DEA Software Packages

It is probably a truism that the lack of simple access to reliable DEA software packages has hampered the diffusion and wider application of DEA analyses. Although, in principle, DEA solutions can be obtained with convential linear programming software, in reality this task can be time-consuming. In principle, DEA solutions require the calculation of as many linear programs as there are DMUs. When using ordinary linear programming software packages, this task can become daunting even for small problems (e.g., 100DMUs). DEA calculations with standard LP software packages are also prone to inaccurate classification of the improperly efficient and nearly efficient DMUs because of the need to calibrate, perhaps by trial and error, the appropriate magnitude of the non-Archimedean infinitesimal1 that introduces lower-bound constraints on all variables (see Charnes, Cooper, and Rhodes, 1979; Lewin and Morey, 1981; see also chapter 4 of this volume and Ali and Seiford (1993) for a comparison of results with different values for the non-Archimedean infinitesimal). Specialized DEA codes eliminate the need to calibrate the non-Archimedean infinitesimal by a preemptive approach. In addition, specialized DEA codes automate the recursive running of LP programs, the scaling of data, and the choice of models (orientation and returns to scale). Most of the empirical DEA papers to date (including many of the chapters in this book) do not provide information on how the DEA calculations were made (e.g., standard LP packages or specialized DEA code). At this stage in its development, the maturation of DEA practice and its wider acceptance will, in our judgment, be facilitated by 1) standardizing DEA notation and reference to models (e.g., as developed in chapters 2 and 3) and 2) providing adequate information regarding the method of computation.