turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Stat Procs
- /
- PROC LOGISTIC parametrization of class variables

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

07-22-2016 09:11 AM

I'm creating a model using proc logistic with a code of the following form:

**proc** **logistic** data=work.MODEL_DATASET;

class cat_var1 cat_var2 cat_var3 cat_var4 cat_var4 cat_var5 cat_var6 cat_var7 /order=freq desc param = glm;

model outcome(event='1')= con_var1 convar2 cat_var1 cat_var2 cat_var3 cat_var4 cat_var4 cat_var5 cat_var6 cat_var4*cat_var6;

I changed the code slightly so that it used reference parameterization i.e. the only change I made was to replace param = glm with param = ref. I was expecting to see almost identical results, but for some of the variables I'm seeing different parameter estimates and odds ratios. Can anyone explain why this is? I'm using SAS 9.4.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

07-22-2016 09:56 AM

I'm assuming the variables that are different are the class variables?

if so, that makes sense as you've changed your parameterization method and the odds ratio is in comparison to a level. If you've changed that level your estimate would change. Your p-values should be the same.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

07-22-2016 09:59 AM

Thanks for the quick response - the variables that are different are some (but not all) of the class variables. However, as far as I can see the same reference level has been used in both parameterizations and so I don't understand why the coefficients have changed. Also, the value of the intercept has changed.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

07-26-2016 09:26 AM

Also, the odds ratios differ between the two parametrizations. With the param=GLM, there is a convergence warning message which states "WARNING: The information matrix is singular and thus the convergence is questionable. Try specifying a larger SINGULAR= value." There isn't a convergence warning message when param=REF is used.

Could this be the reason for the different odds ratios and parameter estimates between the two options? When would there be a convergence issue with one parameterization and not another?

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

07-26-2016 09:34 AM

I wouldn't expect convergence issues.

Ideally you could provide sample data so we could replicate the issue.

If this isn't possible you should contact SAS Tech Support for help.