Generalized multinomial probit Model: Accommodating constrained random parameters

doi:10.1016/j.trb.2018.10.019

Return

Generalized multinomial probit Model: Accommodating constrained random parameters

2018-12-01

PRE

Rajesh Paleti *

DOI：10.1016/j.trb.2018.10.019

Original

Original request for help

Save

Abstract

En 中文

Skewed and constrained distributions for model parameters are better suited to provide realistic taste sensitivity and Willingness-To-Pay (WTP) distributions in many empirical applications. In this context of random taste heterogeneity, the standard multinomial probit (MNP) has seen limited applicability largely owing to the normal distributional assumption of model parameters that will invariably result in (a) counter-intuitive taste sensitivities for a significant proportion of the population, and (b) WTP distributions without finite moments. In this paper, a Generalized MNP (GMNP) model that allows constrained random parameters with multivariate truncated normal distribution was developed. The ability of the maximum simulated likelihood inference method to retrieve the model parameters was demonstrated on synthetic data using the GHK simulator with quasi-Monte Carlo sequences. The bias in the parameters of the MNP model that ignores the constraints on model parameters was also demonstrated. Also, the proposed model was used to analyze car parking preferences in Jerusalem while ensuring (a) negative taste sensitivities for cost and time attributes, and (b) finite moments of the WTP measures associated with walk and search times. (C) 2018 Elsevier Ltd. All rights reserved.

Keywords:

Multinomial probit

Constrained random parameters

Multivariate truncated normal distribution

GHK simulator

Quasi-monte carlo sequences

Parking choice

AI Summary

Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.