Hi Bilal, Can you elaborate a bit on what the advantages of option 2 in the first scenario are? I was unclear on why balance would be important in that situation. Is this because your original sample was balanced and you wanted to avoid the bias that comes with the typical unbalanced cluster RCT estimator? Also, on a related note, if you happen to be looking for subjects for future blog posts, I would be very interested in one on strategies for dealing with missing data in general. This is an issue that has come up in an RCT I am working on and we are struggling with how to deal with it. The strategies that seem to be used in clinical trials don't seem to be used too often in RCTs of social programs and it is not really clear to us why.