Transferability levels analysis of mode choice models (case study: Shopping Trips)

Recently, transportation planning has emphasized on methods that minimize costs of studies. Some researchers have introduced application of a model calibrated based on the data of city A, to a different city B. Although, one cannot expect perfect applicant of observations in the new city B, model transferability analysis can provide useful see sight. This paper introduces different forms and levels of transferability and also measures of transferability of mode choice models. A shopping trip in real scale at four levels of transferability is studied (transferring Tehran mode choice model to Qazvin). Morning peak shopping trips with three modes; private car, taxi and bus are studied, employing multinomial logit model. In order to set the ground for meaningful comparison of results, Tehran model structure was calibrated using Qazvin data. The model was discerned transferable at first, second and third level (except the variable of Central Business District) but not in fourth level. Transferability test statistics show inequality of variable coefficients for the two models at this level. Transfer Index analysis results the inappropriate of Tehran model applied for Qazvin due to lower goodness of fit measures. Moreover, relative error measures indicate relatively large difference between observed and predicted shares of different modes. Different social, economic and traffic condition and the effect of geographic terms and cultural manners on shopping trips would be part of reasons.

Mots clés

Transferability; Mode Choice Models; Spatial; Shopping Trips.