Hi all, sorry to be so "direct" but the product term video is highly misleading. The approach to select only a sub set of (matched) indicators reduced the misfit substantially but still the model is misspecified what the sign. chisquare test shows (and surely not "great" which the person claims). Beyond that the problem is not "misfit" but *misspecification* (which the misfit just reflects). And the misspecification in this model is caused by the failure to model all the error-dependencies between the root indicators and the product term indicators. These will bias the structural coefficients. There is a whole literature about this problem and solutions.
Search for "residual centering". If you allow me a self-citation (because we demonstrated that:
Steinmetz, H., Davidov, E., & Schmidt, P. (2011). Three approaches to estimate latent interaction effects: Intention and perceived behavioral control in the theory of planned behavior. Methodological Innovations Online, 6(1), 95-110. https://doi.org/https://doi.org/10.4256/mio.2010.0030
That being said, I would recommend to start with a manifest path model (which the video showed in the beginning) and only approach a more complicated, latent, version, if the p value of the product term is >.10. The reason why a latent model is beneficial is increased power, surely, but I would not expect that this will do so much of a difference and pull the p value from .10 to