A MONTE CARLO EVALUATION OF SEVERAL TESTS FOR THE SELECTION OF VARIABLES IN DEA MODELS

Data Envelopment Analysis (DEA) is a recently developed methodology that is widely used for estimating relative efficiency scores of Decision Making Units (DMUs) that use several inputs to produce several outputs. Model specification in DEA includes aspects such as the choice of inputs and outputs or the adoption of a returns to scale assumption. As pointed out by many authors, it is obvious that the specification of a model is the key to having reliable efficiency scores. In this paper, we are particularly concerned with the selection of variables in DEA models. To be specific, we investigate the performance of several statistical tests existing in the literature that can be used for the selection of variables. In particular, the behaviour of the well-known tests proposed by Banker2 and the nonparametric tests recently developed by Pastor et al.13 is analyzed in relation to several factors such as sample size, model size, the specification of returns to scale and the type and level of inefficiency. We have drawn some conclusions that will be of help for practical uses, since the observed behaviour of the tests in the different scenarios determined by the specifications of the mentioned factors may provide some useful insight into the choice of an adequate statistical test in the particular context of a given DEA application.