Repeated measures design with moderation PreviousNext
Mplus Discussion > Dynamic Structural Equation Modeling >
Message/Author
 Olivia Fitzpatrick posted on Sunday, December 02, 2018 - 12:02 pm
Hi there,

We are examining the effect of within-patient cognitive change (CC) at Session N on within-patient depression symptoms at Session N+1 across 10 sessions. I would like to test if this repeated-measures effect is moderated by stable interpersonal problems (IP) that were assessed at one time point (i.e., prior to treatment).

I would like to know if there is a way to model the interaction between interpersonal problems and cognitive change across all sessions (i.e., CC_IP = CC1-CC10*IP), rather than at each session (i.e., CC_IP1 = CC1*IP; CC_IP2 = CC2*IP; etc.). For context, we are running our main effect using ML-SEM syntax.

Thank you!
Olivia
 Maia ten Brink posted on Thursday, May 07, 2020 - 5:19 pm
Hello,

I wanted to follow up on this thread with a similar question.

I want to test for moderation (W = negative daily events) at a within-person level in a DSEM with three variables (sleep = X, stress = Y, coping = Z) measured repeatedly. The moderator W can potentially act on all three variables. Of course, multicollinearity is a major concern.

What syntax would be recommended for this kind of moderation?
Are there best practices suggested given multicollinearity (r of 0.2 to 0.3)?

My thinking was to try this:
MODEL:
%WITHIN%
Y_t on Y_t-1 X_t-1 Z_t-1 W_t-1 YW_t-1 XW_t-1 ZW_t-1
X_t on X_t-1 Y_t-1 Z_t-1 W_t-1 YW_t-1 XW_t-1 ZW_t-1
Z_t on Z_t-1 Y_t-1 X_t-1 W_t-1 YW_t-1 XW_t-1 ZW_t-1
%BETWEEN%
Y on X Z W XW ZW YW
X on Y Z W XW ZW YW
Z on X Y W ZW ZW YW

Thank you very much,
Maia
 Bengt O. Muthen posted on Friday, May 08, 2020 - 9:49 am
A critical factor for DSEM is the number of time points; I don't know what your T is. With large T, say at least 50, you can add random slopes, random variances, and random autoregressions.

I am guessing that what you are interested in is a model like:

y on x z xz

with appropriate lags added. I don't see why you want a y*z term. You can think about whether from a substantive point of view, you want y_t on y_t-1 which is DSEM, or if you want the residual of y_t regressed on the residual of y_t-1 which is RDSEM. The x and z variables can be "brought into the model" or not.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a public posting area. Enter your username and password if you have an account. Otherwise, enter your full name as your username and leave the password blank. Your e-mail address is optional.
Password:
E-mail:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: