Measuring Politicians' Charisma with Text
Viernes 29/11, 13.15h
Seminario de Negocios | Rocío Titiunik
El propósito del Seminario de Negocios es convertirse en el lugar donde presentar nuevas investigaciones, así como, también, en un foro para aumentar el conocimiento mutuo entre los miembros del profesorado.
Abstract
We propose a text-based measure of a politician's charisma based on text. Our point of departure is a list of politicians who served in the recent past, for whom we collect a corpus of news articles. We use a supervised machine learning approach to create a measure of charisma based on the language used in these articles. Our approach has the following steps: (i) label a small collection of texts according to whether they employ charismatic terms, (ii) train a model that identifies the most important features of the texts in predicting whether charismatic terms are present, (iii) deploy the trained model to the entire corpus and classify each article according to whether they contain charismatic terms, and (iv) calculate, for every politician, the proportion of the news articles about them that contains charismatic terms. The result is a continuous measure of charisma based on text for each politician, which we validate against vote shares.
Rocío Titiunik is Professor of Politics at Princeton University, where she is also the Director of the Data-Driven Social Science Initiative, and an associated faculty with the School of Public and International Affairs. She specializes in quantitative methodology for the social and behavioral sciences, with emphasis on quasi-experimental methods for causal inference and program evaluation. Her research interests lie at the intersection of political economy, political science, statistics, and data science, particularly on the development and application of quantitative methods to the study of political institutions.