Simple stochastic data envelopment analysis with application to the electric utility industry

In the wake of the deregulation of the electric utility industry and the introduction of competition, utilities are facing a rapidly changing business environment. To stay competitive and ensure their survival in the 21 st century, utilities are searching for ways to improve efficiency and cut costs.; The industry has struggled with the application of regression analysis for more than 30 years and continues to struggle with determining appropriate model specifications. Misspecification has lead to wildly conflicting conclusions that can neither be conclusively accepted nor denied.; Traditional DEA, while not requiring complex model assumptions, implicitly assumes that all of the data is deterministic. Current extensions to DEA to allow for stochastic data have fallen significantly short in terms of simplicity and reliability, and violate some of the fundamental DEA postulates.; The new methodology presented in this dissertation enables accurate evaluation of efficiency by leveraging the strengths of regression (which allows variables to be stochastic and the strengths of DEA (which does not require complex model assumptions). This reformulation of the stochastic DEA model results in a set of simple linear equations that adhere to the DEA postulates, and provides results that are statistically more accurate than other formulations. Validation of this reformulation is presented in two parts: (1) a small illustrative example and (2) a large-scale simulation study that compares the accuracy of historical approaches and the new reformulation. These results are evaluated using traditional statistical hypothesis testing.; Progress in the utility industry depends on its ability to accurately measure technical efficiency, acquire pertinent information on how to improve, and to monitor changes in efficiency over time. In order to illustrate the simplicity and accuracy of the new methodology, it is used to evaluate electric utility companies in the United States. The study confirms that inefficiency due to scale is pervasively present, and shows that incorporating stochastic variability in the load factor does not appear to significantly impact the efficiency score, contrary to what was expected.