Detecting Community-Wide Patterns: Estimating Power Strengthens Statistical Inference

Abstract

Power statistics are drastically underutilized in basic and applied ecological research where they could provide objective measures of the sensitivity of null hypotheses tests and thereby strengthen some statistical inferences. Null models are being used increasingly to investigate the causes of community-wide patterns, yet researchers tend to ignore the risks involved in committing the type II error associated with these models. The three factors that determine power, e.g., critical level of α, sample size, and "effect size" are explained and their effects on power are discussed. In our examples we have attempted to illustrate how power analysis can assist investigators in interpreting their research results. In ecological studies' failing to demonstrate an effect is quite different than implicitly or explicitly concluding that no difference exists. In this situation, assuming low power, the cost of committing a type II error is that of making false claim. Pesticide and drug examples demonstrate that often there can be serious economic, health, and social costs associated with the commission of type II errors. Lastly we describe a situation whereby power analysis can be used to measure the degree to which an effect occurs and thereby expand our conclusions.