New qualitative choice models incorporating individual and choice characteristics
Résumé
In the econometric framework, multinomial and conditional logit models are the most usual regression models for qualitative choices. They differ by their parametrization while sharing the canonical link function. This link function can be decomposed into the reference ratio of probabilities and the logistic cumulative distribution function (cdf). We propose to conserve the reference ratio, appropriate for qualitative choices, but to select the cdf among an enlarged family containing the Student cdf for instance. These new qualitative choice models often outperform logit models in terms of likelihood and error rate of classification and stay easily interpretable. This is illustrated with a benchmark dataset of travel demand between Sydney and Melbourn.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...