THE SPATIAL TRANSFERABILITY OF THE HELSINKI METROPOLITAN AREA MODE CHOICE MODELS

Abstract

The purpose of research was to study spatial transferability of mode choice models for other home-based trips using mobility surveys conducted in Helsinki Metropolitan Area in 1995 and in Turku region in 1997. The Helsinki Metropolitan Area database is used to estimate models that are to be transferred. The database in Turku region represents application context to which estimated Helsinki Metropolitan Area models are transferred. The transfer procedures examined were Bayesian updating, combined transfer estimation, transfer scaling, and joint context estimation. To explore impact of sample size on transferring performance, model transferability was tested using three to four different sample sizes. The model transferability was examined by comparing transferred models to models estimated using entire set of data that can be regarded as best estimate representing the real situation. The results showed that joint context estimation was generally best method. In particular, method was useful, if transfer bias was large or only some of model coefficients could be regarded as precise. The transfer scaling also gave rather good results. The problem with this method was unpredictability of quality of results. The scale factor is usually estimated for travel time and number of transfers. If ratio of these coefficients does not stay constant, model's ability to predict effects of changes in transportation system may become rather weak. The combined transfer estimation procedure was best when there was a large number of observations and transfer bias was small. The Bayesian approach emphasizes coefficients with respect to inverse of variances of each coefficient. Therefore, coefficients based on estimation context data may be emphasized too strongly in model transferring, since number of observations is larger in this context.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 15 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.