Political theorists posit that citizens hand over their natural freedom to the state to ensure collective safety. This logic suggests that voters should reward politicians who ensure their safety and punish those who do not. However, this proposition has received surprisingly little empirical study in American politics. I evaluate four models by which accountability may occur: 1) voters may hold incumbent politicians accountable, 2) they may hold incumbent parties accountable, 3) they may credit or blame the president's party across the whole ballot, or 4) they may vote for one party when crime goes down and the other when it goes up. I develop a new, extensively cleaned dataset of crime at the monthly level from 1960-2019 at the state, county, and municipal levels to test these models. Across all levels of government, I find little evidence for any of the models of accountability. These results are precisely estimated; a 1-SD change in crime rates affects only a very small fraction of statewide and local elections. Testing within-race accountability using county-level data in statewide elections, I find near-zero accountability. I also find no evidence that politicians strategically manipulate crime data as their next election approaches. These findings have troubling implications for the study of voter competence, representation, and democratic elections more broadly.