These are based on the residual (within-centre) variation. The variance (SE2) ofa treatment difference is given by var(ij - tj) = a2(1/n + 1/nj), where a2 is the residual variance and ni and nj are the numbers of patients receiving treatments i and j. When equal numbers of patients, r, receive each treatment at each of c centres, we have ni = rc and the variance can be written var(ti - tj) = 2a2/rc.
This variance will always be less than or equal to that for the model fitting centre-treatment effects as random (see Section 5.2.3).
Was this article helpful?