Nonparametric efficiency analysis with unobserved inputs in multi-output settings

Abstract

Transport service providers are increasingly interested in benchmarking. The digitized nature of railway management, which relies heavily on automation and intangible inputs, conflicts with standard assumptions in the benchmarking literature that maintain full knowledge of all input factors. This creates a potential for omitted variable bias during efficiency measurement. In addition, designing accurate improvement actions requires output-specific rather than overall efficiency scores. To address these concerns we compute output-specific cost efficiency scores for multi-output entities where a subset of inputs remain unobserved to the analyst. We apply the method to highly customized data on Belgian railway traffic control rooms. We document that the magnitude of the omitted variable bias differs across outputs. Moreover, the results of a metafrontier analysis suggest that there are substantial differences in production possibilities across control rooms.