Biology-Related Research Relies Too Much on the Problematic Method of Significance Testing, Whereas There Are Good Alternatives

Abstract

In biology, as in many other empirical sciences, drawing conclusions is often based on statistical analyses. Still, by far, the most common approach for this is null hypothesis significance testing, despite it falling short seriously on deciding what outcomes are biologically relevant. A similar observation can be made for metabolomics, an area in between analytical chemistry and biology. In the present paper, we discuss alternatives that offer an approach for drawing conclusions that actually assess the biological relevance of outcomes. These are based on replacing testing with estimation to begin with. Next, simple Bayesian procedures are offered for actually assessing probabilities of outcomes being or not being practically relevant.