Menu Close

P-values are under discussion because they are often misconstrued…

Forest Plot (7 outcome types of testing effect)statisticalsignificantequivalenceequivalence?equivalence?no equivalenceequivalenceequivalence?equivalence?Caution! Such a statistically signifant effect is not biologically relevant.No clear study result.No clear study result.No clear study result.No clear study result.Clear study result. The effect ist statisticallysignificant and of biological relevance.Clear study result. The effect ist neitherstatistically significant and nor biologically relevant.P < 0.05P ≥ 0.05not statisticalsignificantbiological not relevant0Effect Sizebiological relevant

We addressed this issue within the EU FP7 research project GACE, the EU FP7 research project G-TwYST and other projects, when assessing the potential toxicological effect of a substance in several feeding trials.

Statistical measures like ‘significant’ test results and P values always need interpretation, when one considers what they really mean:

The chance of observing data under the assumption of a null hypothesis (of no correlation or no effect), therefore they only reflect the likelihood that the null hypothesis is true.

Reporting and graphically displaying effect sizes and confidence intervals can help to avoid the yes/no decision trap of statistical tests and to illustrate the size of effects in the context of biological relevance.

Emphasizing the size of effects and the confidence in them avoids the problem of a small, biologically unimportant effect being declared statistically significant and the artificiality of trying to dichotomize a result into a positive or negative finding on the basis of a P value.

Zeichenfläche 7

Zeichenfläche 6

Zeichenfläche 9 Kopie

Zeichenfläche 1