Bus transit technical efficiency using latent class stochastic indirect production frontier

Abstract

Using an Indirect Production Frontier (IPF), this article examines technical inefficiency within a latent class framework while simultaneously accounting for allocative distortions from operating and capital subsidies. It identifies two latent classes of US public transit systems, one characterized by economies of scale with 16.61% technical inefficiency and the other by diseconomies of scale with 14.16% technical inefficiency. It decomposes technical inefficiency among some of its sources and finds that the incentive tier of federal operating subsidies, regulations regarding years of vehicle use, subsidy-induced allocative distortion from labour overuse relative to capital negatively influence technical inefficiency in all transit systems. For the Latent Class 1 transit systems, the sources of lower technical inefficiency are operating speed, purchased transportation and years-of-vehicle-use regulation. For the Latent Class 2 transit systems, these sources are subsidy-induced capital-labour allocative distortion and the incentive tier component of the federal formula grant.

Keywords:

• operating and capital subsidies
• indirect production frontier
• allocative distortion
• technical inefficiency
• latent classes

JEL Classification::

• D24
• H21
• L91
• L92

Notes

¹ For a detailed discussion, see Obeng (2011).

² See Obeng (2011) for a discussion of endogenous transit output and endogenous subsidies as they apply to US operating subsidies and capital subsidies.

³ The homogeneity of degree zero restriction can also be obtained by dividing implied cost by each input's implied price to obtain .

⁴ This is a typical problem in many clustering techniques.

⁵ Because the translog function is an expansion around the mean, it assumes that the mean transit system has no subsidy-induced allocative distortion. Since there is no empirical basis for this assumption, the terms and are not mean centred.

⁶ For in-depth discussion of the decomposition variables in this section, see Obeng (2010).

⁷ In this application, stepwise regression is used as a variable reduction tool before estimating the Tobit regression.