Tuesday 12pm, 11 September 2018


A Text Analysis Approach to Measure Ideological Convergence in Two-Round Elections

Caroline Le Pennec

PhD Candidate - UC Berkeley Economics


To which extent do politicians adjust their discourse and their ideological positions under electoral competition? In this paper, we address this question by exploiting two key features of the French legislative elections: 1) most districts experience natural variation in competition by holding a runoff election; 2) candidates can circulate an individual manifesto before each election round. We construct a unique dataset of about 30,000 manifestos circulated by candidates to each round of the legislative elections between 1958 and 1993. We propose a text analysis approach to scale these manifestos on a left-to-right axis, inspired from the Wordscore approach. Using our constructed ideological score, we show that candidates who make it to the runoff tend to converge toward each other and adopt a more neutral discourse in the second round.


Caroline has a bachelor in Social Science and a master in Economics from SciencesPo in Paris and is now a PhD candidate in Econ at UC Berkeley. She works primarily on topics in Political Economy and combines computational text analysis with standard empirical methods to study both candidate and voter behavior in France and across the world. She also coordinates the D-Lab Computation Text Analysis working group.