Some issues related to the estimation of transport models

This document deals with some issues essential to the estimation of logit models to be implemented in transport model systems. First, we investigate problems caused by measurement errors in variables. This is done by extensive use of Monte Carlo simulations where different types of measurement errors are introduced to essential variables in model estimations and compared to models estimated with "correct" data. The result are very similar to those reported for standard econometric methods. Second, a simple solution to the problem of heteroskedasticity, in terms of varying variance with travel distance, is suggested. There seems to be an improvement in models estimated with utility functions scaled by a function of straight line distance between centroids, compared to models estimated without such scaling. Third, we have shown that most of the well known problems in the Norwegian National Transport Model System, can be amended when addressed by the right amount of resources. The problems addressed are issues like formulation of sentral variables as frequencies and cost/income, problems of segmentation, and problems of model structure.