Monte Carlo Analysis Ignoring Covariates PreviousNext
Mplus Discussion > Structural Equation Modeling >
 Felix Flory posted on Tuesday, May 16, 2006 - 1:32 pm
In a simple two group analysis I want to generate data for y1, a continuous (observed or latent) outcome, that is regressed on z1, an independent continuous (observed and/or latent) covariate, within each group. I want to study what happens if z1, the covariate, is ignored in the analysis. I. e. I want to do a two group meanstructure analysis without the covariate, simply comparing the two group means of y1. Do I have to generate the data externally for this? So far I get warnings that z1 does not covary with any other variable, if I do not mention it in the ANALYSIS option (it is howerver used in the MONTECARLO option to generate y1).
 Linda K. Muthen posted on Tuesday, May 16, 2006 - 2:06 pm
If you want to use a subset of the variables that are generated in the analysis, you need to use the external Monte Carlo facility. See Example 11.6. You would use the USEVARIABLES option to select all by the z1 variable.
 Felix Flory posted on Wednesday, May 17, 2006 - 8:40 am
Thank you so much for you help - that was exactly what I wanted to confirm!
 Daniel J. Beal posted on Wednesday, March 14, 2007 - 12:06 pm

I am interested in running a Monte Carlo study examining moderated multiple regression (interaction terms in a regression model). Although regression is a very basic analysis in Mplus, I cannot seem to figure out a way to obtain an interaction term in a Monte Carlo. In a typical (i.e., non-simulated) model, you can use DEFINE for this purpose, but this does not appear to work for simulations in Mplus. Is there a way to simply multiply two simulated variables together for use in a Monte Carlo?
 Bengt O. Muthen posted on Thursday, March 15, 2007 - 4:57 pm
I have been thinking that you could use "external" Monte Carlo in Mplus for this, that is, in step 1 you use Mplus MC to generate data sets and then in step 2 (the analysis step) you can use Define to create the interaction. But this then requires creating data with this interaction. So perhaps instead this should be done in regular Mplus MC using XWITH where you put a factor behind each of the 2 x's.
 Ron Eldridge posted on Wednesday, April 02, 2014 - 11:02 am

I am new to SEM and Monte Carlo so please forgive this question if the answer is readily apparent.

How would the two analyses differ from one another,

#1) Using Monte Carlo model population to simulate a MIMIC model with two independent covariates (x1 x2), saving the data, then analyzing it through the external Monte Carlo feature with a model using only one of the covariates (x1).

#2) Creating the same data as above, but analyzing it through the internal Monte Carlo feature, specifying the correct population parameters through model coverage, but in the model statement setting the parameter value of x2 to zero (Y on x1 x2@0).

Thank you,
 Linda K. Muthen posted on Wednesday, April 02, 2014 - 11:06 am
I think these should be the same but you can find out for sure by doing the two analyses and comparing them.
 QianLi Xue posted on Friday, June 02, 2017 - 11:49 am
In your response to the post dated March 14, 2007 regarding creating an interaction term in a Monte Carlo. You responded by saying: "... So perhaps instead this should be done in regular Mplus MC using XWITH where you put a factor behind each of the 2 x's."

Could you please elaborate on what you meant by "put a factor behind each of the 2 x's"?
 Bengt O. Muthen posted on Friday, June 02, 2017 - 2:01 pm
Instead use the random slope approach of Table 3.8 in our new book - see postings including the output on our website.
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message