The Indirect Effect of Ethnic Heterogeneity on the Likelihood of Civil War Onset

The article argues that ethnic fractionalization does not play a decisive role in civil war onset. The author argues that including ethnic fractionalization as an independent variable in a research design is not appropriate because its effect is not direct, but indirect. He argues that the degree of fractionalization matters because other significant variables can make sense at a certain degree of fractionalization in a country. 
Literature on civil wars presents a good number of factors that make a country more prone to civil wars. In the early research, it was believed that more heterogeneous societies are more prone to civil wars. Findings of those scholars who did large-N studies and provided empirical data are controversial. Some of such studies show that the more a country heterogeneous the more likely a civil war to break out in that country.  Some scholars, on the other hand, show that homogeneous/heterogeneous composition of the society in a country does not make it more prone to civil war onset. 
He uses Fearon and Laitin’s (2003) data set of violent internal conflicts between 1945 and 1999. He uses probit model in order to determine whether factors affecting civil war onset will have more chance, and lower amount of error, in ethnically diverse countries. 
The statistical results in the article have three major findings.  First, those variables that are expected to have significant effect on civil war onset do not have such effect with very low ethnic fractionalization. Second, most of the hypotheses put forward in Fearon and Laitin (2003) do not hold up in countries that have low level of ethnic fractionalization. Third, highly fractionalized composition of a society counterbalances any advantage in solving collective action problem provided by ethnicity. His results also show that, as long as ethnic fractionalization is held constant, ethnic polarization has a similar effect on the likelihood of civil war onset as ethnic fractionalization. 
In the end, it should be noted that Blimes’s heteroskadastic probit model fits the data more closely than normal probit. However, his model predicted 456 false positives and failed to predict 2 of the 18 civil war onset within the lowest quartile of ethnic fractionalization. As his conclusion states as well, there is a great deal of mystery about civil wars. This study sheds some light but more research is needed.