Hi, now I use the structural missing data. Variables (X1-X10) with 2 groups (G1 / G2)
The point is that the data has two groups, however, one of the binomial variable X1 is completely missing in Group G2. On the other hand, another variable X2 is completely missing in Group G1. And others X3-X10 are all observed in both groups.
I'm trying to model this data with Confirmatory Factor Analysis. So I submitted the code below.
MODEL: F1 by X1* X3-X10(p1-p8); F1@1; MODEL g2: F1 by X2* X3-X10(p1-p8); F1@1;
I want to get same Factor Loadings X3-X10 between the Groups and different FLs X1 and X2 in the Groups. However, the error message "Categorical variable X1 contains less than 2 categories" is returned because X1 is all missing in Group2 and X2 is all missing in Group1.
How can I solve this problem? Any comments and suggestion will be appreciated.
Jon Heron posted on Friday, November 18, 2016 - 5:37 am
Why not create a new variable which is a composite of X1 and X2 and allow all the measurement properties of that item to vary between groups.
Thanks for the comment. That must be nice. However, X1 and X2 are different variables. Even if one of them can't be observed on the dataset, that missing variable is actually existing and just missing. So, if I calculate the factor loadings with composing X1 and X2 as one variable, I'm afraid that the model would be biased.
Jon Heron posted on Friday, November 18, 2016 - 7:40 am
even if the factor loading, intercept and residual variance were allowed to vary between groups?
Thanks again, Dr. Heron. In my opinion (but now I'm not so confident), even if the composed variable is allowed to have different factor loadings across the groups, the variable might be dealt as a same variable in a whole modeling process, such as calculation of covariance matrices. However, if I use different variable, these processing problems must be solved.
Perhaps, can we solve that problem with only allowing different factor loadings, intercept and residuals...?