Statistical concerns in the design and analysis of multispecies microcosm and mesocosm experiments

Abstract Microcosm and mesocosm experiments are useful in the environmental analysis of the effects of toxicants on ecological communities. While there has been much interest in the biological approaches to microcosm design, there seems to have been little thought given to statistical issues in the design of these experiments. Here, some statistical design issues are considered. Some criteria are proposed, and optimal or near optimal designs suggested. Of particular interest is the choice of doses in the experiment, choices that allow for tests with high power and also provide good estimates of effect levels and other parameters. Also, some analytical techniques recently suggested for the analysis of community data are criticized and alternate methods suggested.