Assessing different historical pathways in the cultural evolution of economic development.

A huge number of hypotheses have been put forward to explain the substantial diversity in economic performance we see in the present-day. There has been a growing appreciation that historical and ecological factors have contributed to social and economic development. However, it is not clear whether such factors have exerted a direct effect on modern productivity, or whether they influence economies indirectly by shaping the cultural evolution of norms and institutions. Here we analyse a global cross-national dataset to test between hypotheses involving a number of different ecological, historical, and proximate social factors and a range of direct and indirect pathways. We show that the historical timing of agriculture predicts the timing of the emergence of statehood, which in turn affects economic development indirectly through its effect on institutions. Ecological factors appear to affect economic performance indirectly through their historical effects on the development of agriculture and by shaping patterns of European settler colonization. More effective institutional performance is also predicted by lower-levels of in-group bias which itself appears related to the proportion of a nation's population that descends from European countries. These results support the idea that cultural evolutionary processes have been important in shaping the social norms and institutions that enable large-scale cooperation and economic growth in present-day societies.

Introduction

Economic development is not equally distributed around the world. In 2017, the total GDP of the top 4 countries in the world exceeded the total GDP of the remaining countries despite only representing about 25% of the world's population (IMF, 2018). Economists and other researchers have long debated the proximate causes of development in terms of the technological and policy factors that create economic growth (Spolaore & Wacziarg, 2013). Many researchers argue that institutions that encourage the participation of more of the population in economic affairs, enable markets, and provide appropriate incentives (what have been labelled as “inclusive” institutions (Acemoglu & Robinson, 2012, United Nations, 2016)) are of key importance in explaining why some countries have been more successful economically than others (Acemoglu & Robinson, 2012; Milgrom, North, & Weingast, 1990; North, 1990; Rodrik, Subramanian, & Trebbi, 2004). Researchers have also sought to understand how historical (e.g. Bockstette, Chanda, & Putterman, 2002, Michalopoulos & Papaioannou, 2013, Putterman, 2008, Putterman & Weil, 2010, Spolaore & Wacziarg, 2013) and geographical or ecological factors (e.g. (Bonds, Dobson, & Keenan, 2012, Hibbs Jr. & Olsson, 2004, Sachs & Malaney, 2002) have shaped the development of societies and their economies. However, the causal pathways and relative importance of these various processes are heavily disputed.

Here we employ cultural evolutionary theory (Henrich, 2016; Mesoudi, 2011; Richerson & Boyd, 2005) as an organising framework to examine how these alternative explanations fit together and to test between competing hypotheses. Here we distinguish between the features of current societies that affect economic growth (proximate explanations) and the processes that have occurred in the past that have shaped the modern day situation (historical explanations) (Currie et al., 2016; Wilson & Gowdy, 2013). Rather than treat historical explanations purely as contingent events, an evolutionary ecological perspective leads us to consider how social and economic systems may evolve in response to environmental factors, while the socio-cultural traits that help structure societies can be thought to culturally evolve and are shaped by processes of mutation, selection, and inheritance (Currie, Turchin, Turner, & Gavrilets, 2020; Mesoudi, Whiten, & Laland, 2004). Within the historical and ecological explanations, we distinguish between events or factors that have directly shaped modern day economic outcomes from more indirect pathways where historical processes have shaped the evolution of proximate determinants. For example, a present day society's institutions are inherited, built upon in a cumulative manner, and modified from those of past societies (Currie et al., 2016; Mesoudi, 2016; Nunn, 2012; Rodrik et al., 2004; Spolaore & Wacziarg, 2013; Turchin et al., 2018), while other factors may shape the selective environment that affects what type of institutions become more frequent (Currie et al., 2020; Richerson & Henrich, 2012b). Using this framework, here we collate some of the main hypotheses by which historical and ecological factors have been proposed to affect modern-day economic outcomes, and articulate the potential causal pathways through which these factors may be linked. We then attempt to statistically assess the degree of support for these pathways using cross-national data.

Proximate explanations

A number of different factors have been argued to be important in directly determining economic development. We can divide these into the features of the populations themselves (“social” factors) and the external context in which populations are situated (“environmental” factors). Theories involving these proximate explanations tend to be short-term in focus, arguing that changes in culture or ecology will have immediate knock-on effects for economic performance.

Social: Social rules (institutions) and norms govern social interactions and enable cooperation between individuals and organisations (Currie et al., 2021; Gavrilets & Richerson, 2017; North, 1990; Ostrom, 1990; Ostrom, 2000; Powers, van Schaik, & Lehmann, 2016). Effective institutions are central to the function of markets and governments; enabling contracts to be enforced, providing checks on potentially predatory elites, and shaping incentives for investment and improvement (Acemoglu & Robinson, 2012; Aoki, 2001; Greif, 2006; Rodrik et al., 2004). Acemoglu and Robinson, 2012) refer to such good institutions as “inclusive” because they enable more of the population to take part in economic and political activities. More informal social norms such as trust (Ahlerup, Olsson, & Yanagizawa, 2009; Fukuyama, 1996; Greif, 1994) can help promote economic activity and help support formal institutions. On the other hand, some social norms such as in-group preferences can lead to opportunism and corruption, which stymies economic activities (Gorodnichenko & Roland, 2011; Kyriacou, 2016). Another important factor may be the more general body of knowledge, education, habits, personality attributes (e.g. creativity) known as human capital that enables a population to conduct economic activities and create economic growth {Easterly, 2012 #36;Glaeser, 2004 #105}.

Environmental: Environmental factors such as rainfall, temperature, and soil type may influence economic development due to their effects on the types of crop grown and productivity of agriculture (Galor & Özak, 2016; Lanzafame, 2014; Mayshar, Moav, Neeman, & Pascali, 2015). Infectious disease may also be a major contributor, as high rates of disease reduce labour productivity and raise uncertainty (Sachs & Malaney, 2002). This has been brought home to those of us in the western world due to the Covid-19 pandemic, yet many lower income countries experience the immediate and persistent negative impacts of infectious diseases such as malaria. As another example, the 2014–2016 Ebola epidemic drastically cut the income growth estimates of Guinea, Liberia and Sierra Leone, implying forgone income of $1.6 billion for those countries combined (World Bank, 2015).

Historical explanations

Direct Longer-term, historical events and the experiences of ancestral societies may exert persistent effects on the economies of their descendants. The fact that some societies developed agricultural forms of production (herding and cultivation), centralized states or industrialisation earlier than others may have provided a head-start to those societies (Diamond, 1997; Morris, 2010; Putterman & Weil, 2010) and may have given them an advantage over other societies by allowing them to establish and maintain favourable positions in networks of interactions. In more recent history some European countries profited from establishing colonies in other parts of the world that extracted resources and exploited native populations (Acemoglu & Robinson, 2012; Nunn, 2008).

Indirect Another possibility is that historical processes such as the development of agriculture and the emergence of states have influenced the evolution of modern institutions, norms, technology and human capital. Cultural traits such as norms and institutional rules are inherited across generations and are shaped by those that preceded them (Currie et al., 2016). Complex collective action problems may not be easy to solve directly through conscious forward planning. Instead they may require long periods of experimentation and progressive refinement and accumulation in order to develop the kinds of norms and institutions that lead to positive economic outcomes (Currie et al., 2020; Richerson & Henrich, 2012b). For example, democratic traditions in ancestral societies predict how effective the democratic systems of their descendants will be (Giuliano & Nunn, 2013). In a similar way, the development of institutions that allow more people to participate in economic and political activities such as the rule of law may lead to the development of further such inclusive institutions (Acemoglu & Robinson, 2012).

Differences between countries in how long they have been organized as states may contribute to the development of traits associated with economic success in a number of ways. The centralization of governance is thought to be an important mechanism for facilitating cooperative interactions and enabling the coordination of individuals over large geographical areas (Spencer, 2010; Turchin, Currie, Turner, & Gavrilets, 2013; Turchin, Currie, Whitehouse, Francois, et al., 2018) and centralization is argued to be a prerequisite for effective institutions (Acemoglu & Robinson, 2012). A long history of statehood may also reduce lower-level in-group biases and lead to the emergence and spread of impartial social norms (Hruschka & Henrich, 2013). States also create public goods and infrastructure that lead to increases in human capital and facilitate technological innovation and production (Murtin & Wacziarg, 2014; Sokoloff & Engerman, 2000).

The evolution of politically centralized states may itself have been influenced by the historical development of agriculture, which led to sedentism, specialisation of labour, increased population densities (Currie et al., 2020; Johnson & Earle, 2000; Mattison, Smith, Shenk, & Cochrane, 2016). The development of agriculture may also affect the cultural evolution of other aspects of societies too (see Table 1). Variation in the timing of the development of agriculture and the emergence of states may have in turn been influenced by environmental factors (Currie et al., 2020; Currie & Mace, 2009).

Table 1

Different hypotheses and predicted relationships tested in this study. Arrow number relates to visual representation of these relationships in Fig. 1. The table presents only the hypotheses from the literature that motivated examination of these relationship in this study. Alternative hypotheses or explanations for such relationships are possible.

Predictor variables	Hypothesis	Prediction	Arrow no.
Social predictors
Institutions	Adjudication of contracts and enforcement of law allows large-scale cooperation. Checks on the executive ensure incentives for labour and skill accumulation (Acemoglu & Robinson, 2012, Aoki, 2001, North, 1990).	InQ (+)➔ GDP	1
In-group bias	Differences in standards used to treat in-group and outgroup members introduce risks of opportunism in transactions. Nepotistic aspect of these biases also contributes to political patronage and corruption (Kyriacou, 2016)	IGB (−)➔GDP	3
	In-group preferences may prevent adoption of more inclusive rules or prevent such rules being implemented effectively (Greif, 2006).	IGB (−)➔ InQ	2
Historical and ecological predictors
European descent	A body of knowledge and technologies associated with European populations aids economic activity. Assumes that European colonial settlers brought with them cultural traits and human capital that had developed in the context of the Europeans' societies, which may have aided economic development (Easterly & Levine, 2012).	EA (+)➔GDP	12
	Europeans developed relatively inclusive institutions when they settled in large numbers. Where they did not settle in large numbers, they established authoritarian systems designed to exploit populations and extract natural resources (Acemoglu & Robinson, 2012).	EA (+)➔ InQ	13
	European culture is relatively individualist and impersonal, and Europeans would have brought such cultural traits with them when settling in colonial countries (Schwartz, 2006).	EA (−)➔ IGB	14
State history	Historical experience with central organization is heritable and predicts greater levels of economic development in the present day (Putterman & Weil, 2010, Spolaore & Wacziarg, 2013).	SH (+)➔ GDP	4
	Political centralization is an important pre-cursor for development of inclusive institutions (Acemoglu & Robinson, 2012).	SH (+)➔ InQ	5
	Centralized governance selects for cultures of trust and impersonal treatment (Hruschka & Henrich, 2013).	SH (−)➔ IGB	6
Timing of agricultural transition	Earlier transitions provided a head-start to the development of important technologies associated with economic performance (Diamond, 1997).	Ag (+)➔GDP	8
	Longer histories of features of agricultural subsistence (irrigation, large-scale coordination) suggest more experience with property rights (Baland & Platteau, 1998, Olsson & Paik, 2016), which may aid development of and engagement with centralized institutions.	Ag (+)➔ InQ	9
	Agricultural production benefits from collectivist norms, implying that agriculture selects for in-group biases (Olsson & Paik, 2016).	Ag (+)➔ IGB	10
	Growing population sizes associated with agriculture select for centralized governance to maintain cooperation and coordination (Diamond, 1997).	Ag (+)➔ SH	11
Disease	Disease affects workforce directly and stunts productivity (Sachs & Malaney, 2002).	Dis (−)➔GDP	15
	Increased pathogen risk favours more insular social norms to reduce the probability of contracting diseases from other groups (Fincher & Thornhill, 2012).	Dis (+)➔ IGB	16
	The disease environment influenced the extent of European settlement (McNeill, 1977)	Dis (−)➔ EA	17
	Pathogen stress selects for smaller and more numerous groups (Fincher & Thornhill, 2012), which may inhibit the formation of larger centralized states	Dis (−)➔ SH	7
Latitude	Latitude covaries with climate and natural resources that are important in economic development (Bonds et al., 2012).	Lat (+)➔GDP	18
	Latitude covaries with natural endowments which predict the extent of bias of resources towards elites (Easterly & Levine, 2012, Engerman & Sokoloff, 2012).	Lat (−)➔ InQ	19
	Latitude covaries with the suitability of regions for agriculture (Olsson & Paik, 2013) and may affect the origin and spread of agriculture (Diamond, 1997).	Lat (+)➔ Ag	20
	Environmental factors in higher latitudes may have been more conducive to evolution of centralized societies (Currie & Mace, 2009)	Lat (+)➔ SH	21
	Populations tend to migrate to regions of ecological similarity, with European settlers tending to colonize regions of similar latitude to Europe (Diamond, 1997)	Lat (+)➔ EA	22
	Latitude co-varies with climate, with in-group bias being lower in less-demanding temperate climates (Van de Vliert, 2011)	Lat (−)➔ IGB	23
	Latitude covaries with environmental variables that predict severity of infectious disease (Bonds et al., 2012).	Lat (−)➔ Dis	24

Fig. 1

Full path diagram showing all the theoretically-informed relationship between different variables that we assess in this study. Red arrows relate to direct, proximate explanations of economic performance, blue arrows reflect direct historical explanations, while black arrows indicate relationships that could be involved in more indirect pathways. Dashed lines are used to indicate situations where an arrow runs “behind” another variable in this figure but does not indicate an association with that variable (e.g. arrow number 19 simply reflects the potential relationship between Latitude and Institutional Quality, and has nothing to do with European ancestry). Numbers on arrows relate to numbers in Table 1. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Another indirect pathway involving environmental factors that has been studied involves the processes and effects of European colonialism. Colonization and colonial rule happened in different ways in different regions based on the extent to which Europeans settled in large numbers and replaced indigenous populations, which has had long lasting effects on the composition and features of countries in the present day. Broadly speaking, Europeans colonists are thought to have settled in large numbers in regions where existing populations were at low density with less complex forms of socio-political organization and where there was less exposure to unfamiliar diseases (i.e. North America, Australia, New Zealand) (Acemoglu & Robinson, 2012; Crosby, 2004; Diamond, 1997; McNeill, 1977). It has also been argued that where Europeans settled, they displaced indigenous societies and established colonies that were based on the institutions and organization of the societies from which they came, and settlers brought with them cultural traits and human capital that had developed in the context of the Europeans' societies that may have aided economic development (Easterly & Levine, 2012; Glaeser, Porta, Lopez-de-Silane, & Shleifer, 2004; North, 1990). In colonies where large-scale societies had already existed (e.g. Central America, Peru), where ecological conditions were less similar to those in Europe (e.g. more tropical climates), or disease burdens were high (e.g. malaria in West Africa (Crosby, 2004)), Europeans settled in smaller numbers, normally as an elite, and established institutions that often extracted labour and resources from the native populations. It has been argued that larger European settler colonies established societies that developed more “inclusive” institutions (at least from the from the perspective of settler populations) that were more conducive to economic growth, than the kinds of societies characterized by a smaller, extractive elite (Acemoglu & Robinson, 2012). Under this view, environmental differences are thought to have shaped economic development indirectly by affecting population movements and the evolution of institutions and culture.

Shared history

Another insight that cultural evolutionary theory provides is that societies may share features in common because they share historical connections. Over time societies can diverge and split into separate populations with each population inheriting the features of the original population (Mace & Pagel, 1994), which over longer time periods creates broader and often nested patterns of cultural similarity across societies. For example, societies in Polynesia share many cultural and linguistic features, due to the fact that they are descended from populations that first reached the western Polynesian islands around 3000 years and later spread to other further flung islands (Kirch & Green, 2001). Similarly, many societies across Europe, southwest Asia, and the Indian sub-continent speak languages (“Indo-European”) that are related in a family-tree-like manner, and can be traced back to a common ancestral population that existed around 6000–9000 years ago (Bouckaert et al., 2012). These similarities between languages have also been demonstrated to be related to similarities in other cultural and social features (Fortunato, 2011; Silva & Tehrani, 2016; Welzel, 2013). For economic issues, this is important in understanding how societies come to possess traits that lead to positive economic outcomes. Even in cases where traits are borrowed from another society, traits may spread more readily to societies that are more closely related historically as they share other aspects of culture in common that make the new traits easier to adopt or more effective (Spolaore & Wacziarg, 2013; Spolaore & Wacziarg, 2016).

This is important for practical purposes as cross-country analyses often fail to account for shared cultural history and treat societies as independent data points. However, clustering due to shared history violates the assumptions of standard statistical techniques such as ordinary least-squares regression, and can inflate parameter estimates or lead to spurious estimates of statistical significance (Felsenstein, 1985; Harvey & Pagel, 1991). For example, Fincher and Thornhill (2012) find a correlation between in-group assortativeness and parasite stress across countries. However, there is strong cultural-geographic clustering in these data and analyses have demonstrated that the strength or significance of these simple correlational relationships can diminish when accounting for the non-independence of the data (Bromham, Hua, Cardillo, Schneemann, & Greenhill, 2018; Currie & Mace, 2012). Studies by Matthews, Passmore, Richard, Gray, and Atkinson (2016), and Sookias, Passmore, and Atkinson (2018) have shown that deep cultural ancestry as proxied by linguistic relationships, and more recent cultural connections, are predictive of similarities between countries in socio-economic phenomena, and point to the need to incorporate such information in analyses of economic development. Recent analyses suggest that this issue may also be common to other comparative studies looking at historical effects on modern day economic development (Kelly, 2019).

Testing alternative hypotheses and pathways

The discussion above demonstrates that there are a number of alternative hypotheses and a number of different causal pathways through which different factors may affect economic outcomes. Based on the concepts and factors discussed in the introduction we can identify a large number of potential combinations of variables and pathways that could be tested based on the literature, which we summarize in Table 1. To make all of the theoretically-informed connections between different factors explicit we construct a path diagram (see Peregrine, Ember, and Ember (2007) for an example of this approach as applied to modelling state origins). Being driven by theoretical considerations helps avoid over-interpreting potentially spurious correlations. Having identified the hypotheses to be tested we then statistically assess the strength of support for these different predictions using model selection techniques. The outcome of these analyses is a reduced path diagram that illustrates the pathways that receive support from these analyses, and importantly indicates which pathways and hypotheses are not supported. It is important to emphasize that associations described in Table 1 could be due to other causal processes other than those indicated by these focal hypotheses. After assessing which relationships do receive statistical support given the various controls and features of this analysis, we return to the issue of alternative explanations in the discussion, and assess how future analyses may be able to address these issues.

Methods

Variables

Here we examine the different pathways by which different variables predict variation in nations' Gross Domestic Product (GDP) using the World Bank's measure of GDP per capita for the year 2011. We assess the hypotheses outlined in Table 1 using the following variables taken from previously published datasets (Table 2):

Table 2

Descriptive statistics and information about the variables examined in this study. Values for State History, and Timing of Agricultural Transition have been adjusted using the population ancestry matrix.

Variable	Min	Mean	Max	Notes
GDP	355.6	16,301.5	100,575.1	US$ per person, log10 transformed for analyses
Institution Quality	−1.61	0.11	2.02	Composite “rule of law” variable, standardized score, higher values indicate more “inclusive” institutions
In-Group Bias	−2.32	0.14	2.39	Composite standardized score of different measures, higher values indicate more in-group bias
European Ancestry	0	45.19	100	Percent of contemporary population that derive from European countries
State History	0.06	0.56	0.96	Index (range 0–1) estimating the extent to which there existed governance beyond the tribal level the period 1–1500 CE. Higher values indicate more experience with extensive, centralized forms of socio-political organization
Timing of Agricultural Transition	1224	5559	9568	Number of years before 2000 that populations switched from foraging to food production
Disease	−1.18	0.10	1.20	Standardized score of estimates of historical disease prevalence, higher values indicate greater prevalence of disease
Latitude	0.02	30.43	61.92	Mid-point absolute latitude of countries, decimal degrees

Institution Quality (InQ) is based on the World Bank's Worldwide Governance Indicators database (version VII, Kaufmann, Kraay, and Mastruzzi (2008)). Following Nunn and Puga (2012) we use the composite variable “rule of law”, which is a measure of individuals' confidence in rules of their society and the extent to which they abide by those rules. There is a particular focus on the quality of contract enforcement, property rights, the police, and the courts, which are institutional features that enable more people to take part in economic and political activities in a way that provides security and reduces the likelihood of crime and violence.

As an example of the role that social norms may play in affecting economic performance, in this study we examine measures of In-Group Bias (IGB). Our data on IGB here are taken from Van de Vliert (2011), who used data on compatriotism, nepotism and familism from three different sources (the World Values Survey, the World Economic Forum and the GLOBE study respectively) to calculate an overall national-level estimate of in-group bias using factor analysis. The national level estimates of the three underlying forms of in-group bias were themselves derived from individual level surveys. Attitudes towards in-groups can potentially be affected by a number of different processes, however, our assumption is that this variable is at least partly shaped by socially learned information from others about what is deemed to be appropriate behaviour towards others (Hruschka & Henrich, 2013; Jordan, McAuliffe, & Warneken, 2014; Van de Vliert, 2011).

We include measures of European Ancestry (EA) as a means of testing hypotheses that propose that European settlers brought with them cultural traits, technology, or other aspects of human capital that may have affected economic development. It is important to be clear that the specific aspects of culture or technology implicated in these ideas are not directly measured. Rather the use of this variable is based on the predictions of those hypotheses outlined in the introduction and Table 1. As such any relationship between European ancestry and the social & economic factors demonstrated here may be due to other causal processes, as we expand upon further in the discussion. Our EA variable provides an estimate of the extent to which European colonization has affected the contemporary demographic profile of a country. The variable used in this study measures the percentage of a country's population in year 2000 that is descended from people who resided in Europe in 1500. Our variable was sourced from Nunn and Puga (2012) but was ultimately calculated from the Putterman and Weil (2010) migration matrix (see below).

State History (SH) is taken from Putterman and Weil (2010) and uses historical sources to estimate the extent to which there existed state-level governance in the geographical locations of present-day countries over the period 1–1500 CE. Regions that score higher on this variable have had more experience with extensive, centralized forms of socio-political organization than regions that score lower on this variable.

Timing of Agricultural Transition (Ag) is also taken from Putterman and Weil (2010) and provides an estimate of the number of years prior to 2000 when the first region within present-day countries underwent a transition from subsisting primarily on hunted food to cultivating crops and livestock.

Disease (Dis) is taken from Murray and Schaller (2009) and provides an estimate of infectious disease prevalence at a national level using information compiled from historical atlases showing the incidence and distribution of up to 9 infectious diseases in approximately the middle part of the 20th century. The use of historical estimates attempts to ameliorate the issue of reverse causation for this factor, due to the fact that socio-economic development and associated improvements in health care have reduced disease prevalence.

Latitude (Lat) is calculated based on the geographical centre of the country and is here taken from Nunn and Puga (2012). As we are interested in employing latitude as a proxy variable that could potentially represent different ecological factors (other than the effects of disease which we include as a separate predictor), here we use absolute latitude which measures the environmental differences that occur as one moves away from the equator towards the poles.

Our choice of variables partly reflects pragmatic considerations: the number of countries is limited and each additional variable requires a potential larger increase in the number of parameters (direct and indirect pathways) that need to be estimated. Furthermore, as the data are taken from different sources the nations included in different datasets is not always the same, thus adding another variable can reduce the number of complete cases. The World Bank data has entries for 195 nations (including distinctive non-national territories such as Hong Kong). We have only included cases that have values for all the above variables, and our final sample contains 108 “countries”. While it would be desirable to have more countries as data points Fig. S1 illustrates that the final sample has a fairly widespread geographical distribution representing a good coverage of countries from across the different continents and regions. While more data is available for European countries, which potentially creates a biased sample, our use of phylogenetic information to control for non-independence of the data points (see below) seems likely to mitigate this problem.

Ancestry adjustment

We also follow Putterman and Weil (2010) in adjusting some of the variables to take into account large-scale migration. Earlier nation-level estimates of state history and the timing of agriculture were derived from the geographical locations of historical states and agricultural societies. For example, native Australian populations did not develop large-scale, politically centralized forms of political organization (states), yet due to colonization the current population of the country of Australia is dominated by people of European ancestry (and Europeans brought state forms of organization with them). Therefore, the modern population of this region should be measured as having a longer state histories than their geographic location alone would suggest. To deal with this issue Putterman and Weil (2010) produced a matrix that estimates the proportion of the population of a country in the year 2000 that is descended from people in different source countries in the year 1500. We use this matrix to transform the measures of state history and agriculture timing so that they more accurately reflect the histories of these populations.

Path analysis and model selection

Within the framework outlined above all variables can be considered dependent variables that are potentially predicted at least one other variable (apart from latitude). For each of these dependent variables we start by specifying a full model based on the variables that are hypothesized to explain it. For example, in our analyses institutional quality is linked to several different predictors (i.e. In-group bias, State History, European Ancestry, Latitude, and Timing of Agricultural Transition), while disease is predicted to be potentially affected by only latitude. The model that has European ancestry as a dependent variable also includes an extra dummy variable (“Europe”) that indicates whether the country is part of the continent of Europe or not. This is because the hypotheses being tested here relate to the effect that ecological factors may have had in determining what parts of the world European populations settled in outside of Europe. In estimating the effect of latitude on timing of agricultural transition we also include a squared term for latitude. This reflects the idea that the ecological conditions that most favoured the emergence of agriculture were in the mid latitudes (rather than conditions being ever more favourable as one moves to the poles, as would be implied by a simple positive relationship between latitude and timing of agricultural transition). All variables were scaled to have a mean of zero and standard deviation of one prior to analyses in order to produce standardized parameter values so that the strength of different relationships could be readily compared within models.

We use the lmerTest package to fit linear mixed effects models that estimate parameter values for each of these full models using Maximum Likelihood (Bates, Mächler, Bolker, & Walker, 2015; Kuznetsova, B, & C, 2017) with language family as a multi-level grouping variable (i.e. we estimate random intercepts for each group)(see below for information on language family)(Raudenbush & Bryk, 2002). We report the parameter estimates, standard errors, and p-values from the models. To further assess which predictors are actually substantial and important predictors of each variable we employ model selection criteria (Burnham & Anderson, 2010). We use the dredge function in MuMIn package (https://CRAN.R-project.org/package=MuMIn) to estimate all possible combinations of the specified predictors, and rank the models based on the Akaike Information Criteria (AIC). To avoid examining models with poorly fitting parameter combinations we examine only those models that are within 4 AIC units of the best fitting models. From these models we calculate model weights and use these to produce weighted parameter estimates (parameters that do not appear in a particular model are given a value of zero), and assess the importance as the sum of the weights of each model the parameter appears in. We also calculate approximate R2 values for each full model, and assess the amount of variation each parameter explains by examining the change in R2 for each predictor variable in two ways: 1) removing the variable from the full model, which indicates the unique variance explained by a particular variable, 2) examining a model in which the variable is the only predictor (along with the random grouping variable), which indicates the proportion of variance that is predictable from the simple relationship with the independent variable.

Controlling for shared history using language family affiliation

To account for shared history cross-cultural comparative researchers have incorporated knowledge of historical relationships between societies in the same way that biologists use knowledge of evolutionary relationships between species when conducting comparative analyses (Currie, 2013; Mace & Holden, 2005). Here we conduct our path analysis within a multilevel modelling approach (Raudenbush & Bryk, 2002), and designate countries as belonging to a cultural grouping that reflects the shared ancestry between these countries. The use of phylogenetic techniques in cross-cultural analyses presents its own challenges and comes with a number of assumptions that can be challenged (e.g. Borgerhoff Mulder, Nunn, and Towner (2006)). However, it should be recognized that assigning a linguistic affiliation (or other cultural proxy) presents an additional complication when attempting to adapt these kinds of approaches to cross-national analyses. Many countries have various degrees of cultural heterogeneity and can be home to numerous ethnolinguistic groups. For our global scale analyses we follow other studies in cultural evolution by employing a cultural grouping category based on countries speaking languages that belong to the same language family (Acerbi, Kendal, & Tehrani, 2017; Currie & Mace, 2009; Ruck, Bentley, & Lawson, 2018). One advantage of this approach is that in many cases although cultures within a country speak different languages those languages often belong to the same language family, making it somewhat meaningful to assign countries to language families. Taking this approach, we estimate the strength of relationships between predictor and outcome variables while explicitly modelling the expected covariation within designated groups.

Language families were used as random intercepts in all models to account for shared history. Following previous studies in the field of cultural phylogenetics that use language relationships as a proxy for shared cultural history we assign language family based on the linguistic affiliation of the majority of the population. Due to colonialism the Indo-European family is somewhat over-represented in our dataset, so we therefore use sub-families within Indo-European to capture meaningful patterns of non-independence (Acerbi et al., 2017, Ruck et al., 2018). We explored the effect of using modified versions of the language family classification that a) use the Indo-European family as a whole instead of subfamilies, and b) take into account cases where the main language is classified as Indo-European but the ancestry of the population has a substantial proportion of individuals that are not of European descent (i.e. language family relationships are less likely to be a good proxy for cultural history).

Data availability

All data and R code used for these analyses are included with this submission and will be made available via an open access repository. The original data sources themselves are publicly available and associated with the publications referenced above.

Results

Table 3 presents the results of all main analyses and shows parameter estimates (full model and AIC-weighted) for all pathways tested, associated p-values, parameter importance, and estimated R2 values. Fig. 2 presents a reduced path diagram that summarizes the direct and indirect pathways that receive most support from the analyses. The diagram focuses on those relationships that are statistically significant at p < 0.05 and are found in all the samples of best-fitting models (i.e. weighted importance = 1). It also indicates some relationships that fall short of statistical significance but are still found in large proportion of the best-fitting models (weighted importance >0.7). Pathways that do not meet these criteria lack support from these analyses, have lower standardized parameter estimates, and are not included in the figure. Fig. S2 also illustrates the difference in the estimates of predictive relationships that our analyses produce over simpler analyses that don't account for other variables or control for phylogenetic relationships (see table S1 for raw correlation coefficients).

Table 3

Support for different relationships between GDP, proximate, historical, and ecological factors. Parameter estimates are shown for each relationship indicated by the path numbers from Fig. 1. Results are given for both the full model (which includes all identified predictor variables for a given outcome variable), and the weighted estimates and variable importance from the model selection approach. Change in the amount of variance explained (ΔR2) is shown by either subtracting the variable from the full model, or including the variable as the only (single) predictor, as well as the variance explained by the full model (R2).

		Full model			AIC model average				r2
		β	SE	p	β	SE	p	importance	subtracted	Single predictor	Full model
GDP	Intercept	<0.01	0.08	0.96	<0.01	0.09	1.00	–	–	–	0.74
	Timing of Agricultural Transition	0.10	0.06	0.13	0.07	0.07	0.35	0.62	0.02	0.13
	Disease	−0.10	0.09	0.26	−0.05	0.08	0.54	0.43	0.03	0.24
	European ancestry	0.14	0.09	0.11	0.13	0.11	0.23	0.73	0.03	0.31
	In-group bias	−0.10	0.07	0.17	−0.06	0.08	0.43	0.55	−0.01	0.31
	Institutional Quality	0.49	0.08	<0.01	0.55	0.08	<0.01	1.00	0.14	0.59
	Latitude	0.01	0.09	0.88	0.03	0.07	0.65	0.38	<0.01	0.20
	State History	0.09	0.06	0.17	0.05	0.07	0.44	0.54	0.02	0.12
Institutional Quality	Intercept	<0.01	0.10	1.00	<0.01	0.09	0.98	–	–	–	0.61
	Timing of Agricultural Transition	0.09	0.08	0.26	0.04	0.07	0.56	0.42	−0.01	0.13
	European ancestry	0.15	0.11	0.18	0.14	0.12	0.24	0.73	0.02	0.31
	In-group bias	−0.52	0.08	<0.01	−0.54	0.08	<0.01	1.00	0.23	0.49
	Latitude	0.09	0.10	0.36	0.08	0.10	0.44	0.55	<0.01	0.20
	State History	0.21	0.07	0.01	0.23	0.07	<0.01	1.00	0.06	0.15
In-group bias	Intercept	0.03	0.16	0.84	0.03	0.16	0.86	–	–	–	0.38
	Timing of Agricultural Transition	0.05	0.10	0.64	0.01	0.05	0.89	0.27	<0.01	0.03
	Disease	0.27	0.13	0.04	0.25	0.13	0.06	0.93	0.07	0.23
	European ancestry	−0.36	0.14	0.01	−0.37	0.13	<0.01	1.00	0.08	0.28
	Latitude	<0.01	0.13	0.99	<0.01	0.06	0.99	0.20	<0.01	0.17
	State History	−0.18	0.09	0.04	−0.14	0.10	0.18	0.81	0.03	0.06
European Ancestry	Intercept	−0.58	0.10	0.00	−0.59	0.11	0.00	–	–	–	0.68
	Disease	−0.10	0.07	0.15	−0.05	0.07	0.48	0.50	0.07	0.26
	Europe	0.99	0.13	0.00	0.98	0.13	0.00	1.00	0.32	0.61
	Latitude	0.21	0.07	0.01	0.24	0.07	0.00	1.00	−0.03	0.36
State History	Intercept	0.11	0.17	0.54	0.11	0.17	0.52	–	–	–	0.22
	Timing of Agricultural Transition	0.47	0.09	<0.01	0.48	0.09	<0.01	1.00	0.20	0.23
	Disease	0.07	0.14	0.60	0.01	0.07	0.88	0.28	−0.01	<0.01
	Latitude	0.08	0.13	0.53	0.01	0.06	0.82	0.29	<0.01	0.03
Timing of Agricultural Transition	Intercept	0.35	0.20	0.09	0.31	0.21	0.14	–	–	–	0.10
	Latitude	0.27	0.09	0.01	0.26	0.11	0.02	–	0.06	0.08
	Latitude2	−0.17	0.08	0.04	−0.13	0.11	0.22	–	0.01	0.03
Disease	Intercept	0.08	0.11	0.46	0.08	0.11	0.46	–	–	–	0.47
	Latitude	−0.58	0.06	<0.01	−0.58	0.07	<0.01	–	0.47	0.47

Fig. 2

Reduced path diagram that indicates the pathways which receive statistically significant support in model comparison. Line widths are proportional to the Akaike weighted coefficients of the pathways. Dashed lines are not statistically significant but receive some limited support.

Taken together these analyses indicate the pathways that may connect ecological factors to historical processes and modern day outcomes. Our results indicate one pathway whereby latitude predicts the emergence and spread of agricultural forms of food production, which in turn predicts greater experience of centralized, state forms of political organization. Having a longer State History is predictive of greater institutional quality, which is the strongest, and only statistically significant predictor of modern day economic productivity of the variables considered in this study. The other main type of pathway relates to the impact of European colonization on the modern world. Our results show that European ancestry (outside of European countries) is predicted by latitude. European countries and other countries with higher proportions of European ancestry tend to have lower estimated levels of in-group bias, and this variable is in-turn predictive of better institutional quality. Direct pathways from European ancestry to institutional quality, and from State History and Disease to in-group bias, receive some limited support from these analyses, although these pathways do not reach statistical significance or appear universally in the best-fitting models.

We assessed whether our main results are robust to different modelling assumptions that could be made. Performing path analysis involves having to make several assumptions about which relationships should be assessed. The relationships included in our main analyses are based on pre-existing hypotheses and evidence from the literature, however it is possible to articulate other potential pathways. In the supplementary materials we explore the effect of assuming that the predictive relationship between institutional quality and in-group bias runs in the opposite direction. This has the effect of making institutional quality the only significant predictor of in-group bias (although European ancestry retains some non-statistically significant support), but the strength of State History as a predictor of Institutional quality increases and European ancestry becomes a significant predictor of Institutional quality (Tables S2, S3). We also explored whether the observed relationships between European ancestry and the social factors of institutional quality and in-group bias is driven predominantly by the fact that European countries have high values on this variable. This is not the case and the main results are qualitatively very similar if we include the effects of European ancestry once the contribution of European countries themselves have been accounted for (ESM, Table S4). We also examined the effects of different assumptions about our language family classification on the main results (ESM, Table S5). The results indicate that the models and parameter estimates are broadly similar to our main results whether we a) use Indo-European as a whole language family rather than breaking it down into sub-groups, or b) modify the language classification to reflect cases where language does not reflect the demographic or broader cultural history of the population (Table S6). Overall, the results from these supplementary analyses are qualitatively the same as our main analyses, which increases our confidence in the broad, long-term historical pathways identified. We discuss other issues around modelling assumptions below.

Discussion

In this paper we have assessed many diverse pathways by which different factors may have shaped the nation-level patterns of inequality in economic development we see in the world today. Our framework has enabled us to address ideas not only relating to the immediate, or direct causes of economic performance, but also the varied, social, ecological, and historical processes that may also affect performance, sometimes indirectly. Our analyses indicate that Institutional Quality is the only strongly supported direct predictor of GDP out of the factors considered in these analyses. Our analyses also indicate, however, that cultural factors in the form of in-group bias are strongly predictive of the degree to which a nation's institutional make-up enables its population to take part securely in economic and political activities. These norms, institutions, and other cultural factors can be viewed as features that enable populations to overcome coordination and cooperation problems inherent in economic activities (Henrich et al., 2010). A cultural evolutionary perspective on these issues recognizes that the development of traits that facilitate the coordination and cooperation of large numbers of people are not easy to develop (Richerson & Henrich, 2005). Solutions to such problems are difficult to generate de novo, due to the complex nature of the challenge and the potentially differing incentives and motives or different actors within a population (Gavrilets, 2015; Richerson & Henrich, 2012). The development of effective norms and institutions may need to bring together different elements and may develop in a cumulative manner, potentially requiring long periods of time and several rounds of “cultural experimentation” (Currie et al., 2016; Currie et al., 2020; Wright, 2006).

Our results draw attention to two indirect pathways by which these proximate factors that affect economic growth have culturally evolved over time. One main pathway indicates that ecological factors may have affected the likelihood that different parts of the world adopted food production (Barker, 2006). This change in subsistence strategy is thought to have had many fundamental impacts on the long-term evolution of human societies (Diamond, 1997). A key process is the emergence of more complex societies which is thought to be facilitated by the resource base that agriculture provides as it is more conducive to storage and monopolization (Mattison, Smith, Shenk, & Cochrane, 2016a; Mayshar, Moav, Neeman, & Pascali, 2015). Our results are consistent with the idea that regions that transitioned to agriculture earlier have also generally had a longer history of centralized, larger-scale forms of socio-political organization. This finding is also supported by recent analyses by Currie et al. (2020) that show a similar relationship between timing of the agricultural transition and the development of large-scale states using independent and more detailed historical and archaeological data. Our analyses also support the idea that political centralization, because it provides a structure that facilitates communication and coordination over larger populations and larger areas (Turchin et al., 2018), is a prerequisite for the later development of institutions conducive to economic growth. Although political centralization has been associated with inequalities in wealth and power throughout human history, some centralized societies (but by no means, all) developed more “inclusive” norms and institutions (Acemoglu & Robinson, 2012). The other main pathway that is supported by our analyses is the way in which ecological and cultural evolutionary processes may interact to affect historical population movements. Our analyses support the hypothesis that European populations replaced native populations in greater numbers where ecological conditions were similar to those in Europe. The results are also in line with the prediction of the hypothesis that countries with substantial European-descendant populations have inherited cultural traits (in this case norms associated with the degree of in-group bias) from their European ancestors, and which are associated with institutional quality (but see below for discussion of alternative explanations for this statistical relationship).

The approach we have taken this paper has a number of strengths. Firstly, it has emphasized the importance of simultaneously considering and testing multiple hypotheses, explicitly acknowledging different potential explanations and different routes by which variables may exert an effect (Currie et al., 2016; Platt, 1964). By specifying our hypotheses of interest based on existing plausible ideas in the literature that have some independent lines of evidence, we avoid making too much of spurious relationships that might be discovered just in our particular dataset. By calculating weighted parameter estimates across the range of models we take into account the fact that multiple sets of hypotheses might be plausible given the data. Our approach is in contrast to the kind of null hypothesis testing that is common in economics and many cross-cultural comparative analyses, whereby a single explanatory hypothesis is highlighted and an emphasis is placed on whether effects remain statistically significant after controlling for potential confounding factors. A narrower focus on single explanations has the potential downside of emphasizing factors that have small effects, or don't consider the multiple casual processes that might be involved. As well as highlighting relationships that are supported, our analyses are also important in identifying hypotheses that are not supported as this allows us to reject ideas that do not explain the data well. For example, our results do not provide evidence for strong direct effects of historical or ecological factors on modern-day economic outcomes, rather the effects of these factors appear to be more indirect. Furthermore, previous analyses have argued that traits such as in-group bias are a direct response to ecological factors (Fincher & Thornhill, 2012; Van de Vliert, 2011). However, our results suggest a more-indirect pathway going through European ancestry. Being able to reject ideas that are plausible but lack empirical support potentially is an important part of scientific progress (Dunbar, 1995; Lipton, 2005; Platt, 1964), and may be particularly important in fields, such as History or Cultural Anthropology, where contemporary mainstream approaches tend not to engage with systematic comparative approaches (Turchin et al., 2015).

Another contribution of cultural evolutionary thinking to our approach is the understanding that our units of analysis have shared historical relationships that need to be taken into account when performing these kinds of comparative analyses. The use of phylogenetic information in controlling for shared history is becoming increasingly common in cross-cultural analyses (Bromham et al., 2018; Currie, 2013; Kirby et al., 2016). However, the application of such techniques has not been fully appreciated in cross-national analyses of economic factors and other variables, even though it may have important implications for such analyses as we indicated in the introduction. A challenge for operationalizing this approach is uncertainty about how best to model potential sources of non-independence. We have shown that our broad inferences are robust to different reasonable assumptions about how countries should be classified according to shared ancestry. However, it is important to note that the kind of categorical variable approach to specifying the potential covariance between units of analysis, while better than nothing, only deals with a certain level of non-independence and may not control for this issue completely when the inheritance of traits is tree-like (Matthews, 2019). Sookias et al. (2018) partly minimized this issue by limiting their phylogenetic analysis of the Human Development Index to countries in Eurasia that speak Indo-European languages, where many countries (particularly those in Europe) are relatively homogenous in this sense. Future analyses can employ and explore different ways of modelling cultural relationships between countries. For example, recent analyses developed by Ruck, Matthews, Kyritsis, Atkinson, and Bentley (2020) could provide a useful basis for other future studies. They developed national level linguistic proximity matrices based on combining phylogenetic information for all languages spoken within a country by at least a certain specified proportion of the population. In addition to this role in controlling for non-independence it is worth stating that cultural phylogenetic approaches can also be valuable in more explicitly assessing evolutionary patterns and processes (Currie, 2013)(see below).

While our study has generated potentially interesting insights we recognize the limitations of the current study, and here discuss three particular issues: omission of certain factors, assessing causality, and measurement of the variables that are included. Although we have chosen a number of potentially important factors, and assessed 24 different hypotheses that have been articulated the literature, there are a number of other variables and processes that may be important, either directly or indirectly, in economic development. In terms of proximate predictors we have only investigated the social factors of institutional quality and in-group bias. However, a number of other factors relating to other cultural traits and aspects of human capital, such as specific skills and education, may also be important (Ahlerup et al., 2009; Diamond, 2014; Glaeser et al., 2004). We have also not included other potentially important historical factors. For example, evidence from Schulz, Beauchamp, Bahrami-Rad, and Henrich (2019) suggests that medieval religious practices may be involved in the evolution of the norms and institutions that affect in-group bias, which western Europeans then took with them when the colonized and settled in other parts of the world. Furthermore, modelling and analyses of historical data (Currie et al., 2020; Turchin, Currie, Turner, & Gavrilets, 2013) show that the development of large, complex societies in Afro-Eurasia can be predicted by increased levels of warfare between groups that select for increasingly larger groups in addition to the antiquity of agricultural food production. The use of latitude as a proxy for different environmental factors masks the more specific factors and processes that may be involved in shaping different evolutionary pathways. Addressing the relative importance of other potential hypotheses is an important step for future research, and may reveal more detail about the relationships identified here or additional (rather than alternative) pathways and processes to those covered in this study.

Given the profound effect it has had on shaping the world we live in over the last 500 years or so, it is particularly important to consider the alternative ways that European colonialism has been involved in shaping economic performance. In this study, the main hypotheses involving colonialism considered the potential “positive” effects that European ancestry may have had on present-day institutional and economic outcomes. However, extractive colonial practices may have negative impacts on colonized regions (and positive economic effects for European countries or colonies in which Europeans did settle in large numbers) beyond the kind of cultural inheritance route invoked by those hypotheses. For example, comparative studies have demonstrated empirically the negative impact of the African slave trades on present-day economic performance for African countries (Nunn, 2008). Even where we have identified certain statistical relationships that are in-line with the original hypotheses articulated in Table 1, it is important to point out that alternative explanations for these associations could exist. For example, the association between European ancestry and in-group bias could arise because of colonial practices in places where Europeans did not settle in large numbers (such as “Divide and Rule” policies and rigid territorial separation (Christopher, 1988)) which created conditions that may have created, entrenched, or enhanced in-group biased behaviours. It is also worth emphasizing that this characterization of institutions as “inclusive” in the context of settler colonialism is from the perspective of settler populations. This characterization does not capture the negative effects experienced by indigenous populations, who succumbed to novel diseases, and were killed, or displaced, or the negative effects on those people brought to such colonies as slaves from other colonies, or the continuing legacy of these events in the present day.

A related issue is our ability to distinguish the extent to which the statistical relationships between variables identified in this study are evidence of causal relationships. For some relationships we can be confident about ruling out reverse causality – our historical variables cannot be caused by contemporary variables, and latitude is a measure that is exogenous to both proximate and historical factors. Correlations induced by common relationships with other variables are challenging to rule out completely (particularly for the social factors), but both the inclusion of the other variables as potential predictors, and the modelling of shared history (which helps control for other unspecified factors that may be shared across data points) helps address these points to a certain extent. Measures and approaches that unambiguously distinguish between the effects of contemporary, synchronic institutional rules, social norms, and human capital on economic development are difficult to find in practice (Diamond, 2014). The proximate factors that we have employed in this study are best thought of as culturally-inherited factors that are proposed to have some causal influence on economic development rather than pointing towards strong statements about the relative importance of institutions versus social norms, or human capital more generally, for example. Our supplementary analyses also indicate that making different assumptions about the causal relationships between these proximate factors does not qualitatively affect our inferences about the deeper historical and ecological processes that may have shaped their development.

Although it is possible to dig deeper into any one of these relationships and include additional variables or attempt stronger tests, there are practical challenges to doing so within the statistical approach we have taken here. Many variables of interest have coverage on only a limited number of countries, and which countries are covered can vary greatly across variables. This means that if we want to estimate all relationships using the same countries then adding another variable often comes at the expense of losing incomplete cases, which creates the “double-whammy” of asking more from a smaller sample size. Furthermore, including an additional variable in a path analysis framework means potentially adding multiple additional new relationships with other variables, which may make it more difficult to analyse, interpret, and synthesize the resulting information. These points are not insurmountable though and we return to this point below and suggest ways forward in future research on this topic.

The final limitation of the present study that we consider here relates to the actual measures of the factors we wish to investigate in this study. Our analysis has focussed on national level differences in economic performance and therefore uses countries as the unit of analysis. This presents conceptual and practical challenges. Some countries are more culturally or socially homogenous or heterogenous than others (e.g. there may be different groups within a country with distinct histories, cultural beliefs, and practices, and/or there may be structural inequalities due to things such as gender, race, or ethnicity). This means that some factors, such as in-group bias or timing of agricultural transition, may vary a great deal within countries, which may be masked by national-level agglomerations of individual or regional scores, and raises questions about what a single value for such a country actually represents. A related issue is where variables or indices (such as our in-group bias measure) are made up of potentially distinct measures (e.g. compatriotism, nepotism, and familism), which may have different theoretical or statistical relationships with other variables. In the case of in-group bias measure factor analysis indicates the three underlying components are highly correlated (Van de Vliert, 2011). More conceptual issues are present for variables such as institutional quality, which is actually based on measures of how people perceive certain institutional rules are playing out rather than being objective measures of the rules themselves and may therefore be influenced by other factors. It should also be noted that our current analyses relate to contemporary data on economic performance and social factors from one point in time. Although countries' economic performance and social factors can and do change, we do not expect that the results presented here would be dramatically different if data from another recent year were used as the relative “ranking” of countries in terms of measures like GDP tends not to vary dramatically over shorter time scales. The ability to conduct longer-term time series analyses would be interesting in this regard, but is constrained by the availability of suitable data. Although we do not explore this issue in this paper, due to the already large number of pathways and analyses conducted, it is important to acknowledge that the effect of different historical or ecological factors may be more or less pronounced at different times in history. None of these issues are unique to the analyses in this paper, and below we consider how future research might build on the current study.

Conclusion

In this paper we have attempted to assess the support for a number of different hypotheses that have been proposed to explain the causes of modern-day economic performance. This was motivated by a desire to evaluate which of the many plausible processes are more important than others, and which do not receive support when evaluated against other ideas. Furthermore, scientific progress can come as much from rejecting previously supported hypotheses as it can from finding confirmatory evidence (Lipton, 2005). On the other hand, studies that focus on a more limited set of explanations, or employ different approaches may be able to go further in assessing the strength of support for casual explanations (e.g. mainstream econometrics employs a suite of techniques for making stronger evaluations of causal claims, while experimental studies allow for greater control in distinguishing cause and effect). Holistically, both approaches are needed: reductionism and digging down into particular explanations is an important part of the scientific process, yet it is also important to attempt to see the bigger picture and understand how different explanations fit together (Dunbar, 1995).

Despite its limitations, we feel the current study provides a useful foundation on which future studies and analyses can be built. In this paper, we focussed on analysing the same data within the same statistical framework to assess the proposed pathways. Having highlighted those pathways that receive most support, future studies can evaluate them further using other statistical approaches, or methodological techniques that have been employed in economics, political science, and related fields. Evaluating multiple pathways using panel data would be particularly useful for not only addressing issues of causality, but also for understanding how the importance of different effects may vary over time. Conducting meta-analyses of existing studies as an alternative approach to assessing potential pathways at the same time (including those not assessed in this paper) could also prove valuable. Such an approach could be particularly useful as examining additional pathways would not come at the expense of reducing effective sample sizes, and it would allow for results and insights from many different types of studies to be combined.

Questions of economic development have generally not been a major focus for researchers working in the field of cultural evolution (but see Matthews et al., 2016, Ruck et al., 2018, Sookias et al., 2018, Wilson & Gowdy, 2013), while macroeconomists have generally not appreciated the contributions that evolutionary theory can make (Wilson, Kirman, & Lupp, 2016 but see Nunn, 2012; Spolaore & Wacziarg, 2013). We argue that this is an area to which cultural evolutionary theory and methods can have important insights and add to existing approaches. For example, Spolaore and Wacziarg, 2013, Spolaore and Wacziarg, 2016 have taken inspiration from cultural evolutionary theory to argue that shared cultural inheritance may affect the rate of transfer of traits important to economic productivity (innovations being more easily spread adopted by countries that are similar culturally). They find evidence in support of this idea by incorporating phylogenetic information into their regression analyses. Currie et al. (2021) take a similar approach to examine the spread of democratic forms of governance using an explicit phylogenetic comparative analysis. Evolutionary theory also provides a more dynamic view of historical processes than is generally considered in hypotheses that only consider historical legacy effects. The potential for cultures and cultural traits to undergo selective process creates the possibility of feedback loops that change populations and also affect the response to different selection pressures over time. The path analysis framework we have employed in this study is able to articulate and assess these kinds of hypotheses, but may require additional approaches in order to more fully distinguish between different types of processes. The analysis of time-series data may help reveal these differences. For example, Turchin, Currie, Whitehouse, François, et al. (2018) argue that changes in the historical distribution of social complexity over time contain the signature of driven, selective process due to the fact that not only do we see evidence of larger, more complex societies emerging over time, we also see a reduction in the occurrence of smaller, less complex groups (see also Currie & Mace, 2011, Spencer & Redmond, 2001). Such approaches can be further strengthened by the use of statistical approaches (such as those used in phylogenetic comparative methods)(Currie, Greenhill, Gray, Hasegawa, & Mace, 2010) or computer simulations (Turchin et al., 2013) that explicitly model dynamic evolutionary processes. In a recent study, Currie et al. (2020) examined different types of evolutionary explanations of the development of complex societies relating to historical contingency, selection between groups, and the time available to generate and accumulate cultural traits that support larger-scale organization, and show how the predictive strength of different factors associated with these processes varies at different points in time. By Building on and adapting the kind of approach we have taken in this paper can help develop a better understanding of the cultural evolutionary processes that have shaped the world we live in today. More generally we argue that evolutionary theory provides a conceptual framework that can help connect insights and findings from across a range of disciplines, and facilitate a true intellectual exchange between those interested in investigating economic and social development (Currie et al., 2016; Wilson et al., 2016; Wilson & Gowdy, 2013).

Author contributions

A.F and T.C. designed research; A.F. performed research; A.F & T.C. analysed data; A.F and T.C wrote the paper.

Funding acknowledgment

A.F. was supported by an ESRC South West Doctoral Training Partnership Studentship for his PhD project “The cultural evolution of economic development”. T.E.C. is supported by funding from the (ERC) under the 's research and innovation programme (Project: The cultural evolution & ecology of institutions; Grant Agreement 716212).

Declaration of Competing Interest

The authors declare no conflict of interest.