Random slopes analysis and FIML
Message/Author
 Jakub Mikuska posted on Thursday, December 11, 2014 - 1:27 pm
Hello,

First of all I am a beginner in SEM and a complete beginner in Mplus.
I'm trying to conduct a random slopes analysis (type=random) of meta-analytic data. The predictor variables have some missing data, and I was expecting to get around it using Mplus (version 6) and the FIML approach.
The code I am using is based on Dr. Cheung's 2008 paper hosted on your site, but it leads to listwise deletion.
How can I use FIML approach to estimate parameters on the full sample, rather than only the cases with non-missing data?

Used syntax:
VARIABLE: NAMES interc w Zr x1 x2 x3;
!interc is a constant, w is a sample weight, most predictors have some missing data and are a mix of effect coded categorical and mean centered continuous variables
USEVARIABLES ARE interc Zr x1 x2 x3;
Missing are all (-99);
DEFINE: w2 = SQRT(w); !weighting is to account for different sample sizes of studies/cases, and to equalize the variances
interc = w2 * interc;
Zr = w2 * Zr;
x1 = w2 * x1;
x2 = w2 * x2;
x3 = w2 * x3;
ANALYSIS: Type=random;
MODEL:
[Zr@0.0];
Zr@1.0;
u | Zr ON interc;
Zr ON x1 x2 x3;
[u*];
u*;
 Linda K. Muthen posted on Thursday, December 11, 2014 - 2:08 pm
I believe the listwise deletion you refer to is for missing d ata on the observed exogenous covariates. Missing data theory applies only to observed endogenous dependent variables. To avoid the listwise deletion, mention the variances of the covariates in the MODEL command. They will then be treated as dependent variables and distributional assumptions will be made about them.
 Jakub Mikuska posted on Wednesday, December 17, 2014 - 2:33 pm
Thank you Dr. Muthen for your quick response.

Just to verify that I'm doing this the right way. Would the code for what you suggest be (based on the previous):

ANALYSIS: Type=random;
MODEL:
[Zr@0.0];
Zr@1.0;
u | Zr ON interc;
Zr ON x1 x2 x3;
[u*];
u*;
x1;
x2;
x3;

I tried running this and got results based on the full sample, but I want to make sure this is the right approach.
 Linda K. Muthen posted on Wednesday, December 17, 2014 - 3:59 pm
Yes.
 Jakub Mikuska posted on Saturday, December 20, 2014 - 11:31 am
Dear Dr. Muthen,
I was not able to diagnose why I am getting this error and wanted to ask what are the common causes and remedies for it:

THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES MAY NOT BE
TRUSTWORTHY FOR SOME PARAMETERS DUE TO A NON-POSITIVE DEFINITE
FIRST-ORDER DERIVATIVE PRODUCT MATRIX. THIS MAY BE DUE TO THE STARTING
VALUES BUT MAY ALSO BE AN INDICATION OF MODEL NONIDENTIFICATION. THE
CONDITION NUMBER IS 0.524D-14. PROBLEM INVOLVING THE FOLLOWING PARAMETER:
Parameter 20, ZR ON INTERC