This dissertation addresses two problems that frequently arise in practice when applying Data Envelopment Analysis (DEA)– proportional estimates and longitudinal targets. Periodically data envelopment analysis is conducted on data that include estimated proportions, such as defect, satisfaction, or adverse event rates computed from sample data. These estimates can produce statistically biased and variable estimates of DEA results, even as sample sizes become fairly large. Several approaches are discussed, including Monte Carlo (MC), chance constraint, bootstrapping and optimistic/pessimistic methods. Results of each above method are compared to those if the true proportions were known, with emphasis on the MC approach. Inner 95^th percentile intervals for the MC efficiency scores, weights, and targets are examined assuming different sample sizes and numbers of estimated rates. In most cases, no statistically significant differences were found between the true DEA results and midpoint of the inner 95^th percentile MC interval for efficiency scores. While all methods perform fairly well, the MC approach tends to produce slightly better results and be fairly easy to implement.

The second problem arises from setting targets to an inefficient decision making unit. A typical interpretation of DEA targets is to identify performance goals that if achieved in the future will move an inefficient decision making unit to the best practice efficiency frontier. This only will occur, however, if all other units maintain their same input and output levels, an implicit assumption that is rarely the case in practice. Three alternate approaches for interpreting and setting future performance targets in this context are developed, including forward and backward looking analysis, forecasting, and MC. All methods are demonstrated using several healthcare data sets, including from World Health Organization, and the Commonwealth Fund. Analyses for handling proportional estimates and setting longitudinal targets have been run with the two basic DEA models, input oriented variable returns-to-scale (VRS-I) and output oriented variable returns-to-scale (VRS-O). To facilitate implementation an automated Excel add in– DEA Monte Carlo Solver, which solves all proposed models discussed in this dissertation, has been developed.