Bayesian estimation of Thurstonian ranking models based on the Gibbs sampler

Thurstonian ranking models represent the psychological ranking process by latent random variables that follow a multivariate normal distribution. To evaluate the ranking probabilities and estimate the parameters of the ranking models, traditional approaches such as numerical integration methods are only feasible for ranking problems with a small number of objects. This paper presents a Bayesian approach to the estimation of the parameters of Thurstonian ranking models based on Gibbs sampling methods. Monte Carlo studies demonstrate that the Gibbs sampler is applicable to ranking problems with a large number of objects. To improve the efficiency of the Gibbs sampler for estimating constrained and unconstrained Thurstonian ranking models, two procedures, importance sampling and truncated multivariate normal simulation procedures, are investigated. In an application, rankings of ten objects from a study on compound preferences (McKeon, 1961) are analyzed.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Psychology

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2011

Author dc:creator

• Yao, Kai-Ping Grace

Contributors dc:contributor

• Bockenholt, Ulf

• Statistics
• Psychology, Psychometrics

Statement dc:rights

• Copyright 1995 Yao, Kai-Ping Grace

Language dc:language

eng

Identifier

AAI9624544
(UMI)AAI9624544

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/20979