My colleagues and I are testing a multidimensional model of acculturation using a sample of recent immigrant Hispanic adolescents. We have proposed three domains of acculturation - behaviors, values, and identifications. Within each domain, US and heritage cultural streams are operating.
We have estimated three growth mixture models - one for Hispanic and US behaviors, one for Hispanic and US values, and one for Hispanic and US identifications. Each solution produced 2 classes. We now want to do two things: (1) cross-tabulate these three sets of classes against one another using log-linear modeling; and (2) examine the effects of these three sets of classes on outcome variables (self-esteem, perceived competence, depressive symptoms, and grade point average).
I've read your article with Shaunna Clark indicating that classifying people into their most likely classes is a bad idea. However, how else would we do what we're looking to do? The classes are formed using three separate models, and we do not want the formation of the classes to be affected by the outcome variables.
Thanks in advance for any insights you can provide for us!
With entropy .8 or higher, you could use most likely class membership and do a multiple group analysis using the KNOWNCLASS option. You can still do this but for the group with .7 entropy, it won't be as good.