The sampling effect on the value of travel-time savings: estimation by discrete choice models on Tunisian data

This paper examines how the value of travel time savings changes with the individual socio-economic variables and the transportation system performance. The individuals’ perception of the value of time depends on three key factors: age, sex, income. The Multinomial Logit model underestimates the value of travel time savings unlike the Multinomial Probit model and the MMNL model. In the case of discrete choice models allowing for random taste heterogeneity, a Mixed Multinomial logit (MMNL) is used to estimate the travel-time and the cost coefficients. We discuss the consistency of such estimates where we consider distributions with fixed bounds and unbounded distributions. In each case, we note that the MMNL overestimate the value of travel-time savings. To guarantee the robustness of estimation, it is preferable to use unbounded distributions, especially the Normal distribution (bounds are estimated from data during the calibration). In order to know the effect of the sample criteria on the result of the model fit, we divided the complete sample into many segments and we estimate an MMNL model for each one. Following the segments’ criteria, we compare the advantage of a larger sample and the effect of the segmentation on the performance of the MMNL model.