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 Diagnosis for Logistic Regressio...

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

06-03-2010 02:04 PM

I am running Proc Reg to check multicollinearity for logistic regression models. Almost all the independent variables are categorical variables. I constructed dummy variables and put K-1 dummies in Proc Reg models. For collinearity diagnosis in Proc Reg, there are two options, COLLIN and COLLINOINT. I am wondering if I use the same model for these two options as the later will exclude the intecept from calculation. Should I put all dummies rather than k-1 dummies while using COLLINOINT option? Thanks!

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

06-03-2010 04:17 PM

With more than one categorical variable, I would run the collinearity diagnostics using k{i}-1 dummy variables for the i-th categorical variable AND I would include the intercept. By using k{i}-1 dummy variables for the i-th categorical variable, you do not overparameterize the model with the reference level for any of your categorical variables. Inclusion of the intercept along with the k{i} - 1 dummy variables also does not result in an overparameterized model.

If you were to use k{i} dummy variables for each categorical variable and you have two or more categorical variables, then you will end up with an overparameterized model. So, it is best to use k{i}-1 dummy variables and include the intercept.

If you were to use k{i} dummy variables for each categorical variable and you have two or more categorical variables, then you will end up with an overparameterized model. So, it is best to use k{i}-1 dummy variables and include the intercept.