CMStatistics 2021: Start Registration
View Submission - CFE
Title: Time-evolving text-based ideal point model to infer partisanship in the US senate Authors:  Sourav Adhikari - Vienna University of Economics and Business (Austria) [presenting]
Bettina Gruen - WU (Vienna University of Economics and Business) (Austria)
Paul Hofmarcher - University Salzburg (Austria)
Abstract: Ideal point models analyze lawmakers' votes, speeches, press statements and social media posts to quantify their political positions along a latent continuum. We extend the text-based ideal point model to obtain a time-evolving version to study the evolution of the ideological positions of lawmakers over time and assess the change in the average difference in bipartisanship among representatives from two political parties. We aim to confirm recent findings regarding the increase in partisanship manifested in speeches by Republicans and Democrats in the US Senate during the last years. These findings were drawn using a penalized estimator for measuring group differences in choices with high dimensional data and text analysis based on manually pre-defined topics. By contrast, the time-evolving text-based ideal point model infers topics in a data-driven way and does not use the known party membership to infer the ideological positions and thus is not susceptible to overrating spurious differences in vocabulary use of different party members. Drawing the same substantive conclusions based on the results of two different statistical text analysis methods provides evidence for their robustness.