Abstract
An empirical study of the temporal transferability properties of entropy-type trip-distribution models, applied within a fixed geographic region, is presented. Data for the study were drawn from three travel surveys conducted in 1964, 1986, and 1996 in the Toronto region. Very long transfer periods were examined, and a wide variety of model specifications were tested. These specifications ranged from the simple proportional flow model to an occupationally stratified, doubly constrained entropy model, for which spatial separation was measured by highway travel time. These models were estimated using data from 1964 and 1986. The estimated 1964 models were transferred to 1986 and 1996 to predict the distribution of travel. The 1986 models also were transferred to 1996 to predict the distribution of travel. Results showed that the travel-time parameter was not temporally stable. However, pragmatically, the transferred models were found to provide forecasts very comparable to those generated by models estimated in the respective application contexts. Improved model specification consistently resulted in improved precision of the forecasts obtained. In particular, stratification of the data by worker occupation category resulted in models with the best fit to estimation data, as well as forecasts. However, from a decision-oriented perspective, the improvement in precision obtained from additional complexity in specification did not warrant a recommendation for use of more complex models. Thus, the simple, doubly constrained entropy model with spatial separation measured by highway travel time appears satisfactory for use in practical modeling efforts.
Get full access to this article
View all access options for this article.