Tweets & Completes: Measuring Individual-Level Political Attitudes Using Survey-Matched Tweets

Image credit: Unsplash

Abstract

How do attitudes expressed on Twitter differ from those reported on surveys? Existing work measuring political attitudes on social media relies on comparisons to aggregate public opinion trends (e.g., presidential approval), but understanding political behavior often requires measuring attitudes at the individual level. The absence of benchmark measures of individual social media users’ attitudes has contributed to a lack of understanding about how attitudes expressed online compare to those reported on surveys. In this study we conduct a survey of 1,200 adult YouGov respondents and scrape 360,000 of their tweets to construct text-based measures of political interest and affective polarization. We use supervised machine learning classifiers trained and validated with human-labeled tweets and explore heterogeneity in the relationship between attitudes expressed online and on surveys