Reader Comments
Post a new comment on this article
Post Your Discussion Comment
Please follow our guidelines for comments and review our competing interests policy. Comments that do not conform to our guidelines will be promptly removed and the user account disabled. The following must be avoided:
- Remarks that could be interpreted as allegations of misconduct
- Unsupported assertions or statements
- Inflammatory or insulting language
Thank You!
Thank you for taking the time to flag this posting; we review flagged postings on a regular basis.
closeFundamental statistical problems with this paper
Posted by fubar on 06 Sep 2014 at 03:06 GMT
How did this pass any kind of statistical review?
In particular, I see no evidence that the paired data (repeated measures on the same subjects) were correctly analysed as paired - the anova involved the controls but appears to NOT have included a subject id term, thus falsely increasing the DOF in the denominator of the variances leading to bias away from the null.
A beginner's mistake? Well, ok but there's no evidence of any kind of family wise error rate correction - p values are provided but apparently free from (eg) false discovery rate control. How on earth did this happen in 2013?
Interestingly, a correct (paired) analysis of the baseline-recovery case data using limma finds about 70 significant genes after controlling FDR at 0.05.
Authors' response to comment by fubar
VilmaAho replied to fubar on 25 Nov 2014 at 16:28 GMT
We thank the reader for the comment. Naturally, the analysis was performed paired as there were repeated measures from the same individuals. As we state in the paper, we have used 2-way repeated measures ANOVA with one between subjects (BS) and one within subjects (WS) factor. In this model, the trait (gene’s expression) is modelled by the group (i.e. case or control; BS), the three timepoints (baseline, sleep restriction, recovery; WS), and their interaction. The model does include the subject IDs for pairing the samples taken from the same individuals at different timepoints.
We hope this comment clarified the analysis for the reader.
We are aware that for the individual gene results, the multiple testing correction was not made for the data in the experimental research entity (with N=14), and we state this very clearly in the manuscript. The interpretation of the main results is not based on the results from the individual genes, but from the results of the pathway analysis with permuted p-values. We believe that the additional information on the individual genes, although not significant, can provide the researchers working on the same problems useful information and food for thought. The individual gene results from the epidemiological sample (N=472) were corrected for multiple testing.
The baseline-recovery analysis performed by the reader sounds most interesting – we would be happy to learn more about the details of the analysis (contact: vilma.aho@helsinki.fi, tarja.stenberg@helsinki.fi).