Robust Nonparametric Partial Frontier Input Distance Function Estimates: Testing and Rejecting Convexity

Statistical tests of the convexity assumption for full frontier technologies are still rather rarely applied. We employ existing order-a and order-m partial frontier estimators for production technologies. Since these partial frontier technology estimators have favourable statistical properties, this permits testing convexity using a simple statistical test comparing densities. Using a panel of electric power generation plants in the US we find strong evidence against convexity for most years using both order-a and order-m partial production frontiers. These new empirical results imply that empirical researchers must be cautious about invoking the –mostly implicit– convexity assumption when estimating technologies. Anyway, it is clear that convexity needs to be tested.