I am working on a problem for which I need to do a multiple regression that involves a dependent variable and two independent variables. The dependent variable is a survey statistic for which I have results from a large number of surveys, of very large populations. However, each of the surveys have different sample sizes and thus different errors. It seems like the best way to account for this is to use a WLS regression.

However, I'm a little unclear as to what I should use for the weights. If I was dealing with populations rather than samples I could just weight to the population sizes, but since I'm dealing mostly with samples ranging from n=300 to n=2000 I imagine I should find a way to weight for the differences in estimated variance.

Does anyone have a good idea of what weights to use in a WLS model for survey samples of different sizes?

EDIT: Ideally this would account for the differing error for the statistics themselves and not just the maximum error of a sample of that size.