This thesis is based on research conducted at the University of Lausanne from 2012 to 2017. It focuses on the ways in which violence affects political behaviour, by looking at the voting outcomes (observable behaviour) and the mechanism driving électoral support (underlying attitudes). The voting décision has an inherent ideological component. When choosing to cast their ballot for one candidate instead of another, individuals take into account both the éco¬nomie policy platform and ideology (Lindbeck and Weibull, 1987), The choice on the ideological dimension is determined by the voter's values and beliefs. In recent years, a strand of literature has investigated the impact of violence on beliefs and political attitudes in conflict-torn countries (Bellows and Miguel, 2009; Blattman, 2009). My thesis contributes to this literature by evaluating the effect of violence on in-group identification and political trust in developed countries, thus making a link between violence, political attitudes, and voting outcomes.
Chapter 1 is myjob market paper. Identification with the polity, in the sense of the attachment or loyalty of individuals to the political system, has long been considered a key compo¬nent of political stability and the effective functioning of institutions (Almond and Verba, 1963; Easton, 1967; Schwartz, 1973). Such a political attitude becomes of even greater gravity when taking into account the disastrous effects of political instability and malfunctioning institutions on economic development (Barro and Sala-I-Martin, 1995; Alesina and Perotti, 1996; Acemoglu et al., 2001 ; Rodrik et al., 2004). It is therefore particularly relevant in the state-building procé¬dure of developing countries. Yet, several developing countries have a long, and often ongolng, history of civil conflict. Understanding thus whether the experience of individuals participating in armed conflict leads to political identification or estrangement, the persistence of such attitudes, and their behavioural conséquences is of central importance, since it can refine pol¬icy prescriptions on their demobilization and re-integration, and smooth transition to political stability.
In "The legacy of war exposure on political radicalization" I examine the impact of war participation on identification with the polity and voting behaviour. Borrowing insights frorn a literature in social psychology (Aberbach, 1969; Finifter, 1970), I test the hypothesis that war exposure ieads to low political trust and support for candidates with an anti-establishment discourse, To tackle the threats to identification from self-selection into the army and endogenous allocation in units I take advantage of a particular historical event: the forced conscription of men from the French eastern borderlands to the German military forces during World War II, A discontinuity in the draft rule within the affected régions allows to establish a (plausibly) causal link from war exposure to voting behaviour by comparing localities that are neighbour ing, but were affected to a différent degree by this conscription, In places where more men were conscripted, radical right-wing candidates stiil receive larger électoral shares today. To disentangle the attitude reflected in this voting behaviour I exploit content analysis data. War exposure results in increased électoral support not only for radical right-wing, but for candi¬dates that are more critical towards the polity overall. Survey data confirms that individuels living in places more exposed are less trustful of institutions such as the state, the judicial system, and parties. The paper was awarded with the Best Paper Award in the 13th Augustin Cournot Doctoral Days in France in 2016. 
Chapter 2 isjointly written with Mathieu Couttenieratthe University of Geneva, and Sophie Hatte, and Mathias Thoenig at the University of Lausanne. In recent years, anti-immigration political piatforms have received increasing support in the US and many European countries, Popuiist parties that campaign for such programs build on the notion that communities with growing immigrant population are unsafe, and that migration policy must be evaluated through this lens. Since voters can hardly assess themselves the potential over-propensity of immigrants to commit crime, their beliefs are fuelled with two possibly non-representative samples of crimes: the sample of crimes that voters observe in their local community and the sample of crimes that are reported in the news, If the priming of the popuiist parties is successful, that is, if voters take into account immigrants' criminality when decidingto vote for an anti-immigration platform, news reporting criminality may well affects voting outcome,
In "Seeds of Populism: Crime News Exposure and Anti-Immigration Politics", we study empirically how news coverage of immigrants' criminality impacted voting patterns in the 2009 référendum on "Minaret Ban" in Switzerland, Led by the popuiist party SVP, the campaign has been perceived as highly controversial since it played aggressively on the fears of Muslim immigration and linked Islam with terrorism and violence. The unexpected, and worldwide 
blamed, outcome ofthe référendum was a clearyes (58%) in favour of banning minarets.
The findings of this paper are manifold. First, we document a large distortion in immigrants' over-criminality in the news during the pre-vote period. Comparing the crime over-propensity of immigrants in the détection data and in the news, we find that this distortion ranges from 41% in the six month period before the vote to 69% in the three months preceding the vote. Second, our estimâtes show that proximity between the area where the crime is perpetrated and the area where a newspaper is edited is a key driver of news coverage, Third, we assess the effect ofthe crime over-propensity of immigrants in the news on the Minaret Ban vote. Our analysis uses two sources of variation, We exploit cross-municipality différences in the readership of newspapers, and cross-newspaper variation in the occurrence of crimes in areas where the newspaper has a headquarter. Since the newspapers we cover in this paper are mainly lo¬cal newspapers, and Switzerland is divided in linguistic régions, the newspapers in our sample do not have headquarters in the same places. Our estimâtes imply that, should newspapers report foreigners' crime over-propensity at its true value, this would have decreased the "yes" vote by 4 percentage points.
Chapter 3 is joint work with Johannes Buggle from the University of Lausanne, Unlike the first two chapters, Chapter 3 relates to a more "traditional" literature in political science and économies that looks at the différent means to deliver political information, and their implica¬tions for voters' choices. The use of technological innovations in électoral campaigns allows politicians to reach a broader audience than their électoral base: examples range from the adoption of télévision by Eisenhower in 1952 to Obama's 2008 campaign that made heavy use of social média, New technologies enable candidates to access sections ofthe society that were previously out of political reach. A nationwide speech campaign makes it possible to reach isoiated communities, in the same way as the radio and télévision allow to target women, and social média the youth, Yet, despite the opportunities that technological innovations offer to reach an ever growing audience, traditional means to access the electorate perslst; during the 2016 Presidential élection Donald j. Trump addressed the crowds directly 73 times and Hillary Rodham Clinton 79.
In "Technological innovations in électoral campaigns: Direct canvassing and partisan mobilization" we argue that the main reason direct canvassing persists as a campaign strategy is because it is a very efficient mean to mobilize voters that are already supportive of a candidate's programme. To test our hypothesis we exploit a unique historical context: the first direct canvassing campaign by a presidential candidate in the 1896 élection. In this élection, due to limited funding, the Démocratie candidate William Jennings Bryan adopted an unprecedented campaign strategy by using the railroad to go on a national speaking tour. From September to early November 1896, Bryan gave more than 700 speeches, in 27 states, addressing more than 6 million voters. While Bryan went on to lose the élection, his campaign strategy was already adopted by the Republican party by 1900 and is still used today, in the presence of alternative means to reach larger audiences. 
We start by looking atthe effectiveness ofthe railroad campaign performed by the Démo¬cratie candidate, To overcome the estimation issues that stem from the selection of speech locations we exploit the railroad network. We build on the fact that the Démocratie candidate targeted the largest cities because of the winrier-takes-it-all électoral system in the United States Presidential élection. Our estimâtes imply that the results in the Electoral College could have been overturned if Bryan had targeted localities optimally. In the best case scénario he could have obtained the support of more than the 224 Electors necessary to win the élection, even though he would lose the popular vote by an even larger margin.
We then disentangle the effect of speeches into mobilization and persuasion. More preciseiy, we evaluate whether the gain resulted from persuading Republican voters, or whether Bryan appearances mobilized voters that were already more supportive of his program. To disentangle between these two components we exploit the fact that the House of Représen¬tatives élection (Congress) takes place on the same day as the presidential élection. We thus focus on counties where politica! compétition is extremely high, or extremely low. An increase in the vote share for Démocratie candidates in the Congress élection in congressional districts where political compétition is extremely high should capture persuasion. On the other hand, in districts where there is no compétition, an increase in the vote share of Démocratie candi¬dates should be driven by mobilization ofthe electorate to vote for Bryan. Our findings imply that the électoral gain of the Démocratie candidate can be attributed as 65% coming from increased mobilization, and 35% coming from persuasion.