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
- /
- Can i change the Maximum Likelihood Estimates in l...

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

05-03-2016 12:37 PM

In my solution after the logit-regression the Maximum Likelihood Estimate is the standardmethod. Is it possible to change to least-squares estimate and compare the difference between the two methods in the results?

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

05-03-2016 12:45 PM - edited 05-03-2016 12:47 PM

You may need to specify which procedure you are using. Proc Logistic supports least squares using the LSMEANS and LSMESTIMATE statements.

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

05-03-2016 03:22 PM

The LSMEANS and LSMESTIMATE statements compute marginal (least squares) means and compares them using the maximum likelihood estimates. They do NOT involve estimation using least squares. Logistic models are one in the class of generalized linear models which are all estimated using maximum likelihood estimation.

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

05-03-2016 01:54 PM

I'm not certain what you intend, but PROC LOGISTIC support iteratively reweighted least square solutions, and in fact that is the default solution technique. You can read about the numerical solution algorithms in PROC LOGISTIC in the doc. To compare you would need to run the procedure twice, once with each method.

But as far as I know there is no direct method for using least squares to compute a logistic regression, only iterative methods. The problem is that the conditional probability of the response p_i = E(Y_i | X_i) is not observed. So you have to use some iterative technique to optimize the likelihood function.