Three-level model corrections in mult... PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
Message/Author
 psychresearcher  posted on Monday, June 27, 2011 - 5:21 pm
I have a data that is hierarchically nested. Patients nested within treating clinicians, and clinicians nested within clinics. Using univariate ordinal random intercept models I estimate my ICCs for my 10 items to be in the range of:

Level 2 ICCs range from .150 - .260.
Level 3 ICCs range from .001 - .088.

Level 2 group sizes range from 1-100.
Level 3 group sizes have as many as 100 level 2 clinicians in them with some clinics treating over 1000 patients.

My model is a two-level CFA with WLSMV and categorical data however my data structure 'may' have 3 levels

I appreciate that three-level factor models are not possible in Mplus but I wonder if the level 3 ICCs for some items may be non-trivial (the largest is .088) and given the level 3 group sizes I wondered if there is any way in Mplus in which a clustered sampling correction for level 3 can be used simultaneously with the 2-level CFA to correct the chi-square and standard errors to accommodate the level 3 dependency.
 Bengt O. Muthen posted on Monday, June 27, 2011 - 5:40 pm
You can use TYPE = COMPLEX TWOLEVEL and use
2 cluster variables:

CLUSTER = clinic doctor;

COMPLEX corrects chi-2 and SEs for the nesting in the clinic level and TWOLEVEL models the patient and doctor variation.
 psychresearcher  posted on Tuesday, September 06, 2011 - 10:28 pm
Thanks Bengt,

I notice that this method only provides a loglikelihood (as shown below) however the manual on p.234 and p.501 seems to suggest that the a corrected chi-square is provided.

Am I to assume therefore that this loglikelihood is intended to be used for nested model comparisons only?

Is there anyway to get a chi-square for use in a CFI or RMSEA for example?

MODEL FIT INFORMATION
Number of Free Parameters 74
Loglikelihood
H0 Value -213462.813
H0 Scaling Correction Factor 4.381
for MLR
Information Criteria
Akaike (AIC) 427073.626
Bayesian (BIC) 427659.151
Sample-Size Adjusted BIC 427423.982
(n* = (n + 2) / 24)
 Linda K. Muthen posted on Wednesday, September 07, 2011 - 7:52 am
If chi-square and related fit statistics are not given, they are not available because means, variances, and covariances are not sufficient statistics for model estimation. In this case, nested models can be compared using -2 times the loglikelihood difference which is distributed as chi-square.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: