In testing multigroup CFA invariance, my marker variable approach fails (convergence issues) while a fixed parameter method works. Should I opt for IRT?
# Custom marker method
runModel1: sem (Alpha -> a1 a2 a3) (Beta -> b1 b2), group(Phase) method(marker)
# Alternative fixed parameters method
runModel2: sem (Alpha -> a1 a2 a3) (Beta -> b1 b2), group(Phase) method(fixed)
hey, i lean toward fixed method. marker var issues might be due to identifiability problems. jumping to irt may just complicate things further. try refining your marker approach or stick with the fixed solution
In my experience, the fixed parameter approach tends to be more straightforward after some issues with the marker variable method. I encountered similar convergence challenges with marker variables, which were often resolved by tightening the model constraints and carefully selecting the normalization procedure. Adopting an IRT framework could be interesting, but it introduces additional complexities and assumptions that might not be warranted unless the data structure specifically calls for it. Focusing on refining the marker method or validating the fixed parameter approach seems more appropriate given the present context.