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
- /
- multicollinearity in Logistic Regression

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
- Permalink
- Email to a Friend
- Report Inappropriate Content

08-12-2016 01:05 PM

Dear Team,

I am working on a C-SAT data where there are 2 outcome : SAT(9-10) and DISSAT(1-8). I have approx. 22 predictor variables most of which are categorical and some have more than 10 categories. As with Linear regression we can VIF to test the multicollinearity in predcitor variables. While searching from SAS forum itself i realized we can use "**influence**" as a measure but that helps with outliers. Moreover from this post

https://communities.sas.com/t5/SAS-Statistical-Procedures/Outliers-and-Multicollinearity-for-Regress... there is a link explaining the diagnostics however i do not understand the outcome in detail. Is there any other approach.

Also can we use stepwise/forward/backward regression to remove non signifincant predictors at a given p value. Kindly advice.

Attached is the data for reference. I am using Base SAS.

Regards, Shivi

Accepted Solutions

Solution

08-20-2016
03:24 AM

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

Posted in reply to Shivi82

08-18-2016 09:49 PM

All Replies

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

Posted in reply to Shivi82

08-13-2016 12:16 AM

Unlike proc reg which using OLS, proc logistic is using MLE , therefore you can't check multicollinearity. But SAS will automatically remove a variable when it is collinearity with other variables. Yes. you can use stepwise/forward/backward to remove non signifincant predictors.Like: proc logistic; model y=x1 x2....x40 /selection=stepwise; run; Check documentation to see more examples.

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

Posted in reply to Ksharp

08-18-2016 02:06 PM

Thanks for the help.

I am using WOE & IV to reduce the number of predictors in the model as these can assist with both nominal and continuous variables.

Regards.

Solution

08-20-2016
03:24 AM

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

Posted in reply to Shivi82

08-18-2016 09:49 PM

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

Posted in reply to Ksharp

08-20-2016 03:16 AM

Thank you for assistance again.

Seems like the more you explore SAS the more you realize how vast & robust it is.

As you have suggested i will start witih build stepwise, forward & backward models and will do a comparison as i am not educated on Proc GLM Select and probably may not time as of now. But i will for sure check it in the near future.

Thanks again.