Wandering ideal point models for single or multi-attribute ranking data : a Bayesian approach
Wandering Ideal Point Models for Single or Multi-Attribute Ranking Data A Bayesian Approach
submitted by LEUNG HIU LAN
for the degree of Master of Philosophy
at The University of Hong Kong in December 2003
Ranking data appear in our everyday life. Proper statistical analysis of ranking data helps us to study the individual preference behaviour. In a process of ranking k items, a sample of n judges are asked to rank the items according to a certain attribute / preference criterion, resulting with a set of (single-attribute) ranking data given by the judges. In addition, multi-attribute ranking data are usually collected in which judges are asked to rank the same set of items according to several attributes / preference criteria.
Very often, the first step of statistical analysis is data visualization. The wandering ideal point models, independently developed by De Soete, Carroll and DeSarbo, and B6ckenholt and Gaul in 1986, have been widely used to analyze
paired comparison of items. It visualizes individual preferences and the positions of items on the same graph, so that their relationship can be easily seen. However,
the extension of the model to incorporate single or multi-attribute ranking data is not simple. This is because the calculation of the likelihood function of the model in this case involves high dimensional integration which is numerically unstable and its complexity increases with the number of items to be ranked.
To overcome this problem, in this thesis, we adopt a Bayesian approach via Markov Chain Monte Carlo methods to estimate the wandering ideal point model for displaying single or multi-attribute ranking data. Simulation studies are carried out to demonstrate the computational efficiency of our methodology. The proposed model is applied to a ranking dataset collected from a clinical practice survey in which each respondent was asked to rank potential barriers to the computerization of clinical practice.
School:The University of Hong Kong
School Location:China - Hong Kong SAR
Source Type:Master's Thesis
Keywords:ranking and selection statistics bayesian statistical decision theory markov processes monte carlo method
Date of Publication:01/01/2004