Tobit modeling and SEM PreviousNext
Mplus Discussion > Structural Equation Modeling >
Message/Author
 Erin P posted on Monday, July 09, 2012 - 7:59 pm
I'm hoping you can help me figure out where I'm going wrong...

VARIABLE:
...
MISSING ARE .;
CENSORED ARE outcome (b);

MODEL:

F1 BY v1-v3

F2 BY v4;
F3 BY v5;
F4 BY v6;
F5BY v7;

v4@0;
v5@0;
v6@0;
v7@0;

F6 BY v8;
F6 ON F1-F5;

F7 BY outcome;
F7 ON f6;

outcome@0;
v8@0;

----------------------------
This is the error message I get:
*** FATAL ERROR
THE STARTING VALUE FOR THE VARIANCE OF A CENSORED VARIABLE outcome IS NOT POSITIVE.
 Linda K. Muthen posted on Monday, July 09, 2012 - 9:23 pm
I would imagine it is for setting the residual variance of outcome at zero. It is not necessary to put a latent variable behind an observed variable in Mplus. Simply specify your model using the observed variables.

MODEL:
F1 BY v1-v3
v8 ON f1 v4 v5 v6 v7;
outcome ON v8;
 Barbara O. posted on Tuesday, January 08, 2013 - 6:06 pm
I'm conducting a path analysis using a censored outcome variable (time 3 variable), 3 continuous mediators (time 2 variables), and 3 initial (time 1) predictors. The analysis requires montecarlo integration, and I was able to use the constraint command to estimate the indirect effects.
I'm wondering:
1) Is there a way to use bootstrapping to test for sig mediation (rather than testing the indirect effects)? I get an error message when I try.
2) The output provides AIC, BIC, and my log-likelihood value, but not RMSEA or GIF- is there a way to get these values using path analysis with a censored outcome variable?
 Linda K. Muthen posted on Tuesday, January 08, 2013 - 6:21 pm
1. What is the message?

2. Chi-square and related fit statistics are available only when means, variances, and covariances are sufficient statistics for model estimation. This is not the case when numerical integration is needed.
 Barbara O. posted on Tuesday, January 08, 2013 - 6:46 pm
1. The error message I receive is:
"BOOTSTRAP is not allowed with ALGORITHM=INTEGRATION."

I'm concerned reviewers might ask for RMSEA. If so, is a response similar to yours above something appropriate to give?

Thank you for your time!
 Linda K. Muthen posted on Tuesday, January 08, 2013 - 8:31 pm
1. BOOTSTRAP has not yet been developed for numerical integration.

2. RMSEA does not exist when means, variances, and covariances are not sufficient statistics for model estimation.
 David Peterson posted on Monday, September 07, 2015 - 10:51 am
Given that model fit indices are not provided for censored models, what is the recommended way of assessing/reporting model fit? Thanks!
 Bengt O. Muthen posted on Monday, September 07, 2015 - 1:58 pm
WLSMV gives model fit tests with censored. But with ML you have to work with "neighboring models", that is, add a less restrictive model and see if its extra parameters are significant.
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: