How to interpret the size of weighted mean difference (WMD) in a meta analysis? I am studying Kishi T, Matsunaga S, Iwata N (2015) and they report that the weighted mean difference for the outcome "Total Time in Bed" is -20.16. Is there a rule of thumb to assess whether the effect is small, moderate or large? How do I interpret the value?
 A: A (weighted) mean difference is the difference between effect estimates for intervention and control on a specific scale. Assuming this is the study you are referring to this, your scale is time measured in minutes. So, a mean difference of -20.16 would mean that that the subjective total sleep time (sTST) was 20.16 minutes less in the placebo arm. Normally I would expect this not to be a negative number as we usually set up these analyses as Intervention/ Comparator but in this case it seems they analyzed it as Placebo/ Suvorexant and that's why they have reported statistically significant results that are negative in favor of Suvorexant.
From a clinical perspective, interpretation requires knowing the minimally important difference (MID) for this outcome. This is a clinical judgement, often attained through expert opinion or consensus. It also takes into account the confidence intervals (−25.01 to −15.30). In this case, the lower confidence interval is ~15 minutes. Would this be clinically-relevant? A statistical test won't answer that question, but it's rather clinical judgement.
