Bayesian modeling of group preferences and population utility estimation

In this dissertation, a Bayesian framework is developed for modeling uncertainty about a population utility function. Such a framework is motivated by the need to analyze preference-based measurement data that arise from evaluation of health states by individuals. The Bayesian framework leads to population utility estimation and health policy evaluation by introducing a probabilistic interpretation of the multiattribute utility theory (MAUT) approach used in health economics. In so doing, the approach combines ideas from MAUT approach of Raiffa and Keeney (1976) and Bayesian statistics in order to provide an alternate method of modeling preferences and utility estimation.; Treatment of uncertainty in population utility function is achieved by specifying probability models such as the ordered Dirichlet distribution for attribute utilities and Dirichlet or beta distributions for multiattribute utility function coefficients. Choice of these probability models are motivated by the properties of the utilities and utility function coefficients. For example, the ordered Dirichlet distribution provides a flexible model to describe uncertainty about utility by allowing us to reflect different attitudes towards risk with respect to a given attribute.; Bayesian inference procedures are developed for the above probability models using Monte Carlo Markov Chain (MCMC) methods. Both the additive and multiplicative representations of multiattribute utility function are considered. A reparameterization of the models using a logit transformation is introduced to incorporate the covariate, and random effects on utilities and multiattribute utility function coefficients. Furthermore, an information theoretic approach is presented to assess the expected information gain from additional utility elicitations.; The models and procedures are applied to real world data from a Health Utility 11 survey.