Interaction covariates (intercept & s... PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
Message/Author
 JBP posted on Tuesday, August 23, 2011 - 4:35 pm
Dear Professors,
I'm running a simple LGM (7 repeated measures of single outcome) with two observed predictors (one categorical and one continuous that could be dichotomized if necessary).

I use the following syntax:

USEVARIABLES ARE P1 P2 VD1-VD7 inter;
DEFINE:
inter= P1* P2;

ANALYSIS:
TYPE=MISSING;
estimator=MLR;

MODEL:
i s | VD1@0 VD2@1 VD3@2 VD4@3 VD5@4 VD6@5 VD7@6;
i s ON P1 P2 inter;

However, I first did my analyses with multilevel modeling and I’m trying to replicate it on Mplus to profit from additional features. When I do not enter the interactions I get exactly the same estimates from both approaches. However, when I enter the interaction, the estimates are not the same. So I wanted to see if the syntax I’m using is correct.
Also, in multilevel I first entered the interaction of P1xP2 for the intercept and then for the slope (which corresponds to a three-way interaction with time). Is it ok to fit the two following models or it does not make sense:
i s ON P1 P2 ;
i ON inter;

and a second:
i s ON P1 P2 ;
i s ON inter;

I hope the questions are relevant,

Thanks!
 Linda K. Muthen posted on Wednesday, August 24, 2011 - 1:08 pm
Please send the relevant outputs and your license number to support@statmodel.com.
 Stephanie Smith posted on Saturday, January 07, 2012 - 8:17 pm
Dear Professors-
I am also running a simple growth curve model for my dissertation with 8 time points (violent behaviors across 4 months, a time point exists for every two weeks). I have two observed predictor variables (both continuous) - trait anger and anger rumination. I have hypothesized that the interaction between my two predictor variables predicts an increase in violent behaviors over time so that individuals who tend to ruminate to anger and are also high on trait anger (but not low on trait anger) show an increase in aggressive behaviors over time. I was wondering if I should use the same syntax as above to create the interaction term. If it is significant, I was wondering how I should follow-up on this interaction and how this interaction would be interpreted. Any help is greatly appreciated - thanks so much!
 Linda K. Muthen posted on Sunday, January 08, 2012 - 7:14 am
For observed variables, the interaction can be created in DEFINE. This interaction would be interpreted in the regular way.
 Stephanie Smith posted on Sunday, January 08, 2012 - 12:01 pm
Great! Thanks so much for the quick reply.

How would I follow-up on this significant interaction to determine if individuals who ruminate to anger and are high on trait anger show an increase in aggressive behaviors? Would I use the same methods as I would in a regular regression model - look at people high and low on trait anger (1SD above and below) and see if anger rumination is still a significant predictor of an increase in aggressive behaviors?
 Linda K. Muthen posted on Sunday, January 08, 2012 - 5:04 pm
Yes, you would do this the same way as in regression.
 Sarah Dermody posted on Thursday, January 29, 2015 - 4:03 am
I am examining if the effect of an intervention on change in an outcome over time (linear latent growth curve model with dichotomous indicators and WLSMV estimator) is moderated by a number of observed, categorical demographic covariates. My question is, when examining interaction effects on the latent slope term, is it also important to include the relation between the interaction with the intercept term? My slope and intercepts are significantly correlated (p < .01), and depending on when I include the relation of the interaction with the slope only or slope and intercept (both using an ON statement) I get different results.
 Bengt O. Muthen posted on Thursday, January 29, 2015 - 2:01 pm
First, you need to settle if time point 1 is pre intervention, with intervention starting between time 1 and time 2, where the intercept factor is defined at time 1 by time score = 0 at time 1. If so, I think you should include the main effects of demographics on the intercept factor because otherwise you are saying that its mean is the same across these demographic groups, but you should not include the demographic interaction with treatment because treatment has not happened yet at time 1.
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: