Examining the Inefficiency of Transit Systems Using Latent Class Stochastic Frontier Models

Abstract

This paper estimates a single class stochastic cost frontier model that accounts for heterogeneity by including background variables and a latent class model of the same specification. It is found that both the two-sided random errors and one-sided errors (cost inefficiency) are substantially smaller in the latter model than it is in the former, suggesting possible bias in the estimates of inefficiency from the single class model. Further, 58.9%–68.39% of the calculated inefficiencies are due to differences in technology captured by the latent classes. The paper concludes that using background variables only to capture heterogeneity may exaggerate measured inefficiencies in transit systems and suggests the latent class approach as a solution.