I am trying to estimate a two-level latent growth model in which individuals are nested in neighborhoods but are also grouped into three age cohorts. I am interested in seeing if neighborhood contextual characteristics have a differential impact on individuals' growth factors based on the age cohort of the child.
In order to allow for easier interpretation and possible mediation analysis down the road, I would like to estimate this model with all of the cohorts and no intercept. Is there an option that I can include in my syntax that will avoid issues of multicollinearity in the observed data matrix due to the inclusion of all cohort dummy variables?
If I were doing this in Stata with xtmixed I would use the "nocons" option. Is there something similar in Mplus?
You can treat the cohorts in two ways in Mplus: Using dummy variable covariates or using a multiple-group approach.
Analyzing all 3 cohorts you would simply create 2 dummy variables and use them as covariates.
In Mplus the most flexible way to do this is using a multiple-group growth analysis, where your cohorts form the groups. Then the groups can be specified to have as much or as little equality among any of their parameters as you wish, including the impact of covariates. I think this is a more powerful approach than the Stata approach you mention because you can let any parameter vary across cohorts, including variances. Note,however, that the grouping is on level 1 (subject) not level 2, which means that you have to use approaches described in the Mplus Web Note 16 at