Clustered Political Violence Events as Dirichlet Process Models

• In Event: Innovations in Measuring Strategic Interactions During Violent Conflict

Abstract

Why do violent events cluster in time and space? This paper introduces conflict scholars to Dirichlet Process models. A class of Bayesian nonparametrics, Dirichlet Process models provide several advantages over other approaches, including the ability to explicitly model conflict dynamics and accurately cluster phenomena of interest. Additionally, we show how to use the output of the Dirichlet Process model as an explanatory variable in a more traditional regression, so that it is possible to use how violent events cluster to explain additional conflict features (e.g. duration, outcome, etc). We illustrate the approach with data on protest events and provide validation statistics to demonstrate the benefits of using Dirichlet Process models.

Authors

• Kaitlyn Webster, Duke University

• Margaret J. Foster, Duke University