Error structure in multilevel models PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
 Eric Turkheimer posted on Friday, November 05, 2004 - 7:08 am
Suppose you have a straight ahead linear growth model parameterized as

i s | y1@0 y2@1 y3@2 y4@3

You can then fit a model to the covariances among the errors at the four time points, as has been discussed in the forum.

But you could also fit this model by reconfiguring it as a multilevel model, with three variables, subject (the clustering variable, TIME (1-4) and Y, using something like,



My question is, can you do anything to model the within cluster error structure in this context, and if not, what is the default error structure that is implied?

 bmuthen posted on Friday, November 05, 2004 - 8:15 am
Yes, you can do growth modeling in Mplus either as a multivariate, single-level model or as a univariate, two-level model - this gives exactly the same result. The two-level approach assumes that the level-1 residual variances are the same across time points and that the residual covariances are all zero (reflecting the multilevel model assumption of conditional uncorrelatedness for the outcomes over time given the random effects, i.e. the growth factors). We can think of no trick to avoid those somewhat strong assumptions of the two-level approach.
 Anonymous posted on Thursday, November 18, 2004 - 3:50 pm
Does allowing the residual variances to be unequal across time points compromise the interpretation of the growth parameters. That is, if a model that allows the residuals to be different fits the data better, then should one allow them to be freely estimated?
 Linda K. Muthen posted on Thursday, November 18, 2004 - 3:56 pm
Yes on the second question. No on the first question.
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