It depends on what you want.
If you want to compare C with the average of the two groups A and B, use the 0.5 weights. That would be the case if you want to combine A and B, and you think they should be equally weighted.
If you want to combine A and B, with the one with more subjects given more weight, use the other set of weights. This would be the case if you say to yourself, "OK, now I want to take the same sample of subjects, and disregard the distinction between A and B." That might be the case if you think the numbers of subjects in A and B closely represent the frequencies of such people in the general, unsampled population.
I think most commonly the 0.5 weights will be the right ones, but it's a judgement call.
|