Hi Nick, thanks for your comment. I couldn’t agree more. I think it was beyond the scope of this article to get too neck deep in all the nuances of statistics. In this context, I was merely referring to a general need to have adequate subjects (as opposed to N=1) to ensure some iota of statistical power. Clearly this is also going to depend on the outcome measures being evaluated, %CV of the assay, and what constitutes a clinically meaningful change (enter discussion on effect size and magnitude based inferences vs. sole reliance on P values). And when the smoke clears and the dust settles, there’s always the question of practical applicability of the results and how it applies to the general population. I know I’m preaching to the choir telling you this. Again, in this particular post, it was more a general guide rather than a specific tutorial on statistical robustness of one single factor.

Cheers,
Bill

It’s probably worth clarifying that this be a like-with-like comparison, to avoid it being misinterpreted by someone unwittingly comparing a clinical trial (i.e. a potentially small number of subjects, but under tightly controlled conditions) with an epidemiological study (i.e. a population based study, but less tightly controlled), purely on the basis of the number of subjects.

It’s also worth knowing about the type of study so you can spot inappropriate conclusions in media articles, e.g., you get titles like “Sunscreens questioned as big study reveals Vitamin D deficiency” only to find the study was actually about the effects of Vit D supplementation in aged-care residents in Scotland and sunscreen was not even a factor in the study.