Improved estimation of ecological effects using an empirical Bayes method
The US Environmental Protection Agency is currently developing guidance for assessing environmental impact using biocriteria within a regulatory framework. Of particular interest are statistical methods of design and analysis to test for impairment of biological assemblages in stream ecosystems associated with water pollution. Current recommendations emphasize frequentist statistical methods with the problem expressed as one of classical hypothesis testing. An empirical Bayes approach is proposed here as an alternative and applied for multi-site inference. The advantages of an empirical Bayes approach, in particular the substantive contribution of collateral information from nearby sites, are discussed in contrast to traditional methods that employ site-specific information only. The approach is illustrated in an application concerning trends in the Index of Biotic Integrity (IBI) for the Scioto River in Ohio.