Publication Detail
Multinomial Probit Model Estimation Revisited: Testing of New Algorithms and Evaluation of Alternative Model Specifications for Trinomial Models of Household Car Ownership
|
UCD-ITS-RP-90-01 Presentation Series Download PDF |
Suggested Citation:
Bunch, David S. and Ryuichi Kitamura (1990) Multinomial Probit Model Estimation Revisited: Testing of New Algorithms and Evaluation of Alternative Model Specifications for Trinomial Models of Household Car Ownership. Institute of Transportation Studies, University of California, Davis, Presentation Series UCD-ITS-RP-90-01
Transportation Research Board
In this paper we revisit various important issues relating to practical estimation of the multinomial probit model, using an empirical analysis of car ownership as a test case. To provide context, a brief literature review of empirical probit studies is included. Estimates for a full range of models, including specifications with random (uncorrelated and correlated) taste variation and/or a general covariance structure for alternative-specific errors, are obtained using a recently developed maximum likelihood estimation routine for choice models. The results provide useful insights into specification and testing of probit models, and the MLE algorithm demonstrates many desirable features, including useful convergence diagnostics. Finally, a numerical comparison of Clark's approximation versus numerical integration provides additional evidence against the use of Clark's approximation in probit estimation.
In this paper we revisit various important issues relating to practical estimation of the multinomial probit model, using an empirical analysis of car ownership as a test case. To provide context, a brief literature review of empirical probit studies is included. Estimates for a full range of models, including specifications with random (uncorrelated and correlated) taste variation and/or a general covariance structure for alternative-specific errors, are obtained using a recently developed maximum likelihood estimation routine for choice models. The results provide useful insights into specification and testing of probit models, and the MLE algorithm demonstrates many desirable features, including useful convergence diagnostics. Finally, a numerical comparison of Clark's approximation versus numerical integration provides additional evidence against the use of Clark's approximation in probit estimation.