BMJ statistical errors
http://www.100md.com
《英国医生杂志》
EDITOR—Abbasi in his Editor's choice discusses a study that found statistical errors in 25% of papers published by the BMJ in 2001.1 As statistical advisers to the BMJ we aim to improve the quality of published papers by ensuring that their conclusions are consistent with the data. To this end we hope to identify important errors that affect the interpretation of the findings, but care less about more minor errors. Any stricter policy would be impossibly time consuming. That said, we recognise that important errors do slip through from time to time, and are always keen to improve our performance.
The particular errors flagged in the paper2 were inconsistencies between test statistics and P values. Out of 63 tests seven (11%) were wrong (for example 2 on 1 df = 4.2, P reported = 0.024, P actual = 0.0404). Yet in no case did the error affect the test's interpretation as to whether or not the results could have arisen by chance. This supports our belief that more extreme errors are likely to be weeded out at the review stage. The paper is disappointing in focusing on P values and by implication hypothesis testing. By contrast the BMJ's policy is to present the main findings as confidence intervals where the emphasis is on estimation.3
Tim J Cole, professor of medical statistics
Institute of Child Health, University College London WC1N 1EH tim.cole@ich.ucl.ac.uk
Douglas Altman
ICRF Medical Statistics Group, Centre for Statistics in Medicine, Institute of Health Sciences, Oxford OX3 7LF
Deborah Ashby
Wolfson Institute of Preventive Medicine, Queen Mary University of London, London EC1M 6BQ
Mike Campbell
ScHARR, Community Services Centre, Northern General Hospital, Sheffield S5 7AU
Jonathan Deeks
ICRF Medical Statistics Group
Stephen Evans
Medical Statistical Unit, London School of Hygiene and Tropical Medicine, London WC1E 7HT
Hazel Inskip
Southampton Women's Survey, Southampton SO16 7PX
Julie Morris
Education and Research Centre, Wythenshawe Hospital, Manchester M23 9LT
Gordon Murray
Medical Statistics, Department of Public Health Sciences, University of Edinburgh Medical School, Edinburgh EH8 9AG.
Competing interests: All authors take responsibility for the statistical quality of papers published in the BMJ to the extent that the study design, data, and analysis appear appropriate and internally consistent, and that they support the conclusions drawn.
References
EDITOR's Choice. Do mistakes matter? BMJ 2004;328: 0. (12 June.)
Garcia-Berthou E, Alcaraz C. Incongruence between test statistics and P values in medical papers. BMC Med Res Methodol 2004;4: 13.
Gardner MJ, Altman DG. Confidence intervals rather than P values—estimation rather than hypothesis testing. BMJ 1986;292: 746-50.
The particular errors flagged in the paper2 were inconsistencies between test statistics and P values. Out of 63 tests seven (11%) were wrong (for example 2 on 1 df = 4.2, P reported = 0.024, P actual = 0.0404). Yet in no case did the error affect the test's interpretation as to whether or not the results could have arisen by chance. This supports our belief that more extreme errors are likely to be weeded out at the review stage. The paper is disappointing in focusing on P values and by implication hypothesis testing. By contrast the BMJ's policy is to present the main findings as confidence intervals where the emphasis is on estimation.3
Tim J Cole, professor of medical statistics
Institute of Child Health, University College London WC1N 1EH tim.cole@ich.ucl.ac.uk
Douglas Altman
ICRF Medical Statistics Group, Centre for Statistics in Medicine, Institute of Health Sciences, Oxford OX3 7LF
Deborah Ashby
Wolfson Institute of Preventive Medicine, Queen Mary University of London, London EC1M 6BQ
Mike Campbell
ScHARR, Community Services Centre, Northern General Hospital, Sheffield S5 7AU
Jonathan Deeks
ICRF Medical Statistics Group
Stephen Evans
Medical Statistical Unit, London School of Hygiene and Tropical Medicine, London WC1E 7HT
Hazel Inskip
Southampton Women's Survey, Southampton SO16 7PX
Julie Morris
Education and Research Centre, Wythenshawe Hospital, Manchester M23 9LT
Gordon Murray
Medical Statistics, Department of Public Health Sciences, University of Edinburgh Medical School, Edinburgh EH8 9AG.
Competing interests: All authors take responsibility for the statistical quality of papers published in the BMJ to the extent that the study design, data, and analysis appear appropriate and internally consistent, and that they support the conclusions drawn.
References
EDITOR's Choice. Do mistakes matter? BMJ 2004;328: 0. (12 June.)
Garcia-Berthou E, Alcaraz C. Incongruence between test statistics and P values in medical papers. BMC Med Res Methodol 2004;4: 13.
Gardner MJ, Altman DG. Confidence intervals rather than P values—estimation rather than hypothesis testing. BMJ 1986;292: 746-50.