Limitations of the Wilcoxon matched pairs signed ranks test for comparison studies
- 1Department of Clinical Biochemistry, The Ipswich Hospital, Heath Road, Ipswich IP4 5PD, UK; taptwomeyaol.com
- 2Department of Clinical Biochemistry, Royal Infirmary of Edinburgh, Room S6114 Level 2, 51 Little France Crescent, Edinburgh EH16 4SA, UK
We read the short report by Ellis et al with interest.1 However, we are unsure whether they have adequately proved that no significant difference was detected between the two outlined storage methods.
The hypothesis evaluated with the Wilcoxon matched pairs signed ranks test is whether or not the median of the difference scores equals zero. Let us consider the situation of x measurements tending to exceed y measurements in the low range and vice versa in the high range, with similar values in the mid range. Such results may have a median of the difference scores of approximately zero; that is, there might be no significant differences by the Wilcoxon matched pairs signed ranks test, although there would be differences by linear regression (Deming or Passing-Bablok) and/or difference plots.
However, this short report1 lacks both a regression equation (proportional and constant error) and difference plots. Therefore, we believe that although the IgG anti-rubella activity in frozen serum stored in primary gel separation tubes may not be significantly different from that stored frozen in secondary tubes, this study did not sufficiently prove this. We recommend, in line with others,2 that difference plots3 are used for such comparative studies and that such studies are put into a clinical context.3,4
