Interactions in SEM PreviousNext
Mplus Discussion > Structural Equation Modeling >
Message/Author
 Chantal Hermann posted on Thursday, January 24, 2013 - 7:25 am
Hello,

Could you please tell me if it is possible to test an interaction between a latent variable (based on continuous indicators) and an observed variable that is categorical (0 = condition not present, 1= condition present)in SEM?
Both the continuous latent variable and observed categorical variable are exogenous (not predicted by other variables). If so what is the best way to go about this?
I was thinking I would use the MLR estimator.

Also, do I specify the categorical indicator as categorical in my syntax?

I have a medium sized sample of about 350 participants.

Thank you for your help!
 Linda K. Muthen posted on Thursday, January 24, 2013 - 7:39 am
You would use the XWITH option. You would not put the binary covariates on the CATEGORICAL list. This list is for dependent variables only.
 Chantal Hermann posted on Thursday, January 24, 2013 - 7:46 am
Hi Linda,

Great, thank-you. So can I use the following analysis syntax:

ANALYSIS: ESTIMATOR = MLR;
TYPE = RANDOM;
ALGORITHM = INTEGRATION;

Thank you for your help!
 Linda K. Muthen posted on Thursday, January 24, 2013 - 11:52 am
Yes.
 Maria Carrasco posted on Tuesday, March 03, 2015 - 9:30 pm
Hi Linda,

My sample size is 223 and I have a lot of missing data on the two observed binary variables included in the model. I would like to test the effect of an interaction term in the model. The two terms I am using to create the interaction term are hypothesized mediators. Both terms are continuous latent variables measured by ordinal items (5-item likert scales) with 7 and 8 items each. Most items have high kurtosis (the distribution is very pointy and thin in the tails).

Is it appropriate to try to run a model with an interaction term given the small sample size, missing data, and the fact that the items of the continuous latent constructs are not normally distributed?


ANALYSIS: type = random;

MODEL:
x BY C1* C2 C3 C4 C5 C6 C7 C8 C9;
x@1;
stigma BY S1* S2 S3 S4 S5 S6 S7 S8;
stigma@1;
stigmasw BY sw191* sw192 sw193 sw194
sw195 sw197 sw198;
stigmasw@1;
stigma ON x;
stigmasw ON x;
int | stigma XWITH stigmasw;
CCU_1 ON x stigma stigmasw;
CCU_1 ON int;
CCU_2 ON x stigma stigmasw;
CCU_2 ON int;

OUTPUT:
sampstat tech1 stdyx modindices(all);


Many thanks!

Maria
 Bengt O. Muthen posted on Wednesday, March 04, 2015 - 3:19 pm
You don't mention the degree of missingness - the coverage on the diagonal of what we print, but missing data certainly hurts when max n is only 223.

I don't know if you treat the 5-category Likert indicators as categorical, but if you do the latent variables can still be approx normal when the indicators are not.

I haven't thought through how indirect effects should be evaluated when mediators have an interaction term.
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: