Alcohol data and coding
We purpose-built a dataset of the presence of indigenous alcoholic beverages for the SCCS sample of 186 largely non-industrial societies8. This widely used cross-cultural sample encompasses a range of diverse and well-documented cultures from around the world, each ethnographically described at a specific time (“focal time”, FT), usually prior to intensive acculturation. Data on alcoholic beverages were obtained primarily from ethnographic literature available through eHRAF World Cultures database (https://ehrafworldcultures.yale.edu/). First, we searched the literature using eHRAF’s subject categories: “Alcoholic beverages” (OCM 273) and “Alcoholism and drug addiction” (OCM 733). We supplemented this with keyword searches for words most commonly associated with alcoholic beverages using terms alcohol*, beer* wine*, mead*, ferment*, brew*, intox*, drunk*, liquor*, and distil*. If eHRAF sources did not provide sufficient information, we also used other literature (searched by Google and Google Scholar using the search query: society name + alcohol, or fermented, or beer, or wine). All the references we used are listed in Supplementary Data. Following others29,45,54, we distinguished three general categories of alcoholic drinks: wines, beers and distilled liquors. With respect to the research question, however, we focused only on the presence of wines and beers (combined column Alcohol_WB in dataset in Supplementary Data), as distilled beverages have a significantly higher alcohol content and a different (often more harmful) impact on society5. We defined nine states of presence of alcoholic beverages in each culture’s FT that describe both the assumed origin of the alcohol and data quality:
0 = Alcoholic beverages absent. It is explicitly stated that the society did not consume alcoholic beverages.
1 = Alcoholic beverages prohibited. Alcoholic beverages were absent primarily due to prohibition. This category also includes inferred cases of Muslim societies where alcohol is not mentioned (but other intoxicants are).
2 = Alcoholic beverages probably absent. Absence was inferred from source materials. For example, there are separate sections or a number of continuous paragraphs on beverages and intoxicants in the literature, but there is no mention of alcoholic beverages.
3 = Indigenous alcoholic beverage absent but presence of introduced alcohol not clear. It is not clear from the literature whether alcoholic beverages were present. However, if so, they were probably not indigenous. This category also includes cases where the presence of another type of traded/introduced alcohol (e.g. distilled) is documented, but there is a lack of information for the beverage type (e.g. wines) under examination.
4 = Alcoholic beverages available only through the trade. Alcoholic beverages were present but were introduced (i.e. not indigenous) and only available through the trade (i.e. not produced locally or local production is uncertain).
5 = Alcoholic beverages introduced (<100 years before FT). Local production of alcoholic beverages was present but not indigenous. They learned it from another culture, with the introduction occurring less than 100 years before focal time.
6 = Alcoholic beverages present but their origin is unknown (or introduced >100y before FT). Consumption of alcoholic beverages was present but it is not clear whether their origin is indigenous, or whether their production was introduced, or whether they were only available through trade. This category also includes the introduction of alcohol more than 100 years before the focal time, and cases with unknown date of introduction.
7 = Indigenous alcoholic beverages probably present. Presence was inferred from source materials. For example, when mentioned only briefly without description of their local production, or mentioned only in sources other than S1 or S2 (as defined by eHRAF). This category also includes former production that was abandoned in the post-contact period, and cases where alcohol is present but its consumption is frowned upon.
8 = Indigenous alcoholic beverage present. It is explicitly stated that the society produced and consumed their own alcoholic beverages.
NA = Insufficient information or conflicting information.
Our research question focuses on the role of traditional drinks (i.e. that were produced and consumed before contact with a dominant culture, most often Western civilization), as alcohol introduced into societies without established cultural drinking practices often has socially disruptive effects6. Further, for the purpose of analysis, we simplified the codes into a binary form: indigenous alcohol absent (including categories 0, 1, 2, 3, 4 and 5), or present (categories 7 and 8; Fig. 1a). Categories 6 (“origin unknown”) and NA were excluded from the analyses. Because treating some categories, such as “prohibited” (1), “traded” (4) and “introduced” (5), as absences can be questionable, we also performed all analyses on a second, “conservative” sample as a check. Here, we also excluded the category (7) because it includes cases where indigenous production of alcohol is not certain or where alcohol consumption (albeit present) is frowned upon. In contrast, we retained category “inferred absence” (2) because we assume that there is a low probability of alcoholic beverages being present (in non-negligible quantities) and not recorded in the ethnography. The resulting “conservative” sample therefore includes categories 0, 2, 3 and 8 (Fig. 1b, Extended Data Table 1).
We concede that a more appropriate variable than absence/presence would be intensity of alcohol consumption. However, the latter has a number of limitations, in particular the lack of reliable data (cf. ref.55) due to the fact that most of the reports on traditional drinking behaviour is only incidental6. First, ethnographers often do not mention drinking intensity at all. If they do, it is not an objective measurement but a subjective observation. Further, intensity of consumption is more prone to change over time than the mere presence of alcohol, and likewise one might expect greater internal variability - both between different communities within one culture and between members of the same community (e.g., men vs. women, young vs. old). Using the presence of alcohol as opposed to the intensity of its consumption also has the advantage when modelling causality with political complexity. Although an increase in political complexity may lead to an increase in the intensity of alcohol production/consumption5,7,46, and therefore generate causal back-loops, we are not aware of any theoretical mechanism that it would lead to the actual invention or adoption (i.e. presence) of alcohol, allowing us to focus on unidirectional causality.
Other variables: political complexity, agriculture, environment
The other cultural and environmental variables for SCCS sample were obtained from the D-PLACE database21. As a proxy for political complexity, we chose the “Level of Political Integration” (SCCS157). Originally coded by Murdock and Provost, this five-scale ordinal variable expresses “the complexity of political organization in terms of the number of distinct jurisdictional levels recognizable in the society”23. More generally, this classifies cultures on a scale from “acephalous” lacking any centralised political authority to large “states” organised into several administrative levels. Although there are many other metrics of cultural complexity1,23,56, most of them correlate strongly with each other57,58, with the number of jurisdictional hierarchy levels being one of the most commonly used throughout the social sciences22,25,59,60. As a measure of the presence and intensity of agriculture (i.e. crop cultivation), we used the five-scale ordinal variable “Agriculture” (SCCS151), defined as “the degree of dependence upon agriculture for subsistence and the intensity with which it is practiced”23. For an alternative test, we binarized the variable to express the mere absence or presence of agriculture. To measure the variability of the environmental conditions, we performed a principal component analysis (PCA) on nine environmental variables, following the procedure by Haynie et al.26. The variables we considered were: annual mean, variability, and predictability of temperature and precipitation regimes, annual mean, variability and predictability of net primary productivity, as well as slope and elevation of the terrain. We accessed these data via D-PLACE21 (where data is compiled from the Baseline Historical [1900-1949], CCSM ecoClimate model dataset61, Global Multi-resolution Terrain Elevation Data 201062, and NASA, Terra/MODIS63). We ran a varimax-rotated PCA using the function principal in the R package psych64. We set the number of factors to three, as indicated by the non graphical Cattel’s Scree Test65, which we ran with the function nScree in the R package nFactors66. The first principal component (PC) explained 53% of the variation in the data, and captured a gradient of environmental productivity. High values of PC1 are associated with warm, predictable, and invariable temperatures, humid and seasonal precipitation regimes, and high net primary productivity. The second PC captured variation in predictability of precipitation and variability in NPP, while the third PC mapped onto elevation and slope of the terrain. We chose to include only PC1 in our analyses as its relationship to political complexity, alcohol, and agriculture (i.e., variables of interest in our models) is best grounded in theory (see section Causal models). Environmental productivity links positively with alcohol and agriculture intensity in our dataset (Extended Data Fig. 5 a,b). We do not see a direct effect of environmental productivity on political complexity in our sample (Extended Data Fig. 5c). This may reflect a similar association that Haynie et al.26 identified between environmental productivity and social inequality, in which case the causal relationship is mediated by the use of large domestic animals. That is, the use of large domestic animals correlates negatively with environmental productivity, but positively with the presence of heritable social class. We mark such mediator in our causal structure more broadly as ‘unobserved mediators’, speculating that it could also be the use of large domestic animals, or a belief in moralizing gods25,26,60.
Spatial autocorrelation and phylogenetic non-independence
Although SCCS was designed to minimise cultural relatedness8, there is evidence of autocorrelation in this sample67,68. To account for the non-independence of cultures due to spatial proximity and common ancestry we included two correlation matrices in our models: one based on geographic distances and one based on historical relatedness. We calculated great-circle (haversine) distances between societies using their latitude and longitude coordinates, as recorded in D-PLACE. We took the reciprocal of the distance matrix (1/distance-matrix), and set the diagonal elements to 1. This matrix is akin to a correlation matrix, in which societies that are closer together have higher values, whereas societies that are further apart have small values. To measure relatedness between societies we used phylogenetic distances based on a recent global phylogeny of 6,636 extant languages36. The relatedness among languages is captured in the form of a posterior distribution of possible trees. We first built a maximum clade credibility tree based on 1,000 posterior trees using Tree Annotator, setting node heights to common ancestor69. We then pruned this global tree for the societies in our dataset. Not all SCCS societies were included in Bouckhaert et al.'s tree36. While 142 societies could be automatically matched based on their glottocode, we used proxy languages for 34 societies. Of these, 21 dialects were paired with their parent language (e.g. “Nama” nama1265 to “Nama (Namibia)” nama1264), six with a sister language (e.g. “Gros Ventre” gros1243 to “Arapaho” arap1274), and seven with an otherwise closely related language (e.g. “Twana” twan1247 to “Northern Straits Salish” stra1244). However, we were unable to pair seven societies, including two Creole (Haitians, Saramaccan), which we therefore excluded from the analyses. In addition, we excluded three historical societies (Ancient Romans, Ancient Hebrews, Babylonians), whose temporal distance from modern related languages is more than 1,000 years, which could potentially have a significant effect on phylogenetic tree’s branch lengths. We computed a correlation matrix based on the pruned tree using the function vcv.phylo in the R package ape70 (setting the correlation parameter, corr, as true). We also re-ran our models using 50 trees randomly selected from the posterior of alternative trees, in order to test for the effects of phylogenetic uncertainty in our analyses. These sensitivity analyses show that our results are robust to phylogenetic uncertainty (Extended Data Fig. 6).
Causal models
To understand the association between the alcoholic beverages and political complexity, we compare several causal models, each with different sets of explicit theory-based assumptions (Fig. 2).
Model 1 assumes no confounders and the causal direction of Alcohol → Political complexity, as alcohol is said to have been the great facilitator of civilization and a fundamental element in the state-making process, because it enhances creativity, facilitates social contact, enhances trust and bonding, forges group identity, stimulate and strengthens trade and alliance networks, and reinforces social roles and hierarchies, such as patron-client relationships5,7,45,46. While it is possible that political complexity has reversibly led to an increase in alcohol production and consumption, we are not aware of any theoretical mechanism by which it would lead to the actual invention or adoption (i.e. presence) of alcoholic beverages, which is our independent variable.
Model 2 adds the first potential confounders (spatial proximity and common descent), since it is known that geographically close and historically related societies tend to share many aspects of culture, environment and demography that are likely to affect both the presence of alcohol and the level of political complexity24,71.
Model 3 adds a third potential confounder (Environmental productivity) and assumes the causal direction of Alcohol → Agriculture, as suggested by the “beer before bread” theory5,15,16,72,73. Environmental productivity likely affects both the availability of raw materials necessary for alcohol fermentation55 and the emergence of politically complex societies via its influence on the emergence and spread of agriculture25,74 and/or possible other mediators (e.g. the use of domesticated large animals26).
The remaining two models add intensity of agriculture as a confounder, assuming the causal directions of Agriculture → Alcohol, and Agriculture → Political complexity. Since the evidence for the “beer before bread” theory is not yet conclusive17,43 and recent research points to the variability and complexity of the whole process of agricultural origins40-42, it seems likely that alcohol production was more often a consequence or by-product of crop cultivation rather than a cause of it. Similarly, while it is argued that the causal relationship between agricultural intensification and political complexity is at least bidirectional27,28, the dominant view is that agriculture was a prerequisite for the development of complex and centralised societies1,2,75. Model 4 differs from Model 5 in that the former assumes no influence of the Environmental productivity on Political complexity other than through the mediating effect of Agriculture, while the latter assumes either a direct effect or an indirect influence through another mediating factor. This may be, for example, the presence of large domestic animals or belief in moralising gods; both having a positive effect on the rise of political complexity and more likely occurring in societies living in less productive environments4,26,60.
Bayesian regression models
We ran Bayesian regression models using the R package brms76. Political complexity and agriculture intensity are variables with discrete, ordered categories (each with five different levels). We modelled political complexity as an ordered factor, and set the family argument in our brm() models as cumulative35. We also coded agriculture as an ordered factor, and included it in our models as a monotonic predictor38, using the function mo(). The presence and absence of alcohol was coded as a binary factor, and environmental productivity as a continuous variable. For each relationship in our DAGs, the models estimate a posterior distribution of possible coefficient estimates, which we plot in its entirety to evaluate the strengths of associations between variables of interest. Additionally, we ran two models that mirror models 4 and 5, but excluding alcohol as a predictor. By doing this, we could estimate the total (unmediated) effect of agriculture on political complexity. Lastly, we ran an alternative version of models 4 and 5, in which we binarized agriculture (we code it as presence - original code 1, versus absence - original codes 2-5). We set an exponential prior distribution for standard deviations.
Figure construction
We created Fig. 2 by plotting histograms of posterior coefficient estimates of the association between political complexity and alcohol. For each of our five models, we overlapped coefficient estimates for the “conservative” (blue) and “wide” (orange) sample for direct comparison. We also calculated and plotted the median and interquartile range of each posterior distribution, i.e., the range where the central 50% of the data points are found. The panels all share the same limits for the x-axis for easy comparison. The figure was plotted using the hist function in R. We followed a similar procedure for Extended Data Figs. 2, 4 - 6.
We created Extended Data Fig. 3 by using the function conditional_effects in brms. The plot shows probabilities of values in each political complexity category for each level of the predictor of interest (here, alcohol). We modified the original function from brms to group points by political complexity levels, as opposed to by predictor levels. The plot shows the estimate, as well as the lower and upper interval limits.
Methods references
54 Jennings, J. et al. “Drinking Beer in a Blissful Mood”: Alcohol Production, Operational Chains, and Feasting in the Ancient World. Current Anthropology 46, 275-303 (2005). https://doi.org/10.1086/427119
55 Blum, R. H. A Cross-Cultural Study in Society and Drugs: Drugs 1. Social and Cultural Observations (ed R. H. Blum) 134-186 (Jossey-Bass, 1969).
56 Feinman, G. M. Size, complexity, and organizational variation: A comparative approach. Cross-Cultural Research 45, 37-58 (2011).
57 Chick, G. Cultural Complexity: The Concept and Its Measurement. Cross-Cultural Research 31, 275-307 (1997). https://doi.org/10.1177/106939719703100401
58 Turchin, P. et al. Quantitative historical analysis uncovers a single dimension of complexity that structures global variation in human social organization. Proceedings of the National Academy of Sciences 115, E144-E151 (2018). https://doi.org/doi:10.1073/pnas.1708800115
59 Hamilton, M. J., Walker, R. S., Buchanan, B. & Sandeford, D. S. Scaling human sociopolitical complexity. PloS one 15, e0234615 (2020).
60 Botero, C. A. et al. The ecology of religious beliefs. Proceedings of the National Academy of Sciences 111, 16784-16789 (2014). https://doi.org/10.1073/pnas.1408701111
61 Lima-Ribeiro, M. S. et al. EcoClimate: a database of climate data from multiple models for past, present, and future for macroecologists and biogeographers. Biodiversity Informatics 10, 1-21 (2015). https://doi.org/10.17161/bi.v10i0.4955
62 Danielson, J. J. & Gesch, D. B. Global multi-resolution terrain elevation data 2010 (GMTED2010). U.S. Geological Survey (2011).
63 NASA. Net Primary Productivity (1 month - TERRA/MODIS). Available: http://neo.sci.gsfc.nasa.gov/view.php?datasetId=MOD17A2_M_PSN
64 Revelle, W. psych: Procedures for personality and psychological research v. R package version 2.3.9 https://CRAN.R-project.org/package=psych (2023).
65 Raîche, G., Walls, T. A., Magis, D., Riopel, M. & Blais, J.-G. Non-graphical solutions for Cattell’s scree test. Methodology 9, 23-29 (2013).
66 Raîche, G., Magis, D. nFactors: Parallel Analysis and Other Non Graphical Solutions to the Cattell Scree Test v. R package version 2.4.1.1 https://cran.r-project.org/package=nFactors (2020).
67 Minocher, R., Duda, P. & Jaeggi, A. V. Explaining marriage patterns in a globally representative sample through socio-ecology and population history: A Bayesian phylogenetic analysis using a new supertree. Evolution and Human Behavior 40, 176-187 (2019). https://doi.org/https://doi.org/10.1016/j.evolhumbehav.2018.11.003
68 Dow, M. M. & Eff, E. A. Global, Regional, and Local Network Autocorrelation in the Standard Cross-Cultural Sample. Cross-Cultural Research 42, 148-171 (2008). https://doi.org/10.1177/1069397107311186
69 Drummond, A. J., Suchard, M. A., Xie, D. & Rambaut, A. Bayesian phylogenetics with BEAUti and the BEAST 1.7. Molecular biology and evolution 29, 1969-1973 (2012).
70 Paradis, E., Claude, J. & Strimmer, K. APE: analyses of phylogenetics and evolution in R language. Bioinformatics 20, 289-290 (2004).
71 Bromham, L. & Yaxley, K. J. Neighbours and relatives: accounting for spatial distribution when testing causal hypotheses in cultural evolution. Evolutionary Human Sciences 5, e27 (2023). https://doi.org/10.1017/ehs.2023.23
72 Braidwood, R. J. et al. Symposium: Did Man Once Live by Beer Alone? American Anthropologist 55, 515-526 (1953).
73 Reichholf, J. H. Warum die Menschen sesshaft wurden: Das größte Rätsel unserer Geschichte. (S. Fischer, 2008).
74 Vilela, B. et al. Cultural transmission and ecological opportunity jointly shaped global patterns of reliance on agriculture. Evolutionary Human Sciences 2, e53 (2020).
75 Johnson, A. W. & Earle, T. K. The Evolution of Human Societies: From Foraging Group to Agrarian State. (Stanford University Press, 2000).
76 Bürkner, P.-C. Bayesian Item Response Modeling in R with brms and Stan. Journal of Statistical Software100, 1-54 (2021). https://doi.org/10.18637/jss.v100.i05