Standard practice in connection with route choice is to use a mean VTTS for each user class (e.g. commuters, business, and commercial vehicles). However in the context of route choice models where toll roads are present and compete against non-tolled (free) alternatives, it may be expected that there is a distribution of VTTS within the travelling population and the proportion choosing a particular route will depend on the distribution, not merely the mean value. A distribution will therefore lead to a more reliable balance of traffic on each route.  This paper shows that, using Stated Preference (SP) techniques, a lognormal distribution can be successfully estimated and implemented, and discusses the performance relative to alternative model formulations.

The outcome distribution is sensitive to the exclusion of “non-traders”. This is a topical issue in SP studies and there is no universal agreement on how they should be treated.  For this reason, the paper analyses the non-traders in some detail, in the first place in relation to their real life choices. In additional, a unique feature of this study was the design of two linked SP experiments, the first offering time and cost as the attributes, and the second offering the same attributes but disaggregating the total time into three qualities of time. This enabled an analysis of how the respondents’ stated choices altered when further information was provided about the quality of time, and provided a further check on non-traders. It was concluded after a thorough analysis that the majority of the non-traders were not consistent with utility maximisation and, for the purpose of estimating willingness to pay, should therefore be removed. The resultant impact on the estimated model was a significantly better model fit and more intuitively acceptable results.