An Algorithm for Data Envelopment Analysis

The standard approach to process a data envelopment analysis (DEA) data set, and the one in widespread use, consists of solving as many linear programs (LPs) as there are entities. The dimensions of these LPs are determined by the size of the data sets, and they keep their dimensions as each decision-making unit is scored. This approach can be computationally demanding, especially with large data sets. We present an algorithm for DEA based on a two-phase procedure. The first phase identifies the extreme efficient entities, the frame, of the production possibility set. The frame is then used in a second phase to score the rest of the entities. The new procedure applies to any of the four standard DEA returns to scale. It also imparts flexibility to a DEA study because it postpones the decision about orientation, benchmarking measurements, etc., until after the frame has been identified. Extensive computational testing on large data sets verifies and validates the procedure and demonstrates that it is computationally fast.