- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
I was wondering if there's anyway to check if the equation is correct. I know for linear regression you can check with residuals. Is there something similar for logistic regression?
Accepted Solutions
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Check GOF and LACKFIT option of MODEL in PROC LOGISTIC .
@Rick_SAS wrote a couple of blog about it before.
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
I'm not sure what you mean by "correct" or why you need to check. SAS themselves (or is it SAS itself?) does thorough checking of the results of each procedure, so we don't generally have to do that checking. If the data going into your logistic regression is correct, and your code is correct, then I would be happy to assume that the results are correct. So that's what I would check, the data and the code.
Paige Miller
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
The most important thing for logistic regression is to have 5 to 10 events for each explanatory variable.
Example: if you have a variable to explain for N= 54 patients and 3 explanatory variables:
Var1 = numerical age counts as 1
Var2 = processing yes/no counts as 1
Var3 = occupational category (n = 8 ) counts as 7
So you must have: (1+1+7) x 5 = 45 < 54 patients (limit)
(1+1+7) x10=90 < 54 patients (not good at all) You have to think carefully before validating the model.
Hope this can help you.
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Check GOF and LACKFIT option of MODEL in PROC LOGISTIC .
@Rick_SAS wrote a couple of blog about it before.