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
- /
- HAS PROC REG been out of date?

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

07-18-2016 09:23 PM

Dear All,

Just want to ask a question: I think PROC MIXED, PROC GLM are very powerful in the regression analysis, could I say that PROC REG has been out of date?

What is the main reason that we should still keep such procedure in the SAS package?

Best,

Jack.

Accepted Solutions

Solution

07-19-2016
09:03 AM

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

07-19-2016 07:57 AM

I would argue that you want to use the simplest procedure possible that solves your problem, not the most general. The more specific procedures have additional tests that can be more powerful.

I think logistic regression is the best place to see this. PROC GENMOD can fit logistic models, but PROC LOGISTIC has specific tests (such as the Hosmer-Lemeshow GIF test) and graphics (ROC curves) that are applicable in the special case of a binary response.

PROC REG uses OLS regression, which is faster and gives different results than iterative MLE estimation in the generalized regression procedures. If you have continuous variables and want to run OLS regression, PROC REG provides a direct computation.

To use a math analogy, you could use complex numbers to balance your checkbook, but you don't. Real numbers are simpler, easier to work with, and they get the job done.

All Replies

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

07-18-2016 10:40 PM

proc reg has some unique features that could be implemented in glm or mixed but I guess the real reason proc reg cannot go away is the tremendous legacy of code relying on it.

PG

Solution

07-19-2016
09:03 AM

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

07-19-2016 07:57 AM

I would argue that you want to use the simplest procedure possible that solves your problem, not the most general. The more specific procedures have additional tests that can be more powerful.

I think logistic regression is the best place to see this. PROC GENMOD can fit logistic models, but PROC LOGISTIC has specific tests (such as the Hosmer-Lemeshow GIF test) and graphics (ROC curves) that are applicable in the special case of a binary response.

PROC REG uses OLS regression, which is faster and gives different results than iterative MLE estimation in the generalized regression procedures. If you have continuous variables and want to run OLS regression, PROC REG provides a direct computation.

To use a math analogy, you could use complex numbers to balance your checkbook, but you don't. Real numbers are simpler, easier to work with, and they get the job done.

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

07-19-2016 09:04 AM

Many thanks for your quick and detailed clarification.

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

07-20-2016 01:32 PM

There really is no other procedure available for looking at collinearity diagnostics. For that alone, PROC REG is worth having.

Steve Denham

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

09-29-2016 04:22 AM

And don't forget COOK-D statistic which can check which obs has the most contribution to model.