Will the various selection methods available in PROC REG etc or any other PROC ( ex: stepwise, forward, backward) work if you have more variables than observations?
I have a situation where I am trying to discover something new, with little or no literature or theory to tell me which subset of variables to use. Is there a selection method that would accommodate this?
If I can't use this type of selection method, is there another approach?
Should I first implement a data reduction strategy like factor analysis etc. ( I am sure that many of my potential explanatory variables could be correlated)