Standards of statistical analysis

Q. I am writing to ask you about the minimal standards required by your journal concerning statistical analyses in submitted manuscripts.

I am a biostatistician. A colleague, a biomedical researcher, insists that there are no sample size requirements as long as the minimum statistical power is achieved. To be more concrete, consider a simple two-group t-test where each group is of size n=3 and the power calculations show that the power=80% (a widely accepted standard). My argument is that statistics on sample size of n=3 per group does not make much sense in terms of reliability, but my colleague insists that because the power is 80%, a manuscript will be acceptable to your journal despite the meagre sample size.

Is my colleague right or wrong? Are there any minimal standards on statistics that manuscripts submitted to your journal should satisfy, and how can one find them?

A. The Nature journals do have a statistical checklist, available on Nature’s website as a one-page (44KB) Word document download. We welome volunteers from suitably qualified scientists to act as statistical referees for us. (If any such people read this posting, please drop your address details in the comments.) We also welcome suggestions for improving our checklist — are there other elements we need to add?


Comments are closed.