Benchmarking the benchmarks – Comparing the accuracy of Data Envelopment Analysis models in constant returns to scale settings

Despite the massive use of Data Envelopment Analysis (DEA) models for efficiency estimations in scientific applications, no paper cared about identifying the DEA model, which is able to provide the most accurate efficiency estimates, so far. We develop an established method based on a Monte Carlo data generation process to create artificial data. As we use a Translog production function instead of the commonly utilized Cobb Douglas production function, we are able to construct meaningful scenarios for constant returns to scale. The decision-making units resulting from the generated data are then used to calculate DEA estimators using different DEA models. Finally, the quality of the resulting efficiency estimates is evaluated by five performance indicators and summarized in benchmark scores. With this procedure, we can postulate general statements on parameters that influence the quality of DEA studies in a positive/negative way and determine which DEA model operates in the most accurate way for a range of scenarios. Here, we can show that the Assurance Region and Slacks-Based-Measurement models outperform the CCR (Charnes–Cooper–Rhodes) model in constant returns to scale scenarios. We therefore recommend a reduced utilization of the CCR model in DEA applications.