Two approaches to clustering production facilities with the use of data envelopment analysis

Initially proposed by Charnes, Cooper and Rhodes as a method for comparative efficiency assessment, Data Envelopment Analysis (DEA) eventually got an alternative use. Researchers suggested ways to use it to group (cluster) objects not by the level of their efficiency, but by other parameters, which, from the computational point of view, were secondary results of applying DEA determining the mode used by the object to gain efficiency. The need for such an approach is dictated by two research objectives in strategic management, requiring clustering companies as objects of analysis. First, as companies follow different lines of behavior, finding stable patterns of their actions, and explaining and predicting their behavior is possible only when companies are broken into homogeneous groups. Second, comparative assessment of companies’ success is also possible only within homogeneous groups, because changes in such indicators as unit costs, market share, sales per employee and other similar measures may be assessed quite differently depending on whether the company in question is aspiring to gain the wide market through cost leadership or is following an alternative pathway. Authors undertake a comparative analysis of the two approaches to clustering production facilities based on DEA results. Po, Guh and Yang suggested combining in the same cluster objects with the same production function, when isoquants are determined by the production probability area. Alternative methods based on application of standard clustering procedures to DEA results have been proposed by Kao and Hung, and later by Volkova, Filinov, Titova, Kuskova, Gorny and Nikolaeva. Theoretical analysis and computational experiments show that both approaches (based on finding the edges of the production probability area and based on application of standard clustering procedures to DEA results) yield similar results under certain circumstances but differ in the opportunities offered to the researcher in substantive interpreting of the groups created and performing alternative calculations with the changing number of clusters (groups).