Closed rs4606 closed 1 year ago
Thanks! You are right. In 10.6 W can be considered to be a covariance matrix. Higher weight of an observation corresponds to higher precision (inverse variance) of that observation, and thus the diagonals of W should be the inverse weights. The online version and errata have been updated.
Hi,
I think there's a typo in the definition of weighted regression at the bottom of p.147. Specifically, I think that in the very last line on the page, either W should be the inverse of the weight matrix (so W = diag(1/w_i), not W = diag(w_i)). Alternatively, in equation 10.6, W^{-1} could be replaced by W.
Thanks,
Ravi