Modelfit /Multilevel Analysis
Message/Author
 Martin Geisler posted on Thursday, July 26, 2012 - 2:08 am
Hello,

I would like to ask a question about how to interpret the model fit. In my output I get this information:

Chi-Square Test of Model Fit
Value 0.000*
Degrees of Freedom 0
P-Value 1.0000
Scaling Correction Factor 1.000
for MLR

Chi-Square Test of Model Fit for the Baseline Model

Value 548.239
Degrees of Freedom 6
P-Value 0.0000

CFI/TLI

CFI 1.000
TLI 1.000

Information Criteria

Number of Free Parameters 9
Akaike (AIC) 1454.645
Bayesian (BIC) 1496.068
(n* = (n + 2) / 24)

RMSEA (Root Mean Square Error Of Approximation)

Estimate 0.000

SRMR (Standardized Root Mean Square Residual)

Value for Within 0.000
Value for Between 0.002

 Linda K. Muthen posted on Thursday, July 26, 2012 - 6:28 am
Your model is saturated. You have no degrees of freedom so model fit cannot be assessed for your model.
 Martin Geisler posted on Thursday, July 26, 2012 - 8:24 am
Thank you so much for your answer! But I still don't understand. A saturated model has as many parameters as it has datapoints. In fact, I have 1200 oberservations, one dependent variable and six indipendent variables. It is a twolevel analysis with five predictors on level 1 and one predictor on level 2. So it should not be saturated! Or is there something I don't get?
 Linda K. Muthen posted on Thursday, July 26, 2012 - 11:56 am
 MT posted on Tuesday, August 14, 2012 - 5:37 am
Dear Linda,

I have a question regarding multilevel analysis. We conducted a diary study in which we measured all variables at the within-person level, so there are no variables that should be modeled at the between-person level. However, since the data is nested within the persons (N = 47, and days = 214) we want to use multilevel analysis.
Should we specify a model (the same model) at both levels when using type is twolevel, even though we did not measure anything on the between-level?

 Linda K. Muthen posted on Tuesday, August 14, 2012 - 12:09 pm
You don't need a multilevel model when you have several variables per person. Multivariate modeling takes this into account.
 Anonymous posted on Friday, January 26, 2018 - 6:58 am
Dear Profs. Muthén,
I ran several threelevel random models using FIML. When comparing the fit of the different models using the Satorra-Bentler scaled chi-square difference test, I find the results contradictory: Adding more variables to the model leads to an estimated H0 Value for the loglikelohood that is more negative than without these additional variables. Shouldn't the loglikelihood value be more positive or at least unchanged when I incluce more variables?
I have the impression that it might have something to do with using FIML. Is that right? How would I conduct a loglikelihood ratio test based on FIML results, then?
Thank you very much in advance!
 Bengt O. Muthen posted on Friday, January 26, 2018 - 4:10 pm