Nonparametric and semiparametric estimation of the distribution of the value of travel time

This paper applies nonparametric and semiparametric methods to investigate the distribution of the value of travel time (VOT) from experimental binary choice data using minimal prior distributional assumptions. Using a large dataset, results indicate that the VOT is approximately distributed as lognormal within the range for which observations exist. This implies that the lognormal would be suitable for prediction. The analysis reveals, however, that the experimental design does not allow observation of the right tail of the VOT distribution and it is thus not possible to infer the mean VOT from these data. Different assumptions regarding the right tail have a strong impact on the resulting mean VOT. This problem is likely to occur also with other data and may explain why different specifications of mixed logit models can lead to very different expected values of time.