Modelling the generation of traffic, whether we consider the number of trips or tours made by individuals or households, or the flow of goods requiring transport, remains a challenging problem for transport planners. Thispaper presents some new findings that should help to reduce the problem. The central problem in modelling generation is to represent the impact of accessibility on trip rates (or freight flows). The correlation between accessibility and traffic volumes has been established at the level of socio-economic variation for many years (car owners travel more than non-car-owners) and at the level of geographical variation in recent years. The latter connection is often weak in urban and regional studies, although there is no doubt that such a connection exists, but accessibility can be quite strong in influencing travel rates in specific corridors, particularly when tourism is involved. When choice models are developed for other aspects of travellers' choices, accessibility can be measured quite simply by a 'logsum' taken from those other models, giving the average utility of travelling. The logsum incorporates both socio-economic and geographical variation. In a previous paper by the first author presented at this conference in 1997, it was suggested that there were many advantages in the use for predicting generation of a log-linear model, in which the log trip rate is related to a linear function of accessibility. This model is usually termedexponential, as the trip rate is a function of the exponential of accessibility. However it was not possible in that paper to establish a clear link between the exponential model and utility maximisation, so the theoretical justification for that model form was based only on its approximation to a logit model of the choice to travel or not. In the present paper, the utility theory underlying the exponential model is clarified, so that it can be used as part of a complete structure of utility-maximising choice models. A further issue in modelling traffic generation is to establish consumer surplus. If there is an improvement in accessibility, then current travellers will get the complete benefit of that change - the logsum can be used as a measure of the benefit they will obtain. But newly generated traffic will also obtain benefit, and although the 'rule of a half' can be applied to the logsum improvement to obtain an approximation it is better and simpler to use an exact calculation of the benefit obtained. An exact measure for the exponential model is derived and it is shown that this has the properties that would be expected. Finally, attention is devoted to theissue of estimation of these models. The difficulty of using stated choice data is discussed and methods are suggested by which these problems can be overcome. Issues arising in the use of revealed preference data, such as the relevance of 'Poisson regression' relative to geometric analysis, are also discussed and recommendations are made for future practice. For thecovering abstract see ITRD E135582.
Samenvatting