Amy Hartl posted on Wednesday, March 14, 2012 - 8:16 am
I need to test gender via a multiple groups model for an indistinguishable dyad APIM. However, given the nature of the indistinguishable APIM (all actor and partner paths are constrained), I'm having trouble.
Is it possible to constrain paths separately for each group? I.e. how do I make it so the partner paths for females ('p') are constrained and the partner paths for males ('b') are constrained without constraining 'p' and 'b' to be equal?
I have been running some indistinguishable APIM models, and have tried the suggestion for multi-group analysis.
For one set of variables, the models ran fine. For another set (multidimensional construct), the fit was fine for the overall model, but horrible for the multi-group. I surmise that this means that the fit for one or more of the groups was really bad.
Is it appropriate to run analyses separately on the groups if the multi-group doesn't work? Or on some groups, but not the group that seems to be the problem?
I assume that your multiple-group analysis has some parameter equalities across groups (check Tech1 to see if you have the same number for a parameter in the different groups) - if so, the equalities may have to be relaxed. Also, before you do a multiple-group analysis, you should start with separate group analyses to make sure each group has an acceptable fit.