Hi Faraz

Providing you have used an appropriate statistical model to analyze the data (e.g. a mixed effect model) than this should be accounted for in the analysis. However, if there are profound differences in the variances observed among the centers, you might also want to report this (allowing the reader to make their own mind up as to the strength of your evidence). Also, you might want to consider why there are such large differences in the variation for the different centers.

I agree with Cameron Hurst, in addition to the mixed effects, probability weights associated to the multi-center attributes are needed to take into account in care there is heterogeneity among the clusters.

Hi Faraz, although unfortunately for a long time mutlicenter and cluster randomized trials weren't always analyzed appropriately, in recent years they tend to be analyzed by either mixed or multilevel (aka hierarchical linear) models, which are pretty much the same thing. Documentation will tell you whether there is or is not an assumption of homogeneity of variance at the within-cluster level. Some software will let you run the models under the assumption of heterogeneity and/or test whether the assumption is violated. Assuming you can run the model with heterogeneity of variances, then there shouldn't be a problem. Obviously, if you test for heterogeneity and find it, but can't model it, you are going to have problems with your estimates. Best advice would be to get your hands on software that can model that. Failing that, I'd still try to get your hands on software that can model it, but if not, you'd probably want to try to find advice in the literature on how to weight your sample to get both your parameter estimates and standard errors right given the heterogeneity, but I'd still try to find software that can model the heterogeneity. Bob