摘要: |
An empirical study is presented of the long-range temporal transferability properties within a fixed geographic area of disaggregate logit models of work trip mode choice. The study area is the greater Toronto area, Ontario, Canada. The two temporal contexts are 1964 and 1986, with models estimated from 1964 data being used to predict 1986 travel choices. In addition to the very long transfer period (which does not appear to have been previously examined), a major feature of this study is that a wide variety of model specifications, ranging from the simplest possible market share model to a complex market segmentation model, are tested to investigate the relationship between model specification and transferability. Major findings of the study include (a) as in most transferability studies, model parameters are not temporally stable; (b) pragmatically the transferred models provide considerable useful information about application context travel behavior; (c) in general, improved model specification improves the extent of the model's transferability; (d) an important exception to Point c is the complex market segment model, which appears to be "overspecified" and, in the face of changing contextual factors during the 22-year period predicts 1986 conditions quite poorly; (e) Point c notwithstanding, simple level-of-service models perform very well in terms of their spatially aggregate predictions (which are often of primary practical importance to planners); (f) the models that best fit the estimation context (1964) data do not always transfer the best to 1986 conditions; and (g) "transfer scaling", in which modal utility constants and scales are updated, can significantly improve model transferability. |