Repeated measures logistic regression PreviousNext
Mplus Discussion > Structural Equation Modeling >
 Isabel Thielmann posted on Thursday, January 02, 2014 - 4:18 am
Dear Dr. Muthén,

I want to analyze my data using a repeated measures logistic regression in a SEM framework. I assessed my categorical dependent variable (cooperation: yes vs. no) twice, namely in two different games that vary according to a within-subjects manipulation (independent variable 1, categorical). Now, I want to test whether a personality characteristic (between-subjects, latent factor) predicts cooperation in one game, but not the other. That is, I want to directly test the hypothesized interaction effect between personality and cooperation in the different games.
Can you help me how to model this in Mplus? Unfortunately, I have not yet found a solution on my own.

Thank you in advance!
 Linda K. Muthen posted on Thursday, January 02, 2014 - 11:52 am
You cannot regress a within-level variable on a between-level variable except by using a random slope as shown in Example 9.2. You can regress the between-part of the within-level variable on a between-level variable as shown in Example 9.1.
 Isabel Thielmann posted on Tuesday, January 14, 2014 - 3:20 am
Dear Lina,

Thank you very much for your helpful response. The model worked fine with an observed cluster-level covariate.
However, is there also an opportunity to include a latent cluster-level variable that is modeled based on multiple items?

Thanks in advance!
Best, Isabel
 Isabel Thielmann posted on Tuesday, January 14, 2014 - 3:21 am
Sorry for the typo, Linda!
 Linda K. Muthen posted on Tuesday, January 14, 2014 - 6:25 am
Yes, if the factor is a between-level factor with factor indicators that are between-level variables or the between part of variables measured on the individual level.
 Isabel Thielmann posted on Wednesday, January 15, 2014 - 4:46 am
Thank you very much for the information. It worked!

However, one question still remains: I tried the model with the following specification

y WITH s;

as suggested in the User's Guide Example 9.2. In this case, I get an error message telling me that I have a non-positive definite first-order derivative product matrix. This problem is due to parameter 6 which corresponds to the error variance (psi matrix) of y.
In case I do not allow y to be correlated with s (thus omitting the statement 'y with s;'), this error message does not occur.

Can you tell me what this means for the interpretation of my results?

Thanks again and best regards,
 Linda K. Muthen posted on Wednesday, January 15, 2014 - 6:13 am
Please send the output and your license number to
 bill kowalczyk posted on Wednesday, January 29, 2014 - 9:46 am
Hello I figure I would jump in here as I am trying to recreate an repeated measures logistic regression analysis done with the SAS GLIMMIX Macro in Mplus.

What I have is data gathered at random time points using ecological momentary assessment. My dependent variable is the presence or absence of cues for drug use. The IV is amount of time till the next use of a substance (in hour bins from 1-5). I am examining if there is a linear trend to increased likelihood of the cue being present as one gets closer to the use event. As this is a reanalysis of published data there should be. I've been trying various examples but I keep getting stuck.
 Bengt O. Muthen posted on Wednesday, January 29, 2014 - 11:26 am
I assume GLIMMIX uses a 2-level approach where time is a variable. If so, you can start by looking at UG ex 9.16.
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