One and two sided tests of significance Statsitical hypothesis should be brought into line with clinial hypothesisBMJ 1994; 309 doi: https://doi.org/10.1136/bmj.309.6958.873a (Published 01 October 1994) Cite this as: BMJ 1994;309:873
- R Wolterbeek
- 3584 HL Utrecht, Netherlands
- Department of Obstetrics and Gynaecology, Faculty of Health Sciences, McMaster University, Hamilton Ontario, Canada L8N 3Z5
- Department of Public Health Sciences, St George's Hospital Medical School, London Medical Statistics Laboratory, Imperial Cancer Research Fund, London WC2A 3PX.
EDITOR, - I believe that J Martin Bland and Douglas G Altman's strong preference for two sided significance tests results in one sided advice.1 The wide acceptance of two sided statistical alternative hypotheses in the case of clearly one sided clinical or biological hypotheses has always amazed me. To my mind the statistical alternative hypothesis should be brought into line with the clinical hypothesis more often, even if this means a shift towards one sided tests. If on the basis of scientific considerations the clinical hypothesis is one sided then so should be the alternative statistical hypothesis, which then more precisely reflects the expectation of the clinical investigator.
Although statistical decisions depend on an outcome variable having or not having some critical value, in the case of an extreme result in an unexpected direction one faces the same practical problem regardless of the choice of a one sided or a two sided test and even regardless of the precise P value. The problem is to decide whether a new treatment is harmful or simply ineffective (if it is ineffective the unexpected outcome will be attributed …