A new approach for extreme values in data envelopment analysis

Abstract

Data Envelopment Analysis (DEA) is methodology for relative performance measurement and has been extensively utilised over the past few decades. DEA is however sensitive to the presence of outliers in the data and can cause inaccurate reflections of the relative efficiency score and the projections of inefficient Decision Making Units (DMU) onto the efficient frontier. Stochastic frontier analysis can accommodate for the statistical noise but makes certain assumptions on the data. This dissertation introduces an approach to accommodate for outliers in a DEA model without removing observation that would otherwise affect the results. The results on the proposed model are compared to two deterministic and three stochastic models, and have shown an increase in the efficiency score and the number of efficient DMUs and an increase in the overall efficiency scores.