Researchers often view the production of P-values as an essential component of statistical analysis. But in recent years many statisticians and epidemiologists have become increasingly disillusioned with P-values. Wrong beliefs about P-values, and wrong interpretations of P-values, are encountered on a daily basis. Mis-interpretation of P-values is contributing to the reporting of non-reproducible research results. P-values were the subject of a recent debate at a School of Population Health and Environmental Sciences seminar. Abdel Douiri and I spoke about P-values. There was also a lively exchange of views among those present.
- Why a P-value threshold of <0.05 might result in more than 30% of positive results being false
- Why a P value of 0.90 does not provide evidence of no effect
- Why the American Statistical Association recommends that rules such as “p < 0.05” for justifying scientific claims can lead to erroneous beliefs and poor decision making
- Why transcribing P values from the output of statistical packages may not be a good way to populate tables in a paper
- Why the words ‘significant’ and ‘not significant’ should not be used
The latest issue of The American Statistician (volume 73, supplement 1) includes a series of papers about P-values and how to move to a world beyond ‘p < 0.05’. An accompanying comment in Nature (March 2019), signed by 854 scientists from 52 countries, advocated it is time to “retire statistical significance”.
Martin Gulliford (@MartinG0714) is Professor of Public Health at King’s College London. He qualified in medicine from the University of Cambridge and University College Hospital, London and trained in public health and health services research at Guy’s and St Thomas’ Medical Schools London where he was a Wellcome Training Fellow in Health Services Research. More information on Professor Gulliford, including his research and teaching, can be found on his King’s Research Portal profile.
Abdel Douiri is a Senior Lecturer in Medical Statistics in the School of Population Health & Environmental Sciences at King’s College London. Abdel has MSc in numerical analysis and a PhD in signal processing (applied mathematics). He has been involved in a variety of research projects with world leading research groups, in both academia and industry. He has been a lecturer at King’s since 2009, teaching medical statistics and epidemiology on the undergraduate MBBS the postgraduate MPH. He is a statistical editor for Thorax (a BMJ journal) and a statistical consultant with the Biomedical Research Centre, the Research Design Service London, and King’s College Hospital.