problem getting var to randomize
in [SAS]
|
From: rjs on 11 Oct 2007 10:07 Estimated G Matrix not positive definite getting "." for the std error of some vars in the random statement getting this error and not sure why Some reasons I've been told 1 scaling of vars ran proc means and all vars are close to the same scale 2 vars could be linear combination of each other ran model with proc reg and the highest VIF is about 2.3 3 model contains binary vars of 4 binary vars only one is giving the "." error running model with 100 classes 30 vars and about 100 obs per class get the error ran the same model with only 12 classes get all vars to randomize please any help is greatly appreciated
From: Peter Flom on 11 Oct 2007 10:33 rjs <rjstratton(a)GMAIL.COM> wrote >Estimated G Matrix not positive definite >getting "." for the std error of some vars in the random statement >getting this error and not sure why >Some reasons I've been told >1 scaling of vars > ran proc means and all vars are close to the same scale >2 vars could be linear combination of each other > ran model with proc reg and the highest VIF is about 2.3 >3 model contains binary vars > of 4 binary vars only one is giving the "." error > >running model with 100 classes 30 vars and about 100 obs per class > get the error >ran the same model with only 12 classes > get all vars to randomize > >please any help is greatly appreciated Ummm.... It sounds like this is PROC MIXED, is that right? Could you provide the code and some context of what you are trying to do? What are these 'classes'? I think (but it's a wild guess) that you may be overfitting the model. Certainly, if you had 100 observations and 30 variables, that would be very problematic. Although you gain something by having multiple classes, you don't gain as much as if you had 3,000 observations in one class with 30 variables Also, you say 100 obs. per class, but was it the same in each class? HTH Peter
|
Pages: 1 Prev: About deleting a dataset Next: Functions in PUT statement |