To the Editor.
—In a recent series on peer review, Mr Moher and colleagues1 emphasize the importance of calculating power estimates to assist in the interpretation of randomized clinical trials that have negative results. The underlying point, of course, is that authors should provide estimates of how large a beneficial effect can be ruled out.An alternate approach to this issue, which many epidemiologists believe is more useful, is to provide confidence intervals (CIs) on the estimate of effect. Confidence intervals take into account the data that were actually collected in the study, whereas power calculations, which can be carried out before any data are gathered, do not. A result that is in the expected direction, but just misses statistical significance, carries quite a different message from one that is in the direction opposite to that hypothesized. This difference is easily appreciated from CIs, but is not considered when
Rhoads GG. Reporting of Power and Sample Size in Randomized Controlled Trials. JAMA. 1995;273(1):22-23. doi:10.1001/jama.1995.03520250036018