Stepwise regression data envelopment analysis for variable reduction

In this paper, we develop stepwise regression data envelopment model to select important variables. We formulate null hypothesis to understand the importance of each variable and use Kruskal–Wallis test for this purpose. If the Kruskal–Wallis test does not reject the null hypothesis then we can conclude that all the variables are of equal importance as their presence and on the other hand absence of other variable does not create huge fluctuations in efficiency scores in fact give a complete ranking relative to base model. If the Kruskal–Wallis test does reject the null hypothesis this will imply there is significant fluctuation in the efficiency score relative to base model. And therefore we have to further check the pair of variables that causes the fluctuation in order to determine its importance using Conover–Inman test. The results of the proposed models are compared with the results of previously published models of the same dataset. The proposed models helps understand the extent of misclassification decision making units as efficient/inefficient when variables are retained or discarded alongside provides useful managerial prescription to make improvement strategies.