Measuring Individual-Level Campaign Effects with Latent Models

• In Event: New Approaches to Measurement and Conceptualization

Abstract

How many people experience campaign effects like priming, learning, persuasion, and following in any given period? Most existing research cannot offer an immediate answer because the standard statistical models – regressions with lagged predictors – used to measure these phenomena preclude individual-level countability and sustain an interpretation of campaign effects as population-level properties. But wider availability of panel data and recent consolidation around four major concepts (i.e., priming, learning, persuasion, and following) enable the creation of models that can generate quantities with stronger construct validity. This paper presents a latent variable model intended to classify survey respondents according to the dominant campaign effect they likely experienced between two measurement periods. In brief, the model maps each individual’s configuration of response changes during an interval to one of a fixed number of discrete categories that correspond to ideal types for priming, learning, persuasion, or following. Waves from The American Panel Survey (TAPS) fielded during the 2016 election are used to fit both this latent model as well as conventional linear regressions that serve as comparison specifications. While the two approaches differ in their parameterization and desired outputs, together they provide complementary perspectives for the study of campaign effects.

Author

• William Wen, Princeton University