Tweeting Ideology: Using Text Analysis to Create a Measure of Partisanship

• In Event: 261 - Section on Political Sociology Refereed Roundtable Session and Business Meeting
  In Refereed Roundtable 60min: Table 01. Partisanship and Polarization

Abstract

In this research we demonstrate how to employ automated text analysis on tweets from Members of Congress to estimate their partisan ideal points. Ideal points are central to the study of political partisanship and an essential part of our understanding of the behavior of elected officials. However, current research is limited by the fact that ideal points are based on legislative voting records which can only calculated at the end of a legislative session and cannot be used to compare officials across legislative sessions, branches of government, or levels of government. In this paper, we show how to use naïve Bayes and Support Vector Machine classification to calculate ideal points for Members of Congress and estimate the partisanship of individual words and tweets. We validate these estimates by comparing them against existing ideal point measures. We argue that this approach to measuring partisanship has three key advantages. First, it is more consistent with behavioral theories of partisanship as compared with existing measures. Second, it offers a higher quality measure of partisanship which accurately measures partisanship with relatively few data points. Finally, it enables a wide variety of people, organizations, and institutions to be measured and compared politically.

Authors

• Jason Scott Radford, University of Chicago

• Betsy Sinclair, Washington University in St. Louis