The development of transport demand modelling can be described as a search of flexible models adapting to a greater number of practical situations. However, this search has been characterised by a flexibility-estimability trade off. In one hand, there are the traditional models of the Logit family that offer closed choice probabilities, but with restrictive assumptions that not always are properly justified. On the other hand, the Probit model allows to work with an error structure general in principle, but its estimation is quite complex and subject to identificatio n restrictions. In this context, in addition by technological advances in term of computer's power and numerical methods, the use of simplified models has been questioned and it has appeared with force a new alternative of modelling: the Mixed Logit model. In this paper we study both theoretically and empirically the antecedents that sustain the formulation of Mixed Logit model. Through an analysis of the covariance matrix we discuss how these models are able to model conditions in which independence and homoscedasticity are violated. This analysis is complemented with two numerical applications that allow to verify the real possibility of using this model and its capacity to adapt to practical situations. In the simulation experiments data bases are constructed so that it allows to objectively control the goodness of fit of the model, the reproduction of the calibration sample and the level of answer to changes in the attributes of the alternatives. The application with real data tries to validate the empirical study and to verify the feasibility to apply sophisticated econometric tools. Although its estimation requires simulation, it is observed that in general the model gives to a suitable reproduction of parameters and a good adjustment to the changes of policy. We conclude that Mixed Logit models constitute an interesting and powerful alternative for discrete choice modelling. Nevertheless, as in the case of any flexible model, it is necessary to be rigorous in the construction and implementation of a particular specification, justifying suitably the any assumption done and knowing clearly its consequences previous to the estimation of the parameters.
[1]
D. McFadden.
Conditional logit analysis of qualitative choice behavior
,
1972
.
[2]
Hugh F. Gunn,et al.
Statistical aspects of travel demand modelling
,
1982
.
[3]
H. Williams,et al.
Behavioural theories of dispersion and the mis-specification of travel demand models☆
,
1982
.
[4]
V. Hajivassiliou,et al.
Smooth unbiased multivariate probability simulators for maximum likelihood estimation of limited dependent variable models
,
1993
.
[5]
Marcela Munizaga,et al.
NESTED LOGIT MODELLING: SOME HARD FACTS
,
1999
.
[6]
K. Train,et al.
Forecasting new product penetration with flexible substitution patterns
,
1998
.
[7]
Marcela Munizaga,et al.
ON THE ERROR STRUCTURE OF DISCRETE CHOICE MODELS.
,
1997
.
[8]
C. Bhat.
Quasi-random maximum simulated likelihood estimation of the mixed multinomial logit model
,
2001
.
[9]
D. McFadden,et al.
MIXED MNL MODELS FOR DISCRETE RESPONSE
,
2000
.
[10]
N. S. Cardell,et al.
Measuring the societal impacts of automobile downsizing
,
1980
.
[11]
Marcela Munizaga,et al.
ON THE APPLICABILITY OF THE MULTINOMIAL PROBIT MODEL
,
1997
.
[12]
Paul A. Ruud,et al.
Handbook of Econometrics: Classical Estimation Methods for LDV Models Using Simulation
,
1993
.
[13]
K. Train.
Halton Sequences for Mixed Logit
,
2000
.
[14]
H. Williams.
On the Formation of Travel Demand Models and Economic Evaluation Measures of User Benefit
,
1977
.
[15]
J. Boyd,et al.
THE EFFECT OF FUEL ECONOMY STANDARDS ON THE U.S. AUTOMOTIVE MARKET: AN HEDONIC DEMAND ANALYSIS
,
1980
.
[16]
Staffan Algers,et al.
Mixed Logit Estimation of the Value of Travel Time
,
1998
.