A Dynamic, Ordinal Gaussian Process Item Response Theoretic Model

• In Event: Political Scaling

Abstract

Standard methods for measuring latent traits in political science make strong assumptions about how those traits map onto observed responses, assumptions which may often be violated. However, existing models for measuring latent traits that relax functional form assumptions lack other features that political scientists often need, such as dynamism—measuring political actors' latent traits over time—or handling ordinal responses that can fall into more than two categories. We develop a latent trait measurement model that is simultaneously (1) dynamic, (2) capable of handling multiple ordinal response categories, and (3) avoids strong assumptions about how latent traits map onto observed responses. We illustrate the strengths of this model over existing methods through two applications: responses to a panel survey, and voting data from the Supreme Court of the United States.

Authors

• Yehu Chen, Washington University in St. Louis

• JBrandon Duck-Mayr, University of Texas at Austin

• Roman Garnett, Washington University in St. Louis

• Jacob M. Montgomery, Washington University in St. Louis