Macrohistorical and Evolutionary Dynamics of Between-Group Competition, Sociopolitical Complexity, and Differentiation-Integration Effects

Abstract

Limiting similarity theory (LST) and the principle of competitive exclusion (PCE) affirm that the degree of allowable niche overlap predicts the occurrence of tolerant coexistence between two or more biotic entities. Attribute variation reduces conflict, whereby two biological systems in direct competition for the same type of finite resources are incapable of peacefully existing under conditions of constant population increase. Complex biotic systems often face trade-offs regarding the allocation of relevant bioenergetics resources to different facets of their organization, further increasing the likelihood of attribute differentiation-integration. Evidence in support of the aforementioned perspectives has been found across biological entities, including human societies. Multilevel selection (MLS) theory provides a complementary framework for understanding the evolution of human sociopolitical systems, whereby within-group cooperation is sustained by ultrasocial institutions that can be employed for competition with other groups, such as in warfare. We gathered and analyzed data on sociopolitical complexity, military technologies, and differentiation-integration effects for 360 historical polities (13,000 BC-1895 AD) from the Equinox 2020 database. A cascade model detected a positive effect of Time on the evolution of military technologies. In turn, Military Technology negatively influenced the level of Military Technology Differentiation-Integration, indicating that some polities specialized in developing military technologies in response to local challenges, a finding consistent with LST and PCE. The model revealed that whereas Military Technology increases Sociopolitical Complexity, a result supportive of MLS, Military Technology Differentiation-Integration had a significant and negative influence on this criterion variable. Greater Sociopolitical Complexity also negatively influenced the degree of Sociopolitical Differentiation-Integration.

Appendices

Appendix 1. Macrohistorical hypotheses on the evolution of complex societies

Tainter (1988) provided a summary of the most common lineages of thought for explaining the origins of the state. His account categorized these approaches into four main forms: internal conflict, external conflict, managerial, and synthetic. Theories of internal conflict focus on the evolution of organizational stratification. Such theories attempt to showcase how the elite capture of resources negatively affects social cohesion. Tainter references the work of Fried (1967) and Childe (1951) here, noting that both authors held that the formation of the state allowed for the protection of elite resources from the rest of the population. This view was originally approached via a Marxist framework, where inequality between the classes was posited to yield internal conflict that could drive more advanced forms of social organization, such as the state. Theories of external conflict examine the effects of between-group competition on military innovation and the development of state institutions geared towards defensive fortifications, offensive campaigns, and the administration of controlled territories. Webster (1975) argued, for instance, that infrequent warfare can only offer near-term benefits to a polity; however, continuous warfare decreases the likelihood of within-group competition, compelling people to instead prioritize organizational stability and effective leadership. Managerial theories discuss how states respond to challenges of societal integration caused by external disruptions (such as changes to the availability of resources) or population dynamics (such as rapid growth or loss). Tainter (1988) cites Willey and Shimkin’s (1973) notion that the stresses of the Late Classic era in Mesoamerica were amplified by inadequate responses from the elite; for example, if a polity experiences an ecological disruption (e.g., drought or famine), whether the aristocracy mitigates the problem by either modifying their institutional response to the crisis or by developing a novel solution. Tainter also assigns the fall of the Indus Civilization to the concentration of power in too few hands, arguing (with reference to Erwin, 1966) that polities are more durable when organizing hierarchies with greater dispersion of elite responsibility. Lastly, synthetic theories focus on the possibility that the state originated from multiple influential causes interacting simultaneously. Renfrew (1972) argues that a sequence of developments, starting with agriculture altering social hierarchies and leading to those hierarchies influencing the modes of production, can account for the origin of the state.

Beyond theories of the development of the state are those focused on aggregation, which discuss the ability of the state to increase in complexity. One major emphasis is on the utility of agriculture for population-level growth and the clustering of large groups of people near one another. This is possible through both the increase in food resources provided by agriculture and the concentration of wealth and utility resources such as buildings, tools, land, and farm animals. Additionally, the rise of agriculture contributed to the development of alternative forms of conflict and possibilities for defensive fortification, provided by increased population size and the expansion of urban institutions. Another perspective on the origin of the state was provided by Turchin et al. (2018), who emphasized the importance of cavalry, as well as metals for activities such as smelting, in generating the state. Turchin tested these various hypotheses (from agriculture to those mentioned by Tainter, as well as his own) in multiple publications, finding warfare to be the best fit after quantitative model comparison (Turchin, 2003; Turchin et al., 2018, 2022a). These findings provide strong support for the importance of military activity in the development of the state; however, to this date, a latent dimension of Military Technology has not been generated (via factor analysis and unit-weighted factor estimation instead of principal component analysis to determine the underlying structure). This builds upon Turchin’s work by examining the development of complexity within military systems alongside advancements of the state, rather than focusing on specific aspects of military innovation, such as cavalry, alone.

Appendix 2. Differentiation-integration effects

The interpretation of latent variable modeling ultimately rests on the strength of the correlations among observable variables (Gorsuch, 2014). However, the magnitude of these correlations may vary due to the moderating influence of a third variable. Differentiation-integration effects follow bioeconomic principles wherein an input increase does not guarantee a corresponding rise in a proportional output (Spearman, 1928). For instance, doubling the amount of relevant bioenergetic resources extracted, allocated, and consumed by a system will not ensure the organization will double in its complexity. Consequently, the saturations of sociopolitical and military technological factors, which represent the strengths of the factor loadings (Spearman, 1928), are predicted to be more cohesive at the lower end of their respective distributions, whereby the higher the available levels of bioenergetic resources, the fewer the benefits obtained from further increasing the level of complexity of the various sociopolitical or military technological features.

It is feasible to generate a typology of four categories of sociopolitical systems that arise as a combination of low to high Social Differentiation in conjunction with low and high levels of Sociopolitical Complexity (see Fig. 3). The first type, quadrant 3a, features high sociopolitical values and displays strong integration among its sociopolitical indicators (generalism). The second type, quadrant 3b, arises from high levels of Sociopolitical Complexity in conjunction with a weakening in the magnitude of correlations between the indicators, revealing differentiation effects (specialism). The third type, quadrant 3c, refers to organizations characterized by low Sociopolitical Complexity levels and strong positive correlations among its indicators, hence generalism. The fourth type, quadrant 3d, exhibits low levels of Sociopolitical Complexity and the investment in specific sociopolitical attributes as evidenced by the weaker correlations among the factor’s indictors, hence specialism.

Fig. 3

Typological diagram of hypothetical sociopolitical profiles derived from a similar typology developed by Woodley on differentiation-integration effects (2011). These hypothetical profiles feature six sociopolitical indicators varying in their loadings. The four combinations arise as a function of overall Sociopolitical Complexity and Sociopolitical Differentiation, hence (a) high Sociopolitical Complexity and low Sociopolitical Differentiation; (b) high Sociopolitical Complexity and high Sociopolitical Differentiation; (c) low Sociopolitical Complexity and low Sociopolitical Differentiation; (d) low Sociopolitical Complexity and high Sociopolitical Differentiation. Differentiation-integration effects are structured as a continuum with low differentiation implying high integration and vice versa

Although a methodological alternative would be to dichotomize the data based on an arbitrary cutoff distinguishing high from low scores on the latent variable, and subsequently compare the factor loadings of each group (higher factor loadings found in the lower factor scores group; lower factor loadings found in the higher factor scores group), this procedure suffers from artificially reducing the amount of available variance in the model. Instead, a continuous parameter estimation model (CPEM; Gorsuch, 2005) was used to detect differentiation-integration effects (Figueredo et al., 2013). Traditionally, Pearson’s correlation coefficient of the population is equivalent to the sum of the standardized cross-product between two vectors divided by the sample size (r_xy = Σ(z_x × z_y)/n). A CPE correlation coefficient is equivalent to the standardized cross-product between two vectors (CPE(r_xy) = (z_x × z_y)). Consequently, the cross-product group means automatically becomes the correlation coefficient of each of the groups under consideration. Computing and comparing the group means of these cross-products (z_x × z_y) using ANOVA calculates and compares the correlation coefficients, evidencing how the strength of these associations varies across the various groups of interest.

CPEM provides the unique advantage of using the aforementioned CPE cross-products along with other statistical predictors. Thus, it is feasible to examine what predictor modified the magnitude of the correlation. Usually, it typically requires between 75 and 100 data points per group for the correlation coefficients to reach a point of statistical stability necessary for these statistical tests. However, CPEM does not require the polytomization of a continuous distribution (Figueredo et al., 2015). Moreover, CPEM can also determine the magnitude of the hypothesized factor loadings by operationalizing the latent common variable’s loadings as the correlation coefficients between each standardized indicator with the standardized unit-weighted factor scores. CPEM can estimate the pertinent statistical parameters at the individual casewise level for each of the estimated factors by taking the standardized cross-products of the specific factors with the latent common factor. Therefore, (z_x × z_y) becomes an individual casewise raw score used to test the common factor’s differentiation-integration effects for each group. This procedure has been used to understand the structure of other latent constructs, such as the domain-general cognitive ability factor and life history factor. CPEM does not require gigantic sample sizes often needed for subgroup data stratification based on artificial cutoff points. This procedure allows for examining the continuous covariation of cross-products with the individual casewise differences of any of the indicators’ scores. Furthermore, the CPEs (computed as the cross-product between the standardized values of the indicators and the standardized factor scores) can later be used to calculate a Differentiation-Integration factor loading onto the CPEs of the factor’s indicators.

Thus, the procedure operates as follows. Consider three sociopolitical indicators: population size, territory size, and hierarchical levels (although we used nine sociopolitical indicators in the analyses, the same statistical principles apply to this hypothetical example based on three variables). First, all three variables are usually transformed into z-scores. Second, the unit-weighted Sociopolitical Complexity factor scores are computed as an average across all three standardized variables:

$$\begin{aligned}Unit \text{-}&Weighted\ Sociopolitical\ Complexity\ Factor\ Scores=\\&\left[z(population\ size)+z(territory\ size)\right. \\ & \left.+z(hierarchical\ levels)\right]/3\end{aligned}$$

Third, the factor scores are standardized. Fourth, CPEs are computed as the cross-product between the standardized indicator scores and the standardized factor scores:

$$\begin{aligned}CPE\ population\ size =&\ z(population\ size) \\&\times z(Sociopolitical\ Complexity);\end{aligned}$$

$$CPE\ territory\ size = z(territory\ size)\times z(Sociopolitical\ Complexity);$$

$$\begin{aligned}CPE\ hierarchical\ levels =&\ z(hierarchical\ levels)\\& \times z(Sociopolitical\ Complexity)\end{aligned}$$

Fifth, the CPEs themselves are standardized. Fifth, unit-weighted Sociopolitical Differentiation-Integration factor scores are computed as an average across all three standardized CPEs:

$$\begin{aligned}Unit &\text{-}Weighted\ Sociopolitical\ Differentiation\\&\text{-}Integration\ Factor\ Scores=\left[z(CPE\ population\ size)\right. \\ & \left.+z(CPE\ territory\ size)+z(CPE\ hierarchical\ levels)\right]/3\end{aligned}$$

This factor is operationalized as a continuum, with lower scores (weaker correlations between the factor and the indicators) indicating differentiation and higher scores indicating integration (stronger correlations between the factor and the indicators). Due to the bioeconomic principle of diminishing returns, we predict that as sociopolitical systems increase in complexity, they are expected to allocate their resources to some indicators rather than others. Hence, a negative influence of the Sociopolitical Complexity factor scores on the Sociopolitical Differentiation-Integration factor scores indicates that the greater the complexity is, the weaker the correlations among the indicators in the sociopolitical factor. Figure 4 visually represents the moderating effects of the CPEs as a function of the factor scores. It is also possible to examine whether the differentiation-integration effects are predicted by another variable, for example, another latent common factor (e.g., Military Technology factor). Figure 5 represents his hypothetical scenario.

Fig. 4

Diagram representing differentiation-integration effects as a function of factor scores influencing the model’s factor loadings (represented by the Greek letter λ). The curved arrows represent the individual casewise CPEs

Fig. 5

Diagram representing differentiation-integration effects as a function of factor scores and a second latent factor (II) influencing the first factor’s loadings (represented by the Greek letter λ). The curved arrows represent the individual casewise CPEs

Appendix 3. The principle of Brunswik Symmetry

Mayr (1982) described the elements of most biological systems as being arrayed within a constitutive hierarchy, characterized by multiple ascending and descending levels of organization. Thus, if both the predictor and the criterion variables being modeled within a hypothesized causal relation belong to such hierarchically organized systems, there is a careful selection to me made in establishing the corresponding levels of structural linkages between them. Brunswik developed the principle of symmetry to describe the properties of structural relations among different levels within hierarchically organized systems (Wittmann, 2002, 2011; Wittmann & Klumb, 2006).

The essence of Brunswik Symmetry can be summarized as stating that any structural relation estimated between a predictor and a criterion variable must be carefully specified according to which respective level of psychometric aggregation of the predictor variable is presumed to directly influence which respective level of psychometric aggregation of the corresponding criterion variable. This is because all levels in each hierarchy will be connected indirectly and they will therefore all be mutually inter-correlated. According to the principle of Brunswik Symmetry, the magnitude of the structural relation will be maximized when both the predictor variable and criterion variable are operationalized at the same level of psychometric aggregation, representing their corresponding levels within their respective constitutive hierarchies. Following this recommendation is also wise because different causal dynamics might apply at different hierarchical levels of biological and behavioral organization (Figueredo et al., 2007, 2011).

Thus, when relating our multivariate construct for Military Technology to our multivariate construct for Sociopolitical Complexity, we do not attempt to predict the aggregate for latter from the disaggregated individual manifest indicators of the former (e.g., cavalry), but instead, we predict one latent common factor from the other. Both of these constructs were developed using principal axis factoring (PAF) and estimated using unit-weighted factor (UWF) scoring. This preserves the Brunswik Symmetry between corresponding constructs at the same level of psychometric aggregation. The same is true for the inclusion of the multivariate construct for the Military Technology Differentiation-Integration and the Sociopolitical Differentiation-Integration factors. The Military Technology Differentiation-Integration factor was then used as mediators for the structural relation between the Military Technology and the Sociopolitical Complexity factors. Afterwards, we tested the effect of the Sociopolitical Complexity factor, as well as all the previous latent variables in the cascade model, on the Sociopolitical Differentiation-Integration factor.

Appendix 4. Type 1 multilevel selection and type 2 multilevel selection

Originally advanced by Heisler and Damuth (1987), and most thoroughly articulated by Okasha (2009), MLS has been described as two types of phenomena: MLS1 and MLS2 (Damuth & Heisler, 1988; Hertler et al., 2020; Okasha, 2009). MLS1 posits that phenotypes and their corresponding fitness values are properties of individuals rather than of groups. Consequently, selection operating at the level of groups is analytically defined by MLS1 as produced by the synergistic effects of group membership on individual-level fitness values. MLS1 considers genes to operate as the ultimate replicators and considers both individuals and groups to operate as vehicles for those genes (Hertler et al., 2020; see also Dawkins, 1976). In MLS1, individual fitness gains may be potentiated by group membership, such as those achieved by forming coalitions for the purpose of competing with other similarly constituted groups, where the added strength in numbers facilitates the victory of some groups over others. Nevertheless, MLS1 still assesses any gains in fitness resulting from this kind of cooperative enterprise at the level of the individual group member.

Alternatively, MLS2 posits that phenotypes and fitness values can also be described as properties of groups, in addition to remaining properties of the individuals within the groups. Groups are then thought to spawn other (“daughter”) groups and thus also operate as replicators (Damuth & Heisler, 1988; Heisler & Damuth, 1987; Hertler et al., 2020; Okasha, 2009). Regrettably, although some persuasive empirical evidence has been presented for the existence of MLS1, we are not aware of any conclusive empirical evidence ever presented for the existence of MLS2 (Hertler et al., 2020). Thus, the predictions tested in the present paper are derived exclusively from MLS1 theory and not from MLS2.