Last update:
15-Jul-2010

Hypothesis tests and P values are used wrongly by most researchers for detecting relations and extracting conclusions in biomedical research. Over the last 20 years, systematic efforts are being made for the establishment of evidence-based medicine, which requires health scientists to resort to studies based on a specific scientific hypothesis, in order to make clinical decisions in a rational way. Bayesian methodology can contribute decisively, leading to sure conclusions by an inductive way. Although Bayesian methodology has been developed considerably in the last 30 years, at least in the field of statistics, health scientists have been reluctant to embrace what they perceive as a subjective approach to data analysis. It has been poorly understood that Bayesian methods are at least as objective as P values and hypothesis test, since they rely on data derived from a study for the estimation of evidence. This estimation is achieved with the calculation of the Bayes factor, which in fact is a likelihoods ratio. Bayesian methodology can combine the evidence from earlier studies (by the prior probability of null hypothesis to be true) with the evidence of a specific study (by the calculation of Bayes factor), in order to calculate the posterior probability of null hypothesis to be true (by the application of Bayes theorem). Researchers should in this way not extract conclusions and find relationships, but report in an analytic way the research design of a study and the statistical methodology that is used. In this situation, readers have the opportunity to judge for themselves the magnitude and credibility of the evidence derived from the data of a study. Today, this can happen only with the application of Bayesian methodology, which provides the possibility of estimating the evidence of a specific study and combining this with previously reported evidence, in order to extract credible conclusions.

Key words: Bayes factor, Bayesian logic, Evidence-based medicine, Likelihood, P value.