An investigation of the sensitivity of DEA to data errors

The predictive value of a common measure of efficiency and the robustness of the data envelopment analysis (DEA) technique is examined when statistical noise is present in the data. Inferences are drawn from a hypothetical example regarding the potential limitations of the efficiency measure and pitfalls in both the single- and multi-stage applications of DEA. We propose a simple procedure to investigate the robustness of DEA results. The procedure maintains the relative computational simplicity of DEA and is easy to apply and interpret. Using this procedure, we examine the robustness of the results reported in two published DEA studies and find that, indeed, pitfalls occur in practical applications. We conclude with recommendations for researchers applying the technique and implications for managers.