Assessing the efficiency of hospitals operating under a unique owner: a DEA application in the presence of missing data

Originally developed in the late 1970s to assess the efficiency of comparable operating units, Data Envelopment Analysis (DEA) has since been used in a variety of contexts. Although incomplete data sets are often encountered in practice, the best approach in such situations is unclear in general. This paper explores methods such as multiple imputation, bootstrapping and smart dummy variable replacement, borrowed from similar missing data problems in regression analysis. Each missing data method is tested on a library of DEA problems that are gathered from the DEA literature. These problems are selected in such a way as to represent a thorough cross-section of problem sizes (small, medium, large) and types (type of DEA model, number of decision-making units, number of inputs, number of outputs, etc.). The results are illustrated by comparing the solutions of complete data sets against the simulated versions of the same data sets with missing data. The sensitivity of each method on the efficiency scores and ranking of the decision-making units is presented.