Newbie here. I have a question regarding WLS regression. Specifically, I've come across different ways of estimating weights in WLS regression, the most frequent ones being:
weights <- 1 / abs(fitted.values(model.lm))
weights <- 1 / (fitted.values(model.lm))^2
weights <- 1 / lm(abs(model.lm$residuals) ~ model.lm$fitted.values)$fitted.values^2
weights <- 1 / abs(resid(model.lm)))
weights <- 1 / resid(model.lm)^2
weights <- log(model.lm$residuals^2)
My question is twofold - which of the above is suited to what situation (can you provide some references) and how do you check if the WLS corrected for heteroskedasticity (what do I look at, e.g. R2, residuals etc.)?
Thank you in advance!
