Abstract:
Rigorous measurement of statistical evidence represents a pervasive and unresolved problem in biological and biomedical research. The problem is also largely invisible. Biologists are taught to interpret p-values as reliable indicators of evidence strength, and they are generally unaware that this practice can lead to serious errors in the interpretation of data. But measurement of evidence is also not a mainstream topic within statistics, since it is not required for standard statistical operations such as hypothesis testing or parameter estimation. Thus the topic tends to fall through the cracks between disciplines. In this talk I will illustrate ways in which current approaches to evidence measurement fall short, and I will propose a novel approach to establishing an absolute scale for the measurement of statistical evidence drawing upon basic principles of measurement theory and a precedent from physics.