An Integrated Approach to the Estimation of the Value of Travel Time



An Integrated Approach to the Estimation of the Value of Travel Time

Authors

M Fosgerau, K Hjorth, S Vincent Lyk-Jensen, Danish Transport Research Institute, DK

Description

We present an integrated approach to deal with some issues arising in the estimation of the VTTS: Apply a novel model formulation, test the distribution of the random VTTS, control for experimental design, trip characteristics and socioeconomics.

Abstract

The paper reports on the modelling carried out in the Danish Value of Time Study in relation to a binary route choice experiment using a sample of more than 6000 respondents, covering seven main modes.
The modelling effort had to deal with a number of issues that arise in these circumstances. The aim of the paper is to describe the application of an integrated approach to achieve this.
First, the main driver for individual responses is the individual value of travel time savings (VTTS). Based on recent work by Fosgerau (2005) we specify models that contain a separate parameterisation of the VTTS by background variables in addition to a random component. The log of the VTTS then enters a panel binary mixed logit model corresponding to the choice data such that choice specific errors are multiplicative relative to the VTTS. This specification works extremely well, allowing us to estimate a large number of significant parameters for the parameterisation of the VTTS.
The VTTS is determined up to the random component by three groups of variables, relating to the experimental design, trip characteristics and socioeconomics. Concerning experimental design we find strong influences of gains and losses and the size of the time saving, consistent with prospect theory, but no anchoring effects. Van de Kaa (2005) suggested that responses would be affected by the numerical scale of times and costs, which could explain the low increase in the estimated VTTS between subsequent experiments using the same design. We test whether subjects who are given a larger threshold value in the first choice situation (the anchor) will tend to have a larger VTTS, but find no evidence of such an effect. This is a positive result implying that we are able to control for the experimental design in deriving average VTTS estimates. Among the trip characteristics we examine mode, trip duration and the share of congested time (in car) and argue that the latter should be interpreted as a socioeconomic indicator. Among the socioeconomics, we investigate the effect of income, age, sex, family type, trip purpose, employer-paid trips and interview type on the VTTS. The latter is relevant since 70% of the interviews were carried out over the internet while the remainder were face-to-face.
The second main issue is the distribution of the VTTS. It has been recognised (Fosgerau 2006) that distributional assumptions regarding the random component are crucial for the resulting estimates of the mean VTTS. It may happen that the data fail to identify the VTTS distribution, in which case practically any result may be obtained between a certain minimum and infinity. We demonstrate how we check for this cause of error. It may also happen that the assumed distribution of the random component simply does not fit the data. Again heavy bias may result. We apply a test from a recent paper by Fosgerau & Bierlaire (2005) to test our assumption of the random component of log VTTS.
Finally, as a check on the validity of our results, we compare our estimates of mean VTTS to income data.

Publisher

Association for European Transport