What makes some regions within a country more likely to experience electoral violence than others ? Based on this initial question, this Ph.D. thesis uncovers the mechanisms driving incumbent elites to use violence as an electoral tactic in non-democratic regimes. The local determinants of electoral coercion are investigated by means of a subnational and cross-national research design. The analyses are conducted using a newly collected dataset comprising events of electoral violence at the constituency-level for 71 elections held in 14 countries. Among the key findings of the thesis are that authoritarian regimes are most likely to perpetrate violence where the opposition has an intermediate level of electoral strength, where the populations are economically vulnerable, where there are no alternative means to manipulate the election and where the electoral support for the regime is declining.