Three-level dependence and type complex PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
 psychresearcher  posted on Tuesday, March 01, 2016 - 9:17 pm
Level 1: Patient
Level 2: Clinician
Level 3: Hospital

The design effect is larger than 2.0 for some items at Level 2 and is larger than 2.0 for nearly all items at Level 3.

I am only interested in looking at Level 1 while correcting the SE's/chi-2 for the non-independence associated with Levels 2 and 3.

"TYPE=TWOLEVEL COMPLEX" isn't really an option as this would force me to use a saturated Level 2 model and I wish to avoid this.

Is there any other flexibility within Mplus to deal with this?
 Tihomir Asparouhov posted on Friday, March 04, 2016 - 9:22 am
If you are only interested in looking at Level 1 while correcting the SE's/chi-2 for the non-independence associated with Levels 2 and 3 all you need to do is


There is no penalty for ignoring Hospital (that would matter only if you are making inference on the Clinician level)
 psychresearcher  posted on Friday, March 04, 2016 - 9:54 am
Thanks Tihomir,

I think some simulation evidence suggests ignoring a 3rd level could be problematic for level 1 bias but I can't be sure of this...

could clinician and hospital be controlled by combining these two ID's together such as

NEW_ID = Clinician_ID + Hospital_ID;

cluster = NEW_ID;

which would reflect the unique combination of clinician and hospital?
 Tihomir Asparouhov posted on Friday, March 04, 2016 - 11:55 am
I assumed you have unique Clinician_ID, if you don't you need to make sure you do, possibly with the command you suggest.

Actually I was incorrect in the previous message and indeed you are correct. It easy easy to see that if you consider the case where var on the second level is zero(the number of independent units will be number of hospitals and not number of clinicians). The Hospital variable should be ignored only if its ICC is small otherwise the SE will be incorrect. In fact the best practical thing to do is probably ignore the level (level 2 or 3) with smaller ICC. I can't think of an Mplus method that will give you what you want except
 shonnslc posted on Tuesday, August 27, 2019 - 9:02 am

My data have three levels:
1st level: pre-service teachers
2nd level: programs/departments
3rd level: schools

Since programs/departments are the same across schools (English, Social Science, Math, and Science), can I multiply 2nd and 3rd levels units to create unique 2nd level clusters? Then, I can use

type = complex
cluster = new2ndlevelunits (i.e., 2nd level units*3rd level units)

 Bengt O. Muthen posted on Wednesday, August 28, 2019 - 2:58 pm
This question is suitable for Multilevelnet or SEMNET.
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message