Comparisons within randomised groups can be very misleading

Steve Simon

2014-06-10

In studies with a baseline, examining the decline exclusively within the treated group, or examining the decline in the treated group and then separately examining the decline in the control group is a bad idea, notes two famous statisticians in the British Medical Journal. They explain why you need to look first at comparisons between the two groups, ideally with analysis of covariance.

Bland JM, Altman DG. Comparisons within randomised groups can be very misleading. BMJ. 2011;342(may06 2):d561-d561. doi:10.1136/bmj.d561. Available at http://www.bmj.com/content/342/bmj.d561.

You can find an earlier version of this page on my blog.