The Elephant in the Chamber? Incorporating Tweets about Trump into Congressional Ideal Point Estimates

Image credit: Unsplash

Abstract

Political science has traditionally measured legislator preferences using roll-call voting behavior, but we consider the possibility that legislative votes might fail to adequately capture attitudes toward President Donald Trump. We estimate Congressional ideal points using an alternative expression of preferences: tweeting. Sampling from a population of all tweets authored by members of the U.S. House and Senate related to Donald Trump from November 2016 to February 2018, we incorporate the frequency, timing, and content of these tweets into representations of legislator ideology using a neural net approach. We construct a model in which legislator attitudes towards Trump are represented as vector embeddings, and Donald Trump himself is represented as a vector that is constructed using a neural network from the text of his daily tweets. In our model, the interaction between legislator embeddings and Trump embeddings produces predictions of both the number of times legislators tweet about Donald Trump \textit{and} the sentiment of tweets about Donald Trump. We assess the quality of our learned representations for legislators by comparing to the canonical DW-NOMINATE representations as well as votes on Trump-endorsed legislation.