BIOGEME a free package for the estimation of.pdf

BIOGEME: a free package for the estimation of discrete choice models Michel Bierlaire, EPFL Conference paper STRC 2003 STRC 3rd Swiss Transport Research Conference Monte Verita, Ascona, March 2003 2 BIOGEME: a free package for the estimation of discrete choice models Michel Bierlaire, Dpt of Mathematics, EPFL, Lausanne Phone: +41.21.693.25.37 Fax: +41.21.693.55.70 Email: michel.bierlaire@epfl.ch Abstract Keywords: logit model, cross-nested model, GEV model, random utility, trans- portation demand Swiss Transport Research Conference – STRC 2003 – Monte Verita 1 Introduction Discrete choice models in general, and random utility models in particular, have been extensively used in several fields of applications for the last three decades. The theoretical derivation of these models is well documented in the litera- ture (Luce, 1959, McFadden, 1981, Ben-Akiva and Lerman, 1985, Anderson et al., 1992, Hensher and Johnson, 1981, Horowitz et al., 1986, Bierlaire, 1998, Ben-Akiva and Bierlaire, 1999). Recently, new sophisticated models have been proposed in the literature. Among them, two main categories can be identified: Generalized Extreme Value models, and mixed logit models. Generalized Extreme Value (GEV) models have been proposed 25 years ago by McFadden (1978). It is actually a family of models, consistent with random utility theory. Since then, only a few members have been exploited, the Multinomial Logit model and the Nested Logit model being the most popular (Ben-Akiva and Lerman, 1985). Recently, research on the Cross-Nested logit model (Small, 1987, Vovsha, 1997, Vovsha and Bekhor, 1998, Ben-Akiva and Bierlaire, 1999, Papola, 2000, Bierlaire, 2001, Wen and Koppelman, 2001, Swait, 2001) has extended the number of GEV models used in practice. Also, Daly (2001) and Bierlaire (2002) have proposed new theoretical results providing an operational representation of GEV models. Mixed or hybrid logit models (Ben-Akiva and Bolduc, 1996,McFadden and Train, 1997,Bhat, 2001,Be