I have estimated a model which is a combination of a LCA with 3 classes and a CFA with 3 factors. I want to find out whether the latent means of the 3 factors differ between classes and just want to make sure that I interpret the Mplus estimates correctly. It seems that Mplus establishes full CFA measurement equivalence across classes (so that it makes sense to compare the factor means) and freely estimates the factor means by default (except in the last class were the means are fixed to 0). My questions are:
- is it correct to say that a factor mean in a given class, say class 1, is significantly different from the mean in the last class when factor mean(class 1)/s.e. >= 2 or <= -2?
- would it make sense to compare the model with freely estimated means in class 1 and class 2 to a model in which all factor means are fixed to 0 in all 3 classes? Or how would you test the assumption of equal factor means in all classes?
It is correct to say that a factor mean in a given class, say class 1, is significantly different from the mean in the last class when factor mean(class 1)/s.e. >= 2 or <= -2. This is because the test is a test against zero and the mean of the last class is zero.
Yes, it does make sense to compare the model with freely estimated means in class 1 and class 2 to a model in which all factor means are fixed to 0 in all 3 classes.
Arne Floh posted on Thursday, October 11, 2007 - 5:49 pm
I ran a growth mixture models with 2 classes(n1= 144, n2=592). M+ output shows a latent mean difference of -1.401 (-8.048) indicating that the mean of group1 is significant lower than the mean of group2. However, I don't "trust" the results. I would expect a positive sign. I got even more confused when I had saved the class memberships. Class membership was flipped (n1= 592, n2=144). Additionally, t-tests shows that the mean of oberserved variables are significant higher for the smaller group. Who is right? M+ or me? Follow-up question: Latent means of the last group are fixed to 0. Can I override this? Help appreciated. Thx.
M.O. posted on Monday, January 19, 2015 - 10:05 pm
Dear Dr. Muthen,
I have a question regarding gmodel resultsh for LCA-CFA analysis. When I conduct CFA, as in Example 5.1, I can see gCorrelation between factorsh, gInterceptsh, gVariancesh and gResidual variancesh under gModel resultsh section of output file. However, I could not find those when I conduct LCA-CFA. Are there any ways to obtain those parameters? Here is the script I used: