Colonial legacy and contemporary civil violence: a global study from 1960 to 2018

Abstract

This study assesses whether the legacy of colonialism continues to influence patterns of civil violence in the contemporary era. A large and established quantitative literature attributes civil violence to low levels of economic development and limited political rights, but few quantitative studies consider whether colonial legacy plays an enduring role in such conflicts. This is surprising given the substantial evidence showing that colonialism impeded long-run development in many parts of the world. Drawing on ideas from macro-comparative sociology, institutional economics, and political science, the study develops several theoretical expectations regarding colonialism’s effect on contemporary civil violence. These ideas are tested with a global sample of 152 countries observed annually from 1960 to 2018. Results from logistic regression models support the contention that (1) post-colonial societies are more prone to civil violence than non-colonized societies, that (2) ex-British colonies are especially prone to ethnic-based civil violence while ex-Spanish colonies are especially prone to socio-economic-based civil violence, and that (3) these historical effects change and evolve but never fully abate. This latter finding implies that elevated levels of civil violence are a path-dependent legacy of colonialism. The study ends by running robustness checks and discussing the theoretical implications of the study’s findings, in particular reflecting on our understanding of the long-run consequences of colonialism.

Introduction

This study assesses the historical legacy of colonialism and whether it still shapes patterns of civil violence in the contemporary era. Most armed conflicts now occur within sovereign states rather than between them (Sarkees, Wayman, and Singer 2003). Given the substantive importance of this topic, it is unsurprising that a well-established quantitative literature, spanning the social sciences, seeks to identify the risk factors associated with outbreaks of civil violence (see, e.g., Collier and Hoeffler 2004; Fearon and Laitin 2003; Hibbs 1973; Muller 1985; Muller and Weede 1990). Here the main conclusion is that low levels of economic development, weak economic growth, and limited political rights create persistent grievances with the status quo, lower the “opportunity costs” associated with conflict, and leave would-be combatants with limited avenues for peaceful redress. Under such conditions, the risk of civil violence is relatively high.

By contrast, the possibility that the legacy of colonialism shapes contemporary civil violence receives less attention from quantitative scholars (cf. Blanton, Mason, and Athow 2001; Lange and Dawson 2009; Lange, Jeong, and Amasyali 2021). This is surprising given the large literature concluding that colonialism impeded the development process in many parts of the non-European world (Acemoglu, Johnson, and Robinson 2001, 2002; Kwon 2011; Lange, Mahoney, and vom Hau 2006; Mahoney 2003; Sokoloff and Engerman 2000). Here the broad conclusion is that, especially under Spanish rule, colonialism’s focus on plunder and rent seeking gave rise to absolutist states with coercive powers and ultimately generated entrenched elites whose interests are antithetical to ordinary people. Post-colonial societies inheriting these institutional frameworks have diminished prospects for development, but they may also face ongoing problems related to poor governance and durable inequalities, both of which make them susceptible to outbreaks of civil violence.

These themes are examined in a smaller literature on distinct colonial legacies and their effects on civil violence. Here most research identifies ex-British colonies as being particularly susceptible to civil violence (Blanton, Mason, and Athow 2001; Lange and Dawson 2009). This purportedly occurs because the British system of indirect rule used ethnicity as the constituent unit of colonial governance, thereby leaving behind post-colonial societies with hardened and politicized ethnic divisions. The French system of direct rule caused problems too, as it enabled indigenous urban elites to monopolize the post-colonial state, leaving provincial groups marginalized (Wucherpfennig, Hunziker, and Cederman 2016). Interestingly, the legacy of Spanish colonialism receives less attention in this regard, even though it purportedly created predatory states and highly unequal societies (Acemoglu, Johnson, and Robinson 2001, 2002; Mahoney 2003; Lange, Mahoney, and vom Hau 2006). Furthermore, given that post-colonial societies generally become more peaceful as the duration of their independence grows (Lange and Dawson 2009; Lange, Jeong, and Amasyali 2021), we must account for the 100+ years between the end of Spanish colonialism in the mid-1800s and the end of British and French colonialism in the post-war period. Otherwise, ex-British and ex-French colonies may appear relatively more conflict prone simply because they have less experience as sovereign states.

To assess these and related issues, the present study examines a sample of 152 countries observed annually from 1960 to 2018. The sample includes all states recognized by the United Nations with populations over one million in 2018. My measures of civil violence, drawn from the Major Episodes of Political Violence (MEPV) database developed by the Center for Systemic Peace (CSP; Marshall 2019), allows me to distinguish between ethnic-based and socio-economic-based civil violence. Making these distinctions, even if imperfectly, is important because the literature strongly suggests that some colonial projects entrenched identity-based cleavages (e.g., British colonialism) while others defused them (e.g., French and Spanish colonialism). Despite historical differences in colonial rule, prior research focuses mainly on ethnic violence (Blanton, Mason, and Athow 2001; Fearon and Laitin 2003; Lange and Dawson 2009; Lange, Jeong, and Amasyali 2021; Wimmer, Cederman, and Min 2009). This limits our understanding of colonialism’s long-term effect on the prospects for civil peace in the post-colonial era.

My empirical analysis uses logistic regression models to estimate the chances of civil violence as a function of a country’s colonial legacy and appropriate control variables. The selection of control variables, however, must consider the strong possibility of “endogenous selection bias” (i.e., “post-treatment bias”; Dworschak 2024; Elwert and Winship 2014). Given colonialism’s role in shaping the political economies of post-colonial societies, numerous economic and political factors known to influence civil violence are endogenous and should not be used as control variables. This issue is underappreciated by prior studies on colonial legacy and civil violence (e.g., Blanton, Mason, and Athow 2001; Lange and Dawson 2009; Lange, Jeong, and Amasyali 2021; Wimmer, Cederman, and Min 2009). The present study discusses this methodological complication and ways to address it.

Several findings emerge from my empirical analysis. Results show that, on average, post-colonial societies are more prone to civil violence than non-colonized societies, that ex-British colonies are especially prone to ethnic-based civil violence while ex-Spanish colonies are especially prone to socio-economic-based civil violence. Furthermore, predicted trends show that these broad patterns change and evolve but never fully abate. Additionally, robustness checks assessing the intensity of conflicts, not just their occurrence, find that ex-colonies of Britain and those classified as “other European powers” are the most prone to highly destructive bouts of civil violence. In total, my broad findings are indicative of a limited path dependency whereby colonial-era institutions and practices became entrenched and continue to shape outcomes in the present era.

Literature review

Colonial legacy, development, and civil violence

Dependency theory highlights the significant challenges faced by previously colonized societies as they attempt to develop their economies (Baran 1957; Frank 1966; Prebisch 1950). One of its primary insights is that the development process in colonized societies did not occur endogenously but was imposed through colonialism. As its primary purpose, European colonialism sought to incorporate militarily subjugated colonies into a broad international division of labor whereby the colonies specialized in the extraction of raw materials and the production of basic foodstuffs for export back to the colonial “metropole”. By using trade monopolies and coerced labor, the terms of trade significantly benefited the colonial powers, manifesting as a steady flow of wealth back to Europe (Tadei 2014).

For post-colonial societies, the end of colonialism formally halted their political subjugation and freed them to pursue their own courses of development, but it left many with economic and state structures needing reform. Some post-colonial societies have clearly struggled in this regard. Here scholars document that, despite repeated initiatives from different political leaders, post-colonial states characterized as “neopatrimonial” (Kohli 2004) or “predatory” (Evans 1989, 1995) have made little progress towards reaching their development goals. In these cases, well-connected indigenous elites often succeeded in monopolizing colonial-era institutions and using them for personal enrichment. This not only creates endemic corruption but leaves effected societies under-development. By contrast, other post-colonial societies used their newfound powers in judicious and productive ways, for example by taxing rents from colonial-era enterprises and using the proceeds to fund public goods and to pursue industrial upgrading. These actions accelerated the development of local industry and put these countries on the path toward successful development. In fact, some scholars credit colonial-era institutions with laying the foundations for present-day prosperity. For instance, Kohli (2004) singles out South Korea, claiming that Japanese colonialism reformed its erstwhile corrupt and ineffective state, and Lange (2009, Chapter 8) argues that Hong Kong and Singapore benefitted from British colonial rule in similar ways.

A parallel literature draws similar conclusions. Coming from the perspective of institutional economics, Acemoglu, Johnson, and Robinson (2001, 2002) argue that colonialism reversed the developmental trajectories of many parts of the non-European world, with the new trajectories being decisively shaped by the type of colonialism (see also Dell 2010; Sokoloff and Engerman 2000). “Settler colonialism” arose in areas with relatively small indigenous populations, limited accumulated wealth, and benign disease environments, for example areas that became Canada and the United States. Here European settlers did most of the necessary work and hence built norms and rules that rewarded personal effort and encouraged investment. Eventually, these norms and rules solidified into institutional frameworks supportive of development. By contrast, the dominant form of colonialism, what they call “extractive colonialism,” occurred in areas with relatively large indigenous populations, greater amounts of accumulated wealth, and more deadly disease environments. Here Europeans built extractive institutions, such as systems of coerced labor and methods to plunder pre-colonial stockpiles of precious metals. The resulting economic system gave rise to absolutist states with coercive powers and ultimately generated entrenched elites with interests antithetical to the general population. Post-colonial societies inheriting these institutional frameworks have limited prospects for development and often face problems related to entrenched inequality and poor governance. Similar arguments are made in sociology (see Kwon 2011; Mahoney 2003).

Colonial legacy affects the prospects for civil violence in post-colonial societies through various mechanisms. First, as suggested above, a range of theoretical perspectives contend that colonialism distorted the material foundations required for long-term development, and that these structural disadvantages became entrenched and difficult to overcome. This is clearly important because low development heightens the risk of civil violence by creating strong grievances with the status quo, lowering the opportunity costs associated with rebellion, and leaving the state without the resources needed to maintain order (Collier and Hoeffler 2004; Muller 1985; Muller and Weede 1990). In fact, some low-income countries become stuck in a “conflict trap,” whereby low development gives rise to conflict and conflict further slows development.

Beside issues surrounding inadequate development, some post-colonial societies have been saddled with ineffective political structures inherited from the colonial era. One vexing problem is that the national borders of ex-colonies often reflect the interests and machinations of former colonial rulers rather than indigenous populations (Hughes 2004). In this cases, post-colonial societies can struggle with state building, especially in peripheral areas where state control is often tenuous. Similarly, some post-colonial societies have “weak states” (Tusalem 2016), meaning that their ability to govern is limited by ineffective bureaucracies and insufficient legitimacy. Weak states often struggle with basic functions of governance including the maintenance of law and order. Both problems—colonial-era borders and weak states—should heighten the prospects of civil violence.

In sum, the evidence suggests that the legacy of colonialism had diverse impacts, impeding the prospects for civil peace and long-term development in many but not all cases. Hence, my first theoretical expectation is that, on average, post-colonial societies are more prone to civil violence than non-colonized societies (Hypothesis 1).

Colonial powers and their institutional legacies

Broadly speaking, colonialism was a system of foreign conquest for economic gain, but the major colonial powers developed unique systems for achieving these common ends (Crowder 1968/2023; Grier 1999; Kwon 2011; Lange 2009; Lange, Mahoney and vom Hau 2006; Miles 1994, Chapter 5). On this topic, institutional economics emphasizes the extent to which colonialism established property rights (Acemoglu, Johnson, and Robinson 2001, 2002), but sociology and political science frame these legacies more broadly—emphasizing factors such as the rule of law, the competitiveness of markets, the provision of public goods, access to the state, the autonomy of local populations, and the strength of the state (Kwon 2011; Lange 2009; Lange, Mahoney and vom Hau 2006; Lange, Jeong, and Amasyali 2021; Lange, Jeong, and Gaudreau 2022; Letsa and Wilfahrt 2020; Müller-Crepon 2020; Naseemullah and Staniland 2016). From this perspective, the major colonial powers left behind distinct legacies, arising from the institutions and governing practices that constituted their systems of colonial rule. Consequently, for post-colonial societies, these distinct legacies should retain some influence over important societal outcomes, potentially affecting their prospects for effective governance and ultimately their chances for domestic peace. Below, I offer a stylized description of the modes of colonial rule used by mercantilist Spain, liberal Britain, and statist France, but acknowledge some variation within these broad patterns.

The Spanish empire arose during the mercantilist era and this heavily influenced its approach to colonialism (Mahoney 2003; Lange, Mahoney and vom Hau 2006). In Spanish colonies, the political economy centered on the extraction of precious metals for export back to Spain, with Spanish traders being shielded from market competition by mercantilist policies. These traders, along with administrators, large landlords and mine owners, constituted a Spanish colonial elite. By contrast, a large and marginalized indigenous population was coerced to work in mines and plantations. For example, the mita system used in parts of Bolivia and Peru forced indigenous groups to give one-seventh of their adult male population to work in mercury and silver mines (Dell 2010). This population like others across the Spanish colonial empire was ethnically diverse, but Spanish colonial rulers treated them as an undifferentiated mass. Over time, the core features of Spanish colonialism solidified into predatory states, poorly functioning markets, and highly unequal societies. Quite easily, one could surmise that post-colonial societies inheriting these structures would be prone to civil conflicts, although inter-ethnic conflicts may be rare. Indeed, Lange and Dawson (2009) find that ex-Spanish colonies are the least prone to ethnic-based civil violence. They explain this outcome as arising from the Spanish colonial practice of downplaying ethnic differences (see Lange and Dawson 2009, 806). Overall, this leads to my second theoretical expectation, namely that ex-Spanish colonies are prone to civil violence in general, but it rarely manifests as ethnic-based civil violence (Hypothesis 2).

By contrast, the bulk of the British empire arose in a later era where free commerce and self-regulating markets were accorded greater importance and where the emergent British industrial economy required the importation of raw materials and basic foodstuffs (Lange 2009; Lange, Mahoney, and vom Hau 2006). In its push into Africa and Asia, British colonialism sought to provide the basic legal framework and physical infrastructure required for market-based economies, for example by providing basic public goods and by creating the rule of law and enforcing it through courts and police. These liberal aims, however, often conflicted with entrenched interests from the pre-colonial society, thereby limiting the possibilities for a wholly liberal colonialism. Especially in Africa, Britain is understood to have pursued an indirect or decentralized rule whereby British colonial administrators granted traditional communal leaders significant authority over their local populations, choosing instead to rule the general population indirectly via control over these communal leaders (Crowder 1968/2023; Miles 1994, Chapter 5). This seemingly accorded some deference to the pre-colonial social structure, as indigenous rules and customs were often left intact, but it also hardened and politicized ethnic divisions because communal groups were usually defined by ethnicity. Overall, in broad terms, ex-British colonies inherited the rule of law, decentralized forms of administration, and the capacity for market-based economies, but they also inherited ethnically divided societies and relatively weak states with limited capacities to exercise control and coordinate development.

The French empire also began during the era of liberal capitalism, giving it economic aims and institutional forms similar to its British counterpart. However, inspired by France’s statist approach to governance, French colonial rulers used a distinct administrative style often called direct or centralized rule (see Crowder 1968/2023; Letsa and Wilfahrt 2020; Miles 1994, Chapter 5; Müller-Crepon 2020). Here colonial administrators ceded little authority to indigenous populations or their leaders, but instead subjected the whole colony to a centralized and rule-based system of governance similar to the one used in France. This entailed a forceful reconstruction of pre-colonial social structures, resulting in wholly new institutions and sanctioned identities. Political authority was centralized and colonial administrative boundaries ignored pre-colonial social configurations, instead emulating the standardized administrative practices found in France. Coerced labor was used in some instances, but ethnic diversity was downplayed in favor of common national identities built around French values. As its legacy, French colonialism left strong states capable of coordinating development efforts, although these states may struggle for legitimacy as they reflect French rather than pre-colonial cultural and institutional logics. Furthermore, inter-ethnic tensions should be manageable, because ethnicity was never used as the constituent unit of colonial governance and because strong states should have the resources to maintain stability.

There are, of course, notable variations within these broad inter-colonial patterns. The British system of indirect rule required pre-colonial societies that were hierarchically ordered, thereby making collaboration with indigenous elites possible. However, in parts of West Africa, these social conditions were not present, forcing British colonialists to adopt more direct forms of rule (Müller-Crepon 2020). Likewise, in parts of India, British colonialists found hierarchically ordered societies, but ones organized around land/caste divisions rather than ethnicity. In these regions, indirect rule meant collaboration with upper-caste landlords, which in tun heightened class rather than ethnic divisions (Mukherjee 2018). Indeed, as argued by Lange (2009), British colonialism took fairly diverse forms across India and East Asia, with some regions using more direct forms of rule.

Similarly, the French use of direct rule was not applicable in all cases. Direct rule assumes uniform bureaucratic control over geographic distance, but centralized administration broke down in remote parts of West Africa due to the vastness of this territory. Here the French adopted a de facto system of indirect rule, whereby French colonialists governed in collaboration with local indigenous elites in ways evocative of the British (Firmin-Sellers 2000; Miles 1994, 95). On this topic, Müller-Crepon (2020) finds that, for ex-French colonies in West Africa, nearly 30% of the pre-colonial political titles, such as tribal chieftain, survived into the post-colonial era. The comparable figure for ex-British colonies in West Africa is 70%.

Despite some intra-colonial variations, the unique political economies and administrative practices of the major colonial powers left distinct post-colonial institutions and national power configurations, which should affect the prospects for domestic peace in their former colonies. Here some research identifies ex-British colonies as the most susceptible to civil violence, especially violence between ethnic groups (Blanton, Mason, and Athow 2001; Lange and Dawson 2009). These researchers contend that the British system of indirect rule—also called “divide and rule”—hardened pre-colonial ethnic divisions by granting ethnic groups formal roles in colonial governance. Consequently, political competition in ex-British colonies often pits organized ethnic groups against one another, with the competition potentially spiraling into more serious conflict. To complicate matters, the weak states inherited by many ex-British colonies may lack the capacities needed to maintain social order during periods of unrest. Recent research confirms these broad tendencies but highlights contingent factors that make blanket statements linking indirect rule to post-colonial civil violence problematic (Lange, Jeong, and Amasyali 2021; Mukherjee 2018; Ray 2018). Overall, these ideas lead to my third theoretical expectation, namely that post-colonial societies emerging from British colonialism, compared to those emerging from other major colonial powers, are more prone to ethnic-based civil violence, but not more prone to civil violence overall (Hypothesis 3).

By contrast, Wucherpfennig, Hunziker, and Cederman (2016) contend that ex-French colonies are the most prone to civil violence. Their reasoning is that, under the French system of direct rule, power resided in the capital city and groups in rural areas and provincial towns were marginalized. This meant that indigenous urban elites were far better connected to colonial-era power structures, both in terms of their social and cultural capital but also in terms of their physical proximity. When independence occurred, indigenous elites in the capital moved quickly to gain state power, and once in control, the political architecture of direct rule made power sharing less important. In this way, French colonialism supposedly created systematic inequalities in terms of accessing the post-colonial state, with the related tensions manifesting between groups separated by region rather than ethnicity. On this measure, ex-British colonies fare better, because the legacy of indirect rule provided different communal/ethic groups institutionalized channels for accessing the state.

In sum, the literature suggests that the propensity for civil violence in post-colonial societies varies by colonial legacy. Spanish colonialism stands out as leaving behind predatory states and entrenched inequalities, which may continue to inspire rebellions in the post-colonial era (cf. Lange and Dawson 2009). However, differences between British and French colonial legacies are less clear, as each have characteristics that may spur or blunt civil violence. Combined, this leads to my fourth theoretical expectation, namely that ex-British colonies are the most prone to ethnic-based violence, but ex-French colonies are the most prone to socio-economic-based violence (Hypothesis 4). However, ex-British and ex-French colonies may be similar in terms of their propensities toward civil violence in general.

Possibility of waning temporal effects

A prominent literature suggests that colonialism created path-dependent trajectories in which colonial-era institutions and practices became entrenched and continue to influence outcomes into the present era (Acemoglu, Johnson, and Robinson 2001, 2002; Kohli 2004; Lange 2009; Lange, Mahoney, and vom Hau 2006; Sokoloff and Engerman 2000; cf. Maseland 2018). However, a less recognized possibility is that these long-term effects, especially as they relate to civil violence, may wane over time. This may occur for several reasons. First, the decolonialization process was often politically turbulent, as competing domestic groups vied for control of the colonial-era state (Fearon and Laitin 2003; Ray 2018; Wucherpfennig, Hunziker, and Cederman 2016). Consequently, the years directly following independence should be especially prone to civil violence, but once this periods ends, more normal patterns of domestic stability may prevail. Additionally, some post-colonial societies should make gradual progress towards transforming their economies and resolving their colonial-era grievances, perhaps enabling them to achieve high levels of domestic stability. Indeed, Lange and Dawson (2009) find evidence that ethnic violence in post-colonial societies subsides over time (see also Lange, Jeong, and Amasyali 2021). Combined, these ideas support the theoretical expectation that the risk of civil violence in post-colonial societies declines as their years of independence increase (Hypothesis 5).

Data and methods

Sample

To assess my ideas about the historical legacy of colonialism and its effect on contemporary patterns of civil violence, I compile a global sample of countries observed annually from 1960 to 2018. The sample includes all countries recognized by the United Nations in 2018, except those with populations less than one million people at that time. New countries, such as those arising from the break-up of the Soviet Union, enter the sample when they become independent. In total, this yields 7980 country-year observations from 152 countries, comprising formerly colonized and non-colonized countries. For the latter portion of my analysis, I focus on post-colonial societies only, which reduces the sample to 6655 observations from 130 countries. The countries comprising the full sample, along with their attendant colonial legacy, are shown in the online appendix (see Tables A1 and A2). Note that the study’s timeframe reflects specific data constraints¹.

Since my sample contains nearly all the world’s countries, it more closely approximates an “apparent population” than a random sample (see review by Bollen 1995; see also Lange, Jeong, and Amasyali 2021, 152; Kollmeyer 2024, footnote 3). Under such conditions, the use of statistical significance testing is less applicable, because inferential statistics assume that researchers have drawn random samples from larger populations. This issue clearly applies to my study, because my data constitute nearly all countries in the world with populations greater than 1 million. Consequently, some readers may wish to interpret my regression results as descriptions of the actual population parameters rather than estimates derived from random sampling. Hence, my empirical analysis places less emphasis on tests of statistical significance, even though attendant details are reported in accordance with disciplinary norms.

Measuring civil violence

The study’s measure of civil violence comes from the MEPV database developed by the CSP (Marshall 2019). The CSP/MEPV tracks and records “major episodes of political violence,” which it defines as the “systematic and sustained use of lethal violence,” perpetrated by organized domestic groups and resulting in at least 500 deaths (including combatants and non-combatants). There is no minimum number of annual fatalities, as long as the conflict reaches the 500-deaths threshold over its duration. In particular, I use the CIVTOT variable, which measures the entire range of intra-societal conflicts, including rebellions, coups, and civil wars, but excludes conflicts between states. It is coded along an 11-point scale, ranging from 0 (no major incidence of civic violence) to 10 (systematic and indiscriminate destruction of society). I recode this measure into a dummy variable, with 1 being country-years experiencing civil violence (at any level of intensity) and 0 being country-years with domestic peace. This coding strategy follows the quantitative literature on civil violence, which almost exclusively uses dichotomous measures of this outcome (see Collier and Hoeffler 2004; Fearon and Laitin 2003). Scale measures of civil violence are rarely used, mainly because they capture the intensity of civil violence, a phenomenon driven by separate determinants.

This measure of civil violence can be further refined. Unlike the UCDP/PRIO Armed Conflict Dataset and the Minorities at Risk Project, the CSP/MEPV categorizes each episode of civil violence by its underlying grievances and political aims. This allows me to identify conflicts that predominantly coincide with ascriptive, identity-based cleavages (such as ethnicity, religion, and race), and those that predominantly coincide with non-ascriptive, socio-economic cleavages (such as class, region, ideology, and political orientation). Admittedly, these distinctions blur considerable complexity and minimize the ability of identities and motivations to intertwine, but they provide some indication of the broad nature of these conflicts. Importantly, this distinction allows me to assess whether the types of civil violence systematically differ by colonial legacy. This is highly germane because some colonial projects entrenched identity-based cleavages (British colonialism) while others defused them (French and Spanish colonialism).

In sum, my analysis employs three dependent variables, each coded as a dummy variable. Civil violence (total) captures all major episodes of intra-societal violence regardless of the underlying aims and motivations of the combatants. These cases are subdivided into two mutually exclusive categories, namely civil violence (ethnic-based) and civil violence (socio-economic-based).

Measuring colonial legacy

To measure the colonial legacy of the countries in my sample, I consult and cross-check data from several sources (Lange and Dawson 2009; Olsson 2009; Ziltener, Künzler, and Walter 2017; also CIA 2021). In most cases, a country’s colonial history is straightforward, but occasionally it is more complicated. For these latter instances, three ideas guide my coding decisions. First, following Ziltener et al. (2017), I note that some countries were never formally colonized but suffered colonial domination nonetheless (e.g., Afghanistan, Bhutan, China, Ethiopia, etc.). The form of colonial domination, I suggest, is less important than the actual control foreign powers exerted over subjugated countries. Often colonial domination was formal (e.g., India was an overseas colony of Britain), but sometimes it was informal (e.g., China signed “unequal treaties” with Britain after defeat in the Opium Wars). Hence, I group formal and informal forms of colonial domination together, rather than treating countries in the latter group as escaping colonialism altogether. Second, some countries were officially colonized/dominated by multiple colonial powers (e.g., China was first dominated by Britain and then by Japan). I treat these cases as instances of “multiple colonization.” Third, ex-British settler colonies were officially colonized but their historical experiences were notably different and their long-term consequences unique (Acemoglu, Johnson, and Robinson 2001, 2002). Hence, I code ex-British settler societies as non-colonized for the main analysis. However, as a robustness check, I use a measure consistent with Lange and Dawson (2009), which treats ex-British settler colonies as having been colonized. The results of my coding process are shown in Table A1 and A2 of the online appendix.

Guided by this coding framework and using the data described above, I create several broad measures of colonial legacy. My simplest measure is Post-Colonial Society, a dummy variable identifying countries that were either formally or informally colonized. Again, ex-British settler societies are coded as not colonized. This variable adheres closely to the coding logic of Ziltener et al. (2017). Next, Former Colonizer identifies the colonial powers exerting control over the colonized countries. To simplify this measure, I use the following categories: Britain, France, Spain, other-European powers, non-European powers, multiple powers, or never colonized. The other-European powers include Belgium, Holland, Germany, Italy, and Portugal, and the non-European powers include Japan, Turkey, Russia/Soviet Union, and the United States.

Years independent (log) captures the number of years post-colonial societies have been sovereign states. It equals the year of the present observation minus the year of independence, with the difference being logged to account for non-linear effects. If there was no formal granting of independence, as occurred under informal colonialism, I use the most relevant political milestone. For example, China’s colonial domination ended in 1949 with the Chinese Communist Revolution. Hence, in 1960, China was independent for 11 years (i.e., 1960–1949 = 11;²). The introduction of this variable, however, reduces the sample size because it only pertains to post-colonial societies.

Post-treatment bias and control variables

Ideally, I could estimate the direct and indirect effects of colonial legacy on civil violence. To begin, I would estimate raw odds ratios by fitting logistic models in which civil violence is a sole function of colonial legacy. These results would depict the actual differences in the chances of civil violence by colonial legacy. Then, I could partial out the effect of economic development, democratization, and related factors by introducing them as control variables. Comparing the two sets of results would reveal whether the observed differences in civil violence are driven by variations in the control variables, or whether colonialism left an enduring effect, which goes beyond these indirect effects.

Unfortunately, this cannot be done without creating “post-treatment bias” (see Dworschak 2024; Elwert and Winship 2014). For many topics in macro-comparative research, the key explanatory variable, that is, the “treatment” variable in the language of causal analysis, influences not only the outcome of interest but other explanatory variables in the model. Since the “post-treatment” variables temporally follow the main explanatory variable in the causal sequence, they are endogenous and should not be used as standard control variables. For my study, colonial legacy clearly shaped economic development and democratization, yet these factors are themselves well-known covariates of civil violence (Collier and Hoeffler 2004; Fearon and Laitin 2003; Hibbs 1973; Muller 1985). Prior research overlooks this complication by using post-treatment covariates as control variables in their models (Blanton, Mason, and Athow 2001; Lange and Dawson 2009; Lange, Jeong, and Amasyali 2021; Wimmer, Cederman, and Min 2009).

How can this complication be overcome? To address the resulting endogeneity, Dworschak (2024) recommends abandoning attempts to partial out the net effects of the post-treatment covariates and instead only controlling for pre-treatment covariates. In my case, pre-treatment covariates are those control variables not directly influenced by colonialism. This model specification will provide unbiased estimates of the total effects, although the total effects cannot be subdivided into direct and indirect components.

With this in mind, the following pre-treatment covariates are used as control variables. Population size accounts for the fact that populous countries have greater exposure to separatist movements and face higher odds of civil conflict occurring somewhere within their borders simply due to their larger populations. Data comes from the World Bank (2022). Mountainous terrain measures the percentage of a country’s landmass covered by mountains and similar terrain. Such topography can facilitate rebellions by providing cover and hideaways to rebel groups. Data from Fearon and Laitin (2003). Latitude measures a country’s geographic center point and accounts for climatic effects that may hinder development in equatorial and tropical regions. Data from the CIA (2021). Cold War accounts for the proxy wars between the Soviet Union and the United States, which spurred intra-societal conflict in many countries. It is coded “1” for the years 1960 to 1990. Another dummy variable, ongoing international war, identifies instances in which a country is engage in a major conflict with another country. Such conflicts can weaken the state internally, creating opportunities for rebellion, but they sometimes create a “rally around the flag” effect. Data come from the MEPV database (Marshall 2019).

Lastly, years of peace equals the number of years since the last civil conflict and captures the tendency of post-conflict countries to revert to civil violence (Collier and Hoeffler 2004). This measure is derived from the dependent variable starting with the year 1960. Its relationship with civil conflict should be negative, indicating that the chances of conflict are highest in the period directly following a prior conflict and then decline over time.

The following post-treatment covariates are used in robustness checks, even though this risks post-treatment bias. GDP per capita (log) controls for economic development, with the measure being adjusted for inflation and purchasing power parities and logged for non-linear effects. Development should reduce incidences of civil violence by lowering grievances, by imposing higher opportunity costs on would-be combatants, and by giving states more resources to foster social stability (Collier and Hoeffler 2004). Economic growth should reduce civil conflict through similar channels. Data for these variables come from the Penn World Table (Feenstra, Inklaar and Timmer 2019). Finally, political systems should matter too. Democracies are usually peaceful, but autocracies can create peace by curbing dissent with force. Hence, civil violence should disproportionately occur in anocracies, regimes that are neither fully autocratic nor fully democratic (Hegre et al. 2001). Data come from the Polity IV Project (Marshall, Gurr, and Jaggers 2019).

Analytic strategy

My estimation strategy follows standard modeling techniques used to study civil conflict when the dependent variable is dichotomous and the data exhibit a panel structure (see Fearon and Laitin 2003; Kollmeyer 2024; Lange, Jeong, and Amasyali 2021; Mihalache-O’Keef 2018; Wimmer, Cederman, and Min 2009). Following this literature, I employ logistic regression analysis coupled with robust standard errors to mitigates against heteroskedasticity and serial correlation in the error terms. I also include years of peace (expressed as a cubic polynomial) to account for temporal dependence in the data, which can be problematic for logistic regression (Carter and Signorino 2010). Furthermore, I consider the possibility of non-stationarity (Box-Steffensmeier et al. 2014,125–149). Especially when panel data have long temporal dimensions (such as mine), annual observations can trend over time rather than vary randomly. This creates the potential for “spurious regression.” To safeguard against this possibility, I include a time-trend variable as a regressor (i.e., t = 1, 2, 3… n). This variable is logged to account for non-linear trends.

Results

Analysis of global sample

Table 1 reports the logistic regression results from my global sample of post-colonial and non-colonized societies. The results are reported as odds ratios and assess the relative chance that given societal characteristics are associated with civil violence. Model 1 is a bivariate model in which my broadest measure of colonial legacy is the sole predictor of civil violence. This result can be interpreted as a raw odds ratio because there are no control variables. As expected, the resulting odds ratio for post-colonial society is large, indicating that previously colonized societies are 5.7 times more likely to experience civil violence than non-colonized societies. This difference is statistically significant at the P < .000 level.

Table 1

Global sample: Odds ratios of civil violence by colonial legacy, 1960 to 2018.

	(Civil violence yes = 1, no = 0)
	(1)	(2)	(3)	(4)
Colonial legacy
Post-colonial society (1 = yes)	5.72***	3.04***
Former colonial power
Britain			7.94***	3.65***
France			3.72***	2.85***
Spain			3.79***	3.18***
Other-European powers			14.54***	3.00***
Non-European powers			1.34	2.95***
Multiple powers			15.00***	2.24**
Never colonized (ref cat.)			--------	--------
Control variables
Population size (log)		1.35***		1.37***
Cold war (1 = yes)		2.70***		2.64***
Ongoing international war (1 = yes)		1.33		1.33
Mountainous terrain (log)		1.12**		1.14**
Latitude (log)		1.05		1.06
Prior years peacea		0.38***		0.38***
Constant	0.03***	0.02***	0.03***	0.02***
Observations	7989	7831	7989	7831
Number of countries	152	152	152	152

Note: Logistic regression with robust standard errors. Results reported as odds ratios. Time trend included but not reported.

^aSquared and cubed terms included but not reported.

^***P < .001, ^**P < .01, ^*P < .05, ^†P < .10.

Model 2 introduces controls for the pre-treatment covariates. The odds ratios for these control variables are largely consistent with theoretical expectations and show that population size, mountainous terrain, the Cold War, and ongoing international wars heighten the chance of civil violence, whereas the duration of continuous peace lowers the chance of civil violence. A country’s latitude shows no effect. As expected, the introduction of these control variables lowers the odds ratio associated with post-colonial society. Now, post-colonial societies are 3.0 times more likely to experience civil violence than comparable non-colonized societies, with this difference being statistically significant at the P < .001. This generally means that post-colonial societies remain moderately more prone to civil violence even after partialing out the effects of the pre-treatment covariates (supports Hypothesis 1).

Models 3 and 4 re-estimate the same models but now using a finer-grained measure of colonial legacy. Former colonial power not only distinguishes between post-colonial and non-colonized societies, but also identifies the previous colonial ruler (if applicable). Model 3 shows the raw odds ratios depicting the actual chances of civil violence by former colonial power. As expected, non-colonized societies (the reference category) are the least prone to civil violence, as evidenced by the odds ratios for all other categories being well above 1. The results also indicate that countries previously colonized by multiple powers are the most prone to civil violence (OR = 15.0) while those colonized by non-European powers are the least (OR = 1.3). For the major colonial powers, ex-British colonies are roughly twice as prone to civil violence as compared to ex-French or ex-Spanish colonies. This outcome is consistent with prior studies (Blanton, Mason, and Athow 2001; Lange and Dawson 2009).

Model 4 introduces the pre-treatment covariates. The results continue to identify non-colonized societies as the least prone to civil violence, but the relative differences drop notably. Interestingly, net of the control variables, the odds ratios for ex-British, ex-French, and ex-Spanish colonies are now quite similar. They range from 3.6 for ex-British, to 3.2 for ex-Spanish colonies, to 2.8 for ex-French colonies. These differences are not statistically significant. Although not shown here, I re-run this model with Britain as the reference category and this generates p-values well above p = 0.10 for the French and Spanish odds ratios.

Analysis of post-colonial sample

Table 2 extends the analysis in two important ways. First, these models introduce the variable years independent (log), which enables me to assess whether post-colonial societies are becoming less conflict prone as they gain more experience as sovereign states. This factor is potentially important because years of independence differs greatly by former colonial ruler. Indeed, according to averages computed from my sample, ex-Spanish colonies have been independent for 146 years, compared to 33 years for ex-British colonies and 39 years for ex-French colonies. Note that the introduction of this variable reduces the sample size as non-colonized societies drop out.

Table 2

Post-colonial societies only: Odds ratios of civil violence by colonial legacy, 1960 to 2018.

	Predominant type of civil violence
	Ethnic violence (5)	Socio-economic violence (6)	All civil violence (7)
Colonial legacy
Former colonial power
Britain	3.12***	0.38***	1.55*
France	0.94	1.15	1.22
Spain	0.28***	2.37***	1.51†
Other-European powers	1.39	0.72	1.26
Non-European powers	1.93**	0.63†	1.18
Multiple powers (ref cat.)	--------	--------	--------
Years independent (log)	1.12†	0.74***	0.89†
Control variables
Population size (log)	1.59***	1.08†	1.35***
Cold war (yes = 1)	1.18	2.97***	2.71***
Ongoing international war (yes = 1)	2.32***	0.57**	1.45†
Mountainous terrain (log)	1.16***	1.01	1.17***
Latitude (log)	1.01	0.91	1.05
Prior years peacea	0.31***	0.51***	0.38***
Constant	0.05***	0.06***	0.04***
Observations	6528	6528	6528
Number of countries	130	130	130

Note: Logistic regression with robust standard errors. Results reported as odds ratios. Time trend included but not reported.

^aSquared and cubed terms included but not reported.

^***P < .001, ^**P < .01, ^*P < .05, ^†P < .10.

Second, these models examine whether the predominant types of civil conflict vary by colonial legacy. Prior research identifies ex-British colonies as being particularly prone to ethnic-based civil violence, purportedly because British colonialism used local ethnic groups as the constituent unit of colonial governance, something that politicized ethnicity. However, French and Spanish colonialism supposedly created durable inequalities, but ones that align with socio-economic and regional differences rather than ethnicity. Whether these colonial-era differences still influence patterns of civil violence can be tested using the CSP/MEPV data.

Models 5, 6, and 7 regress each of my three measures of civil violence on the same set of explanatory variables. This allows for a direct comparison of colonial legacy’s effect on the different types of civil violence examined in this study. The new reference category is “multiple powers,” because non-colonized societies drop from the sample, and because this facilitates comparisons across the British, French, and Spanish cases. According to these regression estimates, ex-British colonies are clearly the most prone to ethnic-based civil violence, but not necessarily to civil violence overall (supports Hypothesis 3). Additionally, ex-Spanish colonies are the most prone to socio-economic-based violence, but the least prone to ethnic violence (supports Hypothesis 2). Finally, ex-French colonies are slightly more prone to socio-economic-based violence than ethnic-based violence (supports Hypothesis 4). All of these findings are consistent with my theoretical expectations and generally support the notion that unique colonial legacies affect the patterns of civil violence experienced in post-colonial societies.

These three models also shed light on the possibility that post-colonial societies are becoming less conflict prone as they gain more experience as sovereign states. This is evaluated by the years independent (log) variable. Interestingly, for this variable, the results vary across the types of civil violence. In particular, the prevalence of ethnic-based violence worsens slightly as the years of independence increase, although the effect is not statistically significant (fails to support Hypothesis 5). However, the prevalence of the other two measures of civil violence improves modestly and both effects are statistically significant (supports Hypothesis 5). Taken together, these findings imply that the worsening of ethnic-based violence is more than offset by improvements in socio-economic-based violence, causing the overall prevalence of civil violence to ebb downwards as years of independence increase.

These temporal effects are examined further by re-running Models 5, 6, and 7, but now including interactions between years independent (log) and the former colonial ruler. These results, used to generate figure 1, allow me to assess whether the effect of years of independence is the same across all colonial legacies, and to consider a counterfactual scenario in which years of independence is equalized across ex-British, ex-French, and ex-Spanish colonies. This latter point is important because Spanish colonialism ended nearly a century before British and French colonialism ended. To make these assessments, I use Stata’s margins command to predict the propensity for civil violence when years independent (log) varies across its sample range.³ Recall that this variable is in log format, so the x-axis of figure 1 is expressed in non-linear terms. Hence, the mid-point on the x-axis is approximately 20 years of independence (i.e., ln(20) = 2.99) and the end-point is approximately 150 years of independence (i.e., ln(150) = 5.01;⁴).

Figure 1

Predicted probability of civil violence by former colonial ruler and years of independence. Note: Predicted probabilities are calculated with Stata’s margins command. They represent the effect of colonial legacy on civil violence when the other covariates are held constant at their sample means and when years of independence is statistically equalized across all post-colonial societies.

Figure 1 shows the predicted probability of civil violence under a counterfactual scenario in which years of independence is statistically equalized across all colonial legacies. The results generally show that the risk of civil violence is highest in the decades immediately following independence and then ebbs downward over time but never completely vanishes (supports Hypothesis 5). This is apparent from the various trendlines, which mainly but not always grow closer together as years of independence increases. Given the non-linear measurement of years of independence, the trend toward less civil violence occurs rapidly at first, but never reaches the zero point. Notably, ex-Spanish colonies now emerge as being highly prone to civil violence in the years following independence, something that had been masked by the nearly 100-year difference between the end of Spanish colonialism and the end of British and French colonialism. However, the situation for ex-Spanish colonies improves markedly over time, giving rise to patterns of civil violence that become similar across the three colonial legacies.

Figure 1 also sheds light on whether the patterns of civil violence change and evolve over time. The predicted trendlines indicate that ex-French and ex-Spanish colonies are prone to socio-economic-based violence, especially in the years directly following independence, but as time passes, the risk of ethnic-based violence spikes notably upwards for ex-French colonies and slightly upwards for ex-Spanish colonies. In fact, ex-French colonies are projected to eventually become more prone to ethnic-based violence than socio-economic-based violence. Ex-British colonies experience some countermovement as well, with their risk of socio-economic-based violence drifting upward over time, even as ethnic-based violence ebbs slightly downwards. Combined, many of the predicted patterns of civil violence change and evolve. This seems especially true for ex-French colonies, something that is discussed in the conclusion. Again, these are predicted trends based on the regression results, and they may fail to materialize as shown in figure 1.

Robustness checks

My analysis ends by performing several robustness checks. The first check revisits my handling of post-treatment effects. Prior research on civil violence and colonial legacy routinely controls for economic development, democratization, and related factors, but this creates post-treatment bias because these factors are themselves influenced by colonial legacy (Dworschak 2024; Elwert and Winship 2014). For this reason, I exclude post-treatment covariates from my original models, but some readers may wish to see them included.

With this in mind, I re-run the models presented in Table 2, but now include measures of economic development, economic growth, and democracy. The new estimates, shown in Table 3, do not alter my substantive conclusions. Ex-British colonies remain the most prone to ethnic-based violence and ex-Spanish colonies remain the most prone to socio-economic-based violence and to civil violence overall. In fact, the effects attributed to colonial legacy have now become larger (cf. Table 2 and Table 3). This suggests that the “post-estimation bias” is inflating the revised estimates. I also try alternative measures of democracy but find nearly identical results⁵.

Table 3

Post-treatment variables included: Odds ratios of civil violence by colonial legacy, 1960 to 2018.

	Predominant type of civil violence
	Ethnic violence (8)	Socio-economic violence (9)	All civil violence (10)
Colonial legacy
Former colonial power
Britain	5.88***	0.18***	1.18
France	1.14	0.76	0.77
Spain	0.19***	4.00***	1.63*
Other-European powers	2.01**	0.48**	0.94
Non-European powers	1.91*	0.63	0.78
Multiple powers (ref cat.)	------	------	------
Years independent (log)	1.22*	0.62***	0.85*
Post-treatment variables
GDP per capita (log)	1.03	0.79***	0.83**
Economic growth	0.98	0.96***	0.95***
Political regime type
Autocracy	0.53***	1.18	1.23
Anocracy	0.89	1.26	1.69***
Democracy (ref cat.)	------	------	------
Control variables
Population size (log)	1.52***	1.25***	1.37***
Cold war (yes = 1)	1.32†	3.43***	3.07***
Ongoing international war (yes = 1)	1.97**	0.73	1.88*
Mountainous terrain (log)	1.39***	0.85	1.09*
Latitude (log)	1.12†	0.90	1.16*
Prior years peacea	0.30***	0.52***	0.40***
Constant	0.01***	0.49	0.11**
Observations	6036	6036	6036
Number of countriesb	124	124	124

Note: Logistic regression with robust standard errors. Results reported as odds ratios. Time trend included but not reported.

a. Squared and cubed terms included but not reported.

b. Six countries drop due to missing data for the post-treatment variables.

^***P < .001, ^**P < .01, ^*P < .05, ^†P < .10.

The second robustness check considers whether my results hold under an alternative measure of colonial legacy. Recall that my original measure treats informally dominated countries (such as Iran and China) as having been colonized but excludes ex-British settler colonies (such as Canada and the United States) from this designation. To see whether these measurement choices affect my substantive conclusions, I recode my colonial legacy variable, making it consistent with Lange and Dawson (2009, Table A1 and A2).

The new results, shown in Table 4, continue to support my original conclusions. Ex-British colonies are still the most prone to ethnic-based violence, although the pertinent odds ratios fall from 3.12 to 1.53. This likely reflects that, under this measure, ex-British colonies now include the relatively more peaceful ex-settler colonies of Australia, Canada, New Zealand, and the United States. By contrast, ex-Spanish colonies are still the most prone to socio-economic-based violence, with the pertinent odds ratio increasing from 2.37 to 3.46.

Table 4

Alternative measure of colonial legacy: Odds ratios of civil violence by colonial legacy, 1960 to 2018.

	Predominant type of civil violence
	Ethnic violence (11)	Socio-economic violence (12)	All civil violence (13)
Colonial legacy
Former colonial power
Britain	1.53†	0.37***	1.00
France	0.35***	1.57†	0.88
Spain	0.13**	3.46***	1.51†
Other-European powers	0.60†	1.38	0.84
Non-European powers	------	------	------
Years independent (log)	0.97	0.76**	0.78**
Standard control variables included?	Yes	Yes	Yes
Post-treatment variables included?	No	No	No
Observations	6296	6296	6296
Number of countries	130	130	130

Note: Logistic regression with robust standard errors. Results reported as odds ratios.

^***P < .001, ^**P < .01, ^*P < .05, ^†P < .10

A third robustness check measures civil violence as a scale variable. Recall that the CSP/MEPV data range from 0 (no major incidence of civic violence) to 10 (systematic destruction of society). Following the literature on civil violence, I recode this measure into a dummy variable (see Collier and Hoeffler 2004; Fearon and Laitin 2003). Scale variables are rarely used because they capture the intensity of civil conflicts, something with its own set of determinants. Nonetheless, I now re-run my models use the original 11-point scale.

The results, shown in Table 5, reveal slightly different patterns between colonial legacy and types of civil violence. As expected, ex-British colonies are the most prone to highly destructive outbreaks of ethnic-based violence, but unexpectedly ex-colonies of “other-European powers” emerge as the most prone to highly destructive outbreaks of civil violence overall. This latter finding is consistent with anecdotal evidence. For instance, Angola and Democratic Republic of Congo (ex-colonies of Portugal) and Mozambique and Rwanda (ex-colonies of Belgium) have experienced deadly civil conflicts over recent decades. By contrast, the ex-Spanish colonies fare much better by this measure. Their lower scores suggest that, relative to other post-colonial societies, ex-Spanish colonies have experienced more numerous but less destructive episodes of civil violence.

Table 5

Random effects estimates using scale rather than binary measures of civil violence, 1960 to 2018.

	Predominant type of civil violence (scale from 0 to 10)
	Ethnic violence (14)	Socio-economic violence (15)	All civil violence (16)
Colonial legacy
Former colonial power
Britain	0.45†	−0.26†	0.33†
France	−0.05	−0.14	−0.03
Spain	−0.14	0.09	0.12
Other-European powers	0.19	0.45	0.72†
Non-European powers	0.05	0.06	0.29
Multiple powers	------	------	------
Years independent (log)	−0.03	−0.09**	−0.09**
Standard control variables included?	Yes	Yes	Yes
Post-treatment variables included?	No	No	No
Observations	6527	6527	6527
Number of countries	130	130	130

Note: Generalized least squares regression with random effects and robust standard errors.

^***P < .001, ^**P < .01, ^*P < .05, ^†P < .10

A final robustness check extends my sample’s timeframe. The CSP/MEPV data on civil violence start in 1946, but my study’s timeframe begins in 1960 largely because the CSP/MEPV database are missing all observations from the mid-1940s and 1950s for many developing countries, including nearly all sub-Saharan African countries. These missing cases will skew my sample for the pre-1960 period, causing it to over-represent countries that are more stable and more developed, and to bypass a heavily colonized parts of the world. This makes projecting historical trends difficult.

Nonetheless, as a robustness test, I extend my sample back to 1946. This is done by collecting the non-missing data from the CSP/MEPV database for the mid-1940s and 1950s and by extrapolating the missing data for relevant control variables over this same period. This latter step is done with Stata’s extrapolation command, ipolate/epolate, which estimates the missing observations from the observed values from 1960 to 2018. Control variables with fixed values were unchanged and extended back to 1946.

Table 6 shows the results from this robustness check. The new results are fairly consistent with the original ones, although the magnitudes of the odds ratios are reduced (cf. Table 2 and 6). Ex-British colonies are still the most prone to ethnic-based violence, and ex-Spanish colonies are still the most prone to socio-economic-based violence. However, years independent (log) now shows either no change over time or a worsening effect. This contrasts with my original estimates, which generally indicate a move toward less civil violence. This divergent outcome, I believe, reflects the altered composition of the extended sample. When conflict-prone countries enter the sample in the 1960s and beyond, it now makes these decades seem more troubled by comparison.

Table 6

Extended sample: Odds ratios of civil violence by colonial legacy, 1946 to 2018.

	Predominant type of civil violence
	Ethnic violence (17)	Socio-economic violence (18)	All civil violence (19)
Colonial legacy
Former colonial power
Britain	1.98***	0.33***	1.23
France	0.36***	0.86	0.73†
Spain	0.10***	1.50**	0.57**
Other-European powers	0.99	0.73†	1.18
Non-European powers	0.81	0.40***	0.58*
Multiple powers	------	------	------
Years independent (log)	1.26**	1.03	1.14**
Standard control variables included?	Yes	Yes	Yes
Post-treatment variables included?	No	No	No
Observations	7356	7356	7356
Number of countries	130	130	130

Note: Logistic regression with robust standard errors. Results reported as odds ratios.

^***P < .001, ^**P < .01, ^*P < .05, ^†P < .10

Discussion and conclusion

A large and highly developed quantitative literature seeks to explain the patterns of civil violence found in the world today. The dominant explanation emphasizes factors related to development and democratization, contending that poor countries with stagnant economies and unresponsive political systems are the most prone to civil violence. Interpreted largely through a rational-choice perspective, these conditions are thought to heighten the incentives but lower the constraints for civil violence, creating societies in which armed conflict becomes a relatively appealing means of economic gain and political redress. By contrast, a smaller literature considers the historical legacy of colonialism and whether it continues to influence the prospects for domestic peace today. The present study puts forward and empirical tests several ideas about colonial legacy’s effect on domestic peace.

More specifically, the present study grapples with several interconnected points regarding the nature of causation between colonial legacy and civil conflict. The first relates to the possibility of path dependency. As used in this study, path dependency occurs when distant historical conditions continue to affect contemporary outcomes, largely through an inertia effect whereby the initial institutions and social practices become deeply entrenched (see Mahoney 2000). The literature on comparative development, both from sociology and al economics, highlights the strong possibility that colonialism introduced economic structures and political institutions that retain their influence over societal outcomes today. The present study reaches equivalent conclusions, finding that various measures of colonial legacy help to explain contemporary patterns of civil violence, even after controlling for known covariates.

However, the literature on comparative development, as well as the more specific literature on colonial legacy and civil violence, insufficiently addresses the duration of path-dependent effects. Are they permanent, or do they wane over time? Perhaps their magnitude increases in some cases? On this subject, the present study finds that the patterns of civil violence associated with unique colonial legacies change and evolve but never fully abate. This highlights the importance of considering temporal dynamics when studying path-dependent effects.

The present study also raises questions regarding the capacity of post-colonial societies to change their historical trajectories. A particularly important question is why do unfavorable conditions originating from colonialism retain their troublesome effects, especially when post-colonial societies have sovereign policymaking powers? Or posed in broader sociological terms, what causal mechanisms hold the initial social practices and institutions in place? Hopefully, future research can address this question more fully, but the literature reviewed here offers some possible answers. One possibility is that the indigenous elites inheriting the post-colonial state may engage in corruption and rent seeking rather than reform (see Evans 1989, 1995; Kohli 2004). In these cases, the inertia upholding path dependency is the narrow self-interest of entrenched post-colonial elites. Another possibility is that, operating in a competitive global economy, not all post-colonial societies will equally succeed in upgrading their economies (see Baran 1957; Frank 1966). Surely some countries will lag behind in this process, and possibly the initial starting point matters for who eventually succeeds. Hence, under this scenario, the inertia upholding path dependency is market competition and its tendency to create and reinforce stratified outcomes.

A less anticipated findings is that ex-French colonies are becoming increasingly prone to ethnic-based civil violence. This may reflect the gradual erosion of centralized state authority in ex-French colonies and the concomitant re-emergence of local authority, potentially now built around pre-colonial ethnic identities. Indeed, as mentioned above, French colonialists did not always succeed in refashioning diverse ethnic identities into universal republican values (Müller-Crepon 2020). Given that direct rule limits the political representation of peripheral areas (Wucherpfennig, Hunziker, and Cederman 2016), perhaps aggrieved groups from these regions are now mobilizing around pre-colonial ethnicities rather than colonial-era identities. Similarly, the worldwide demise of communism and its attempt to build political movements around socio-economic inequalities may affect these outcomes too. Clearly more research is needed to explain the rise of ethnic-based violence in ex-French colonies.

My study has some methodological limitations. Foremost, it uses former colonial ruler as a proxy for the distinct colonial-era institutions and governing practices thought to influence civil violence in the contemporary era. By contrast, some researchers use finer-grained measures of colonial legacy (e.g., Lange, Jeong, and Amasyali 2021; Lange, Jeong, and Gaudreau 2022; Ziltener, Künzler, and Walter 2017). These measures, however, result in low sample sizes and limit comparisons to those between ex-British or ex-French colonies. Invariably, there are trade-offs between finer-grained and broader measures of colonial legacy. The benefit of my approach is that the requisite data are available, enabling me to use global rather than regional samples. Perhaps, in the future, researchers will extend their finer-grained measures to a global sample, allowing for broader studies such as mine.

About the author

Christopher Kollmeyer is a senior lecturer in sociology at the University of Aberdeen. His most recent publication, which examines whether economic globalization promotes civil peace in developing countries, is forthcoming in Cooperation and Conflict. His research on changes in the social structures and political economies of advanced capitalist countries appears in well-known journals such as American Journal of Sociology, European Sociological Review, and Social Forces.

Funding

This work was supported by the Leverhulme Trust (grant number RF-2021-530).

Conflicts of interest

None declared.

Data availability

The data and Stata codes needed to replicate the results of this study are available at Harvard Dataverse: https://doi.org/10.7910/DVN/URWOVE.

Endnotes

1

At the time of publication, the CSP/MEPV database covers 1946 to 2018. Many observations from the mid-1940s and 1950s are missing, including nearly all observations from African countries. Similarly, data for key control variables start in 1960 for prominent databases such as the Penn World Tables and the World Development Indicators. Combined, these data constraints result in the timeframe of 1960 to 2018.

2

Taking the log of Years of Independence is complicated by the nature of logarithms. The log of zero is undefined, and the log of one is zero. Thus, I lose one observation per country by logging this variable. Also, given the temporal boundaries of my sample, countries gaining independence after 1960 receive negative scores for this variable. In these cases, I recode the initial observation as zero.

3

Stata’s margins command allows users to make post-estimation predictions based on actual or counterfactual scenarios. Predictions use the regression coefficients from selected models along with the sample’s actual data. I use margins to predict the propensity for various types of civil violence when years independent (log) ranges from 0 to 5 for ex-British, ex-French, and ex-Spanish colonies. All predictions hold the other covariates constant at their sample means and include interaction effects between years independent (log) and former colonial ruler. Since years of independence is measured in logs, a value of 5 on this variable approximates 150 years of independence. This value corresponds to the high end of my sample (i.e., many ex-Spanish colonies have been independent for more than 150 years).

4

The x-axis in figure 1 is expressed in logged format because the relationship between years of independence and the likelihood of civil violence is non-linear. Since figure 1 is generated from regression models using years independent (log), the logged format for the x-axis must be retained.

5

I try a scale measure of democracy (ranging from −10 to 10) based on the same Polity Project data. I also try this scale measure along with its squared term to capture purported non-linear effects. Neither specification alters the main substantive findings.