Using statistical power to optimize sensitivity of analysis of variance designs for microcosms and mesocosms

Controlled laboratory and field studies are often used in the risk assessment process to evaluate the fate and effects of chemicals on the structure and function of populations and communities. These studies have evolved, over the past 20+ years, from field studies to outdoor simulated ecosystems (microcosms and mesocosms). A review of pertinent literature indicates that analysis of variance is most often used to evaluate those data. Assumptions of homogeneity of variance are often violated, however, and standard normal-theory methods should not be used. Counts of macroinvertebrates collected from a mesocosm and microcosm study were analyzed using log-linear regression models (general and robust) written in the S-Plus statistical language. The statistical power of the analysis is also provided. Dose-response experiments may result in elimination of sensitive taxa for all replicates of a given condition. The program proposes a new method to deal with these zero counts. This program and directions for its use are available from an anonymous file transfer protocol (FTP) site.