This thesis discusses possibilities and limitations of describing and planning relevant aspects of commuters' modal split through mathematical modelling, especially using multinominal logit models. The relevant theory is discussed, and data sets are analysed, both collected by means of a journey-to-work survey (comprising on-board interviews on train and bus, respectively, and licence plate numbers annotated on the motorway) and produced with Monte Carlo simulations, according to predetermined choice strategies, mainly based on attributes sampled from the data set obtained in the travel survey. Logit models were estimated from the simulated data sets, and the conclusions from the choice strategy and the estimated models were compared. A bias was found that seems to be systematic in overestimating the absolute value of parameters, even under perfect modelling conditions. The effect increases with decreasing sample size, but is quite large enough to be significant already at sample sizes as large as n=210. (A)
Abstract