(with D. Kselman)

[pdf]

[older version]

Local brokers are essential in the implementation of clientelist politics, but their efforts on parties' behalf are not fully observable. A growing literature studies how parties address this agency problem, highlighting two distinct reward schemes: allocating payments, promotions, or prizes based on observed vote shares, or doing so based on inferred effort allocations. This paper develops a formal model to examine the conditions under which one or the other of these reward schemes is optimal for minimizing brokers' rent-seeking. Intuitively, the effort-based reward mechanism is optimal as long as broker performance is inferred with relative precision, while the vote-based scheme is optimal if these inferences are highly `noisy'. Less intuitively, the vote-based (effort-based) mechanism will tend to be optimal when a parties' supporters are evenly (unevenly) distributed across regions, and when the prize $\beta$ is relatively large (small).