|
|
 |
| MIMIC Model with Mean Structure? |
 
|
| Message/Author |
|
|
|
|
I am estimating a MIMIC model using MLR with 11 indicators (continuous), 2 latent factors, and one exogenous covariate with directional paths to both latent factors. The factor disturbances are also correlated with one another. Including the mean structure, I am calculating 12*13/2=78+12=90 unique parameters in the variance-covariance matrix. With 36 freely estimated parameters (9 loadings, 11 residual covariances of the observed variables, 11 intercepts of the observed variables, 2 direct paths, 1 covariance between factor disturbances, 2 factor disturbances), I should have 90-36=54 degrees of freedom.
My output does indeed indicate 36 freely estimated parameters, but says that I have 52 (not 54) degrees of freedom. How could this be? I have included my model syntax below.
Analysis:
estimator=mlr;
Model:
N BY a1sf4c a1sf4h a1sf4m a1sf4s a1sf4e;
E BY a1sf4a a1sf4f a1sf4k a1sf4w a1sf4o a1sf4t;
E ON A1PAGE_M2;
N ON A1PAGE_M2;
OUTPUT: STANDARDIZED TECH4 TECH1 RESIDUAL MODINDICES; |
|
|
|
|
| You count 12 variables for the covariance matrix but Mplus makes a distinction between DVs and covariates. This is explained in the FAQ on our website: Covariates – analysis conditioned on covariates. The covariates' means and variances are not counted. So Mplus considers H1 as the model with an 11x11 covariance matrix and 11 means plus 11x1 covariances between the DVs and the covariate. |
|
|
|
|
| Thanks very much, this makes sense! Are the variances of covariates not included in free parameter calculations in MPlus? |
|
|
|
|
| That's right - no parameters related to the covariates are included in the count. |
|
|
|
|
| Thanks very much! |
|
|
| Back to top |
|
|
