So, I am testing a theory, proposed in the literature, and I want to do a multigroup analysis on it. One of the HOCs(reflective) of the model has three LOCs(all reflective); when I run the tests on the model I cannot reach compositional invariance(smart PLS); however, when I remove the LOCs and just make it a first order reflective construct with all indicators attached to it, the model gives full measurement invariance.

The question is that am I allowed to do this removal and change as a way to reaching invariance?

Thanks.