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
- /
- Need Help with Proc MIXED, nested random factor re...

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

07-06-2016 12:27 AM - edited 07-06-2016 11:22 AM

Hello,

I run an analysis by using PROC MIXED as implemented in the EG 7.1 software. I have the code below:

PROC MIXED DATA = WORK.SORTTempTableSorted covtest

PLOTS(ONLY)=ALL

METHOD=REML;

CLASS IsolStr IsolRepeatStr IsolVialStr InfStaStr;

MODEL TotFec=InfStaStr;

RANDOM IsolStr IsolRepeatStr(IsolStr) IsolVialStr(IsolRepeatStr IsolStr ) / TYPE=VC;

RUN; QUIT;

As can be seen in the code, I have a fixed factor, "InfStaStr," and several nested random factors. The random factors "IsolStr" and "IsolVialStr(IsolRepeatStr IsolStr)" had a varainace zero when the full model was run, but the nested term "IsolRepearStr(IsolStr)" did not a have a variance of zero (see attached file), and I get the "NOTE: Estimated G matrix is not positive definite."

I read in the SAS manual that in a case like this, random factors with variance zero can be romoved from the model. If I follow that advise, I would be left with a model with a single nested factor, "IsolRepearStr(IsolStr)," where the factor it is nested whitin, "IsolStr," is not present.

So, my question is: is it statistically sound to remove "IsolStr" in this case? Thank you for any help on this matter.

Igforek

Accepted Solutions

Solution

07-16-2016
02:13 AM

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

07-07-2016 03:39 AM

All Replies

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

07-06-2016 02:10 PM

For nested variance component (VC) effects, you can leave the 0-variance terms in the model. The results will be the same with or without them. Also, there is nothing wrong with leaving in a single random effects term as you described.

Solution

07-16-2016
02:13 AM

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

07-07-2016 03:39 AM