Selecting Models for Latino Partisan Attachment: An Application of Bayesian Model Averaging

Abstract

Evaluating the determinants of Latino partisan attachment based on p-values risks misleading results, especially if reported in the context of a model with numerous independent variables and in the absence of an accompanying evaluation of model uncertainty. We address model uncertainty by applying Bayesian Model Averaging to multiple national surveys of Latinos felded from 1989 to 2010, and evaluate the extent to which independent variables are candidates for foundational determinants of Latino partisan attachment. We find that some predictors that are viewed as foundational in explaining Latino partisanship are not at all robust, while some predictors that often play second-fiddle are better candidates for foundational determinants of Latino partisanship.