Date of Award

2015

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Public Health Sciences

College

College of Graduate Studies

First Advisor

Andrew Lawson

Second Advisor

Christel Faes

Third Advisor

Russell Kirby

Fourth Advisor

Elizabeth Garrett-Mayer

Fifth Advisor

Bethany Wolf

Abstract

In disease mapping where predictor effects are to be modeled, it is often the case that sets of predictors are fixed and the aim is to choose between fixed model sets. In this dissertation, I focus on this dimension reduction objective by applying the Poisson data model commonly used for disease mapping of small area health data. I begin with a software comparison of the recently developed R package INLA (Integrated Nested Laplace Approximation) to the MCMC approach by way of the BRugs package in R, which calls OpenBUGS. This software comparison leads to choosing the appropriate platform for carrying out the second portion of this work: a methodology comparison of my proposed non-spatial and spatial approaches of Bayesian model selection to Bayesian Model Averaging. Following that, for the third and final aim, I extend my Bayesian model selection methodologies to the spatio-temporal setting and evaluate the benefit and usefulness of four different modeling approaches. These explorations demonstrate the importance of altering the defaults in INLA and the flexibility of the BUGS software. Additionally, they offer a novel way of determining appropriate linear predictors in the context of non-spatial, spatial, and spatio-temporal small area health data in disease mapping.

Rights

All rights reserved. Copyright is held by the author.

Download

Share

COinS