Multi-criteria site selection using DEA profiling

We are given comparative data for various criteria in order to assist in the choice of a location for a particular facility. An obvious approach is for experts and other stake-holders to apply weights to the criteria to obtain overall scores for the purposes of simplifying the comparison. Reaching agreement on such weights can be difficult, if not impossible, particularly if there are many criteria to be assessed. Data envelopment analysis (DEA) presents an interesting and non-subjective method for obtaining weights. It uses optimization to discover a separate set of weights for each candidate location; these are chosen so as to maximise its score subject to the condition that these weights do not cause any of the locations to exceed a score of 100%. Unfortunately when DEA is applied to such problems a large proportion of the sites obtain a score of 100% so that it provides little help in reducing the list of sites for further consideration. This occurs because there are too many criteria for the method to work well. We present an application of DEA called ‘profiling’ in which groups of related criteria have DEA applied to each of them in turn. This produces for each site a profile of scores (smaller in number than the original raw data) which greatly eases the site selection process. The profile scores provide much greater discrimination than conventional DEA because they deal with fewer criteria each time. We illustrate the method using real data for the siting of a freeway interchange.