I have a complex sample with weights, strata, and clusters. I want to test measurement invariance for factor and IRT models between subgroups within a subpopulation. When doing so I get the following error messages:
The SUBPOPULATION option is not available for multiple group analysis.
The SUBPOPULATION option is not available in conjunction with the KNOWNCLASS option.
Can you direct me to any literature or documentation explaining this limitation?
Which of the following alternatives would you recommend (or others):  Analyses with and without weights ignoring the grouping. If they are approximately the same, proceed without weights (but still include the cluster variable and TYPE=COMPLEX).  If possible, physically subset the data and reweight the cases in subpopulation.  Regress the factor indicators on the grouping variable. If measurement invariance holds, I believe we would expect the coefficient to be zero. However, if the regression coefficient is not zero, loading and intercept/threshold non-invariance would not be distinguishable.  Some other approach?
We have not yet implemented it in the multiple group/known class case. In our experience the SUBPOPULATION option makes little if any difference from using USEOBSERVATIONS. I would suggest running the analysis in the separate groups using SUBPOPULATION and then using USEOBSERVATIONS. If you see no important differences as I would expect, then I would use USEOBSERVATIONS in the multiple group analysis.