1.1 Patterns of war and social complexity

There has been extensive debate around the extent to which war was present among small-scale foraging societies thought to be representative of early human history (Ember & Ember, Reference Ember, Ember, Martin and Frayer2014; Glowacki, Reference Glowacki2024; Hames, Reference Hames2019; Otterbein, Reference Otterbein1999). Some of this disagreement can be attributed to differing definitions of war, difficulty with assessing effects of pacification in the ethnographic record, and uncertainty about the representativeness of archaeological evidence for war in the Pleistocene (Ember & Ember, Reference Ember and Ember1992, Reference Ember, Ember, Martin and Frayer2014; Ferguson & Whitehead, Reference Ferguson, Whitehead and Ferguson1992; Hames, Reference Hames2019; Kissel & Kim, Reference Kissel and Kim2019). Overall, extensive surveys of ethnographic literature as well as archaeological evidence have indicated that war – as defined above – does occur among mobile foragers and other small-scale, non-state societies, although it varies in scale, frequency, and severity (Ember & Ember, Reference Ember, Ember, Martin and Frayer2014; Kim & Kissel, Reference Kim and Kissel2018).

Although this research tends to focus on frequency, causes, and mortality, it can point to certain patterns of violence and consequences for enemies in nonstate war. Broadly, such societies do not have the formal military organizations that come with centralized political authority. Instead, they rely on widespread voluntary participation in fighting by adolescent and adult males. War is conducted through raids and ambushes on living spaces of other groups with little separation between combatants and non-combatants (Keeley, Reference Keeley1996). Such tactics could facilitate greater levels of indiscriminate violence and in some cases extermination of enemy groups (Keeley, Reference Keeley1996; LeBlanc & Register, Reference LeBlanc and Register2003). These patterns are corroborated by archaeological sites with skeletal remains, particularly Jebel Sahaba (∼13,000–18,000 years ago), which provides evidence for repeated intergroup violence encompassing male and female children and adults (Crevecoeur et al., Reference Crevecoeur, Dias-Meirinho, Zazzo, Antoine and Bon2021), and Nataruk (∼10,000 years ago), purported to show the massacre of a foraging community (Mirazón Lahr et al., Reference Mirazón Lahr, Rivera, Power, Mounier, Copsey, Crivellaro, Edung, Fernandez, Kiarie, Lawrence, Leakey, Mbua, Miller, Muigai, Mukhongo, Van Baelen, Wood, Schwenninger, Grün and Foley2016). However, it is unclear how representative sites like Jebel Sahaba and Nataruk are of small-scale warfare in the late Pleistocene/early Holocene. Raids and ambushes may not have left visible archaeological traces such as the fortifications and specialized weapons apparent from later settled societies (Kissel & Kim, Reference Kissel and Kim2019). Alternatively, these sites could represent exceptional events. There is clearer evidence for intensified war with early farming societies corresponding with settlement consolidation and population growth in multiple regions (Arkush & Tung, Reference Arkush and Tung2013; Haas, Reference Haas, Feinman and Price2001; Kim et al., Reference Kim, Kusimba and Keeley2015; Underhill, Reference Underhill, Arkush and Allen2008). Mass graves are present at several sites in Neolithic Europe with the development of agricultural lifeways (Fibiger et al., Reference Fibiger, Ahlström, Meyer and Smith2023; Meyer et al., Reference Meyer, Kürbis, Dresely, Alt, Dolfini, Crellin, Horn and Uckelmann2018). While the presence of all ages and sexes at some sites illustrates indiscriminate violence, other sites have demographic variation that may indicate captive-taking of women and children (Meyer et al., Reference Meyer, Kürbis, Dresely, Alt, Dolfini, Crellin, Horn and Uckelmann2018). Factors implicated in such studies include resource competition in conjunction with increased inequality and social stratification (Fibiger et al., Reference Fibiger, Ahlström, Meyer and Smith2023), possibly exacerbated by climatic disturbances (Gronenborn, Reference Gronenborn2006).

With the development of more centralized political structures, coercive processes of expansion may have led to intensified violence on frontiers, particularly with technological advancement and military specialization (Kim et al., Reference Kim, Kusimba and Keeley2015; Otterbein, Reference Otterbein1970). Based on a cross-cultural sample of 42 societies, Otterbein (Reference Otterbein2000) concludes that centralized societies were more likely to kill women and children than were uncentralized societies, and that despotic early states, chiefdoms, and ‘dependent native peoples’ were most likely to kill captives and non-combatants. He writes that these findings contradict those of an earlier study by Hobhouse et al. (Reference Hobhouse, Wheeler and Ginsberg1915), which found decreased killing of captives with higher agricultural stages. In contrast, his findings are somewhat consistent with Ember et al.’s (Reference Ember, Adem and Skoggard2013) findings in eastern Africa that state societies were more likely to commit atrocities during internal war. However, Ember et al.’s results indicated states were not significantly more likely to kill combatants and non-combatants in external war (Ember et al., Reference Ember, Adem and Skoggard2013). Both works note how the social meanings of wartime violence and the resulting treatment of enemies changed with increased political consolidation among state societies. This is consistent with analyses of the integration of military and political authority and greater social importance of militaristic norms in early states (Arkush & Tung, Reference Arkush and Tung2013; Haas, Reference Haas, Feinman and Price2001; Kim et al., Reference Kim, Kusimba and Keeley2015). In particular, Ember et al. write that a connection between ‘military glory’ and behaviours like killing non-combatants, torture, and trophy-taking in war exists in state but not non-state societies (Reference Ember, Adem and Skoggard2013, p. 51). Otterbein (Reference Otterbein2000) similarly links the killing of captives and non-combatants, as well as the use of torture and mutilation, with strategies of terror used by hierarchical ‘despotic states’ to consolidate power. It is possible that reduced interdependence and increased social distance from enemies in state societies could facilitate killing non-combatants (Ember et al., Reference Ember, Adem and Skoggard2013), although this is complicated by other incentives – taking over territory for land could have more exterminatory results than taking it for political conquest and/or slaves (Otterbein, Reference Otterbein2000). Therefore, specific material and political objectives of war could determine the expedient treatment of people residing in a territory desired for conquest by an invading polity.

Ember et al. (Reference Ember, Adem and Skoggard2013) note that their results of greater atrocities in state societies may in part be due to the tendency of ethnographically recorded state societies to be autocratic and dependent on violent coercion, in line with Otterbein’s description of despotic states. It is possible that among less autocratic states with professional military structures, increased social and physical distance between combatants and non-combatants might decrease exposure of populations to external violence. This contrasts with what is effectively total war among small-scale societies (Keeley, Reference Keeley1996). There may also be pacification of conquered areas through increased political control and centralization (Glowacki, Reference Glowacki2024; Snyder & Arkush, Reference Snyder and Arkush2024). These arguments are combined with claims about shifts in moral norms by Pinker (Reference Pinker2011). He argues for a decline of warfare (and other forms of violence) first during the development of agricultural societies and later in seventeenth- and eighteenth-century Europe, emphasizing the role of organized states in reducing the severity of war and in promoting greater reason and empathy. However, there have been extensive critiques of Pinker’s data and conclusions: researchers have questioned the criteria used to group together deaths occurring across widely varying amounts of space and time to illustrate how deaths from violent conflict have supposedly declined (Falk & Hildebolt, Reference Falk and Hildebolt2017; Kim, Reference Kim2012; Mann, Reference Mann2018). His estimates for deaths in historical warfare have been described by relevant experts as highly exaggerated with unclear or inappropriate sources (Butler, Reference Butler2018; Dwyer & Micale, Reference Dwyer, Micale, Dwyer and Micale2021; Ferguson, Reference Ferguson and Fry2013; Fibiger, Reference Fibiger2018). As noted above, there is archaeological evidence for increased intergroup violence among early farming societies in comparison with foragers (Fibiger et al., Reference Fibiger, Ahlström, Meyer and Smith2023). With regards to the formation of modern organized states, ideologies and institutions regulating internal violence could occur alongside extreme external violence during colonial projects (Kim, Reference Kim2012; Mann, Reference Mann2018). Two studies (Falk & Hildebolt, Reference Falk and Hildebolt2017; Oka et al., Reference Oka, Kissel, Golitko, Sheridan, Kim and Fuentes2017) apply scaling relationships to estimates of conflict deaths in relation to population size. Falk and Hildebolt (Reference Falk and Hildebolt2017) find that war deaths scale with population size regardless of social organization and conclude that ‘people living in small-scale societies are not inherently more violent that those living in “civilized” states’. Oka et al. (Reference Oka, Kissel, Golitko, Sheridan, Kim and Fuentes2017) model war group size and population size and find that lethality is not higher among small-scale than state societies, whereas among large societies efforts to eliminate enemy populations may result in higher levels of non-combatant deaths. Although mortality estimates are a valuable measure of the scale and severity of war, they do not necessarily track moral attitudes about the limits of wartime violence in the participating societies. Below, more qualitative work on social norms and potential drivers of indiscriminate violence is reviewed.

1.2 Norms of restraint

There is historical evidence for mutually understood norms of conduct between centralized small polities which frequently fought one another in multiple regions, such as in pre-colonial western Africa (Smith, Reference Smith1989) and in early Greek and Roman warfare (Raymond, Reference Raymond and Hensel2010). These codes of conduct may have become more common and detailed with the formation of complex polities, and often centred around views of a natural or divinely ordained state of relations between human groups. However, these did not necessarily provide any protection for non-combatants or limits on other forms of violence. Instead, rules of conduct during war were often used to defend the divine order of rulers and their territories from chaotic outsiders (Brekke, Reference Brekke2005; Cox, Reference Cox2017). For instance, the concept of ma’at in ancient Egyptian war characterized foreigners as barbarians rebelling against the order and justice of the pharaoh. This legitimized offensive war, including widespread captive-taking and slaughter of defeated enemies, as self-defence and law enforcement (Cox, Reference Cox2017). While the reasons and means of waging war may have taken place within a system of moral thought centred on order, civilization, and divine rule, this appears not to have resulted in consideration of the morality of how enemies were treated. Some scholars have argued for a decrease in these extreme displays of violence with the development of moralizing, broadly prosocial ideologies with major religious traditions (Bellah, Reference Bellah2011). In particular, there are similarities in institutions and aspects of moral thought across many of these traditions, including concepts of just war (Abou El Fadl, Reference Abou El Fadl2001; Brekke, Reference Brekke2005; Johnson, Reference Johnson1981).

While less studied in this context, small-scale societies often had parallel forms of restraint. These include protection for some classes of people based on their supposed incapacity to fight (women, children, the elderly or disabled) or social occupation/class marking them as peaceful (religious officials, diplomats, peasants). Among the Mae Enga of Papua New Guinea, norms included immunity for women and children, the wounded, refugees, or people from a clan with whom one was not at war (Wiessner, Reference Wiessner2019). Similar limits to violence in war can be found across many societies with different emphases and inclusions/exclusions of particular individuals or actions (Chirot & McCauley, Reference Chirot and McCauley2010; Hasluck, Reference Hasluck, Hutton and John1954; Straight, Reference Straight2017). Other common restrictions included forbidding the use of certain types of weapons, killing wounded enemies or captives, or bodily mutilation (Keeley, Reference Keeley1996). Norms based on social distance were also common and delineated applications of the other restraints (Solometo, Reference Solometo, Arkush and Allen2008). For instance, cultural familiarity was central to Mae Enga codes, which developed as part of an interclan exchange and reparations system that maintained a balance of power and provided an avenue for status-seeking through constrained war and peacemaking (Wiessner, Reference Wiessner2019). Such codes could of course be highly variable. Many societies were recorded as regularly killing men, women, and young children without much distinction unless they were taken as captives or slaves (Straight, Reference Straight2017). Beliefs about proper conduct could also vary within societies, as illustrated by a recent study on Turkana pastoralists that found high individual variation in beliefs about the acceptability of killing individuals of different ages or genders (Zefferman & Mathew, Reference Zefferman and Mathew2020). Unfortunately, as Keeley notes, ‘ethnographers have seldom asked individuals – men or women – about their attitudes toward and reactions to war’ (Keeley, Reference Keeley1996, p. 146), so the justificatory mechanisms surrounding these three types of combat can in most cases only be inferred indirectly from the actions themselves and records of the cultural systems surrounding them.

1.3 Mass killing and civilian victimization

Research on mass killing and deliberate targeting of civilians offers an additional perspective on social complexity, in the form of state capacity, ideology, and resultant forms of violence. This work indicates that exclusionary ideologies, through which societies premise their identity on excluding unwanted groups from their circle of moral consideration, are a consistent contributing factor in genocide (Bellamy, Reference Bellamy2012; Harff, Reference Harff2003; Nyseth Brehm, Reference Nyseth Brehm2016). This can be manifested through a group’s belief in its higher status and responsibility to civilize others. For instance, norms developed in Europe for protecting civilians, eventually codified in the Geneva conventions, were generally understood as not applying to the ‘uncivilized’ peoples living in colonized areas (Bellamy, Reference Bellamy2012). A common theme in extermination attempts is a desire to return to an idealized past utopia of racial purity, often tied to a romanticized view of pastoral life and a right to conquer and cultivate lands wasted on their current-day, non-farming inhabitants (Kiernan, Reference Kiernan2007). These exterminatory worldviews are also consistent with arguments for greater externalization of violence among complex societies engaged in conquest or colonization (Kim, Reference Kim2012). It is therefore difficult to claim that violence in war decreased with the increase in codified restraints in state societies of any formulation. The present-day need to justify rather than glorify mass slaughter may document a shift in norms of acceptable behaviour in a globally interconnected human community more than actual changes in violence.

Another focus of this research is how extreme violence may be used instrumentally to achieve political goals; namely, when actors view mass killing as less costly than any alternatives (Chirot & McCauley, Reference Chirot and McCauley2010; Straus, Reference Straus2012). In particular, insecure power structures may use mass violence in response to perceived existential external or internal threats (Downes, Reference Downes2006; Straus, Reference Straus2012; Valentino, Reference Valentino2014). This reasoning can apply in different senses to both ancient polities as described above and contemporary nation-states. It is also consistent with the use of terrorism by relatively weak insurgent groups as part of an overall pattern of asymmetric war (Valentino, Reference Valentino2014). These descriptions indicate that levels of violence in war would not necessarily track levels of social complexity (either positively or negatively) but be more pronounced among somewhat centralized political systems lacking security and stability.

Centralized states have greater organizational ability and physical resources to carry out extermination campaigns. However, this capacity for destruction is in absolute rather than proportional terms and does not necessarily indicate more exclusionary morality on the part of states. (Likewise, greater destruction in proportional terms among nonstates does not necessarily indicate the reverse.) Based on this literature, a possible conclusion is that mass killing as an end in itself applies more to social groups with political goals beyond resource acquisition or revenge. This contrasts with the collateral, incidental forms of indiscriminate killing often described in the ethnographic literature on small-scale societies. There are instances of exterminatory warfare among small-scale societies as well (Burch, Reference Burch2005; Karsten, Reference Karsten1935; Keeley, Reference Keeley1996). However, driving factors such as fear of pollution by ideologically or ethnically impure groups, perceived threats to a centralized regime, or desire to achieve a utopia are less likely to be present among these groups. Instead, the motives for war described as exterminatory in intent cited in ethnographies include preventing the enemy from returning to enact revenge (Jivaros, Karsten, Reference Karsten1935), taking possession of the defeated’s territory (Chuuk, Fischer, Reference Fischer1958), and fear of neighbouring population growth impinging on a society’s land and resources (Crocker & William, Reference Crocker and William1990). Based on this data, motives for exterminatory war resemble those for warfare overall among small-scale societies, centring around revenge and resource acquisition.

Multiple factors may promote or restrain indiscriminate violence, including a society’s past experiences of conflict, its internal social structure, norms, and ideologies, and perceptions of threat from other groups or the environment (Chirot & McCauley, Reference Chirot and McCauley2010; Glowacki et al., Reference Glowacki, Wilson and Wrangham2020). Even if the evidence ultimately supports one position or another regarding the historical trajectory of war, attitudes towards outgroup members, norms of combat, and levels and discrimination in the use of violence may not necessarily correlate with the scale, frequency, or effectiveness of war as it has coevolved with social complexity over history. It is therefore possible that no linear relationship will appear between measures of social complexity and indiscriminate killing in war due to overwhelming variation across cultures (or, more pessimistically, a consistently high level of indiscriminate killing).

3.1 Data set construction and sampling

For this study, data were coded from ethnographic documents accessible in the eHRAF World Cultures database, other ethnographic sources including the Ethnographic Atlas (Murdock et al. Reference Murdock, Textor, Barry III, White, Gray and Divale2025) and Standard Cross-Cultural Sample (Murdock & White, Reference Murdock and White2025) accessed through D-Place (Kirby et al., Reference Kirby2016), and secondary historical sources, as well as data stored in the Seshat Global History Databank (Turchin et al., Reference Turchin, Brennan, Currie, Feeney, François, Hoyer and Peregrine2015). The final sample consists of 73 societies documented as cultural units in these sources. Although this is clearly a tiny fraction of the breadth of human history and culture, an attempt was made to create a sample somewhat representative of different geographic regions and levels of social complexity for societies that engaged in war and for which sufficient descriptions of treatment of enemies were found in the literature. However, it should be noted that this is not a probabilistic sample such as the eHRAF Probability Sample Files (PSF; Naroll, Reference Naroll1967), as the main objective was to find relevant descriptions of war in the available materials, so generalizability may be limited. For instance, there may have been biases around which societies had ethnographic records describing the conduct of war. A world map showing the geographic distribution of societies is shown in Figure 2.

Figure 2. Geographic distribution of the societies in the data set.

To construct the data set, eHRAF subject categories, particularly those on armed forces and war, as well as keywords were used in initial searches to find societies for which there was some description of warfare. Materials in eHRAF have been indexed according to the Outline of Cultural Materials (OCM), which contains a range of broad subjects ranging from religion, family life, subsistence, and other aspects of human societies (Ember & Ember, Reference Ember and Ember2013). Specifying relevant subject categories when searching eHRAF is an established method for finding relevant ethnographic records to compile data sets for cross-cultural studies (e.g. Ember et al., Reference Ember, Adem and Skoggard2013). For the purposes of this study, the OCM categories 700 (armed forces) and 720 (war) were primarily used. Keywords potentially contained in relevant descriptions may also be used to narrow eHRAF search results (Fischer & Ember, Reference Fischer, Ember and Chen2018). For this study, keywords related to violence against enemies in war were used to search across and/or within subject categories. A full list of the keywords is in the Supplementary Material. The keywords used would sometimes be adjusted based on the amount of material on warfare for different societies as well as the content of that material: for societies with extensive descriptions of warfare, fewer, more direct keywords were used whereas societies with limited descriptions of warfare were searched with additional, broader keywords to find any relevant descriptions. Those descriptions were then used to code the target variable of indiscriminate killing.

To expand the data set to include historical and archaeological societies, Seshat polities that had been coded with some degree of certainty (present/inferred present or absent/inferred absent) for variables of interest in the ‘Warfare Intensity’ section (e.g. general massacre and extermination) as of 2020 were also added. Although different variables on wartime behaviour were created for this study, these Seshat variables provided an indication of available data and sources on indiscriminate killing. The sources and evidence for the codes on massacre/extermination were checked for each polity included; those for which the evidence was found to be insufficient and no additional sources were found were removed. Societies which could not ultimately be coded for the outcome variable of indiscriminate killing due to insufficient descriptions in the primary/secondary sources were removed. Overall, 54 societies of 73 total were primarily coded from ethnographic sources, generally accessed through eHRAF, 13 from secondary historical sources, and 6 from papers with archaeological data. Fifty-nine societies have a start date after 1500 bce, and 41 of these have a start date after 1900 ce. As ethnographies tended to provide greatest detail for such treatment and these were carried out in more recent times, a majority of the sample, 56 societies, have an end date after 1900 ce.

It should also be noted that this study does not include data for nation-state conflicts from World War I onwards (societies dated after this period are non-state, ethnographically recorded societies). Extensive quantitative work on genocides and civilian targeting in the context of modern nation-state war has been carried out in political science and genocide studies (e.g. Downes, Reference Downes2006; Eck & Hultman, Reference Eck and Hultman2007; Harff, Reference Harff2003). Instead, this study attempts to take a less detailed but broader historical perspective on the relationship between wartime violence and social complexity across different cultures.

3.2 Variables

The predictor variables were social complexity as measured through a principal component analysis of population estimates, degree of centralization, and hierarchical governance levels as coded in the above databases for each society; presence of a formal military structure, and violent political conquest/expansion (full descriptions of the variables in Supplementary Material).

The outcome variable of enemies killed in external war was defined as whether enemy individuals in the following age/gender categories tended to be killed during war: male or female infants/toddlers; children; younger adults; older adults. One point was added for each age/gender category with the minimum being 1 and the maximum being 8 (everyone). Ranges were used for some societies when there was ambiguity or overlap across categories. An overview of these codes is shown in Table 1.

Table 1. Possible codes for enemies killed in war

For instance, the following description of the Aleut was used to code them as 6–8 (all killed except for some who are taken as wives or slaves): ‘The Aleut was a ruthless warrior. He attacked other villages for revenge, new wives, rare stones for charms and weapons, and perhaps most significantly, to divert aggressive impulses to outsiders. War parties, usually led by a chief or one of his relatives, conducted surprise attacks on other villages (Aleut and Eskimo), frequently murdering every person in the village. Prisoners brought back to the villages were accorded diverse treatments. Some became wives and were integrated into the community. Others remained in slave status to be used or abused according to the whim of the owner’ (Jones, Reference Jones1970, p. 35). Codes and descriptions for each society are in the supplementary file ‘inkill_dataevidence.xlsx’.

Age and gender are not necessarily the most salient reasons for why enemies would be killed or spared in a conflict, but scoring in this manner ideally captures variation in scales of violence less arbitrarily than a more undescriptive categorical or numerical scale. There was a lack of evidence for any norms constraining the extent of violence or protection for enemy individuals in most of the societies surveyed. However, it cannot be assumed that for all societies where no limits of war are found in the sources that indiscriminate violence was normal or desirable – the presence or absence of such evidence may result from the vagaries of historical records or the interests of specific researchers. With these caveats, this variable is intended as a proxy for the regular and presumably accepted means of conducting war in these societies. Generally, societies coded 1–3 are those that tended to target adolescent to adult males participating in combat, while those with higher scores, at least based on available descriptions, tended to direct violence more broadly against enemy populations.

3.3 Processing of data pre-analysis

Political centralization, hierarchical levels, and population were treated as continuous variables; military organization and political expansion were treated as binary. The first factor from a principal components analysis (PCA) (using the prcomp() function, stats package in R) on the population, centralization, and hierarchical levels variables was used as a measure of social complexity (cumulative proportion of variance explained = 0.88). Political centralization, hierarchical levels, and population were standardized (mean centred at 0 with a standard deviation of 1) prior to the PCA. Missing data were imputed using the R package mice (van Buuren & Groothuis-Oudshoorn, Reference van Buuren and Groothuis-Oudshoorn2011) using the methods provided for continuous data (predictive mean matching). This was a total of seven datapoints prior to conducting the PCA: five societies missing data for population and two for hierarchical complexity. To ensure basic assumptions of the hypotheses were met, logistic regression models on political expansion and military organization as predicted by the social complexity measure (henceforth PC1) were run. Both variables were predicted by PC1 (log odds = 1.51, p < 0.001 and log odds = 1.40, p < 0.001 for expansion and military, respectively).

3.4 Statistical models

Analyses were conducted using the brms package (Bürkner, Reference Bürkner2017). A set of multiple regression models were used to test effects of military organization, social complexity, and expansion on the outcome variable. They were designed to test the relationships implied by the graph in Fig. 3. The models were constructed based on the adjustmentSets() function in the dagitty package (Textor et al., Reference Textor, van der Zander, Gilthorpe, Liśkiewicz and Ellison2017) that, based on the provided graph, returns the necessary covariates for which to adjust when estimating an effect between two variables. For this graph, the following covariates were returned (for predicting the total effect of the given exposure): social complexity when killing is predicted from military organization; social complexity when killing is predicted from expansion; and none when killing is predicted by complexity. Another model was also run with all three predictors to assess effects of complexity while accounting for the other two variables and to check consistency of estimates.

Figure 3. Directed acyclic graph showing hypothesized relationships between the described variables, with measurement error on the outcome. C = social complexity, E = political expansion, M = military organization, K_true = enemies killed (true value), K_obs = enemies killed (observed value), e_k = measurement error for observed killing.

Models were specified as follows, with one model with all three predictors and then dropping one term each. Priors were chosen to be weakly informative (Prior Choice Recommendations n.d.). To incorporate uncertainty in the outcome variable of enemies killed, it was modelled as an unobserved true value distributed normally with a mean of the observed value (equal to the mean of the range for each society) and standard deviation calculated from the difference from the maximums/minimums and means (based on the model incorporating measurement error in McElreath, Reference McElreath2020, chapter 15 as implemented in brms by Kurz, Reference Kurz2019):

K_obs ∼ Normal(K_true, K_se) [distribution for observed values of enemies killed]

K_true ∼ Normal(µ, σ) [distribution of unobserved true values]

µ = α + β_C*C + β_E*E + β_M*M [linear model with complexity, expansion, and military as predictors]

α ∼ Normal(4, 1.3) [intercept prior]

β ∼ Normal(0, 1) [slopes prior for complexity, expansion, military]

σ ∼ Exponential(1)

Given the ways in which the outcome variable might vary depending on context, it seems unlikely to be influenced significantly by cultural phylogeny or spatial proximity (perhaps excluding cases where two societies fought one another). However, a simplified model using the score ‘kill_mean’ without the error term was run using world region as a varying intercept and compared to a model without it. As there was also a possibility of bias due to the large number of societies in the data set dated to the nineteenth and twentieth centuries when much ethnographic fieldwork was conducted, another model was run with all three main predictors as well as the mean date (standardized) for each society as another predictor.

3.5 Preregistration and data availability

The analyses were preregistered prior to conducting the main analyses (after the PCA and imputation) on the Open Science Framework site (https://osf.io/nskct). The data set with sources and code for analyses are available at https://github.com/kcbasava/historical-war-practices/tree/main/indiscriminate-killing.