A method is proposed in this paper for more systematic validation of traffic models which tests the assignment and trip matrix separately. While it is recognised that traffic models derive from a process of observation and analysis based on sound statistical design, current practice gives inadequate attention to the problems of validation. Two stages are needed, validating the assignment and then the trip matrix. Assignment does not raise statistical issues. Adequacy, rather than accuracy, of the underlying assumptions is sought. Sensitivity analysis - varying parameters of generalised cost, link costs and variance of perceived cost - is the requisite tool. Existing assignment methods are unsuitable for this purpose and further development is indicated. A set of flows, unaffected by uncertainties in the assignment, can then be chosen to provide statistical measures of possible deviations from the modelled trip matrix. The two technical difficulties in this programme - biased flows and insufficient information to specify the trip matrix fully - are the standard problems of using judgemental samples and matrix estimation. They can be tackled simultaneously, using jack-knife estimation techniques to obtain confidence intervals for a number of statistics: from total trips and mileage, disaggregated to origin and destination zone sub-totals and finally to goodness of fit for individual trip matrix cells. (A) This paper is also published in Sminar E, (Volume P307) p. 1-11. (see IRRD 842404).
Abstract