Identification of Outliers in Data Envelopment Analysis

Data Envelopment Analysis (DEA) is a deterministic method for the aggregation of multidimensional measures and subsequent efficiency analysis. Due to its inherent determinism, however, it reacts sensitively to outliers in datasets. Existing methods for identifying such outliers have two main disadvantages. First, from a more conceptional point of view, a uniform definition of an outlier is missing. Second, there are technical disadvantages of each method. For instance, arbitrarily limited values have to be set by the user, like the amount of efficiency value from which on a decision making unit is regarded as an outlier. This paper initially presents a definition of outliers, which explicitly takes the specifics of DEA into account. Based on this definition, an approach for identifying outliers in DEA is introduced which explicitly tackles the technical disadvantages and takes them into account in the developed algorithm. The plausibility of this approach is validated on the basis of empirical examples from performance measurement at the university level.