I am trying to create an AMOS SEM model between Childhood Traumas (CT) and Borderline Personality Disorder (BDL), and the Behavioral İnhibition System (BİS) will be moderating this relationship. It will be a full structural model, so I will add CT as a latent variable predicting its 6 subscales (trauma types) and BDL as a latent variable predicting its 8 scale items. Moderator variable BİS will join the model as a latent variable , predicting its 7 scale items. To draw this moderation analysis, I need to create interaction terms. Since CT and BİS both latent, to create the interaction terms, I need to multiply all of their observable variables, meaning there will be a latent variable called CTxBİS that is predicting a total of 6*7=42 observable variables, interaction terms like CT1xBİS1, CT1xBİS2, CT1xBİS3 and so on. The model fit will probably be poor since it is too complex.
Doing moderation analysis in AMOS shouldn't be this complicated and hard.
I could use the matched pairs method to simplify the model, but it would require me to create all 42 interaction terms and then select the top 6 of them (I want to have at least 1 interaction term for each subscale of CT) via choosing the interaction terms with the top standardized regression weight. I am following Collier's video on the Matched Pairs Method.
If I were to go a bit too far, I would first use the parceling method to reduce the observed variables of BİS, reducing them to 3 observable variables (BİSf1, BİSf2, and BİSf3). After that, I will create CTxBİS latent variable that predicts 6*3=18 interaction terms. To reduce the interaction terms' count, I will proceed with the matched pair method and will choose the interaction terms with the top standardized regression weight. I will be careful to choose one interaction for each subscale of CT, so I will still have 6 interaction terms (this part is debatable; maybe I can go for 3 interaction terms for each subfactor of BİS (BİSf1, BİSf2, and BİSf3). See the picture under:
So the main question is, "Is there a better and easier way to do moderation analysis with latent moderation variables in AMOS?" and the second one is, "Using the matched pairs method with the parceling method is too much?"
