Grantee Research Project Results

2000 Progress Report: Assessing the Effects of Multiple Stressors in Environmental Monitoring Programs

EPA Grant Number: R827953
Title: Assessing the Effects of Multiple Stressors in Environmental Monitoring Programs
Investigators: Smith, Eric , Ye, Keying
Institution: Virginia Tech
EPA Project Officer: Hahn, Intaek
Project Period: October 1, 1999 through September 30, 2002
Project Period Covered by this Report: October 1, 1999 through September 30, 2000
Project Amount: $266,388
RFA: Environmental Statistics (1999) RFA Text |  Recipients Lists
Research Category: Aquatic Ecosystems , Environmental Statistics , Human Health

Objective:

This research will investigate the use of Bayesian Model Averaging as a tool to aid in interpretation and modeling relationships between stressors, covariates and biological endpoints. The methods will be applied to a set of multivariate methods based on multivariate regression analysis including principal components analysis, correspondence analysis, canonical variate analysis, canonical correlation analysis, redundancy analysis and canonical correspondence analysis. The analysis will also focus on the use of Bayesian Hierarchical Modeling as a tool to combine information across different regions of stress.

Progress Summary:

Bayesian model averaging has been applied to principal components analysis. In principal components analysis of environmental data, variables are combined to form composites that are often interpreted as different types of stressors. The variables are typically environmental variables such as concentrations of pollutants. The composites are used to plot scores for the sites that data were collected from. In this way, the sites may be ordered in terms of stress. In Bayesian model averaging, multiple models for the composites and scores are used. Thus, there is no single model but the results are based on combining information in good models. A variation of the Bayesian information criterion (BIC) was used to determine the importance of a particular model. Scores may be estimated and graphed. The analysis allows for interpretation of the variation in the scores in terms of the within model variation as well as between model variation. Between model variation allows for evaluation of the impact of the variable that are selected on the scores. Importance of the variables is assessed through an activation probability that measures the probability the variable should be included in the interpretation. The approach results in a novel way to interpret the components.

Future Activities:

The research will focus on other multivariate methods such as canonical variate analysis and canonical correspondence analysis. Programs will be developed in SAS and EXCEL to implement the analyses.

Supplemental Keywords:

principal components analysis, correspondence analysis, canonical variate analysis, canonical correlation analysis, redundancy analysis, canonical correspondence analysis, stressor effects, ecological statistics., RFA, Ecosystem Protection/Environmental Exposure & Risk, Scientific Discipline, Economic, Social, & Behavioral Science Research Program, Environmental Statistics, Mathematics, Ecological Indicators, Ecological Risk Assessment, Ecosystem Protection, Ecology, Ecosystem/Assessment/Indicators, Ecological Effects - Environmental Exposure & Risk, Applied Math & Statistics, Ecological Effects - Human Health, Ecology and Ecosystems, canonical correlative analysis, computer models, multiple stressors, aggregation, data analysis, Bayesian method, correlative analysis, ecological statistics, environmental risks, innovative statistical models, risk assessment, statistical methods, statistical models, co-pollutant effects, ecological exposure, data models, regression analysis, environmental stressor, environmental monitoring networks

Progress and Final Reports:

Original Abstract

2001

Final