On Sun, 31 Mar 2013 12:44:09 -0700 (PDT), SChapman <email@example.com> wrote:
> > >In the chapter on Analysis of Deviance on Generalized Linear Models I came across 'Saturated Model' - a model that includes as many parameters as the number of observations. > >1) I am having difficulty visualising Saturated model. Can someone please give me the Saturated model in terms of Yi, X1i, X2i etc? > >2) Since I am unable to visualise the saturated model, I can't appreciate the fact that the for the saturated model the fitted value is in fact the actual response value.
I don't think of "staturated model" in any ANOVA context.
Where I have encountered the term is in multidimensional, log-linear model building. In that case, the number of parameters is equal to the number of d.f. rather than the number of observations.
For instance, a 2x2x2 table has 7 d.f. There are 3 main effects (means), 3 interactions of two terms, and one 3-way interaction in the saturated model. The use of the saturated model is for computing the deviance from the less-full model, instead of just saying "from zero". I guess it seems more proper to do it that way.