The Estimation of Aggregate Modal Split Models

In the last thirty years, there has been a considerable increase worldwide in the use of disaggregate models. These range from the seminal work of McFadden and Ben Akiva (Domencich, McFadden, 1975; McFadden, 2000; Ben-Akiva, Lerman, 1985) to more recent advances involving the use of mixed multinomial models (Bhat, 1997, 2000) whose application has been facilitated by the availability of a number of software packages. However, despite this undeniable progress, the disaggregate approach still requires a number of hypotheses and, in particular, a complex estimation process (Bonnel, 2002). Consequently, the aggregate approach is still widely used in many countries, in particular France where there has been little development of disaggregate modal choice models (there have been few really significant studies for urban areas: Abraham et al., 1961; CETUR, 1985; Daly, 1985; CETE de Lyon et al., 1986; Bouyaux, 1988; Hivert et al., 1988; RATP, Cambridge Systematics, 1982; Rousseau, Saut, 1997; CERTU, 1998b). Although the development of aggregate modal choice models is fairly simple as far as formalisation is concerned, the same cannot really be said to apply for their estimation. The samples in the travel surveys that are used for these estimations are generally too small for the needs of aggregate models. It is therefore usually necessary to conduct zonal aggregations which result in a high degree of uncertainty about the quality of the estimations. In response to this situation this paper proposes an estimation method which allows very small-scale zoning to be retained. The paper starts with a description of the estimation problem for aggregate models (Section 1) and then describes the model we have developed (Section 2). We then present an application of this method for estimating an aggregate modal choice model for the Lyon conurbation. This empirical analysis allows us to study the benefits of our method in comparison with more conventional methods (Section 3).