A Multilevel Approach to Measuring Efficiency

Abstract

The conventional single level stochastic frontier model ignores the structure of the data resulting in the falsification of efficiency scores. Accordingly, the multilevel models, a special case of random coefficients models, take into account, the structure of data but are not frontier. These models are extremely uncommon to researchers, more so, their applicability. Researchers in both theory and practice have attempted to incorporate these models in their respective fields but with little success. In a bid to show the effects of ignoring the structure of data in efficiency analysis, clustered data is generated, partly in Rstudio and in partly in Limdep. A multilevel stochastic frontier model is proposed. The proposed model is run in Limdep using simulated dataset. The Three versions of the conventional single level stochastic frontier model are also run in the same software and with the same simulated dataset. Results of the proposed model are compared to those of the conventional models. The comparisons carried out show that ignoring the structure of data in efficiency analysis distorts the efficiency levels as well as the rankings of the subjects in the analysis. Efficiency levels from the developed model are significantly higher than those of any of the three versions of the conventional model, suggesting that efficiency levels being estimated in hierarchical organizations like municipalities are underestimates of reality. The adoption of the proposed model will in these situations will rectify this anomaly.