Hello, I would like to have some advises concerning checking glms quality for prediction. I've a glm with Gamma family and link=log, and I would like to compute a Press statistic to check the predictivity of my model. Since my link function is not "identity", I wonder which "Hat" matrix I should take ... Is the expression of this matrix unchanged (ie equal to sqrt(W)%*%X%*%solve(t(X)%*%W%*%X)%*%t(X)%*%sqrt(W)) even if my link is equal to log. Should I make a development of log ??? Thanks in advance for your help, Regards Isabelle Zabalza-Mezghani