Bayesian Spatial Survival Models for Political Event Processes

Authors


  • A previous version of this article was presented at the 2006 annual meeting of the American Political Science Association. I thank Sudipto Banerjee, Katherine Barbieri, Neal Beck, Janet Box-Steffensmeier, Brad Carlin, Rob Franzese, Andrew Gelman, Darren Greenwood, Jude Hays, Andrew Lawson, Dave Lunn, Lyndsey Young Stanfill, Tracy Sulkin, Chris Zorn, and three anonymous reviewers for helpful conversations and comments on this project. Any remaining errors are my own.

David Darmofal is assistant professor of political science, University of South Carolina, 350 Gambrell Hall, Columbia, SC 29208 (darmofal@mailbox.sc.edu; darmofal@gmail.com).

Abstract

Research in political science is increasingly, but independently, modeling heterogeneity and spatial dependence. This article draws together these two research agendas via spatial random effects survival models. In contrast to standard survival models, which assume spatial independence, spatial survival models allow for spatial autocorrelation at neighboring locations. I examine spatial dependence in both semiparametric Cox and parametric Weibull models and in both individual and shared frailty models. I employ a Bayesian approach in which spatial autocorrelation in unmeasured risk factors across neighboring units is incorporated via a conditionally autoregressive (CAR) prior. I apply the Bayesian spatial survival modeling approach to the timing of U.S. House members' position announcements on NAFTA. I find that spatial shared frailty models outperform standard nonfrailty models and nonspatial frailty models in both the semiparametric and parametric analyses. The modeling of spatial dependence also produces changes in the effects of substantive covariates in the analysis.

Ancillary