Un mondo senza significatività statistica?
A word without statistical significance?

AUTHORS

ABSTRACT

A word without statistical significance?

A recently published comment (Nature, 2019), proposing to give up the use of the P value in scientific literature, spurred several contributions on the topic. The main target was the need to avoid the dichotomization of P, with p<0.05 identifying the statistically significant results. A first proposal was to lower the threshold value to 0.005, labeling as “suggestive” results previously classified as significant but not meeting the new threshold. A more radical suggestion was to suppress the use of P, allowing its presence only in a descriptive sense. At the time of writing, only one journal took such a radical position, and this choice gave rise to problems in the interpretation of studies’ results. To avoid p-hacking and other inappropriate uses of P, the most sensible strategy would be to mandate the pre-publication of the study protocol, including the statistical analysis. The authors should then be required to adhere to their original published plan. This rule could be of great help for pragmatic trials, but does not apply to exploratory studies, which are more frequent in life sciences. It could also be imposed to report the P value only for sufficiently large sample sizes, reporting otherwise only descriptive statistics. Moreover, the term “statistical significance” could be replaced by “statistical accuracy,” in order to avoid the common confusion with “clinical significance”. This debate probably will not lead to the abandonment of P, but it may help to improve the quality of the statistical analysis of trials’ results.

REFERENCES