Between 1987 and 1984, the Dutch Ministry of Transport conducted research studies aimed at developing a predictive model system able to produce detailed forecasts of travel demand. The model was multimodal, analysable by market segment, and applied to assignment in car and rail. The model system covered the whole of The Netherlands, and is called the National Model System (NMS). The model system was later extensively extended and refined, and is still in use. It now has published aggregate statistics, against which to assess results from the 1985 projections. This paper reports research which uses these statistics to: (1) assess the accuracy of the first predictions; (2) analyse the major factors contributing to the discrepancies between forecast and outcome; (3) assess how far later improvements of the NMS would have improved the predictions; and (4) identify the areas where future research could resolve the remaining problems. The five factors, for which adjustments should be made, are scale effects, car ownership effects, cost effects, travel time effects, and `system shocks'. The paper aims to formulate a structure in which predictive tools can be monitored for shortcomings, from which the potential for incremental improvements can be identified. It also quantifies errors from different sources.
Abstract