On the first run of this mode, the variance of f0
was 1.063 in the TECH4 output. It should be exected that the constraint on u01
or sad
derived from the one time point model would not perfectly produce a 1.0 variance in the longitudinal model. The longitudinal model we have forces equal residual variance on the time-specific latent traits (f0
, f3
, and f6
) and these will come out as a weighted average. It is not reasonable to assume this parameter would be the same as the model that only included f1. Also, the f3
and f6
factors are predicted by latent variable S
, but f0
is not.
The implication of this is we could probably have skipped the three step procedure using the baseline model to get a latent variable variance of 1 in the MILGCM and just run two MILGM models (as was done here).
You will see in this model, the TECH4 variance estimate for f0
is 1.0.