Interactions: cross-product vs. multi... PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
 Thomas Olino posted on Wednesday, July 20, 2005 - 11:05 am
Is multiple group SEM always the preferred manner to examine moderation?

Are there any references that discuss the limitations or problems with including cross-product terms of observed variables to predict latent variables (as would be done in growth modeling)?

 bmuthen posted on Wednesday, July 20, 2005 - 11:29 am
Multiple-group SEM is only applicable when the moderator is observed and categorical.

The Mplus Version 3 User's Guide gives a table of interaction (moderation) modeling approaches.

I am not aware of any references specific to predicting growth factors. The West & Aiken interaction book is good at explaining interactions in regression.
 Lisa M. Yarnell posted on Friday, April 08, 2011 - 1:53 pm

If one uses the DEFINE command to add an interaction between two continuous, observed variables to the model, and finds the interaction to be significant, how does one probe the interaction (i.e., for "regions of significance")? I am familiar with probing interactions for regions of significance in OLS regression, but not in SEM.

Thank you,
 Lisa M. Yarnell posted on Friday, April 08, 2011 - 2:01 pm
I am thinking that I can probe an interaction in SEM similar to how I would do so in OLS regression: by trichotomizing one of the variables then graphing scores on the other variable in a plot.

However, the difficulty arises in that in the SEM model, the interaction term and its components can be correlated with covariates, as well as regressed on other exogenous variables. So, would the approach of trichotomizing one variable then creating a plot be too simplistic, when there are actually more complex relationships among the variables?
 Bengt O. Muthen posted on Friday, April 08, 2011 - 5:28 pm
Is it an interaction between latent variables?
 Lisa M. Yarnell posted on Friday, April 08, 2011 - 5:36 pm
No, both variables are measured variables.
 Lisa M. Yarnell posted on Friday, April 08, 2011 - 6:20 pm
Here is a link that displays the model:

The Parental Education X Young Adulthood Education interaction term has four significant effects, according to Mplus results--specifically, on three of the mediators and on health.

How do I probe the interaction to make tentative conclusions about how effects of Young Adulthood Education on Young Adulthood Health "depend" on levels of Parental Education, for example? I am just unsure about approaches to probing measured variable X measured variable interactions in Mplus or in SEM more generally.
 Bengt O. Muthen posted on Friday, April 08, 2011 - 8:51 pm
For that you want to study texts such as the Aiken-West regression interaction book.
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