The current version of the Norwegian National Transport Model (phase 4) is estimated on travel survey from 1991/92. The model is based on a series of disaggregate choice submodels including driver licence models, car ownership models, short distance travel models and long distance travel models. The forecasting system uses a prototypical sample procedure. The geographical unit of the phase 4 model is Norwegian municipalities. In 2000, the fifth phase of model development commenced. As part of this programme new long-distance models were estimated using survey data collected in 1997/98. The estimation of the new long-distance models used new and more detailed network data, both for roads and public transport. Part of the network improvements was the specification of more detailed zones: the 435 municipalities in the phase 4 model were increased to approximately 1400 zones in the revised models, where the zones aggregate into municipalities. The 1997/98 survey data contained detailed zonal information about the observed trip (tour) origin; however, the destination choice information was only known at municipality level. In the study, three approaches were tested to incorporate the detailed destination choice information into the models: using a weighted average of the level-of-service (LOS) to each zone within each municipality; sampling a random zone within each municipality and using the LOS for the sampled zone for that municipality; extending the mode and destination tree structure such that the zonal destination choice was below the municipality and where the choice was specified at municipality level. The third approach with the extended tree structure resulted in the best model fit and was therefore recommended as the best way in which to incorporate the detailed level-of-service data in the model structure. It is interesting to note that the first approach, using averaging, gave a better model fit than did the second approach using sampling. Another difficulty that came up during the model re-estimation was differences observed between the trip length distribution for the long-distance choice data, which was used for the estimation of the long-distance travel models, and trip diary data. It was felt that the divergences could be caused by differential reporting rates between the two surveys, and that the trip diary data was probably more reliable, and that these differences would cause bias in the resulting models. In order to compensate for these differences a Weighted Exogenous Sampling Maximum Likelihood (WESML) correction was applied in the estimation of the travel models. The model results suggested that the trip length distribution from the travel diary data better represented the true trip length distribution. In summary, the study highlights how advanced modelling techniques can be used to overcome data shortcomings. For the covering abstract see ITRD E124693.
Abstract