Selection of Variables in Data Envelopment Analysis Using the Li Test

The Li Test is a nonparametric test which was originally introduced to evaluate the closeness between two unknown distribution functions and was later adapted to compare vectors of technical efficiency scores in production theory. In this paper, we adapt the Li Test to be used for the selection of variables (inputs and outputs) in production processes. If the removal of a variable results in statistically significant differences in the scores, it is deemed relevant to the production process, whereas those variables which can be excluded without significantly affecting the efficiency scores are removed as irrelevant to the production process. The performance of the method is evaluated using a computational experience where it is compared to other methods in the literature, successfully identifying the set of relevant variables in up to [Formula: see text] of the simulated scenarios.