Temporal transferability of work trip mode choice models in an expanding suburban area: the case of York Region, Ontario

This paper presents an investigation of the transferability of home-based work mode choice models in the context of a rapidly growing suburban area: the Regional Municipality of York in the Greater Toronto Area. Between 2001 and 2006, York Region experienced a rapid change in population and saw the introduction of new transit mode. With a wealth of revealed-preference household survey data from the Transportation Tomorrow Survey, there is an obvious opportunity to investigate whether there were any structural changes in travel behaviour amongst the region's residents. Three heteroskedastic generalised extreme value (GEV)-class choice models are estimated: one for 2001, one for 2006 and a model estimated using data pooled from the 2001 and 2006 data sets. Disaggregate and aggregate transferability tests are conducted. Disaggregate transferability refers to the ability of a model to predict the individual choices observed in the context of application and is measured, in absolute terms, by the transfer log-likelihood. Aggregate transferability refers to the ability of a model to predict overall trends in the data (e.g. mode share). It becomes clear that the sets of estimation parameters are statistically different before and after the new bus transit system introduction. This implies that even advanced heteroskedastic GEV models are not fully transferable. However, interestingly, when assessing aggregate transferability, it is found that the transferred models perform quite well; in some cases, the transferred models fit the data better than the original estimated model. This suggests that disaggregate choice models capable of addressing both the systematic and random effects of transportation and land-use changes on choice-making behaviour should be developed.