Notes on Ethnic and Nonethnic Civil Wars: Part 1

OVERVIEW

• This article (Sambanis, 2001) investigates whether ethnic/religious (identity) civil wars have the same causes as nonidentity (revolutionary and other) civil wars.
• Core claim: identity wars are driven more by political grievance and institutions than by purely economic conditions, and they show different relationships with ethnic heterogeneity than nonidentity wars.
• Uses a new dataset: 161 countries observed annually from 1960–1999 (40 years) to compare onset of ethnic wars with revolutionary/other wars, and to integrate theories from economics and international relations (neorealism and neoliberalism).
• Finds evidence of systemic (international) effects on ethnic war onset (e.g., neighbors’ war status, neighborhood democracy), and emphasizes the importance of political variables (democracy) and ethnic fragmentation for identity wars.
• Contributes by disaggregating civil wars by type, showing differences in determinants, and highlighting policy implications for prevention and termination that differ by war type.

DEFINING AND CODING ETHNIC VS NONETHNIC WARS

• Ethnicity definition (Horowitz-inspired): ethnicity = an inclusive, ascriptive difference in color, language, religion, etc., representing groups with a sense of common origin and identity. Ethnic wars involve conflicts over the power relationship between ethnic communities and the state.
• Ethnic war vs other wars: not all wars involving ethnic groups are ethnic wars; core issues must be integral to the concept of ethnicity (identity, borders, citizenship, autonomy).
• Coding source and rules:
  • Relies on the State Failure Project for coding ethnic wars: episodes of violent conflict between governments and national/ethnic/religious minorities seeking major changes in status; rioting/competition within groups not coded as ethnic war unless tied to political power or policy.
  • Civil war coding (to ensure comparability across datasets): combines data from Correlates of War (Singer & Small), Wallensteen & Sollenberg, and the State Failure Project.
  • Civil war onset defined by: more than 1,000 war-related deaths in the entire war and in at least one year; challenge to sovereignty of an internationally recognized state; within-state conflict; principal combatants; rebels able to mount organized military opposition.
• Important note on robustness and classification: three extensions planned include robustness tests, prevalence estimation, and post-World War II classification criteria; this article focuses on war onset with the above coding.

INTEGRATED THEORY OF ETHNIC CIVIL WAR

• Theoretical landscape: many IR and economic theories explain civil war broadly, but not whether causes differ by type (identity vs nonidentity).
• Primordial, ethnic-clathic, and cultural theories: various explanations include old enmities, ethnolinguistic ties, cultural divides, ethnic security dilemmas, modernization effects, and mobilization by ethnic entrepreneurs.
• International relations theories: neorealism highlights system-level anarchy and external power dynamics but struggles to explain ethnic division within a state; neoliberalism emphasizes domestic institutions, legitimacy, nonstate ethnic networks, and economic motives.
• Economic theories (Collier & Hoeffler; Fearon & Laitin): war as a rational choice based on expected gains from rebellion, influenced by grievances, opportunity costs of rebellion, and rebel financing; many studies aggregate all civil wars, which may mask type-specific differences.
• Sambanis’ synthesis: combine ethnic-conflict theories with IR theories and economic theory to produce a parsimonious model that includes system-level variables (Cold War, diffusion/contagion, neighboring wars) and domestic determinants (democracy, development, ethnic heterogeneity, war history).
• Key claims from integrated theory:
  • Systemic factors can shape ethnic war occurrence (e.g., neighboring states’ conditions affect risk).
  • Democracy and development are important but may affect ethnic wars differently than nonidentity wars.
  • Ethnic heterogeneity has a positive association with ethnic war onset, potentially more robust than in nonethnic wars.

HYPOTHESES TESTED

• Hypothesis 1: The probability of ethnic war should be an increasing function of ethnic heterogeneity (EHET). Rationale: in ethnically diverse societies, coordination costs and cross-group coordination barriers can differ; for ethnic wars, a single-ethnic-rebellion dynamic may lower coordination costs, yielding a positive relation with EHET.
  • Competing findings from other work (e.g., Collier & Hoeffler’s ELF results; Fearon & Laitin) are discussed and contrasted because they aggregate all civil wars.
• Hypothesis 2: Greater economic development reduces ethnic war risk by increasing opportunity costs of violence; higher development lowers the incentive to rebel.
• Hypothesis 3: Lack of democracy (low political rights) significantly raises ethnic war risk because democracy protects ethnic/political rights and reduces grievances being translated into violence.
• Hypothesis 4: Democracy in neighboring countries lowers ethnic war risk in a country (good neighborhood effect); more democratic neighbors deter spillovers and contagion.
• Hypothesis 5: The presence of war in a neighboring country increases the likelihood of ethnic war in the country under study (contagion/diffusion effects).
• Hypothesis 6: The Cold War should be negatively correlated with ethnic war onset (end of Cold War changes system-level dynamics that previously contained ethnic conflicts), yet time trends (development and democratization) may counterbalance this effect.

DATA SET, VARIABLES, AND METHODS

• Data: annual observations for 161 countries from 1960 to 1999.
• Dependent variables:
  • EWARST: onset of ethnic war (1 if onset; 0 otherwise; missing for ongoing wars after first observation).
  • RWARST: onset of revolutionary/other (nonethnic) wars (same coding as ethnic onset).
• Core proxies and variables (with lags to address reverse causation):
  • LOG of energy consumption per capita (LIENCL1): proxy for economic development; preferred over GDP-based measures to avoid endogeneity with democracy;
  • Real per capita income (IRGDPL1): imputed values; used in some specifications.
  • Polity-related measures (POLL1; POL2L1): democratic rights; POL is democracy index minus autocracy index; nonlinear specification includes POL squared (POL2L1).
  • Ethnic heterogeneity (EHET): Vanhanen index (0 to 177) capturing racial, linguistic, and religious division.
  • Ethnolinguistic fractionalization (ELF): index (0 to 93), a measure of ethno-linguistic diversity; used as a robustness proxy.
  • Neighbor at war (NATWARLI): indicator if a directly bordering country is at war.
  • Cold War (COLDL1): dummy (1 for the Cold War period; 0 after).
  • Geographical region (GEOL1): nine-region dummy to capture regional effects.
  • PW10: war in the previous 10 years (contiguity of conflict historically).
  • LOGPOPL1: log of population size; controls for scale effects and reduces endogeneity concerns.
  • NMDPOLLI: neighbors’ median polity index (regional/diplomatic influence).
  • DEMCHGL1: democratic change (annual) to capture shifts in democracy.
• Estimation method and robustness checks:
  • Primary model: probit estimations for onset probability; control for duration with a Behrens-Katz-Tucker (BKT) approach to handle duration dependence in BTSCS data (Beck, Katz, Tucker 1998).
  • Alternatives: random effects panel probit; models with peace-year controls; duration-splines tested but not reported when nonsignificant.
  • Endogeneity: Rivers and Vuong exogeneity tests conducted; instrumenting energy use and polity with region/density-based controls; exogeneity not rejected in the core models.
• Data coding and quality notes:
  • Civil war onset coding aligned with Doyle and Sambanis (2000) and the State Failure Project; using a composite of CODs from COW, Wallensteen/Sollenberg, and State Failure data; adjustments made to start/end dates for consistency.
• Sample sizes and descriptive statistics (from tables):
  • Observations: 5,378 with no war; 818 with war (218 revolutionary, 600 ethnic/religious); 32 revolutionary starts; 77 ethnic starts.
  • Summary statistics in Table 1 show the means, standard deviations, and ranges for core variables; Table 2 shows their correlation matrix.
• Interpretation of results uses statistical significance (one-tailed tests noted in the tables) and model fit statistics like log-likelihood and Wald chi-square.

DESCRIPTIVE TESTS AND MODEL COMPARISONS

• Equality of means tests (ethnic vs nonethnic): strong evidence that ethnic and nonethnic wars differ on key variables (e.g., EHET, ELF, war duration, democracy indicators).
• Implication: supports treating identity and nonidentity wars as distinct phenomena with potentially different causal mechanisms.
• Table 3 (Probit models of core economic theory of civil war onset):
  • Core finding: energy consumption and real per capita income are typically negatively associated with war onset when treated in aggregate models; polity (democracy) effects often small or nonlinear.
  • ELF often not significant in these aggregate models; COLDWAR typically nonsignificant.
  • LOGPOPL (population) tends to have a large, robust effect, consistent with opportunity costs and scale considerations.
• Table 4 (Probit models of ethnic war onset):
  • EHET (ethnic heterogeneity) is robustly positive and significant, indicating higher heterogeneity increases ethnic war onset risk; interaction EHET × LOGPOPL lagged once remains significant, suggesting heterogeneity effects persist when controlling for population size.
  • Polity and its square show a parabolic relation in some specifications, indicating nonlinearity in how democracy relates to ethnic war onset.
  • Energy proxy (log energy consumption) remains negative but often less robust after regional controls are added.
  • Neighbor at war (NATWAR) is highly robust and large in magnitude, supporting diffusion/contagion theories.
  • Peace years and other controls show varying significance depending on specification; the model with regional/demographic controls improves fit for economic variables.
• Figure 1 (illustrative dependence): compares estimated probability of ethnic war onset under different democracy and development scenarios across EHET levels, illustrating that higher democracy and higher development both reduce risk, particularly when ethnic heterogeneity is high.
• Table 5 (Revolutionary/other war onset results):
  • Economic variables become more important when onset is revolutionary/other rather than ethnic; polity and EHET tend to be less predictive for RWARST.
  • War in the previous 10 years remains a strong predictor in some regressions for RWARST, indicating persistence of conflict in non-identity wars.
• Overall interpretation: political factors (especially democracy) have a stronger and more robust association with ethnic war onset than with nonethnic conflicts; economic variables still matter but are comparatively less predictive for ethnic war, once identity is disaggregated.

ONSET OF ETHNIC WAR: KEY RESULTS AND INTERPRETATIONS

• Ethnic heterogeneity (EHET) as a robust predictor:
  • Positive relationship with ethnic war onset—more heterogeneous societies are more likely to experience ethnic war starts.
  • Interaction with log population suggests that heterogeneity effects persist in larger populations, though the exact form varies by specification.
• Democracy and development:
  • Democracy (POLL1, and its square POL2L1) exhibits a parabolic relationship in some models, with high democracy reducing risk but suggesting nuanced effects at intermediate levels.
  • Energy consumption (proxy for development) is negative but not consistently significant when regional factors are included; energy often loses significance with better geographical controls, indicating regional factors may mediate development effects.
• Neighborhood effects (contagion/diffusion):
  • Neighbor at war (NATWARLI) shows a strong, robust positive association with ethnic war onset; the magnitude is large, suggesting cross-border spillovers or diffusion mechanisms.
  • Neighbors’ democracy (when included as a regional average) can dampen domestic risk, indicating diffusion of democratic norms can mitigate ethnic tensions, though this effect is less robust in certain specifications.
• Regional and systemic variables:
  • Cold War dummy tends to be nonsignificant or negative for ethnic war onset, indicating the end of the Cold War did not consistently increase ethnic conflicts on onset in the sample; time trends in development and democratization may offset any such effect.
  • Geographical region fixed effects help account for climatic and regional-level determinants; controlling for region can alter the significance of energy and political variables, underscoring the importance of controlling for regional heterogeneity.
• Endogeneity checks and robustness:
  • Rivers-Vuong exogeneity tests indicate that core regressors (log energy consumption and polity index squares) are not definitively endogenous given the chosen instruments; results remain robust when using BKT duration controls.
• Practical implications of findings:
  • Policy design should differentiate between ethnic and nonethnic civil wars; reducing ethnic grievances may require strengthening political rights, ensuring representation, and supporting regional democracy, rather than focusing solely on economic development.
  • The strong neighbor effect implies regional diplomatic and mediation efforts, cross-border institutions, and contagion-aware preventative strategies could be effective.
  • The results highlight the potential value of regional norms and institutions that increase transparency and protect minority rights to mitigate ethnic conflict onset.

ONSET OF REVOLUTORY/OTHER WARS: DIFFERENCES FROM ETHNIC WARS

• In nonethnic (revolutionary/other) wars, economic variables (energy consumption, real per capita income) tend to be stronger predictors than in ethnic wars.
• Polity and EHET show weaker or inconsistent effects for RWARST; the relationship between democracy and nonidentity war onset is less robust than in ethnic war onset.
• The presence of a neighboring war and regional dynamics still matter, but their effects differ in magnitude and significance compared to ethnic wars.
• Democratic change may be associated with a higher risk of revolutionary onset in some specifications, suggesting political changes can spur nonidentity mobilizations, but this pattern does not appear robust for ethnic wars.

ROBUSTNESS CHECKS AND ENDOGENEITY DISCUSSION

• Endogeneity concerns addressed via Rivers and Vuong exogeneity tests with instrumental variables (twice-lagged RHS variables, density, region income) and alternative specifications.
• Exogeneity cannot be rejected for polity and energy proxies in core models, supporting interpretation of the estimated effects as not driven by reverse causation in these specifications.
• BKT approach: duration dependence tested; cubic splines generally nonsignificant; peace-years control is robust and helps capture ongoing time effects.

CASES, OUTLIERS, AND INTERPRETATION NOTES

• Outliers and predictive successes:
  • Iceland and Japan: strong, transparent institutions and homogeneous societies yield near-zero predicted risk of ethnic war, aligning with expectations.
  • Cyprus: predicted risk is high in the model despite historical contexts; data limitations or structural model misspecification may explain the discrepancy.
  • Nigeria and Turkey (1970s–1980s) and Ethiopia: among the highest predicted risks, and actual onset occurred, illustrating model usefulness for certain high-risk cases.
• Practical caution:
  • Data quality and coding choices for ethnic heterogeneity and regional/demographic proxies can influence findings; time-variant measures of diversity could improve understanding of intensity and perception of ethnic tensions over time.

CONCLUSION AND MAIN TAKEAWAYS

• Identity wars (ethnic/religious) show distinct determinants compared with nonidentity wars: political grievances and institutions play a larger role, while purely economic determinants are less predictive when wars are disaggregated by type.
• Ethnic heterogeneity is positively associated with ethnic war onset, challenging some economic-only explanations that predict a weaker or nonmonotonic relationship.
• Democracy and development matter for ethnic war onset, with democracy showing particularly robust associations in many specifications; higher development lowers risk by raising opportunity costs, but democratic strength also matters as a stabilizing force for ethnic rights.
• Systemic effects are real and significant: neighbors’ war status and diffusion effects, regional democratization, and Cold War dynamics all influence ethnic war onset in meaningful ways.
• Implications for policy: tailor prevention and termination strategies to war type; strengthen democratic institutions and minority rights; address cross-border spillovers; and develop more nuanced measures of ethnic diversity and regional dynamics.
• Areas for future research highlighted by Sambanis: refine data on ethnic heterogeneity (time-variant measures, subcomponents of EHET), explore time-varying regional effects, and incorporate detailed case studies to better understand elite mobilization and ethnic networks.

KEY DEFINITIONS AND NOTATION (FOR QUICK REFERENCE)

• Ethnic war: defined as war among communities (ethnicities) over power relations between those communities and the state; core issues tied to ethnicity, identity, and political status.
• Ethnic war coding criteria:
  1) >1,000 war-related deaths in entire war and in at least one year;
  2) state sovereignty challenged by an internationally recognized state;
  3) conflict occurs within state territory;
  4) state is a principal combatant;
  5) rebels mount an organized military opposition.
• Ethnicity (Horowitz definition): an inclusive concept of ethnicity based on ascriptive differences (color, language, religion, common origin).
• EHET: Vanhanen’s ethnic heterogeneity index, range 0 to 177.
• ELF: Ethnolinguistic Fractionalization index, range 0 to 93; proxy for cross-group coordination costs.
• EHETPOPL1: interaction of EHET with log(population); captures how heterogeneity interacts with country size.
• EHET squared: EHET^2; tests for nonlinear effects of heterogeneity on war onset.
• EHET log of population: interaction term used in some specifications.
• ELF and EHET sometimes show different predictive patterns depending on war type and model specification.
• NATWARLI: whether a directly neighboring country is at war (lagged).
• NMDPOLLI: neighbors’ median polity index.
• COLDL1: Cold War dummy (1 before 1990; 0 after).
• PW10: war in the previous 10 years (lagged).
• LOGPOPL1: log population size (control variable).
• DEMCHGL1: democratic change (annual change) to capture shifts in democracy.
• EWARST: onset indicator for ethnic war.
• RWARST: onset indicator for revolutionary/other wars.
• BECK-KATZ-TUCKER (BKT 1998) methodology: used to address duration dependence in BTCS data, with tests for cubic splines and duration effects; peace years as controls when independence cannot be rejected.
• Data sources and coding references: Singer & Small (Correlates of War), Wallensteen & Sollenberg, State Failure Project, Doyle & Sambanis (2000), Elbadawi & Sambanis (2001), and others cited in the article.

REFERENCES FOR FURTHER READING (SELECTED)

• Collier, Paul, and Anke Hoeffler. 1998/2000. On the economic causes of civil war; Greed and grievance in civil war.
• Fearon, James, and David Laitin. 1999/2000. Weak states, rough terrain, and large-scale ethnic violence; Ethnicity, insurgency, and war.
• Doyle, Michael, and Nicholas Sambanis. 2000. International peacebuilding: A theoretical and quantitative analysis.
• Elbadawi, Nicholas, and Nicholas Sambanis. 2000/2001. Why civil wars in Africa; How much war will we see?
• Horowitz, Donald L. 1985. Ethnic groups in conflict.
• Lake, David, and Donald Rothchild. 1996. Containing fear: The origins and management of ethnic conflict.
• Tilly, Charles. 1978. From mobilization to revolution.
• Vanhanen, Tatu. 1999. Domestic ethnic conflict and ethnic nepotism: A comparative analysis.