This paper examines the impacts of devolving authority for public resource allocation to local governments in a setting of limited electoral control. Such a setting differs from that assumed by seminal formal models of devolution, but describes many developing countries. This study presents a formal model of this setting and tests it using unique data from a natural experiment in rural Ethiopia whereby half of the country’s regions were decentralized but not the other half. Employing a spatial regression discontinuity design, this article shows that decentralization strongly improved delivery of agricultural public services, which are of high priority to the central government. In contrast, it did not impact drinking water services, on which the central government places lower priority but citizens place high priority.