Date of Award

2018

Embargo Period

8-1-2024

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Public Health Sciences

College

College of Graduate Studies

First Advisor

Brian Neelon

Second Advisor

Lane Burgette

Third Advisor

Leonard Egede

Fourth Advisor

Kelly Hunt

Fifth Advisor

Andrew Lawson

Sixth Advisor

Paul J. Nietert

Abstract

Motivated by recent work exploring cluster-level confounding in multilevel observational data, we develop methods specifically addressing geographic confounding, which occurs when measured or potentially unmeasured confounding factors vary by geographic location. Accounting for this source of confounding achieves spatially-balanced global estimates of the treatment effect of interest, allowing researchers to compare individuals as if they were residentially similar and leading to policy decisions that benefit patients and areas most in need. This dissertation consists of three aims: 1. To develop a hierarchical spatial doubly robust estimator in propensity score analysis framework; 2. To develop spatial propensity score matching methods for hierarchical data; 3. To apply spatial propensity score matching to more complex analyses of spatially varying, zero-inflated outcomes. Each of these aims strives to explore the issue of geographic confounding and contribute to its resolution. Aim 1 seeks to build upon multilevel propensity score methods through augmentation of modeling with spatial random effects to create a spatially balanced estimator that is demonstrated in simulation to exhibit favorable performance under various sample sizes and levels of spatial heterogeneity. Aim 2 seeks to develop methods in a propensity score matching framework, allowing for a more complete understanding of geographic confounding remediation techniques and extensions to additional applications. Finally, as modeling non-binary, spatially varying outcomes can prove challenging, Aim 3 seeks to incorporate spatial matching to alleviate geographic imbalance to allow for a minimally confounded analysis. We apply the spatial matching approach to the analysis of zero-inflated count outcomes.

Rights

All rights reserved. Copyright is held by the author.

Download

Share

COinS