Message/Author |
|
|
Hello, I'd like to understand the Variance-Covariance matrix structure of my Growth Mixture Model. The model I've fitted is as follows: ANALYSIS: ESTIMATOR = MLR; PROCESSORS = 4; TYPE = MIXTURE; MODEL: i s | t1@0 t2@4 t4@9; i WITH s; What sort of variance-covariance matrix would this result in (if possible to determine from this alone)? Unstructured? Any guidance on how I can determine this would be greatly appreciated - I couldn't find variance-covariance matrices in Chapter 8 of the user guide, but this might be because I'm not familiar with all the terminology that can be used in reference to this issue. Alice |
|
|
We discuss the estimated covariance matrix from growth models in our Short Course Topic 3 video and handout. Perhaps you instead refer to the covariance matrix for the residuals - which is by default uncorrelated residuals allowed to have different variances. You can correlate the residuals over time if you like using WITH statements. See also UG ex 6.17. With only 3 time points, however, there is little freedom to add parameters. |
|
|
Thank you, I think I'm a little clearer now. So in relation to the above code for the model I've fitted, would it be accurate to say that: 1. The intercept and slope residuals within each latent class are correlated ("i WITH s;"); 2. Otherwise, the residual variances between each latent class are not fixed, and are therefore allowed to be class-specific (i.e. the residual variances are freely estimated and not fixed across latent classes); and 3. The variance-covariance matrix is also not constrained across classes (i.e. each latent class has its own independently estimated variance-covariance matrix)? |
|
|
1. They are not residuals because you have no predictors instead say that i and s are correlated. 2. All variances and covariances are held equal across the classes as the default. They can be freed if you like it to be. The i and s means are free by default. 3. It is constrained as the default but it can be freed. |
|
Back to top |