Continuous time and unbalanced measur... PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
 Tom Hildebrandt posted on Monday, October 17, 2011 - 8:53 am
I have a complex data questions and am wondering if mPlus can handle the modeling.

I have 3 days worth of behavioral data collected where individuals recorded their mood at the moment of a specific behavior. The complexity of modeling happens because the number of recorded events differs per individual, so I can't seem to find a way to model slope without having to assume a large amount of missing data for individuals who didn't engage in the behavior frequently.

Any suggestions on how to handle this?
 Bengt O. Muthen posted on Monday, October 17, 2011 - 9:50 pm
Can a two-level version of growth modeling handle this? That is, is the time dimension of growth relevant for studying change? You would then take care of individually-varying times of observation. See UG ex9.16. This also has the advantage over the usual wide, single-level approach of handling many time points.
 Tom Hildebrandt posted on Tuesday, October 18, 2011 - 10:12 am

Thanks for the quick response. Time does matter as the second part of the complexity is that there is a nonlinear relationship between time and mood characterized by oscillations of different frequency and amplitude. Frequency is equivalent to the number of observations per individual and oscillations reflected in alternating sign of the DV (mood), from negative to positive or vice versa.

So in the example ex9.16, I would model the within-subject change using the nonlinear constraints and have the ID be a cluster variable to account for different number of assessments?

Thanks for your help.
 Bengt O. Muthen posted on Thursday, October 20, 2011 - 9:52 am
 Jason Payne posted on Friday, July 12, 2013 - 12:18 am
I too have a complex data question, but not sure where on the forums it belongs.

I've been happily running LCGA models in MPlus for a while now, but I am trying to operationalise ZIP LCGA with exposure adjustment in the WIDETOLONG format because I now have missing data on both the response and the covariates (in this case the exposure). I was hoping not to lose the cases with missing data on the exposure, which led me to the use of WIDETOLONG in ex9.16.

Anyway, I'm a bit lost now on how to specify the model and I keep getting an error which i'm sure is my fault. I wondered whether you had any other examples similar to ex9.16 but where the time function is quadratic for the classes within the LCGA? I've been also running the models in Stata's TRAJ module which can handle missingness on the exposure, but i am keen to use Mplus if possible.

I believe i have set up the WIDETOLONG part of the code correctly - at least im not getting any errors here. I am however at a loss about how to specify the WITHIN and BETWEEN components.

Its a pretty simple two class model but I know i've completely misspecified it which is why an example would be very helpful. I also have no idea how to constrain the within class variances to zero consistent with the LCGA.
 Jason Payne posted on Friday, July 12, 2013 - 2:13 am
As an aside, I also used the description on this thread to guide me:

However, when in my case I specify:

i s q | conv on t;
ii si qi | conv on t;

i s q | conv on t;
ii si qi | conv on t;

I get the following error:

*** ERROR in MODEL command
Only one cross-sectional random effect can be defined at a time.
Definition for the following: I S Q

Any help you can provide would be much appreciated!
 Linda K. Muthen posted on Friday, July 12, 2013 - 10:20 am
Create a time squared variable and say

s1 | y ON time;
s2 | y ON testsq;

Please limit posts to one window.
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