Longitudinal Assessment of Transportation Planning Forecasts

Abstract

Transportation planning applications typically calibrate models for a base year and then make forecast year projections. However, modelers rarely evaluate the accuracy of their forecasts by using past data sets to predict present conditions. This is complicated by the fact that longitudinal data sets for a geographical area exhibit data incompatibility, shifts in planning emphasis, temporal changes in travel characteristics, and added modeling complexity. In addition, if new data elements are needed for model improvement, these will not be available with the historical data set. In an effort to test the feasibility of using older planning data to predict the present, a case study area for which transportation planning data were available was selected at three points in time over a 25-year period. This facilitated comparison of longitudinal data sets, development of base year models, and subsequent testing of their performance for a forecast year application. This paper discusses the experience of creating a longitudinal data set for such testing and illustrations how one can use longitudinal data to ascertain the sustainability of model structures over time. Some simple workable approaches are illustrated; although these are more aggregate than desired, they convey how one may devise models that can function both in the base year and then, later, in the forecast year, in spite of changes in transportation and land use activity.