"Persuasive Partisans: The Informational Content of Policymaking with Application to US Governors" - Hassan Sayed, Princeton University

Date
Oct 21, 1:20 pm2:35 pm

Details

Event Description

Abstract

Incumbent political leaders want to persuade voters of their skill, but are constrained by legislatures. Skilled leaders may be better at pushing more ambitious partisan reform, but such bills may still fail in a divided legislature. I study a model where the partisanship of a policy agenda generates different signals of ability. I show that as leaders become more popular, they first implement partisan policies, then bipartisan policies, then extremely partisan policies. Voter welfare is nonmonotonic with respect to popularity, legislative alignment, and competition.

I test the model’s predictions using a panel of US gubernatorial agendas: governors’ annual “State of the State” speeches from 1990-2020. I use handcoded data to fine tune a “Bert” LLM model that isolates policy proposals in the text, and then measure the partisanship of each governor’s policy proposals. I document an upward creep in aggregate partisanship starting in 2000 followed by a spike in 2017. Nevertheless, I show that the predictions of the model are borne out in the data: governor speech is 0.5-1 standard deviations more partisan in the lowest and highest quintiles of popularity than in the middle. I estimate the model until 2005 and show that it predicts out of sample aggregate partisanship well, suggesting reputational concerns can explain the recent uptick in partisanship.