I'm running PROC REG usimg a model with about 125 variables. What I'd like to do is to be able to estimate the relative contribution of each variable to the Model Sums of Squares. The Type I SS add up to the MODEL SS, but I believe it is dependent on the order of the variables in the model statement. Does anyone know a way to get the variance components that is not affected by the order of variables in the MODEL statement?
Sadly, I don't think the question has a meaningful answer.
Unless your 125 variables are statistically uncorrelated with one another (I doubt you can do this), then they are correlated with one another, and you cannot treat them as uncorrelated and say this variable accounts for 5% of the response, and this variable accounts for 3.7% and so on.
This is not a problem with SAS. This is a problem with the question that is being asked in this situation.
> If what you are interested in is contribution of each
> variable when all other variables are included in
> the model, then you can use Type III SS.
Using the Type III SS to tell you the contribution to total variability explained of each individual variable is a nice idea, but it doesn't really make a lot of sense when you have 125 variables which are correlated with one another. In fact, I wouldn't recommend PROC REG in this case at all; instead I would use PROC PLS.