REGRESSION - AUTOMATICALLY CHOOSE THE BEST MODEL

[AlexeyS]

Hi All,

I don't remember the name of SAS Procedure that run a lot of different regressions(transformations) based on specific variables(model) and choose the best.

It can be linear regression or quadratic regression, it can contain different transformations of variables and so on.

Thank you

[PeterClemmensen]

I think you're looking for PROC GLMSELECT.

If not, consult the same documentation and choose from the list on the left.

[PaigeMiller]

I will add that PROC GLMSELECT will select a model for you, it generally cannot be considered as selecting the BEST model. Each method in PROC GLMSELECT will likely choose a different model, and it may be that none of them are BEST in any global sense ... if there is even such a thing as the global BEST model, which I think there is not.

Also, the methods of forward selection, backward elimination and stepwise selection are considered problematic by many statisticians and generally I avoid them.

[AlexeyS]

Can you add please any example how proc glmselect will select model for me? Because in proc reg, if specify method, it also knows to select model(stepwise, backward...). But what i mean by selecting model, that procedure knows to do transformations on variables or create not linear but quadratic model by itself?

[PaigeMiller]

There are plenty of examples at the link given by @PeterClemmensen 

I don't think you can have the program fit smoothing/transformation or polynomial models unless you specifically tell it to do so. It does not figure out that you need a transformation by itself.

[SteveDenham]

Hmm. That sounds a bit like PROC TRANSREG, but much of the model "selection" will have to be done manually based on output. However, you can get optimal transformations of the variables which may be what you are looking for.

SteveDenham