Skip to main content

Assessing the Effects of Measurement Error in Cross-National Social Research

Fearon and Laitin (2003) have published an important recent article challenging the conventional wisdom that the root cause of most civil wars lies in ethnic and religious diversity. Instead, they argue that state strength is the primary predictor of civil war onset. There are two reasons to believe that measurement error might complicate Fearon and Laitin's findings. First, the theoretical story that Fearon and Laitin tell revolves around a latent -that is, unmeasured - variable we term "state strength." In their story, the risk of insurgency increases as the capacity of the state decreases. Despite the centrality of this latent variable to their analysis, Fearon and Laitin do not posit a rigorous model relating the observed indicators to the underlying concept of state capacity. Second, as much past literature suggests, the indicators of ethnic and religious diversity used in the analysis are all measured with considerable error. From a statistical perspective, these two issues both pose the problem of measurement error. In general, it is both possible (and empirically common) for measurement error in multiple predictor variables to bias coefficients in unknown directions (possibly changing the signs of coefficients). Furthermore, maximum likelihood estimators are no longer consistent in the presence of covariate measurement error. We use the same data as Fearon and Laitin to estimate models of the determinants of civil war onset with one principal difference: our models take into account measurement error in the independent variables.