Basic longitudinal model and intracla... PreviousNext
Mplus Discussion > Structural Equation Modeling >
 Holly Orcutt posted on Friday, May 20, 2005 - 2:45 pm
Hi. I am resubmitting a grant in the next few days. I am using college females who will be assessed in person at T1 and then every 3 months via internet survey (4 time points - T2 through T5). The primary questions of interest have to do with sexual revictimization so at each follow-up time point, I will be obtaining reports of trauma exposure in the previous 3 months. I have proposed reasonably sophisticated analyses using Mplus for modeling longitudinal data. All 3 reviewers were generally on board with my analytic approach. Except one comment that I really don't know how to respond to this:

"In the Analyses Section, within-class clustering (intra class correlations) are not taken into account."

I have already consulted with one person about this reviewer comment and she said, "The
intraclass correlation they are referring to is probably the correlation between scores on the same measure across time -- in repeated measures ANOVA, cross-time correlations between scores should be the same across each pair of time points. I don't know of any such assumption in SEM models. So basically, my response to this point would be to do nothing."

Would love some insight on this issue. I don't want to allocate more than a sentence or 2 responding to the point, but I'm at a loss here. Any help would be appreciated. Thanks!!
 Linda K. Muthen posted on Friday, May 20, 2005 - 5:05 pm
Is your data a random sample? Perhaps the reviewer is responding to the fact that you have lack of independence of observations that you need to take into account.

If they are referring to repeated measures, the random effect growth model does take this into account.
 bmuthen posted on Friday, May 20, 2005 - 5:37 pm
Just to add to Linda's response, if your data are not obtained by cluster sampling (such as sampling of schools within which students are observed), then the reviewer must refer to the correlations across time within a person and that is modeled by the standard growth model (and the correlations of adjacent pairs of time points need not be the same - the model can estimate them).
 Holly Orcutt posted on Friday, May 20, 2005 - 7:35 pm
thanks for the helpful responses - i am drawing a sample each semester for four semesters of females from the same university - repeating it for 4 cohorts because my phenomenon of interest is low base-rate. but there are no group-level variables in my study (such as different schools or organizations) - they are all drawn from the same pool and followed for 1 year (and this procedure will be repeated for 4 semester, 250 women per semester). it is a convenience sample of students from an introductory psychology subject pool. unless this information changes the issue (and i don't think it will), i will go with the idea that the correlations across time within a person can be modeled. thank you!!
 bmuthen posted on Saturday, May 21, 2005 - 8:21 am
So I assume this means that you observe the same group of females at 4 time points during a year? The fact that you have several cohorts is just an added wrinkle that may lead to mutiple-group analysis. Given that you are working with only one university, this then means that the intraclass correlation for each person arises because of repeated measures on that same person - and growth modeling handles this as we said earlier.
 Holly Orcutt posted on Sunday, May 22, 2005 - 11:01 am
thank you. very helpful!
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