Civil War as a Hidden Markov Model


This paper estimates a model of civil war prevalence with a hidden Markov model (HMM) to account for measurement error in classifying civil wars. Since there exist multiple coding rules for civil war with substantial disagreement, each of these coding rules is treated as a noisy indicator of a latent concept. A HMM is used to classify country-years into latent states of civil war and peace and to model the probability of civil war onset and continuation with these latent states. The estimated classification supports an inclusive concept of civil war with a low threshold for what constitutes a civil war. The only statistically significant covariates in the civil war transition models are population and GDP on civil war onset.