I read in some books and web pages:
"GLM are fitted with MLE (maximum likelihood estimators) estimators rather than OLS (ordinary least squares)"
I have searched a little bit more and have found something like:
_ When error structure is not normal, LS technique can be difficult.
_ When link function is not identity, LS technique can be difficult.
_ In some GLM different from normal standard one, estimates from MLE and those from OLS are different in general, why ? and why do we prefer estimate parameter of a GLM with those from maximum likelihood?
---> I would like deeper explanations than these ones :(
Thanks !