I am using survey data from different countries, and I pool them together for part of the analysis. Now, weights puzzle me for two reasons:
1) some of them are below 0 (i.e. the probability of being picked in the sample would be above 1), which would tell me weights have been somehow been rescaled. Is that true? Or in other words, why doesn't the rescaling matter for regression analysis?
2) Because I am pooling data, and therefore weights that are potentially scaled differently, from different surveys, what is the normal procedure in this cases? Should I rescale them to represent their actual share in the population?