07-17-2012 09:42 AM
quick question about proc logistic, do i need to define class variables ??
i have seen i few examples on the sas online docs some do and some dont .....
and just to make sure if i do define the class variables do i need to add the predicted to it ??? ( i dont think so .. )
07-17-2012 02:05 PM
You must define any nominal predictor variable (a predictor whose value as no numerical meaning, e.g. SEX, STATE) as a class variable. The dependent variable (predicted) is implicitly defined as class. - PG
07-17-2012 03:08 PM
To add to PG's comment, if your class variable is numeric and coded 0 and 1 (binary), then you can treat it as a continuous predictor variable (that's what we had to do before the CLASS statement was added to LOGISTIC). You will get identical p-values to a binary CLASS variable (like SEX), but the coefficients might be reversed depending on the parameterization.
Sometimes ordinal variables are treated as class variables and sometimes not (PROC LOGISTIC does not recognize a specific data type of "ordinal"). If you are willing to consider the ordered values to be equally spaced on the real line, then you could treat it as continuous and (generally) get reasonable answers. When I am trying to decide which way to go, I'll do a lot of plots ahead to understand the behavior of the ordinal variable relative to the outcome.