I am scoring my dataset using a logistic model. For getting the betas (coefficients) I used
proc logistic y = x1 x2 x3.
To tell the model which are class variables I am using
class statement. Now, lets say if I have a categorical variable (with name ppsc), which has 4 categories, the betas are generated for top 3 categories (ppsc1, ppsc2, ppsc3 ) and I guess the fourth category is taken as reference. Now to score, I am using
proc score to score, therefore I have to generate 3 binary variables ppsc1 ppsc2 and ppsc3 such that
if ppsc =1 then ppsc1= 1 else ppsc1 = 0.
... ans same for ppsc2 and 3.
Now my questions are:
If my category 4 is most important (number vise) and the logistic is generating betas for ppsc1, ppsc2 and ppsc3, what do I do? So for 4th category my logit(p) will be zero for that variable if the value of ppsc is 4? How do I handle this problem? Would this hold for other categorical variables too?
I dont want to make those ppsc1, ppsc2 and ppsc3 binary variables. Can't I specify at the time of scoring that my following variables are categorical variable by some class statement just like we do in
I hope my problems are clear... I am BTW more concerned about 1st problem as that is something fundamental to understanding how the score is getting generated.