Dear Linda and Bengt, This forum is amazing - so much valuable information. I was, however, not successful in finding out how to treat the following problem:
moderation analyses in which the moderator is a 4-level categorical variable, outcome is continuous, predictor variables continuous and categorical. major issue: these data are weighted
I have tried using type=mgroup, estimator mlr with weight=smplwght specified but the data were not weighted as could be seen in the output. I also tried using type=mgroup complex but this requires stratification and cluster, which our data don't have. I know I could try mixture if I categorized the outcome, but I don't really want to do that.
In short, do you have any suggestion as to how to tackle a multigroup comparison of a regression with weighted data?
excellent! this worked! we compared the estimates mplus 4.2 and 5.21 produced and they were slightly different although the exact same syntax was used. can you explain the reason for this? thanks again, tina
The default in Version 4.2 was listwise deletion. The default in Verson 5.21 is TYPE=MISSING. They use different information matrices which could account for the differences. If you look at the summary of the analysis, you will see which information matrix is used in each analysis.
that explains a lot of slightly different results i obtained recently with different versions. thanks a lot again, also for making the workshop in berlin recently so much fun!
Djangou C posted on Tuesday, February 27, 2018 - 3:46 pm
Dear Mplus team, I am running an SEM model with a weight and a clustered variables. I want to do invariance testing (loading, intercept, variances…) . It seems to me that the SUBPOPULATION and the GROUPING options are not using the weight in the same manner. Given that the SUBPOPULATION is performing the weighting appropriately for each group, I wonder if we can do invariance testing accurately in this case. If yes, could you please tell me how? Thank you in advance.
You would perform the invariance testing the same way you would without the weights. I would use the grouping command rather than subpopulation. For more information send your example and data to firstname.lastname@example.org