 |
Tuesday, May 19, 2009 - 10:30 AM
171
Comparison of multiple regression and propensity score methods for analyzing biomonitoring data: An example using the Minnesota biological monitoring program dataset
Laurie C. Alexander1, Lester L. Yuan2, Amina I. Pollard1, Joel W. Chirhart3, Scott L. Niemela3, and Robert E. Murzyn4. (1) US Environmental Protection Agency, Office of Research & Development, National Center for Environmental Assessment, 1200 Pennsylvania Avenue (P-8623), Washington, DC 22202, (2) Office of Research and Development, U.S. Environmental Protection Agency, 1200 Pennsylvania Ave NW, Mail code 8623P, Washington, DC 20910, (3) Biological Monitoring Program, Minnesota Pollution Control Agency, 520 Lafayette Road, St. Paul, MN 55155, (4) Environmental Data Management, Minnesota Pollution Control Agency, 520 Lafayette Road, St. Paul, MN 55155
Stream biomonitoring programs compare concurrent samples from test and reference sites to assess stream condition and identify environmental causes of impairment. Inferences based on paired observational data assume that (1) similar types and levels of background variation are present across stream sites, and (2) the selection of “test” and “reference” conditions is unbiased. Here we use the Minnesota Biological Monitoring Program dataset to compare results from multiple regression and propensity score analysis, two methods for reducing confounding error and selection bias in the analysis of observational data. We present estimates from both methods of causal effects on fish and macroinvertebrate taxa richness, EPT relative abundance, and relative abundance of insect functional groups (e.g. clingers) in Minnesota streams. We discuss the merits and shortcomings of each method for establishing causal relationships from stream biomonitoring data, and evaluate the use of “ecoregion” for grouping stream sites.
Web Page: biomonitoring, causal analysis, propensity score