Loren Post posted on Tuesday, May 15, 2012 - 12:25 am
I have a basic question. Does Mplus automatically correlate dependent latent variables even if it is not written as a command? If so, does that mean the model then includes a disturbance correlation between the latent variables? ( I thought disturbance correlations were something to be avoided)
Mplus covaries as the default the residuals of final outcomes but not mediating outcomes. I am not sure where you heard disturbance correlations are something to be avoided. If they are large and not included, the model is misspecified. It is desirable for them to be small because if they are not, there are covariates that have been left out.
Loren Post posted on Tuesday, May 15, 2012 - 6:28 pm
Loren Post posted on Tuesday, May 15, 2012 - 8:01 pm
I'm sorry, I have yet another question. If I received a message that the latent variable covariance matrix is not positive definite and I want to report the problem do I refer to the y2 WITH y1 standardized model results (estimate = 1.047) or the TECH 4 estimated correlation between y2 and y1 (.985)? What is the difference between the two?
Loren Post posted on Wednesday, May 16, 2012 - 6:53 pm
Fantastic! Thanks so much!
Steve posted on Wednesday, August 28, 2013 - 9:10 pm
It was my understanding SEM models only include the correlation of exogenous latent variables - however it seems that Mplus is estimating correlation between dependent latent variables as well.
F1 by X1 X2 X3; F2 by X4 X5 X6; F3 by X7 X8 X9; F4 by X10 X11 X12; F3 on F1 F2; F4 on F1 F2;
The STDYX output lists correlation F3 with F4 (and they are theoretically correlated). So:
1) is this really a correlation as traditionally defined - or is this some kind of 'disturbance correlation'? 2) if 'disturbance correlation' - could yo please explain what that is? 3) does the inclusion of this estimate then somehow make this a feedback loop and the model non-recursive?