Priming Self-Reported Partisanship: Implications for Survey Design and Analysis

Can features of surveys, such as question ordering and informational stimuli, affect respondent self-reported partisanship? We report the results of two studies to examine how the survey environment affects the probability that a respondent identifies with a particular party and presidential approval conditional on self-reported partisanship. In an original experiment, we find that, under some circumstances, a question-ordering treatment increases Republican partisanship. The estimated effects are statistically different from zero in unweighted specifications where leaners are excluded from the definition of Republicans and also among respondents who identified as Republicans in previous survey waves. In our second study, we show that an informational intervention reduces the probability that a respondent identifies as an Independent and that including post-treatment partisanship in a regression changes the estimated treatment effect of informational interventions on presidential approval. Our analysis shows that seemingly minor features of survey design can affect self-reported partisanship, average outcomes for different partisans, and treatment effect estimates. We discuss the implications of our findings for survey design and theories of partisan identification and presidential approval.