Quantifying electoral behavior using applied econometrics and big data[Quantifier le comportement électoral à l'aide d'économétrie appliquée et de big data]

This thesis studies electoral behavior by creatively combining big data, causal econometric models, and different statistical programming languages. This results in the following three chapters. The first chapter studies the short-term effects of exogeneous events on voters´ perceptions of their political representatives. It uses data from the British Election Study (BES), exogeneous variation in football results, bookmakers´ betting odds to adjust for expectation formation, geospatial data from the social media platform Twitter to match respondents and scores, and OLS panel data regressions.The second chapter analyses the short-term impact of presidential TV debates in the US on voters´ polarization under peer-effects. It uses big data from the social media platform Reddit, machine learning algorythms via Google BERT to extract users´ political views, and OLS panel data event study regressions.The third chapter studies the long-term effect of US military supply operations and war crimes in France during WWII on local voting patters for far-right parties. It uses a novel geocoded dataset, reconstructed based on archival documents, election outcomes for presidential and legislative elections, as well as OLS and 2SLS regression models.