Browse by author
Lookup NU author(s): Professor Elisabetta Cherchi
Full text for this publication is not currently held within this repository. Alternative links are provided below where available.
The random coefficients Logit model allows a more realistic representation of agents’ behavior. However, the estimation of such model may involve simulation, which may become impractical when the number of random coefficients is large, because of what is known as the curse of dimensionality. In this article, we compare the traditional Maximum Simulated Likelihood (MSL) method with the alternative Expectation-Maximization (EM) method that does not require simulation. Previous literature had shown that, for cross-sectional data, MSL outperforms the EM method in terms of ability to recover the true parameters and estimation time, and that EM has more difficulties in recovering the true scale of the coefficients. In this article, we extend the analysis from cross-sectional data to the less volatile case of panel data to explore the impact in the relative performance of the methods of having several realizations of the random coefficients. Using a series of Monte Carlo experiments, we found evidence suggesting four main conclusions. (1) Efficiency increases when the true variance-covariance matrix becomes diagonal. (2) EM is more robust to curse of dimensionality both in terms of efficiency and estimation time. (3) EM does not recover the true scale neither with cross-sectional nor with panel data. (4) EM systematically attains more efficient estimators than the MSL method. This implies that, if the purpose of the estimation is only to determine the ratios of the model parameters (e.g. the value of time), the EM should be preferred. For all other cases, MSL should be used.
Author(s): Cherchi E, Guevara CA
Publication type: Article
Publication status: Published
Journal: Transportation Research Record
Year: 2012
Volume: 2302
Issue: 1
Pages: 65-73
Print publication date: 01/01/2012
Online publication date: 01/01/2012
ISSN (print): 0361-1981
ISSN (electronic): 2169-4052
Publisher: Sage Publications, Inc.
URL: https://doi.org/10.3141/2302-07
DOI: 10.3141/2302-07
Altmetrics provided by Altmetric
