We combine a model of symmetric information with selfish and office-motivated politicians and an Regression Discontinuity Design analysis based on close municipal elections to study partisan bias in the allocation of drought aid relief in Brazil. We identify a novel pattern of distributive politics whereby partisan bias materializes only before municipal elections, while it disappears before presidential elections. Furthermore, before mayoral elections, it fades for extreme (high or low) aridity levels while persisting for moderate levels. Our empirical results show that in this case alignment increases the probability of receiving aid relief by a factor of two (equivalent to 18.1 percentage points).