Bayesian Inference in Non‐Replicated Ecological Studies

• Published in: Ecology (Durham) 1990-12, Vol.71 (6), p.2053-2059
• Main Author: Reckhow, Kenneth H.
• Format: Article
• Language: English
• Subjects: Acids; Animal, plant and microbial ecology; Bayesian analysis; Biological and medical sciences; Ecology; Fundamental and applied biological sciences. Psychology; Histograms; Hypothesis testing; Lakes; Null hypothesis; P values; Probabilities; Sampling distributions; Special Feature: Statistical Analysis of Ecological Response to Large-Scale Perturbations; Statistical inferences; Statistics
• ISSN: 0012-9658; 1939-9170
• DOI: 10.2307/1938619

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.