Hospitals Evaluation Using Restricted Data Envelopment Analysis With Canonical Correlation Analysis Limits

This study presents an approach for the definition of weight restrictions in Data Envelopment Analysis (DEA). To this end, the results of a Canonical Correlation Analysis (CCA) developed with the same DEA variables are used as restrictions for DEA weight calculation. Thus, in this approach the limits of the Wong-Beasley DEA method are chosen without interference from a decision maker. As an example, weight restrictions for a DEA model (Constant Returns to Scale (CRS)—model) were obtained through the Wong—Beasley method, which was applied to a dataset of 12 general hospitals in the city of Rio de Janeiro—Brazil, 2016. The DEA had as inputs the variables number of deaths and days of stay; and as outputs number of admissions and number of beds, and the rankings of a “restricted” and an “unrestricted” model were compared by a Spearman correlation procedure. The CCA had R = 0.92; and the Spearman correlation between methods was 0.82; p < 0.01. No “zero weight” coefficient was obtained by the new procedure. In conclusion, a better discrimination was achieved, and a problem that would have arisen in the classical CRS implementation (variables receiving zero weight) was avoided, allowing for a better classification procedure for the evaluated units.