Hi All; I'm planning a data analysis with the following parameters: 3 time points 2 groups, 1 is flat across all 3 time points, the other group is quadratic (so the plotted data would look like a triangle). I'm predicting that people in the quadratic group will converge on the linear group (thus making a perfect triangle). I'm also predicting heteroskedasticity at the third timepoint, where people who were high at intercept in the quadratic group will remain higher than the rest at t3 and those lower at intercept in quadratic group will be lower than the rest at t3.
My 2 questions are: 1) Does this have to be run as a multi-group longitudinal growth model, or can it be 1 model with an indicator variable for which group is which onto the latent growth parameters?
2) How do you plan for heteroskedasticity in one group while also testing the intercept/slope relationship in one group only?
Is the automatic WITH statement built in between i and q the heteroskedasticity parameter I'm looking for? It seems it would be under-estimated because that would be across both groups but I am only predicting such a relationship in the quadratic group.
1) Multi-group approach gives more flexibility in what can be different across the groups but a dummy covariate approach can also be used.
2) Residual variance can be modeled (for one of the groups) as a function of an observed variable such as your outcome at time 1. See the input for the Table 1.15 run in our book Regression and Mediation Analysis using Mplus at