I am using BSEM to estimate a model with latent interactions. It seems the built-in method for latent interactions (LMS) described in the Mplus manual (Example 5.13) will not work with ESTIMATOR= Bayes. Please correct me if I'm wrong but I get the following error:
*** ERROR in MODEL command Interaction variables are not allowed with ESTIMATOR=BAYES.
Can you recommend another way to do latent interactions in BSEM? Would you recommend the unconstrained method by Marsh, Wen & Hau (2004, 2006)?
I tries an approach to estimating a latent interaction using Bayesian estimation. This approach requires specifying a single non-linear constraint, following this example (code snippet is from Coenders, Batista & Saris 2008):
model: eta1 by y1@1; eta1 by y2* (p1); eta2 by y3@1; eta2 by y4* (p2); eta3 by y1y3@1; eta3 by y2y4* (p3);
model constraint: p3=p2*p1;
I get the following error: *** FATAL ERROR THIS MODEL CONSTRAINT IS NOT AVAILABLE WITH BAYES ESTIMATION.
However, I was able to do the following in Bayesian estimation to test an indirect effect following an example in the manual: NEW(indirect1 indirect2); indirect1 = a*b1; indirect2 = a*b2;
Am I doing something wrong? or are non-linear constraints not supported with Bayesian estimation? If not, is there another approach? Another posts mentions phantom variables but I don't know what that is and can't find anything about this term in the manual.
Tibor Zin posted on Thursday, December 06, 2018 - 4:19 am
Dear Dr. Muthen,
I would like to ask a question about how to inspect the effect of IV on DV on different levels of moderating variable(Johnson-Neyman technique) using Bayesian estimator. Would it be correct approach to consider the value -1 as a minimum, 0 as a medium, and 1 as a maximum value?
Typically, you use 1 SD below and 1 SD above the mean for the moderator. Or, the 20th and 80th percentiles.
Tibor Zin posted on Saturday, December 08, 2018 - 11:04 pm
Thank you for the answer, Dr. Muthen. Please, could you tell me how to obtain SD and mean of a latent variable using Bayesian estimator or 20th and 80th percentiles? I know that the question may be trivial but I do not know how to proceed.
TECH4 gives you the mean and variance of a latent variable. The latent variables are assume to be normally distributed, so you can get the percentiles from a standard normal distribution table (subtracting the mean and dividing by the SD).
Tibor Zin posted on Monday, December 10, 2018 - 1:20 am
Thank you for the advice. The problem is that when I use ML estimator, the model is not correctly estimated. When I use Bayesian estimator, TECH4 output is unavailable. My goal was to estimate an interaction between two latent variables, type = random and algorithm = integration. Please, do you know what could be a problem or whether there is another way how to approach this problem?