes to shape culture. They suggest perhaps that some of ecology's effects on culture may occur at a lag, or in response to chronic conditions and thus may involve cultural transmission in addition to aggregated individual evoked responses. However, this is a speculative interpretation at present and future research can hopefully explore this question further.

Another contribution of the present work comes from examining links between temporal ecological variation and culture. There is a growing body of work exploring the effects of ecological variability and unpredictability on individual-level behaviour [73,76,83]. In the present work, we extend this approach to the cultural level. Although, as previously noted, we find that these linkages are fairly modest, these measures of temporal variation in ecology still hold some predictive value. Future research should delve deeper into these relationships as there may be some aspects of culture which are more closely (or differentially) linked to ecological variability or predictability than to current levels ecological conditions.

Our work also suggests that some cultural variables may be more strongly linked to ecology than others, and that features of the ecology also differ in the extent to which they are related to cultural variation writ large. Namely, well-being and innovation were strongly related to ecology, whereas other cultural features, such as national pride showed weaker links with ecology (electronic supplementary material, table S4). In terms of ecological features, temperature, population density, income inequality and disease threat were the most consistent predictors of cultural variables (see electronic supplementary material, table S6).

Although the primary aim of the present work was not to dive into relationships between specific ecological and cultural variables, it is worth noting that we do replicate some past seminal findings regarding such individual linkages. For example, we find that gender egalitarianism was negatively correlated with historical averages in disease threat [89]; GDP per capita was positively correlated with innovation [90]; and population density was negatively correlated with individualism [71,91]. Future researchers may also wish to mine this rich dataset further as it may reveal novel and theoretically interesting relationships, such as a link between variability in disease threat and tightness (see electronic supplementary material, tables S19–S29 for details).

(a) . Limitations and caveats

Rather than providing an exact point estimate of the relationship between ecology and culture, we believe our results may perhaps represent a range of lower bound estimates. For example, societies that are less able to buffer the effects of meteorological conditions through wealth or technology might be more susceptible to such ecological influence [18,23,60]. Thus, one might expect that ecological conditions might have had a stronger impact upon human culture in the preindustrial era, and earlier, than they do today. It is also probably the case that ecology may explain additional variance if interactions between ecological variables (and between metrics of such variables) are modelled as well. For example, high resource inequality paired with occasional famine has been linked to greater prevalence of polygyny [92], and more extreme temperatures are associated with more innovation in wealthy countries but less in poorer countries [81]. Finally, there are other ecological variables which have previously been linked to variation in behaviour and personality, such as sex ratio [68,93,94], which we did not explore in the present work due to the absence of systematic cross-cultural time series. Thus, the true comprehensive effect of ecology on cultural variation may be somewhat larger than that suggested by the current analyses. Future work should build on the present findings by attempting to quantify the relationship between ecology and cultural variation in the more distant past and by exploring the extent to which interactions between key ecological dimensions may be linked to this variation.

In the present work, we attempted to capture a broad set of ecological variables, cultural variables, statistical metrics and societies to assess the relationship between ecology and human cultural variation. Although this may be the most systematic effort in this vein to date of which we are aware, we do not claim that the dataset is exhaustive. It may be that inclusion of additional variables would yield different results. Again, we wish to reiterate that the goal of the present work is not to provide an exact point estimate of the relation between ecological and cultural variation. Rather, our aim was to provide a sense of the range in which such an effect is likely to exist. Future work will hopefully help to refine such estimates, and we also welcome efforts to broaden the EcoCultural Dataset to make such estimates more comprehensive.

Finally, as we note in the introduction, ecology is but one of many explanations for human cultural variation. Indeed, although our work may represent a lower bound estimate, given the substantial amount of variation in culture unaccounted for by ecology in our analyses, other factors probably play a role over and above that of ecological conditions. We hope that the present work might be used as a template to begin to systematically address other such explanations and their contributions relative to ecology, and to each other, as well as putative mechanisms by which cultural variation comes to be.

Data accessibility

All data are publicly available on the Open Science Framework at https://osf.io/45am7/ and Dryad at https://doi.org/10.5061/dryad.f7m0cfz1x [95]. Data should be cited as: [78].

The data are provided in electronic supplementary material [96].

Authors' contributions

A.W.: conceptualization, data curation, formal analysis, funding acquisition, visualization, writing—original draft, writing—review and editing; J.Y.K.: conceptualization, formal analysis, writing—original draft, writing—review and editing; M.B.: conceptualization, writing—original draft, writing—review and editing; M.E.W.V.: conceptualization, supervision, writing—original draft, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

The authors declare they have no competing interests.

Funding

A.S.W. was funded by the National Science Foundation's Graduate Student Fellowship.