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
- /
- How to conduct a netsted 3-level multilevel model ...

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

06-15-2017 12:50 PM

Hello, everyone. I want to conduct a nested 3-level multilevel model where my response variable is a polytomous variable with 3 ordered categories (please see the attachment) with all the covariates being continuous. I have no level-1 covariates, X, B are level-2, whereas C-F are level-3 covariates. I am interested to see the relationship between X-Y controlling the other covariates. I was trying the syntax:

proc GLIMMIX data=Sample Data1 METHOD=LAPLACE NOCLPRINT; CLASS level2 level3; MODEL Y (order=internal ref=LAST) = X B C D E F/ CL DIST=MULTI LINK=CLOGIT SOLUTION ODDSRATIO (DIFF=FIRST LABEL); RANDOM intercept / SUBJECT=level3; RANDOM intercept / SUBJECT=level2(level3); RUN;

My intention was to see how the random intercept varies between level3, as well as between level2 within level3. However, I couldn't run this code as sas stops due to insufficient memory. To be honest, I am not really sure if this is the right way to write this code, I was following the syntax for the continuous response. Then I tried this next syntax:

proc GLIMMIX data=Sample Data1 METHOD=LAPLACE NOCLPRINT; CLASS level3; MODEL Y (order=internal ref=LAST) = X B C D E F/ CL DIST=MULTI LINK=CLOGIT SOLUTION ODDSRATIO (DIFF=FIRST LABEL); RANDOM intercept / SUBJECT=level3; RUN;

where I basically focused on the variation between level3. This code ran and provided some interesting results.

So these are the questions I am looking for an answer of,

1. Is the last piece of code I have written is correct? For a 3-level model, can I define my level-3 as the class ignoring level2?

2. Centering the covariates is a piece of the puzzle I can't seem to fit. all of my continuous covariates have meaningful zero so I would rather not center them. But all the literature I find on the multi-level model make arguments on group or grand mean centering, but no one says you don't need to center if you have meaningful zero or something.

Any suggestions, recommendations is very much appreciated! I really appreciate the help I got from this community in another question, it was of tremendous value. I could really use your help in this one as well. Thank you!