Perturbations on the uniform distribution of p-values can lead to misleading inferences from null-hypothesis testing

Publication date: Available online 16 October 2017 Source:Trends in Neuroscience and Education Author(s): László Zsolt Garamszegi, Pierre de Villemereuil Null-hypothesis testing (NHT) based on statistical significance is the most conventional statistical framework, on which neuroscientists rely for the analysis of their data. However, this approach can provide misleading results if p-values are wrongly interpreted, as often done in practice. Misconceptions can arise, in particular, when i) wrong null-hypothesis is chosen for reference; ii) the assumptions of the statistical model are not met; iii) p-values are interpreted as the probability of the null- or alternative hypotheses or as the measure of the importance of findings; iv) statistical thresholds guide scientific conclusions and decision making; v) one applies multiple testing or p-hacking. In this commentary, we address these issues by bringing into the focus the uniform distribution of p-values with the hope of enhancing the appreciation and proper use of the NHT approach among neuroscientists. We propose guidelines for the correct interpretations of p-values that brain and behavioral scientists may adopt to improve both the transparency of statistical reports and the value of scientific conclusions drawn from them.

Source: Trends in Neuroscience and Education - October 16, 2017 Category: Neuroscience Source Type: research