Tuesday Seminar – 3 November

Partisan Semantic Overlaps: Floor-speeches and Ideological Position

Benjamin Guinaudeau (University of Konstanz)

Estimating the ideological position of Members of Parliaments (MPs) remains a challenge for political scientists. Different approaches have been developed including surveys, roll-call votes and floor speeches. Inspired by the measure of polarization proposed in Peterson and Spirling (2018), we present a new unsupervised strategy to extract ideological positions from speeches. We rely on partisan semantic overlaps (PSO), defined as language patterns indistinguishably used across parties. We train artificial neural networks to predict party labels given text and expect these semantic overlaps to be mapped by the partisan probabilities. The higher the overlap between two MPs, the smaller is their ideological distance.
We use three decades of parliamentary speeches in six countries (Canada, France, Germany, New Zealand, Spain, United Kingdom) and estimate, in each of these countries, partisan probabilities with a convolutional network. We show party-level positions are accurately captured by the measure (high correlation with CMP). In the absence of any broadly accepted individual ideological measure, we use a new expert survey designed to capture MPs’ position to validate our ideological scores at the individual level. For now, intra-partisan heterogeneity is not accurately captured. We discuss the potential origins of these results and propose possible ways to address these in the future.

Contact Semih Çakır if you would like to participate in the seminar.

This content has been updated on 29 October 2020 at 14 h 45 min.