Origins of Politicized Ethnic Differences

A large literature analyzes why ethnic differences sometimes create an important political cleavage, a precursor to understanding how ethnic differences can trigger violence.Footnote ²⁷ Recent research devotes considerable attention to horizontal political inequalities, meaning that some ethnic groups access power in the central government whereas others do not. This research primarily concentrates on three historical factors to explain inequalities in political power among ethnic groups: (1) historical warfare, (2) modernization and the spread of nationalism to the Third World, and (3) strategies of European colonial rule.Footnote ²⁸ First, the precolonial era in Africa predated the emergence of nationalist sentiments in the region. Historically, before nationalist ideas spread outside Europe, warfare was an important cause of group-level inequality and domination.Footnote ²⁹ Second, later in the precolonial era and into the colonial era, emergent nationalism hardened ethnic identities, facilitated interethnic inequalities, and made groups cognizant of grievances toward other groups. Whereas many premodern societies lacked the ability to translate macro-cleavages into political action, modernization and nationalism created politically relevant differences among ethnic groups.Footnote ³⁰ Most postcolonial countries in Africa combine many nations—specifically, ethnic nations—into a single political unit. These nascent distinctions engendered differential political power among groups. Third, colonizer identity as well as distance between the colonial center and an ethnic group's territorial settlement also influenced central power access.Footnote ³¹ Britain often granted political authority to indigenous political units whereas France sought more centralized colonial rule. Indirect British ruling strategies enabled groups located farther from the coast to access power at independence.

Strategic Causes of Coups and Ethnic Civil War

Building upon their historical discussion, Cederman, Gleditsch, and Buhaug emphasize that the postcolonial state does not provide a neutral arena for competing group interests, and ethnic groups instead view the state as a prize.Footnote ³² Only by entering and exerting influence in the political arena can ethnic groups articulate their social, cultural, and economic interests. This institutional logic generates their key hypothesis: if the incumbent regime blocks access for challenging groups, then the excluded groups face incentives to violently rebel.

Roessler builds on this theory by addressing a key question: why would a ruler exclude ethnic groups if this strategy raises the likelihood of a violent rebellion?Footnote ³³ Focusing on postcolonial Africa, he argues that weak political institutions undermined rulers' ability to commit to sharing power at the center with rivals, for example, inability to commit to letting a challenger retain lucrative cabinet positions. This commitment problem created incentives for rivals to use violence, which generated a tradeoff for governments between preventing coups and preventing civil wars. Given the more imminent threat posed by potentially treasonous insiders, many rulers responded by prioritizing co-ethnics in central government positions and by excluding members of other ethnic groups, triggering the incentives for outsiders to launch civil wars that Cederman, Gleditsch, and Buhaug discuss.Footnote ³⁴ Roessler posits that these leadership security fears often provoked conflict specifically along ethnic lines because ethnic identity provides an easy information shortcut in countries where actors believe ethnicity is politically salient. This logic yields his core hypothesis that ethnopolitical exclusion substitutes civil war risk for coup risk.Footnote ³⁵

However, this strategic rationale for ethnopolitical exclusion is not the only possible motivation for exclusion. Even when facing low coup risk, rulers may still exclude groups for a distinct opportunistic rationale.Footnote ³⁶ If a ruler believes that an ethnic group is unlikely to rebel if excluded, then regardless of its coup threat, the ruler may exclude to maximize control of rents and political power—implying that not all excluded groups pose a rebellion threat.

Summary of Logic for Coup/Civil War Tradeoff

Table 1 summarizes key aspects of this logic by considering a stylized interaction between a ruler and a rival that can be either coercively weak or coercively strong. It is most natural to conceive of these actors as members of different ethnic groups, although parts of the logic apply to intraethnic rivalries as well. If the ruler shares power, then the rival may be able to launch a coup. By contrast, if the ruler excludes, then the rival may be able to initiate a civil war.Footnote ³⁷ The government's ability to commit to any promises to share power and spoils affects the rival's motives to fight (rows in the table) and the rival's coercive strength affects its opportunity to fight (columns). The left column highlights that a coercively weak rival is unlikely to fight (either a coup or civil war) regardless of its ethnopolitical inclusion status or government commitment ability. Consequently, the ruler prefers to exclude weak rivals for opportunistic reasons—why would the government share rents with a rival that will not punish exclusion by rebelling?

Table 1. Coups, civil wars, and inclusion/exclusion

If instead the rival is strong (right column of Table 1), then government commitment ability determines the outcomes. High commitment ability—perhaps in the form of credible democratic power sharing or an institutionalized interethnic party—implies that the ruler can commit to deals such as lucrative cabinet positions for the rival (bottom row). The credibility of deals lowers the rival's coup incentives if included in power, but high coercive strength implies the rival would pose a rebellion threat if excluded. Constrained by this threat, the ruler prefers to share power.

But if the rival is strong and government commitment ability is low (top-right box in Table 1), then the government faces a coup/civil war tradeoff. Because the rival is strong, the ruler's inability to commit to deals implies high coup risk under power sharing, but ethnopolitical exclusion yields high civil war risk. Connecting low commitment ability to conflict is a foundational result in formal conflict bargaining models,Footnote ³⁸ and following Roessler's logic,Footnote ³⁹ the more imminent threat of a coup implies that the ruler prefers to exclude the rival for strategic reasons. This logic also follows from a large literature on the tradeoff that rulers face between loyalty and efficiency. Rulers often choose to coup-proof their militaries by promoting unqualified generals—perhaps consisting of co-ethnics and family members—and by hindering communication among the officer corps, even if these strategies diminish battlefield efficiency in war.Footnote ⁴⁰

Why Do Coups Occur?

Given the strategic logic summarized in Table 1, it is somewhat puzzling that coups would ever occur—given rulers' preference to exclude threatening groups. However, even without appealing to factors such as incomplete information or misperceptions, this behavior is less puzzling when considering constraints that rulers face to excluding rivals, highlighted by the “can exclude” and “cannot exclude” language in the top right box of the table. Two types of constraints are particularly relevant for the present substantive context. The first are historically rooted. Some African rulers inherited “split domination” regimes at independence in which one ethnic group dominated civilian political positions and a different group dominated the officer corps.Footnote ⁴¹ In these and related situations, the process of trying to exclude the rival may itself trigger a coup attempt.Footnote ⁴²

Second, the broader coup literature implies that general circumstances of political instability create permissive conditions for both interethnic and intraethnic coup attempts by constraining the ruler to rely on potentially disloyal—and therefore rival—generals. Strong rebellion threats by outsiders create a “guardianship dilemma” for rulers: the ruler needs to empower the military to defeat the threat, but greater dependence on the military raises its capacity to stage a coup.Footnote ⁴³ Ongoing war may improve opportunities for a coup by “expand[ing] the influence of the military,”Footnote ⁴⁴ and increase motives for intervention by creating severe costs for members of the military and by increasing uncertainty about who will control the state in the future.Footnote ⁴⁵ Even without an ongoing war, a general perception of instability raises the stakes of the political game and can trigger coups by generals who perceive a greater threat from outgroups than does the incumbent ruler.

Amid such political instability, ethnic identity may no longer serve as a reliable marker for loyalty or disloyalty because even the rulers' co-ethnics face heightened coup incentives. Because no dictator can govern alone, rulers cannot purge all potential rivals, which can result in “reshuffling coups” by co-ethnics as distinct from “regime change” coups by members of other ethnic groups.Footnote ⁴⁶

Pre-Colonial Statehood Acronyms

The remainder of the article uses the following acronyms to distinguish ethnic groups. Countries that contain at least one PCS group are “PCS countries” and those that do not are “non-PCS countries.” PCS countries contain precolonial state ethnic groups (“PCS groups”) and stateless groups (“SLPCS groups”). All ethnic groups in non-PCS countries are stateless (“SL groups”). Figure 1 illustrates the abbreviations. The Baganda in Uganda were organized under the Buganda kingdom before colonization and are a PCS group (black). Therefore, any stateless ethnic groups in Uganda, such as northern groups, are SLPCS (light gray), and Uganda is a PCS country. By contrast, there are no PCS groups in Kenya, and therefore all its groups—such as the Kikuyu—are SL (dark gray), and Kenya is a non-PCS country.

Figure 1. Acronym examples

Historical Background on Precolonial States

Two Foundational Assumptions

The historical background motivates the empirical applicability of two foundational assumptions about precolonial states used to derive the hypotheses.

Hypotheses: Combining the Strategic Logic and Historical Facts

Combining the historically based assumptions with the logic of the coup/civil war tradeoff proposed in Roessler and related research yields the hypotheses,Footnote ⁶⁶ and Appendix section B.4 presents a simple game that formalizes the logic. Assumption 1 implies that any ethnic group in a PCS country should engage in political violence—either civil wars or coups—more frequently than a group in a non-PCS country because governments in PCS countries should have low commitment ability (top row in Table 1). Assumption 2 explains the form that violence should take within PCS countries, differentiating PCS and SLPCS groups. PCS groups' historical privileges imply that, when they engage in violence, they should frequently use the insider technology of coups d'état (and, related, they should also enjoy a favorable position to succeed at coups). By contrast, SLPCS groups' lack of such privileges should make their members easier to dislodge from power—given strategic incentives for exclusion in PCS countries—which should associate SLPCS groups with civil wars.

Crucially, the posited within-country spillover effects of PCS groups—raising conflict propensity for all groups within their country—implies that the relevant comparison for PCS groups and for SLPCS groups is to SL groups. Therefore, theoretical considerations require disaggregating stateless groups by whether or not any PCS groups reside in their country. If the theory is correct, then comparing PCS groups to SLPCS groups will yield the incorrect implication that precolonial statehood exerts pacifying effects with regard to civil wars by ignoring within-country spillover effects. This logic yields the two main hypotheses:

The theoretical building blocks for the two main hypotheses yield secondary hypotheses. SLPCS groups should face two disadvantages for accessing power at the center relative to SL groups. First, low government commitment ability in PCS countries creates strategic exclusion incentives (assumption 1). Second, SLPCS groups tended to be disadvantaged in the colonial hierarchy, which restricted their access to power (assumption 2).

Comparing PCS groups to SL groups does not not yield a corollary to H3. Combining the strategic exclusion incentives in PCS countries (assumption 1) with PCS groups' historical advantages (assumption 2) yields ambiguous expectations for whether PCS or SL groups should more frequently access power at the center.

Appendix section B.3 presents additional hypotheses that predict coups and civil wars conditional on ethnopolitical representation status, explaining why both PCS groups and SLPCS groups should more frequently initiate civil wars than SL groups conditional on exclusion (H4 and H5), and why PCS groups should more frequently launch coups than SL groups conditional on inclusion (H6).

Sample

The unit of analysis in the main regression specifications is ethnic group-years. The sample includes ethnic groups from most countries in Sub-Saharan Africa between independence and 2013. I draw ethnic groups from the Ethnic Power Relations database (EPR),Footnote ⁶⁷ which provides panel data on politically relevant ethnic groups and their access to power in the central government.Footnote ⁶⁸ Appendix section C.1 lists African countries missing EPR data, most of which are small countries that do not meet key scope conditions of the theory by not having an indigenous population or by lacking multiple ethnic groups that can compete for power.

Robustness checks analyze two theoretically relevant subsamples: ex-British colonies, and the Cold War era. The posited colonial mechanisms are most relevant for ex-British colonies because Britain was more likely to rule indirectly through PCS groups' established political hierarchies. Furthermore, the focus on events leading up to independence suggests that the mechanisms should most likely operate earlier in the postcolonial era. Other appendix tables stratify on ethnopolitical inclusion status.

Precolonial States

Defining states has posed difficulties for social scientists, including anthropologists who have long debated how to classify states in precolonial Africa.Footnote ⁶⁹ Even precolonial African polities that exhibited hierarchical organization pose classification challenges when considering “stateness” traits used by social scientists to explain historical state formation in other regions. African rulers and the African state system as a whole placed lower emphasis on territorial sovereignty than did early modern Western Europe.Footnote ⁷⁰ Even highly centralized states by African standards usually possessed rudimentary political institutions compared to contemporary Asian agrarian empires.Footnote ⁷¹

Coding PCS Groups

Acknowledging these difficulties, I coded an EPR ethnic group as having a precolonial state if co-ethnics governed a substantial percentage of members of the EPR ethnic group through a single or small number of political organizations that exhibited some degree of centralized rule on the eve of colonization. Although the operational definition of a state is minimal, the paucity of reliable historical information for most precolonial African political organizations makes it difficult or perhaps impossible to operationalize a conceptual definition that requires more nuanced information about the degree or origins of centralization for each group. I restrict attention to states that existed at the eve of colonization because the theory posits that precolonial states' influence on colonial policies is a key persistence mechanism.Footnote ⁷²

Constructing the data set proceeded in three main steps that Appendix A details along with country-by-country coding justifications. First, I consulted eleven continent-wide maps of historical states and other sources to generate a list of candidate states. Second, I used additional secondary sources to match candidate states with EPR ethnic groups to generate a list of candidate PCS groups. Third, I examined four conditions that correspond with the operational definition of precolonial states to determine which candidate groups to code as PCS: (1) co-ethnic governance requires the state was independent rather than tributary; (2) some evidence of central authority such as acknowledged hierarchy of authority in regions outside the capital and centralized tax collection, as opposed to nomadic confederations or trading centers; (3) one (or a small number of) states governed a substantial percentage of members of the EPR ethnic group, as opposed to groups such as Yoruba in Nigeria or Bamileke in Cameroon that were fractured into dozens or hundreds of mini-states; and (4) these conditions held on the eve of colonization. In total, I consulted over one hundred secondary sources to code the PCS variable, which Figure 2 depicts. It uses the same color scheme as Figure 1: PCS groups in black, SLPCS groups in light gray, and SL groups in dark gray. Table A.2 lists every PCS group and PCS country.

Figure 2. PCS groups in Sub-Saharan Africa

Robustness checks evaluate alternative PCS measures that narrow the statehood definition. First, the main measure codes groups like the Tswana in Botswana that exhibited characteristics of centralized rule but were not organized into a single state as PCS if a small number of states governed members of the modern ethnic group, especially if one state was clearly ascendant. The first alternative PCS measure codes groups like the Tswana as non-PCS by changing the third operational criterion. Table A.2 denotes these groups with an asterisk. Second, the eve-of-colonization criterion is ambiguous for territories that experienced early major colonial interference (pre-1870), such as Egypt's takeover of central Sudan, Portuguese penetration of the Kasanje and Matamba states in Angola, and France's incursion into Senegambia. Table A.2 denotes states that experienced early interference with a dagger, which the second alternative PCS measure codes as non-PCS by changing the fourth coding criterion (also see the detailed coding rules in Appendix A).

Two reasons motivate coding a binary variable rather than an ordinal or continuous PCS measure. First, conditional on a PCS group containing enough members to potentially disrupt creating nationally oriented political organizations,Footnote ⁷³ the theory does not address PCS groups' size. Perhaps larger PCS groups monotonically imply less ability to cooperate, or perhaps there is a nonmonotonic effect in which very large PCS groups achieve sufficient support to mitigate the posited sources of bargaining failure. The mostly peaceful PCS countries Botswana, Guinea, Senegal, and Zambia (see the qualitative evidence section) each feature a PCS group composing at least 40 percent of its country's total population. Although intriguing to consider in future research, I prefer a simpler measure that directly assesses the present hypotheses. Second, at the country level, having at least one PCS group or not is the key theoretical distinction. Conditional on having at least one, the number of PCS groups should be unimportant because even one PCS group should create conditions for ethnic violence. The theory does not suggest that multiple PCS groups would necessarily cause more violence.

Dependent Variables

Most regression tables examine two civil war measures. I coded ethnic group-level civil war onset by assigning Fearon and Laitin's major civil wars (at least 1,000 battle deaths during the conflict) to EPR ethnic groups primarily using ACD2EPR, which codes ethnic wars as involving ethnic-specific recruitment and war aims.Footnote ⁸³ Appendix section C.2 elaborates upon the coding procedure. I additionally examine Roessler's and Roessler and Ohls's civil war onset variable,Footnote ⁸⁴ which also measures group-level major conflicts but does not require ethnic-specific rebellion aims (hence, “ethnic participation”). Robustness checks examine ACD2EPR's measure, which uses a lower death threshold.Footnote ⁸⁵

The regression tables also analyze successful and attempted coups d'etat.Footnote ⁸⁶ A robustness check examines participation in irregular interethnic regime changes. Successful coups are appropriate to analyze because the theory anticipates that PCS groups' greater access to power at the center should enable succeeding at coups in addition to launching them. Successful coups can also be measured with less error than failed coups, which Appendix section C.3 discusses.

Origins of Precolonial Statehood Covariates

No process randomly assigned precolonial states. Although no silver-bullet research design can completely solve this fundamental causal inference concern, the statistical models control for posited causes of state formation in precolonial Africa and for alternative explanations of civil wars and coups, in addition to numerous robustness checks. Appendix section C.4 details every covariate.

A growing social scientific literature examines causes of state formation in precolonial Africa. Much of this research applies factors proposed in research on European state formation—population density, trade, and warfare—and analyzes their applicability to Africa.Footnote ⁸⁷ Studies using ethnic group-level data have shown that precolonial states were more likely to emerge in territories exhibiting greater ecological diversity, which promoted trade,Footnote ⁸⁸ and with lower tsetse fly prevalence, which facilitated higher population density.Footnote ⁸⁹ At the country level, Nunn demonstrates a negative correlation between slave exports and political centralization.Footnote ⁹⁰ Others examine long-term effects of historical warfare in Africa, albeit without directly analyzing effects on precolonial state formation.Footnote ⁹¹ In a broader sample, Putterman shows a strong relationship between timing of a territory's neolithic revolution and historical state development.Footnote ⁹² Each regression table accounts for these possible determinants of precolonial statehood by including specifications with and without the five precolonial covariates: ecological diversity, tsetse fly, slave exports, historical warfare, and neolithic timing.

Standard Conflict Covariates

Each table also evaluates specifications that include the following seven covariates, which are commonly examined in studies of civil war and coups: at the country level, income per capita, population, democracy level, and geographic constraints to broadcasting power; and at the ethnic group level, share of the national population, distance from the capital city, and presence of a giant oil field. Although some state formation and standard conflict covariates raise posttreatment bias issues, each table includes specifications with and without different combinations of covariates to show the results do not hinge on a single set of covariates.

Statistical Models

The main regression tables estimate logistic regressions:

(1)$${\rm ln}\left( {\displaystyle{{Y_{it}} \over {1{\rm -}Y_{it}}}} \right) = \beta _0 + \beta _P{\rm \cdot} P_i + \beta _S{\rm \cdot} S_i + {\bi X}_{it}^{\rm ^{\prime}} \beta _X + {\bi T}_{it}^{\rm ^{\prime}} \beta _T + \varepsilon _{it},$$

where Y _it is an indicator variable for ethnic civil war onset (with years of ongoing civil war dropped) or successful coup attempt, P _i indicates PCS groups, S _i indicates SLPCS groups, β _P and β _S are the main parameters of interest, and X_it is a vector of covariates that differs by column. The vector T_it contains standard event history controls for civil wars or coups—years since the last with civil war incidence or years since last coup, and cubic splines—plus lagged country-level civil war incidence in the civil war regressions. I cluster the standard errors at the ethnic group level.

Importantly, the main specifications do not include country fixed effects. A key premise of the theory is that PCS groups caused within-country spillover effects that should raise violence propensity for all groups in PCS countries, which necessitates comparing groups in PCS countries to stateless groups in non-PCS countries (SL groups), as the hypotheses state. The main civil war hypothesis (H1), for example, concerns stateless groups in PCS countries because PCS groups tended to exclude SLPCS groups from power. Therefore, within-country comparisons generated by modeling country fixed effects would not provide a valid test of the main hypotheses. This distinguishes the present analysis from existing statistical studies on precolonial statehood and civil war that use a subnational unit of analysis and include country fixed effects in every regression model.Footnote ⁹³

Main Results

Tables 2 and 3 support the main hypotheses. Every column contains PCS and SLPCS indicators—leaving SL groups as the omitted basis category—and event history controls. Table 2 assesses H1 by examining ethnic civil war onset in columns 1 through 4 and onset of ethnic group participation in civil war (the Roessler measure) in columns 5 through 8.Footnote ⁹⁴ Table 3 assesses H2 by examining successful coup attempts in columns 1 through 4 and all coup attempts in columns 5 through 8. The first column for each dependent variable (1 and 5) does not add additional covariates.

Table 2. Main civil war results

Notes: Logistic regression estimates with ethnic group-clustered standard errors in parentheses. * p < .1; ** p < .05; *** p < .01.

Table 3. Main coup results

Notes: Logistic regression estimates with ethnic group-clustered standard errors in parentheses. * p < .1; ** p < .05; *** p < .01.

The estimates from column 1 of Table 2 show that SLPCS groups participated in civil wars 4.9 times more frequently than SL groups, in 0.70 percent of ethnic group years compared to 0.14 percent. PCS groups participated in civil wars 2.3 times more frequently than SL groups, in 0.32 percent of ethnic group years. Analyzing raw trends in the data demonstrates that, quite simply, SL groups have rarely initiated civil wars. The first major ethnic civil war in a non-PCS country occurred in 1989 in Liberia. In total, only six major ethnic group-level onsets occurred in non-PCS countries between independence and 2013. This compares to forty-nine group-level onsets in PCS countries, including thirty through 1989. The relative onset frequencies are similar for the Roessler civil war variable in Column 5: 1.16 percent for SLPCS groups, 0.86 percent for PCS groups, and 0.34 percent for SL groups.

The estimates from column 1 of Table 3 show that PCS groups participated in successful coups 2.2 times more frequently than SL groups, in 1.56 percent of ethnic group years compared to 0.70 percent. Column 5 shows respective frequencies of 2.41 percent and 1.28 percent for coup attempts. By contrast, SLPCS groups participated in both successful and all coup attempts less frequently than SL groups, which corresponds with SLPCS groups' infrequent access to political power (see Table 6, presented later). Although PCS groups and SL groups participated in roughly the same number of successful coups between independence and 2013—thirty-one versus twenty-nine, respectively—the percentage difference arises because PCS groups infrequently appear in the sample relative to SL groups (see Table C.2).

Adding different sets of control variables does not change the main findings in Tables 2 and 3. Columns 2 and 6 add to the baseline specification the five predictors of African state formation discussed earlier, columns 3 and 7 add to the baseline specification the seven standard conflict covariates, and columns 4 and 8 include all controls. The relationships between SLPCS groups and civil wars, and between PCS groups and coups, are statistically significant at 5 percent in most specifications, although the p-value for PCS groups rises above 0.05 in two coup attempt regressions.

Intriguingly, although the coefficient estimate for PCS groups is positive in every civil war specification, it is statistically significant only when analyzing ethnic participation in war (columns 5 through 8) rather than rebellions with ethnic aims (columns 1 through 4). Although only a small number of war onsets differentiate the two measures, cases such as Baganda in Uganda (two war onsets in the 1980s, but neither proclaimed ethnic-specific aims) suggest that after being forced from power, PCS groups might become willing to forge broader coalitions to regain power. Overall, however, the theoretical expectations for PCS groups and civil wars are ambiguous. Although PCS groups should rebel frequently when excluded (see Appendix section B.3 for H4), their historical advantages for gaining access to power (assumption 2) should diminish reliance on outsider rebellions, which is why the theory does not produce an analog of H1 for PCS groups.

Regarding other covariates, democracy and GDP per capita enter negatively, and statistically significantly in many specifications, consistent with existing arguments. Group percentage of the population positively covaries with coups, which is sensible because group size and political inclusion positively correlate (see Table 6). Finally, a longer period since a country transitioned to agricultural production positively covaries with civil wars and coups, which may be intriguing to analyze in future research because the typical assumption in related research is that neolithic transition timing influenced modern outcomes by affecting precolonial statehood.Footnote ⁹⁵

Additional Robustness Checks

Appendix section D.1 conducts numerous robustness checks including jackknife sample sensitivity analysis, estimating the bias of selection on unobservables from selection on observables, changing the dependent variables and PCS measures, and changing the statistical models. Appendix section D.2 addresses debates about border formation and “dismembered” ethnic groups in Africa by showing similar results even when limiting the sample to ethnic groups partitioned across international boundaries and modeling transnational ethnic group fixed effects.Footnote ⁹⁶ Appendix section D.3 disaggregates two theoretically relevant subsamples, showing the main results are particularly strong among ex-British colonies and during the Cold War era.

Ethnic Parties at Independence

A key posited historical legacy is that PCS groups promoted ethnic parties at the expense of interethnic and nationalist-oriented political parties, which motivates assumption 1. Tables 4 and 5 statistically support this argument. In Table 4, the unit of analysis is country and the dependent variable is ethnic parties' vote share in the final decolonization election.Footnote ⁹⁷ Harkness codes this variable using Chandra's coding rules for ethnic parties.Footnote ⁹⁸ The main explanatory variable is an indicator for PCS countries. Column 1 runs a bivariate regression, and column 2 adds every country-level covariate from Tables 2 and 3 plus an indicator for British colonialism. In both columns, the coefficient estimate for PCS country is positive and statistically significant. The coefficient estimates are also large in magnitude: ethnic parties received 52 percent of the total vote share in the final pre-independence elections in PCS countries, compared to 16 percent in non-PCS countries. The PCS country indicator also nearly perfectly predicts countries in which an ethnic party governed at independence: the only counterexample is non-PCS Sierra Leone, compared to eight PCS countries (Angola, Benin, Chad, Nigeria, Rwanda, Sudan, Uganda, and Zimbabwe).

Table 4. Ethnic parties during decolonization: country-level evidence

Notes: OLS regressions with robust standard errors in parentheses, using a cross-section of countries with available ethnic vote share data. * p < .1; ** p < .05; *** p < .01.

Table 5. Ethnic parties during decolonization: ethnic group-level evidence

Notes: Logistic regression estimates with robust standard errors in parentheses, using a cross-section of ethnic groups with available ethnic party data. The covariates in different columns correspond with Tables 2 and 3. The sample in panel B includes only ethnic groups that compose at least 10 percent of their country's population. * p < .1; ** p < .05; *** p < .01.

Table 5 changes the unit of analysis to ethnic groups. For each group, the dependent variable indicates representation by an ethnic party. Operationally, this requires an ethnic party that represents the group to have received a positive percentage of votes in the final pre-independence election.Footnote ⁹⁹ Panel A includes all politically relevant ethnic groups in each country's first year of independence (see Tables D.6 and D.7), except groups from the four countries that Table 4 excludes, and panel B includes only groups that compose at least 10 percent of the country's population. PCS groups' distinctiveness is stark: in column 1 of panel A, the predicted probability of an ethnic party is 52 percent for PCS groups, 22 percent for SLPCS groups, and 13 percent for SL groups. Stateless groups are somewhat more heterogenous than PCS groups in terms of their size and political clout, which perhaps explains the null coefficient estimate for SLPCS groups in panel A. However, when stratifying on larger ethnic groups that could more reasonably have sought to exert considerable influence in the political arena (panel B), SLPCS groups exhibit statistically distinguishable behavior from SL groups—consistent with the argument that actions by PCS groups to undermine nationalist-oriented parties triggered responses by other groups within their country. In column 1 of panel B, the predicted probability of an ethnic party is 59 percent for PCS groups, 48 percent for SLPCS groups, and 18 percent for SL groups. Related, in ten of thirteen PCS countries in which a PCS group formed an ethnic party (out of eighteen PCS countries in the sample), at least one SLPCS group also formed an ethnic party.

Predicting or Conditioning on Ethnopolitical Inclusion

Statistical evidence supports the secondary hypotheses 3 through 6, which either predict or condition on ethnopolitical inclusion. These results relate to three findings from the literature. First, few studies endogenize ethnopolitical representation status. Wucherpfennig, Hunziker, and Cederman examine distance from the coast and colonizer identity,Footnote ¹⁰⁰ and my findings contribute to this small but important research agenda. Second, whereas Roessler shows that ethnopolitical exclusion covaries positively with civil war onset and negatively with successful coups,Footnote ¹⁰¹ my findings show that precolonial statehood explains variance in ethnic violence even when stratifying on access to central power. Third, the results yield an opposing finding from Wig by showing that, conditional on exclusion, PCS groups rebel more often than SL groups.Footnote ¹⁰² Earlier I discuss that these differences arise because (1) I do not compare PCS groups to stateless groups in their own country and (2) the present PCS variable exhibits less measurement error than combining Murdock's jurisdictional hierarchy variable with EPR data.Footnote ¹⁰³

Table 6 analyzes ethnopolitical inclusion (H3) using EPR data.Footnote ¹⁰⁴ A group-year is coded as 1 on the ethnopolitical inclusion variable if it scores “monopoly,” “dominant,” “senior partner,” or “junior partner” on EPR's political status variable—indicating that the group commands either the presidency or at least some cabinet positions in the administration—and 0 otherwise.Footnote ¹⁰⁵ Using the same covariate groupings as in Tables 2 and 3, the four columns in Table 6 support H3 by demonstrating the rarity of ethnopolitical inclusion for SLPCS groups relative to SL groups. In column 1, the predicted probability for inclusion is 41 percent for a SLPCS group and 70 percent for a SL group (and 76% for PCS groups). The difference between SLPCS and SL groups is statistically significant in all specifications.

Table 6. Ethnopolitical inclusion

Notes: Logistic regression estimates with ethnic group-clustered standard errors in parentheses. No specifications contain event history controls because the logit models do not converge, although unreported linear models with years since last change from inclusion to exclusion or vice versa (akin to civil war onset and coup onset) and cubic splines yields qualitatively identical results. * p < .1; ** p < .05; *** p < .01.

In appendix section D.4, I assess the three conditional civil war and coup hypotheses derived in section B.3. I provide evidence that both PCS and SLPCS groups rebel more frequently than SL groups conditional on exclusion, and PCS groups are more likely than SL groups to participate in coups conditional on inclusion. Although the primary purpose of these regressions is to assess the conditional hypotheses, they also rule out a particular confounding concern for Tables 2 and 3. Even if unmodeled factors affected PCS groups' coup behavior by raising their propensity for ethnopolitical inclusion (independent of affecting PCS status), and affected SLPCS groups' civil war behavior by lowering their propensity for inclusion (independent of affecting SLPCS status), these results show similar patterns even when conditioning on ethnopolitical access.Footnote ¹⁰⁶

Modal Violence Paths in PCS Countries

All but one PCS country fits into one of four modal paths based on ethnopolitical inclusion patterns and ethnic violence. The present analysis focuses only on events up to 1989 under the premise, discussed in Appendix section D.3, that changes following the Cold War opened new political cleavages in Africa.

Table 7. Summary of cases

Note: Author's calculations.

Among cases in the first path, a PCS group dominated the government at or shortly after independence. Specifically, a PCS group had achieved either “dominant” or “monopoly” status at the center within five years of independence (according to EPR) in Angola, Burundi, Ethiopia, Mali, Nigeria, and Sudan. The PCS group also either dominated or was heavily overrepresented in the officer corps of each country except Nigeria (see Table B.5). In five of the six countries, at least one excluded SLPCS group initiated a civil war by 1975, and the same occurred in the sixth, Mali, in 1989. Additionally, in every case except Angola, the dominant PCS group staged a successful coup. Sometimes, coups struck against rival ethnic groups, as in Nigeria in 1966. In other cases, such as Sudan, the coups rotated power among members of the PCS group. The theory section discussed how riverine Arabs in Sudan launched numerous coups in reaction to the incumbent's perceived poor handling of ongoing civil wars with non-PCS groups in the south.

For example, in Nigeria, a PCS group controlled power at the center, retained power via a military coup, and an excluded SLPCS group launched a civil war. Nigeria's federal formula—a legacy of Frederick Lugard's invention of the Native Authority System for the Sokoto Caliphate (precolonial state) in the north—dictated that each of the country's three regions would be apportioned seats in the national legislature based on population share. As a result, the constitutionally mandated dicennial census in 1962 carried huge consequences for the distribution of power, especially considering the distinct ethnic parties that represented each region. “The Northern Region's political strength, marshaled by the NPC, had arisen in large measure from the results of the 1952–53 census, which had identified 54 percent of the country's population in that area.”Footnote ¹⁰⁷

Despite conducting the census twice, experts estimated the total count for northerners was wildly inflated and Igbo (SLPCS group) leaders publicly charged the northern (PCS) government with fraud. Igbo officers led a successful coup attempt in 1966. Despite stating that they aimed to create a unitary government without ethnic bias, northern leaders perceived the coup “not so much as an effort to impose a unitary government as a plot by the Igbo to dominate Nigeria” because of deep-seated regional cleavages.Footnote ¹⁰⁸ This led to a northern-dominated countercoup in 1966, followed by ethnopolitical exclusion of Igbo and an Igbo secession attempt in 1967.

In the second path, an SLPCS group dominated the government at or shortly after independence. In all four cases, an excluded PCS group participated in a civil war either shortly after independence (Chad, Rwanda, Zimbabwe) or after further ethnic narrowing at the center (Uganda). The theory correctly anticipates political violence—specifically, PCS groups often rebel against the government when excluded—although these cases deviate from the trend of PCS groups controlling the government.

The third path features ethnopolitical inclusion and infighting at the center. The theory correctly anticipates the coups that occurred in these PCS countries (Benin, Ghana, Madagascar), although without amendment cannot explain why their leaders usually did not resort to ethnopolitical exclusion to prevent future coups.Footnote ¹⁰⁹ PCS groups either participated in or were central to the violence that occurred. In Benin, two of the three groups that rotated into and out of power via a series of coups in the 1960s were PCS: Fon, and Yoruba/Nagot and Goun. In Ghana, the leader of the country's first coup in 1966 explicitly denounced president Kwame Nkrumah's attempts to undermine traditional Asante (PCS) organizations—a rivalry that began during the colonial era after Nkrumah attempted to undermine Asante planter-chiefs. EPR codes the PCS group in Madagascar, Highlanders, as excluded. However, they retained a strong presence in the military, which EPR does not reflect by primarily using information on executive and cabinet positions to code political status. Highlanders launched a successful coup in 1972.

The Democratic Republic of the Congo is an idiosyncratic case that does not fit neatly into any of these three modal patterns of ethnic violence, although broadly supports the theory because two PCS groups initiated civil wars. The country, in essence, had no central government during its first five years of independence. Correspondingly, EPR codes all politically relevant ethnic groups in DRC as enjoying regional autonomy between 1960 and 1965. Within the first year of independence, PCS groups Luba Kasai and Lunda-Yeke initiated civil wars to secede, the latter led by their king.

Of the eighteen PCS countries, only four are theoretical anomalies by exhibiting relative peace. Botswana, Guinea, Senegal, and Zambia provide a fourth modal path by experiencing long periods of stable civilian rule in ethnically inclusionary regimes, although in the 1980s a successful coup occurred in Guinea and a civil war began in Senegal. PCS groups such as the Barotse in Zambia and Peul in Guinea created regional parties during the decolonization era to contest nationalists, but ultimately failed to prevent a dominant nationalist party from taking power at independence.

Evidence from Uganda

Uganda exemplifies PCS groups' privileges and the difficulty of forming broad nationalist parties during the decolonization era in colonies with a PCS group, and consequences for political violence after independence. When Britain colonized Uganda, it bestowed the powerful state of Buganda—which frequently warred with and raided slaves from neighboring groups also incorporated into modern UgandaFootnote ¹¹⁰—with significant self-governance privileges. “The special status of Buganda in Uganda was the most important legacy of the colonial era” and their founding treaty with Britain in 1900 “appeared to the Baganda as in some sense at least an agreement between equals.”Footnote ¹¹¹ Therefore, Buganda's later “integration within the rest of Uganda posed serious problems first to colonial officials and subsequently to nationally oriented African politicians [because] Buganda could not be dethroned from its dominant position without seriously compromising the viability of Uganda as a whole.”Footnote ¹¹² In response to Britain's attempt to unify colonial administration after World War II, Buganda attempted to secede from the rest of Uganda to “safeguard the traditions, Kabakaship, and the customs of Buganda in an independent Uganda.”Footnote ¹¹³ The king—known as the kabaka—cited Buganda's distinct status in the Uganda Agreement of 1900 to promote his claim.

Consequently, “the power of traditional groups … precluded the success of a centralized, ideological mass party” among all Ugandans.Footnote ¹¹⁴ Supporters of the kabaka instead created the Kabaka Yekka party—meaning “king only”—after the kabaka led a highly successful boycott of the 1961 Legislative Council elections in which less than 2 percent of eligible Baganda voted. Kabaka Yekka provided “a practical avenue through which Buganda could enter national politics and yet preserve its own autonomy and unity.”Footnote ¹¹⁵ The ethnically oriented party received 26 percent of parliamentary seats in the final pre-independence elections in 1962.Footnote ¹¹⁶ In response to political deadlock created by the sizable vote share of Buganda's ethnically oriented party, an interethnic ruling coalition formed at independence that composed an “alliance of complete opposites”Footnote ¹¹⁷ between Kabaka Yekka and a major party led by a member of an SLPCS group, Milton Obote's UPC party. Consistent with the theoretical expectation that such circumstances facilitate an internal security dilemma, “it is hard to determine at what stage Prime Minister Obote made up his mind to confront the Kabaka and the State of Buganda … but it is tempting, from the small amount of evidence available and his careful preparing of the ground, to think that he had intended it all along.”Footnote ¹¹⁸

A key event occurred in 1964 that foreshadowed future violence. To provide background, Buganda's initial treaty with Britain awarded territory, subsequently called the “Lost Counties,” to Buganda that a British-Baganda alliance had recently captured from rival precolonial state Bunyoro (also incorporated into Uganda). In 1964, Obote attempted to undermine Buganda by allowing a referendum to occur over control of the Lost Counties. After the counties' residents voted to join Bunyoro, a political crisis occurred. “As President, the Kabaka should have ratified the transfer; as Kabaka, such an act was an impossibility” [emphasis in original] because the transfer would erode the king's support from his main constituency.Footnote ¹¹⁹ The Lost Counties referendum soon “shifted the balance of power back from Buganda to the central government.”Footnote ¹²⁰ Capitalizing on his improved bargaining position, Obote arrested key Baganda ministers before they could appeal for British military assistance. The crisis culminated in 1966 when Obote militarily suppressed a Baganda secession attempt, staged a coup to become the undisputed head of state, and unilaterally terminated the Baganda monarchy.Footnote ¹²¹ After continued turmoil at the center including Idi Amin's kleptocratic reign in the 1970s, the then-politically powerless Baganda participated in civil wars that removed Amin and, subsequently, Obote after he became president for a second time.