Speeches, Not Votes: New Way to Map Lawmakers' Positions

🔍 Problem and Contribution

Existing approaches to measuring political disagreement from text perform well only on narrowly selected texts that discuss the same issues in the same style. This article presents the first viable approach for estimating legislator-specific position scores from the entire legislative speech corpus and for producing detailed, time-varying measures of speech polarization and politically loaded language.

🛠️ How positions are measured from every speech

A whole-corpus method is applied to legislative debates to recover legislator-level scores and track changes in speech over time. The approach is designed to work across varied topics and styles rather than requiring narrowly filtered texts. Comparisons are made to external benchmarks such as roll-call-based scales and campaign donation patterns.

📚 Evidence From Two Legislatures

• Ireland (Dáil):
• The dominant dimension of speech variation separates government and opposition.
• Ministers are more extreme on that government–opposition dimension than backbenchers.
• A second, distinct dimension separates establishment parties from anti-establishment opposition parties.
• United States (Senate):
• The recovered speech dimension shows moderate within-party correlations with scales based on roll-call votes and campaign donations.
• Speech-based positions show greater overlap across parties than roll-call positions do.
• Partisan polarization in speeches responds more visibly to major political events than does roll-call polarization.

📈 Why This Matters

• Enables measurement of individual lawmakers' positions from raw speech even when debates cover many topics and styles.
• Produces dynamic indicators of polarization and politically charged language that can be linked to events and institutional roles.
• Offers a tool for scholars and analysts to complement vote-based measures and campaign-data scales when studying representation, party conflict, and the dynamics of legislative discourse.