A close look at second stage data envelopment analysis using compound error models and the Tobit model

• Published in: Socio-economic planning sciences Vol. 65; pp. 111 - 126
• Main Authors: da Silva, Aline Veronese, Costa, Marcelo Azevedo, Lopes, Ana Lúcia Miranda, do Carmo, Gabriela Miranda
• Format: Journal Article
• Language: English
• Published: Oxford Elsevier Ltd 01.03.2019; Elsevier Science Ltd
• Subjects: Correlation analysis; Data envelopment analysis; Electricity; Environmental variables; Mathematical models; Power; Regression analysis; Second stage analysis; Statistical analysis; Stochastic frontier analysis; Tobit regression; Second stage analysis; Stochastic frontier analysis; Tobit regression; Environmental variables; Data envelopment analysis
• ISSN: 0038-0121; 1873-6041
• DOI: 10.1016/j.seps.2018.04.001

In 2014 the Brazilian Electricity Regulator (ANEEL) evaluated the efficiency of power distribution utilities using Data Envelopment Analysis (DEA). Estimated efficiencies range from 22.46% to 100%. Although environmental information is available in the data set, corrected efficiencies were not investigated. Different second stage models can be applied to adjust for environmental heterogeneity. Although statistical correlation among efficiencies and environmental variables can be easily estimated, corrected efficiencies are subject to the underlying structure of the second stage model. Therefore, different second stage models may achieve different corrected efficiencies. We provide a detailed statistical analysis of the Tobit model and compound error models for second stage analysis. Limitations are described and the corrected efficiencies using these models are evaluated. Potentially, Brazilian power distribution utilities may achieve substantial changes in estimated efficiencies if second stage analysis is used. •Statistical properties of DEA second stage analysis are presented.•Second stage models with compound error are evaluated.•A second stage analysis applied to Brazilian benchmarking model is presented.•Results show significant changes in efficient operational costs using second stage analysis.