Bayesian Inference in Non‐Replicated Ecological Studies

Abstract

Classical hypothesis testing is founded on a long—run frequency perspective that is the basis for error rates and P values used in classical statistical inference. Thus in ecological studies involving formal hypothesis testing, it is common practice to report the P value as the summary result from the test of a point null hypthesis. However, many important ecological studies concern single, non—replicated events in which the P value has no clear interpretation. For the non—replicated study, Bayesian statistical inference provides an attractive alternative to classical statistical inference, as the results from a Bayesian analysis either may assume a long—run frequency interpretation or may be expressed as a probability of a unique event. An example concerning trends in lake acidification is used to show that the Bayesian approach is more compatible with scientific needs and scientific judgment than is classical hypothesis testing.