A number of RSS fellows, including RSS president Deborah Ashby and past president Sir David Spiegelhalter, have co-signed a Nature editorial that calls for the ‘the entire concept of statistical significance to be abandoned’. The editorial, which had more than 850 co-signaturies, follows a special edition of the American Statistical Association’s journal, The American Statistician, which has compiled articles all discussing ‘A World Beyond p < 0.05’.
The RSS welcomes the debate moving forward; we have hosted several meetings to discuss the issue, including a session at last year’s Conference featuring David Cox, Deborah Mayo, Richard D Morey and Aris Spanos. Our flagship magazine Signficance has also featured an editorial on the subject.
What RSS fellows say about the Nature article:
Deborah Ashby, director of the School of Public Health at Imperial College London and RSS president
‘I understand the desire for a ‘simple’ rule of thumb- but a naïve interpretation of p-values can lead to seriously wrong conclusions such as whether medicines are effective or not. This is well-understood by many, but misused by many more. I’ve signed to help draw attention to the dangers of this outdated practice and promote the wider use of better alternatives.’
David Spiegelhalter, Winton professor for the public understanding of risk at the University of Cambridge and RSS past-president
‘I like p-values, but feel they are delicate things and should not be crudely split into “significant” and “not-significant”. I signed this article because I am fed up with researchers claiming a discovery when p < 0.05, and claiming there is no effect when p > 0.05.’
Guy Nason, professor of statistics at the University of Bristol and RSS vice president for academic affairs
‘I signed the article because I agree with it! I am often surprised by how the outcomes of statistical methods are used to communicate results with unwarranted definiteness, based on assumptions, sometimes hidden, for which there is also usually considerable uncertainty. I particularly liked the article's idea to talk about results' compatibility with the data.’
Stephen Senn, consultant statistician and former RSS Council member
‘Information is rarely dichotomous but decisions often are. Significance versus non-significance as a qualitative absolute distinction is ridiculous. As a threshold for action it may sometimes be justified but the appropriate standard will differ according to context.’
RSS fellows: Let us know what you think and we’ll publish some of the responses in this article, or tweet us your views @royalstatsoc with the hashtag #pvalues.