Inference in Gaussian Process Models for Political Science

• In Event: Political Methodology Virtual Poster Session

Abstract

Political scientists often seek to perform inference in settings where knowledge about the functional form mapping predictors to outcomes is imperfect or the traditional assumption of conditional independence of observations does not hold. Recently Gaussian process (GP) models, a family of machine learning techniques, have been used to study politics in such settings; however, many inferential quantities of interest to political science have either not been derived in the statistical and machine learning literature the models hail from or have not been employed in the political science literature yet. I provide practical guidance for applied researchers to implement GP models for more accurate inference in their research, including how to obtain quantities of interest to political scientists, the derivations of which are novel to the GP model literature.

Author

• JBrandon Duck-Mayr, Washington University in St. Louis