Statistical Methods for Panel Studies with Applications in Environmental Epidemiology
Pollution studies have sought to understand the relationships between adverse health effects and harmful exposures. Many environmental health studies are predicated on the idea that each exposure has both acute and long term health effects that need to be accurately mapped. Considerable work has bee...
| Main Author: | |
|---|---|
| Other Authors: | |
| Language: | en_US |
| Published: |
Harvard University
2013
|
| Subjects: | |
| Online Access: | http://dissertations.umi.com/gsas.harvard:10049 http://nrs.harvard.edu/urn-3:HUL.InstRepos:10121973 |
| Summary: | Pollution studies have sought to understand the relationships between adverse health effects and harmful exposures. Many environmental health studies are predicated on the idea that each exposure has both acute and long term health effects that need to be accurately mapped. Considerable work has been done linking air pollution to deleterious health outcomes but the underlying biological pathways and contributing sources remain difficult to identify. There are many statistical issues that arise in the exploration of these longitudinal study designs such as understanding pathways of effects, addressing missing data, and assessing the health effects of multipollutant mixtures. To this end this dissertation aims to address the afore mentioned statistical issues. Our first contribution investigates the mechanistic pathways between air pollutants and measures of cardiac electrical instability. The methods from chapter 1 propose a path analysis that would allow for the estimation of health effects according to multiple paths using structural equation models. Our second contribution recognizes that panel studies suffer from attrition over time and the loss of data can affect the analysis. Methods from Chapter 2 extend current regression calibration approaches by imputing missing data through the use of moving averages and assumed correlation structures. Our last contribution explores the use of factor analysis and two-stage hierarchical regression which are two commonly used approaches in the analysis of multipollutant mixtures. The methods from Chapter 3 attempt to compare the performance of these two existing methodologies for estimating health effects from multipollutant sources. |
|---|