topic Re: PROC LOGISTIC - only missing values for P_1 and P_0 in SAS Programming

PROC LOGISTIC - only missing values for P_1 and P_0

miriam93 — Thu, 17 Sep 2020 13:40:15 GMT

Dear SAS community,

I am having a personal crisis due to PROC LOGISTIC.

I am training my model using a data set scoringsatz, and I want to test it using testsatz, like this:

PROC LOGISTIC DATA=&scoringsatz
          outest=work.Model_Parameters
        outmodel=work.model_out
          PLOTS(ONLY)=ALL
          ;
class &varlist_model
;
     MODEL &ziel (Event = '0')=
&varlist_model
     /
          gof
        SELECTION=none SLE=0.05 SLS=0.05
          CORRB
          RSQUARE
          CTABLE
          LINK=LOGIT
          CLPARM=WALD
          ALPHA=&alpha.
          
     ;
     OUTPUT OUT=&scoringsatz._out (LABEL="" DROP=__FLAG) 
          PREDPROBS=INDIVIDUAL;
    score data=&testsatz out=ergebnis_out  ;

But here's the problem: ergebnis_out only shows missing values for P_1 and P_0. And I have been trying to figure out for 2 days now why that is.

- Scoringsatz and Testsatz have exactly the same number of columns and column names.

- If I test the model using the trainingset, it works.

- If I split the trainingset using surveyselect and then use 1 of the sets for training and the other for testing it also works.

So, it should be due to the testset? But except for different entries it has the exact same structure??

I cannot post parts of the datasets as they are sensitive data with personal info.

Thanks for any hints,

Miriam

Re: PROC LOGISTIC - only missing values for P_1 and P_0

PaigeMiller — Thu, 17 Sep 2020 13:46:33 GMT

The only reason I know of to get missing values for P_1 and P_0 is because there are missing values in your predictor variables. So you need to look in the data sets involved to see if there are missing values there for the predictor variables.

Re: PROC LOGISTIC - only missing values for P_1 and P_0

PGStats — Thu, 17 Sep 2020 16:35:38 GMT

You say "Scoringsatz and Testsatz have exactly the same number of columns and column names." That's good but not enough. The levels of class variables must also be the same. If a variable is coded as YES or NO in Scoringsatz and Yes or No in Testsatz, the procedure will not calculate the probs.