We compare experimentally two contest designs. In the between-group design (BGD), cohorts compete with one another in stage 1 and a single cohort is then advance to stage 2. In stage 2, members of this cohort compete for a single prize. In the withingroup design (WGD), the order of the competition is reversed. Our findings support the theoretical conclusion that the WGD is superior to BGD in generating more expenditure. They also show that if the cohorts are fixed, small, and interact repeatedly over time, then tacit collusion in the WGD, but not the BGD, leads to suppression of expenditures.