Modelling and Forecasting with Latent Variables in Discrete Choice Panel Models

Modelling and Forecasting with Latent Variables in Discrete Choice Panel Models


M.F. Yáñez, S. Raveau, M. Rojas and J. de D. Ortúzar, Pontificia Universidad Católica de Chile, CL


Based on the Santiago Panel survey, we identified three LV: reliability, accessibility, and comfort. We obtained superior models in comparison to those where LV are not considered. Moreover, we discussed about using the new models in forecasting.


Modelling the choice of transport mode is a central element in travel demand
estimation. Traditional modelling has used mainly objective modal attributes,
such as travel times and cost, as explanatory variables. Nevertheless, it is well
known that attitudes and perceptions may also influence users? behaviour. The
use of hybrid discrete choice models constitutes a good alternative to
ncorporate the effect of these subjective factors, as they allow the inclusion of
both tangible alternative attributes and latent variables associated with attitudes
and perceptions. Data available to us, in a panel context, included users?
perception about some aspects of their available transport modes and allowed
us to generate indicators of these perceptions. With this data we estimated
hybrid discrete choice models in a panel context. This paper analyses the
results of applying hybrid discrete choice models to a real urban study case,
with several alternatives, and also an approach to forecast using these models.
Our results clearly show that hybrid models are superior to traditional models
hat ignore the effect of subjective attitudes and perceptions.


Association for European Transport