Hi all, I have some difficulties to understand how the dispersion parameter is estimated in GLM models. Suppose I want to fit a quasibinomial model and mydata$W are the weigths for this model. I suppose that I have p parameters in model myModel and n observations. Model is estimated with : res<-glm(myModel,family=quasibinomial,data=mydata,weights=mydata$W) For my data set, summary(res) indicates : (Dispersion parameter for quasibinomial family taken to be 25.85539) I tried to find the value 25.85539 with the command (1/np) * sum( res$prior* (res$y-res$fitt)^2 / (res$fitt*(1-res$fitt)) ) The value I obtained with this estimation (estimation based on the Pearson Khi2) is near the value returned by summary(), but they are not equal. I read that dispersion parameter can also be estimated with deviance or by maximum likelihood... So my question is, what is the estimation returned by the command summary when I specified a quasi family? and what is the estimation if I only use the function quasi() ? Thanks for your help, olivier