What is the discussion around p < 0.05 about?
The classification of study results into "statistically significant" versus "statistically non-significant" is no longer tenable. This is not only the opinion of the three speakers at the Update Session on Clinical Research Day 2020, but the consensus of many scientists worldwide. Last year, the American Statistical Association (ASA) - the world's largest society of statisticians - published a "special issue" with 43 articles on "Statistical inference in the 21st century - A world beyond p < 0.05" (Wasserstein et al. 2019). At the same time, more than 800 scientists signed a call to retire statistical significance (Amrhein et al. 2019).
Why has the concept of statistical significance become obsolete?
Too often, "statistically non-significant" results have been described as "negative", resulting in an overall conclusion that "no effect, no difference or no correlation" exists. Such results are usually not published or discussed further. This dichotomization based on the p-value and an arbitrary threshold value (usually 0.05) is misleading and does not allow a meaningful interpretation. Instead, observed effect sizes should be communicated and their uncertainty interpreted in context, e.g. by discussing the upper and lower limits of confidence intervals.
Are p-values no longer allowed?
The current discussion is not about abolishing the p-value, but about the practice of labelling study results as "statistically significant" or "statistically non-significant". This classification is based on the p-value and an arbitrary threshold value (usually 0.05). The main problem are the common misinterpretations associated with this practice. A p-value can be given as supplementary information together with the estimated effect size and its uncertainty. However, the p-value is meaningless on its own.
What are the alternatives?
Statistics never provide a clear yes/no answer, but an estimate that is subject to uncertainty. We have to get used to deal with this uncertainty by taking it into account in our interpretation. The alternatives to "statistical significance" are manifold. The statisticians of the Department of Clinical Research (DKF) will be happy to help you. For your personal questions and further information, please arrange an appointment:
Do you know "p"? Michael Coslovsky, PhD, Teamleader Data Analysis/Statistics, DKF. Recording of the Clinical Research Update Seminar, September 22, 2021
Andrea Wiencierz, PhD, Senior Statistician, DKF. Recording of the Clinical Research Update Seminar, September 22, 2021
P-value - what else? Fabrice Helfenstein, PhD, Statistician, DKF. Recording of the Clinical Research Update Seminar, September 22, 2021