This thesis critically assesses the ’greed and grievance’ model of civil war developed by Collier & Hoeffler (1998, 1999, 2001, 2002). Their model and theoretical argument has gained prominence in the quantitative field of the study of civil war, and has provided valuable insights—in particular into the role of economic factors generating civil war. In brief, Collier & Hoeffler’s model present two competing and complementary explanations for civil war: atypical levels of grievance or atypical opportunities for forming a rebel organization. Their rational choice-based analytical model focuses on conditions that favor the formation of rebel organizations, and their econometric model predicts the probability of a civil war being initiated in a country during a five-year period. While inequality, political rights, ethnic polarization and religious fractionalization were insignificant, natural resource dependence emerges as a very significant factor in CH’s study of civil war from 1960 to 1999. Economic development and access to finance, through the form of foreign contributions or control over natural resources, supports a combined greed/opportunity explanation for civil war.
The analysis in this thesis was conducted in three stages. The first replicated the CH model on a five-year period dataset. Although there were some discrepancies between the published CH results and my replication, the findings remained much the same: economic development and growth, social fractionalization and peace reduced the propensity for conflict. A large population and primary commodity dependence increased the risk. CH found no effect of the regime measurement ‘Polity’ in their analysis. Adding an alternative regime-type variable, ‘Polyarchy’, did not produce significant results either. Breaking up primary commodity exports into oil and mineral dependency did not produce the hypothesized relationship. Neither the Sub-Saharan Africa nor the Middle East dummies were significant, and no particular, undefined regional effect was detected. Comparing the two regions, however, the model predicted widely different conflict risks for the two regions. From 1990 to 1995, the model predicts a sharp increase in the onset of conflict for Sub-Saharan Africa and a substantial decrease for the Middle East and North Africa.
I argued that the findings of Collier & Hoeffler might be influenced by the research design they had chosen. I therefore developed an annualized version of the CH dataset, and applied it on a dataset including civil war and armed civil conflict. For civil war, the analysis of the annualized dataset produces rather different results than the findings in CH’s analysis. Most interestingly, social fractionalization and ethnic dominance ‘switched places’. While social fractionalization was significant in the five-year analysis, it was no longer so in the annualized version. Ethnic dominance was not significant in the five-year analysis, but strongly significant in the annualized. This contrasts CH findings on mobilization (weakens their opportunity model) and political exclusion (strengthens the grievance perspective). Regime type displayed a curvilinear relationship with civil war, suggesting that opportunity as well as grievance is important. For civil war, primary commodity exports displayed an inverted U-relationship with conflict. Breaking down primary commodities to oil and mineral dependency also revealed a curvilinear pattern, but also here the coefficients were not statistically significant. Some interesting changes also occurred when the threshold of conflict is lowered to include conflict with a minimum of 25 battle-related deaths per year. For armed civil conflict the social fractionalization continued to be insignificant while ethnic dominance was positively related to conflict. I draw the conclusion that the mobilization aspect is less important, and that grievance is more important than what CH argue. Quite surprisingly, primary commodity display a positively, linear relationship with armed civil conflict. Breaking down primary commodities into oil and mineral dependence led to a loss of around 1,300 observations and 35 conflicts. Oil and mineral dependency displayed only curvilinear relationships, although not at statistically significant levels. CH dropped regime type (as a linear term) from their model because it was not significant. On the annualized dataset and using both the linear and squared term of regime type the hypothesized inverted U-relationship was recurrently present. In conclusion, applying the CH model to an annualized dataset was a useful experience. It shed some light on the fact that their very research design might have influenced their results. Using an annualized dataset design, both the opportunity (rather than the greed) and grievance aspects are strengthened. This study shows that regime type is relevant for an analysis of civil armed conflict. Regime type makes a difference. In some of the three-dimensional figures, the conflict curve is tilting towards the authoritarian side, possibly hinting that also democracy makes a difference.