Is unified or divided authority optimal for voter welfare? We study this question in a political agency model where a politician’s task-specific competences are correlated. The model highlights trade-offs both within and across facets of voter welfare. Regarding incentives, unified authority yields higher total effort, but an allocation of that effort across tasks less aligned with voter preferences. Regarding the selection of good types, unified authority yields more voter information but constrains voters to use that information less flexibly. Our comparative static analysis highlights a fundamental trade-off between determinants of the optimal institution—factors that make divided authority more attractive for incentives (e.g., voters focused on one task, highly correlated competences) make unified authority more attractive for selecting good types. For some parameter values there is nonetheless an unambiguously optimal institution. For other parameter values, the overall optimal institution depends on the heterogeneity of politician competences.