Selection Models and the Study of Politics
Patrick T. Brandt, Christina J. Schneider
December, 2007
Abstract
Selection models are now widely used in political science to model conditionally observed processes (e.g., conflict onset and escalation, democratization and foreign direct investment, voter turnout and vote choices). We argue that many applications of selection models are poorly identified since the same predictors are used for predicting selection and the outcome of interest and only few additional exogenous regressors are included to predict selection. The paper shows that this can lead to biased inferences and incorrect conclusions about the presence and effects of selection. We then propose methods to evaluate the costs and consequences of poorly specified selection models and give guidance about how to identify selection models and when to abstain from estimating such models. A replication of Reed's (2000) analysis of conflict onset and escalation further illustrates our results in a substantive application.Keywords:
none.