The Contribution of Bayesian Methods in Solving the Paradoxes of Classical Statistical Tests in Biomedical Research

Almost all publications in biomedical literature have employed statistical tests, with p-values being considered of particular importance in the assessment of the presence of a link between two variables. However, these tests and p-values have been the subject of considerable criticism. It may appear paradoxical that tools utilised by the scientific community for nearly a century could possess all the flaws attributed to them. This paradox can partially be explained by the counterintuitive nature of p-values and the fact that the test that generates them is the result of a combination of two tests that were developed to answer statistical questions of a very different nature. The respective characteristics of these two tests are essentially unknown to the majority of users of p-values. The aforementioned paradox can be partially explained by the paucity of publications that seek to elucidate these concepts for users of p-values, the majority of whom are not statisticians. The recently introduced Bayesian methods have properties that enable us to understand the limitations of traditional methods. In Bayesian methods, the use of a specific interpretation of probability allows for better exploitation of clinical research data. The aim of this article is to highlight the limits of non-Bayesian methods and explain the principles and functioning of Bayesian methods to a non-statistical audience.