Solving the Problem of Monotone Likelihood in Duration Modeling

• In Event: Generalized Linear Models in Political Science

Abstract

Duration models have become ubiquitous in the political science literature, with applications to topics as diverse as polity survival, the length of inter- and intra-state wars, the fate of political leaders, and the effectiveness of international peacekeeping missions, among others. As with any empirical strategy, however, the application of duration modeling to real-world data can be vulnerable to methodological complications. Recent research has highlighted some of the problems associated with duration models, but to date, one issue in particular has escaped scrutiny in the political science literature---namely, monotone likelihood. This article provides an accessible introduction to the phenomenon of monotone likelihood in duration modeling, with applications to political science research. It explains the nature of the problem, identifies its causes, and describes an easy-to-apply corrective. Using simulation techniques and examples drawn from the existing literature, the article illustrates the advantages of Firth penalized likelihood estimation in resolving the methodological complications underlying monotone likelihood. It concludes with guidance for political scientists applying duration modeling techniques in their empirical research.

Authors

• Noel Anderson, University of Toronto

• Benjamin E. Bagozzi, University of Delaware

• Ore Koren, Indiana University Bloomington