

Bayes test of indirect effect in MSEM 

Message/Author 


I am running a 2level MSEM and want to test indirect effects using Bayesian credible intervals rather than the tests provided by MODEL INDIRECT because my understanding is that the MODEL INDIRECT tests, like Sobel tests, assume that the indirect effects are normally distributed over repeated sampling, and they can produce biased significance tests as a result. However, I also have equality constraints placed on some parameters. When I try to run the model, I receive this error: *** FATAL ERROR VARIANCE COVARIANCE MATRIX IS NOT SUPPORTED WITH ESTIMATOR=BAYES. WITHIN A VARIANCE COVARIANCE BLOCK TWO OR MORE PARAMETERS ARE EQUAL. ONLY FULL VARIANCE COVARIANCE BLOCKS WITH UNEQUAL PARAMETERS ARE ALLOWED. USE ALGORITHM=MH TO RESOLVE THIS PROBLEM. Algorithm=MH doesn't work for a twolevel model. Do you have a suggestion for how to obtains Bayes CIs for indirect effects in a 2level MSEM featuring equality constraints? Thank you! 


You can express the indirect effect in Model Constraint. This will give you the nonsymmetric Bayes credibility intervals. Model Indirect is expanded for Bayes in the upcoming Mplus version 7.4. It does not rely on normality assumptions for the indirect effect estimate distribution. Regarding the Fatal Error message, please send input, output, data, and license number to Support. 


Dear dr. Muthen, I am running a two level MSEM model with indirect effects at both levels. I specified the indirect effects using the Model Constraint. I do, however, receive the same FATAL ERROR as described above. I think that this error is caused by two lines in my code that I use to obtain covariances: a with c var1 var2 var3; c with var1 var2 var3; When I remove these two lines, my model runs. I already tried to run my model with the ALGORITHM=GIBBS(RW), but also this does not seem to work. Is there another way to get my model running with the covariances in there? I use Mplus version 7. Thank you! 


Send the output for the Fatal error run to Support along with your license number. 

Back to top 

