Making the most of assignment models or is iteration boring?

This paper discusses how to get the best out of strategic traffic assignment models. However, while there are many sources of inaccuracy in making traffic forecasts with strategic models, this paper shows that inadequate use of the assignment sub-models contributes as well. It is expected that an assignment model will return the same result if repeated (unless it is a Stochastic form of model with random choice cost variables) but this paper will show that it may be quite different if repeated with one more cycle, some traffic volumes being more than 25 per cent different. It focuses on the need for two forms of modelling iteration. They are (a) Iteration within the assignment process, which is called cycles in this paper, and (b) Iteration between trip generation and traffic assignment. The main questions are how many iterations or cycles are necessary ? to get the best results and how do we tell when the best results are achieved ? A secondary question posed in the paper is which is the best form of assignment model ? These questions are addressed using a congested model of future Sydney, which possibly exaggerates the conclusions but clearly illustrates the principles involved. (a) For the covering entry of this conference, please see ITRD abstract no. E214666.