Citation

Badoe DA, Miller EJ. Transp. Plann. Tech. 1998; 21(4): 235-261.

Copyright

(Copyright © 1998, Informa - Taylor and Francis Group)

DOI

10.1080/03081069808717611

PMID

unavailable

Abstract

The prevailing practice in travel demand modelling is to estimate disaggregate models of mode choice with data from the most recent cross?sectional travel survey available on an urban area for forecasting purposes. Very often, however, most urban areas have available data from older cross?sectional surveys, which are often entirely ignored in the modelling effort. This paper explores the possibility of pooling data from two independent cross?sectional travel surveys on the same urban area for model estimation and forecasting by applying a model structure which allows for transfer?bias, referred to as the joint context estimation procedure. This procedure consists of joint, full information maximum likelihood estimation of a related set of logit choice models for the contexts which are based on the following two assumptions: (1) differences in model parameter values between contexts are expressible in terms of differences in the contexts? alternative?specific constants and overall scale of the contexts? utility functions; and (2) aside from these differences in alternative?specific constants and scales, model parameters are common across contexts. An empirical case study is presented, involving the use of two datasets, gathered 22 years apart (1964 and 1986) for the Greater Toronto Area (GTA), to estimate morning peak period work trip mode choice models. The estimated models are applied in prediction tests on the 1964, 1986 and a third independent data set, the 1991?data, also collected in the GTA. The performance of the joint context models is compared to that of an independent model, estimated on the 1986 data only. The results clearly demonstrate that joint context estimation dominates the independent 1986?model in predictive performance. The paper concludes by briefly discussing the possible roles which joint context estimation might play in the development of improved transferability of disaggregate choice models.