Mavericks Revisited : A New Index to Identify Maverick Units in Data Envelopment Analysis

In recent times Data Envelopment Analysis (DEA) has been extensively used in order to assess efficiency levels in organisations. DEA has many advantages: it can deal with several inputs and outputs; inputs and outputs do not have to be measured in the same units; there is no need to specify a production function. But DEA has shortcomings. The technique is based on multiple comparisons, and the assessed units can choose the basis of the comparison by setting a set of weights. Unfortunately, some weights can take zero values with the consequence that a particular input or output is excluded from the comparison. Therefore, some units can achieve technical efficiency while being price inefficient.In order to tackle some DEA shortcomings, including the problem of technical and price efficiency, cross-efficiency method was suggested about three decades ago. As one of its goals, the method tries to identify the problematic units which achieve their efficiency in an unacceptable way. These units are called mavericks in the literature, and in order to identify them some maverick indexes have been proposed. However, the literature does not agree on exact characteristics of maverick units, and the commonly used maverick identification methods do not accurately deliver what they are supposed to deliver.The purpose of this paper is to address these issues in order to improve the highlighted drawbacks in both theoretical and practical aspects of maverick literature. To do so, a new maverick detection method is suggested. From theoretical perspective, the method is based on some adjustments and refinements of maverick assumptions and definitions, and from the practical perspective, it is based on multidimensional unfolding of cross-efficiency matrices in high dimensional space. The paper starts with a critical appraisal of the maverick literature, following by some refinement to address the drawbacks. Afterwards, the new maverick index is explained, and its application on a real data set is presented