When does division matter? Revisiting the optimal contest architecture

We compare the aggregate effort between centralized and decentralized contests. Employing a relatively general setting that admits the ratio-form and additive-noise contests as special cases, we identify the population size elasticity of marginal winning probability as a key determinant of this comparison. •We compare total effort across centralized and decentralized contests.•The population-elasticity of marginal contest success arises as a key factor.•Convexity of marginal effort cost per marginal contest success also matters.•Decentralized contests elicit greater effort with sufficient luck in performance.