The baseline-inflated multinomial logit model for international relations research

Abstract

International relations scholars are often interested in nominal dependent variables, and commonly analyze such variables with multinomial logit (MNL) models that treat status quo outcomes (e.g. “peace”) as a homogeneous baseline-choice category. However, recent studies of zero-inflation processes within international relations suggest that these baseline cases may often arise from two distinct sources. Specifically, some status quo responses are likely to correspond to observations that actively opted for this choice over all others, while the remaining status quo outcomes are likely to arise from observations that were unable to realistically register a non-status quo choice under any reasonable circumstances. Including both sets of responses within an MNL model’s baseline category can bias the estimated effects of covariates, leading to faulty inferences. As a solution to this problem, this study considers a recently proposed baseline-inflated MNL (BIMNL) model that explicitly estimates and tests for heterogeneous populations of status quo observations. After discussing the model and its theoretical underpinnings, I demonstrate the BIMNL’s utility through replications of two existing studies of political violence and cooperation within the areas of international relations and civil war.