ESTIMATING VALUE OF TIME AND FORECASTING TRANSPORT CHOICE IN ROAD FREIGHT WITH A NONLINEAR PROFIT. THE LOGIT MODEL VERSUS NEURAL NETWORKS
ers Bergkvist · 2000
The main aim of the work presented here is to estimate the value of time (VOT) for road freight in Sweden with methods that will also be used to forecast transport choices. We have reasons to assume that the VOT among companies may be non-linear related to transport time and other attributes of a transport mode. We compare estimates obtained from two-logit model specifications, and two neural networks (NN). The VOT may be derived analytically in the logit case, but not in the case of the neural network. For the latter the inherently nonlinear structure of the data is exploited. The NN are first trained and then used in numerical simulations to estimate VOT. Given these, the models can be compared. The estimated VOT from the two logit models are quite close. For the NNs and the nonlinear logit model, it is not possible to derive just one VOT or to argue for which should be used. Neither are there any conventions that could be used for the nonlinear freight transports. VOT decreases as the model fit improves. Finally, the non-global convergence of the used NNs is commented on. For the covering abstract and the URL of the conference see ITRD E205769. (A)