Abstract: Previous models of corruption (Persson and Tabellini, 2000; Ferraz and Finan 2005, Alesina and Tabellini 2006) assume that informed voters value accountability: that all else equal, they express a low tolerance for corruption, which often translates into an anti-incumbent bias. This assumption is based on the expectation that honest or accountable politicians are likely to deliver more effective policies than their dishonest counterparts. Absent from these models, however, are transfers (public jobs, local policies, pork barrel politics, poverty alleviation policies, etc.) from corrupt incumbents to the public at large, transfers that might effectively compensate for reputation loss. In this paper, we specify an agent-based computational model to examine the conditions under which public goods are used for private gain, albeit with net public benefits. Specifically, if the incumbent has sufficient resources to compensate informed voters, even corrupt incumbents are likely to be re-elected. As the value placed on private goods or direct transfers by voters increases, the threshold for re-electing corrupt incumbents declines non-monotonically. To state this alternatively, individuals who stand to gain the most from such transfers are likely to vote, even for a corrupt incumbent, far from their ideal point.