Search Climate Insights Directory

The author develops a Focused Information Criterion (FIC) for dynamic multinomial logit models, specifically designed to select models that provide the most precise estimate of a specific focus parameter rather than the model closest to the true data-generating mechanism. The thesis establishes the asymptotic theory for maximum likelihood estimators under model misspecification for these models, proving they are consistent and approximately normally distributed. Through simulation studies, the author verifies that the FIC effectively identifies models with the lowest true mean squared error for various focus parameters. The methodology is applied to the Militarized Interstate Disputes (MID) dataset (19502010), focusing on the probability of conflict escalation. The analysis indicates a tendency for higher democracy levels to lower escalation probabilities, although these findings are not statistically significant.