The SAS Output Delivery System and reporting techniques

Scoring Logistic Regressions

Reply
N/A
Posts: 0

Scoring Logistic Regressions

Hi all,

I was wondering if there was a way to score a test dataset based upon the results of a logistic regression I had run already, without writing out the equation myself.

It seems like the 'score' procedure works well, but is limited to linear combinations.

Thanks!
SAS Super FREQ
Posts: 8,866

Re: Scoring Logistic Regressions

Posted in reply to deleted_user
Hi:
This is not a BASE SAS or reporting procedure (PRINT, REPORT, TABULATE) question.

For more help with SAS/STAT procedures or your statistical questions, your best bet is to contact SAS Tech Support. To send a question to Tech Support, go to http://support.sas.com/ and in the left-hand navigation pane, click on the link entitled "Submit a Problem".

cynthia
N/A
Posts: 0

Re: Scoring Logistic Regressions

Posted in reply to Cynthia_sas
I haven't put my foot in my mouth this year yet, so I'm not being brave enough. Time to change that

ODS for most stat procedures, and I'm not clear which procedure you used, will write the results of the analysis to a table. By specifying a suitable target table, you should be able to capture the results of the regression and then insert those terms, either with macro coding, or by using the data set to write the sas code to a file which you then include.

There are many examples on the SAS web site for capturing data from a procedure to a data set and then reinserting it to a later step.
Super Contributor
Posts: 260

Re: Scoring Logistic Regressions

Posted in reply to deleted_user
Hi.
The LOGISTIC procedures offers a SCORE statement (new in version 9). It applies the current model to a different dataset. If you don't want to run the modelling step each time you want to score a new dataset, save your model with the OUTMODEL option when building the model, then use the INMODEL option in the scoring step.
And if you already used the SCORE procedure, just remember that the result has only to be transformed as Pred Prob(Y=1) = 1/(1+EXP(-1 * procScoreResult)).
Regards.
Olivier
N/A
Posts: 0

Re: Scoring Logistic Regressions

ooo, that's exactly what im looking for

thanks for that, i'll give it a shot!
Ask a Question
Discussion stats
  • 4 replies
  • 197 views
  • 0 likes
  • 3 in conversation