Outliers in data envelopment analysis
Purpose The purpose of this paper is to improve the estimation of the production frontier in cases where outliers exist. We focus on the case when outliers appear above the true frontier due to measurement error. Design/methodology/approach The authors use stochastic data envelopment analysis (SDEA) to allow observed points above the frontier. They supplement SDEA with assumptions on the efficiency and show that the true frontier in the presence of outliers can be derived. Findings This paper finds that the authors’ maximum likelihood approach outperforms super-efficiency measures. Using simulations, this paper shows that SDEA is a useful model for outlier detection. Originality/value The model developed in this paper is original; the authors add distributional assumptions to derive the optimal quantile with SDEA to remove outliers. The authors believe that the value of the paper will lead to many citations because real-world data are often subject to outliers.