Incumbent accountability is central to democratic elections, which are expected to discipline poor performers by removing them from office. However, officeholders might survive even when voters withdraw support from them. I examine this disconnect in the context of economic downturns, where prior work has largely treated vote loss and office loss as interchangeable measures of accountability, even though punishment may reduce incumbents' vote share without crossing the threshold for turnover. Drawing on the 2014 Petrobras scandal and the global oil crisis, which decreased royalty inflows to oil-dependent Brazilian municipalities, I analyze whether economic shocks produce punishment in terms of both vote loss and reduced survival in office. Using a differences-in-differences design, I provide causal evidence that mayoral incumbents lose votes in municipalities that suffered royalty losses, but their likelihood of being removed from power is not altered. I argue that the economic shock weakened incumbents' perceived advantage, encouraging more challengers to enter the race and dispersing anti-incumbent votes across multiple candidacies. Using post-shock royalty loss as an instrument for competition, I show that the increase in challengers reduces the incumbent's support, yet this does not translate into higher rates of office turnover. I further show that the shock significantly increased vote fragmentation among opposition candidates in affected municipalities. Overall, the study shows that although voters do punish leaders, the structure of electoral competition can limit the extent to which elections serve as mechanisms of accountability. More broadly, fully understanding incumbent accountability and its mechanisms requires distinguishing between electoral punishment and electoral removal.

To what extent does partisanship bias copartisans' responses to corruption? While existing research shows that partisanship can shield politicians from the consequences of corruption, the conditions under which this varies remain unclear. I argue that partisanship's effect on corruption voting depends on whether corruption is targeted (when the corrupt actor belongs to the voter's preferred party) or systemic (when the electorate perceives corruption to be widespread). These contexts activate different processes: copartisan bias in candidate evaluation and the weakening of partisanship's behavioral shielding effect. Using a survey experiment in Brazil and public opinion data from 17 Latin American countries (2008--2018), I find that copartisans express significantly greater support for a candidate accused of corruption when their party is implicated. However, as systemic corruption increases, they become more likely to withdraw support and defect to opposition parties. These findings show that copartisan bias in favor of accused politicians is conditional and can become fragile under certain circumstances.