Duration models and proportional hazards in political science

A key assumption of nearly all widely used duration models is that the hazard ratios (i.e., the conditional relative risks across substrata) are proportional to one another and that this proportionality is maintained over time. Estimation of proportional hazards models when hazards are non-proportional results in coefficient biases and decreased power of significance tests. Techniques for relaxing this assumption allow scholars to test whether the effects of covariates change over time and also permit a more nuanced understanding of the phenomenon being studied. We address the potential problems with incorrectly assuming proportionality, illustrate a number of tests for non-proportionality, and conclude with a discussion of how to accurately and efficiently estimate these models in the presence of nonproportional hazards. We investigate the proportionality assumption for Cox's semiparametric model in the context of the "liberal peace" debate, using data on international conflict in the postwar period.