On the Approximation Bias to Benefit Measures in Discrete Choice Models

Abstract

Bias estimation associated with use of different approximations for transportation project benefit calculation is presented by the authors. A simulated databank was generated by the authors for real revealed preference sample behavior reproduction on work trip mode choice. There was exact measure computation for three transportation policies differing in mode choice impact degree. Simpler but less precise measure application results were then compared against these values, showing significant bias existence with classical explanatory model consideration. Correct methodology application for marginal changes in the presence of non-marginal policies also detected significant biases. Proportional error, in the first case, does not depend on policy size, but, in the second case, grows with their impact. There was another databank that the authors also simulated with non-linear income utilities and some welfare measures were again calculated. That no significant biases exist if there is no consideration of income effects was found by the authors, as was that there is no systematic relation between policy size (in terms of change) and these biases.