I–O stepwise method for selecting variables in data envelopment analysis models

This work develops a new procedure for selecting input–output variables to be incorporated to a Data Envelopment Analysis model, named I–O stepwise. Monte Carlo simulation is explored as a tool for developing the method. Diversely from that proposed by Norman and Stoker, the procedure presented here uses correlation analysis adopting different criteria as the entering variable is an input or output. Two case studies are presented, illustrating the advantages of the method: the first one applied to offshore oil production efficiency in Rio de Janeiro, the second one to assess the performance of Brazilian states in public resources utilization. We stress the importance of deciding about which pair should be the starting one in I–O stepwise procedure.