SEM: Latent Interaction Terms

This video demonstrates how to do moderation in SEM using latent interaction terms.
I use Lavaan for the SEM and semTools to create residual-centered items for the interaction term.
This video simplifies the procedure somewhat by using the matching procedure to create the interaction items.

[@toshitakahamamura2996]

Thank you for creating this helpful video!

[@yiwenlin7119]

Please keep uploading these videos! And do a follow up on non matched pair! You literally just saved me from despair! I cannot describe how hard it is for me to learn all these by myself especially when my native language is Chinese and all the accessible instructions are papers in English. Your instruction is so clear! Thank you so much!!!

[@nikitarao1606]

Thanks so much for this tutorial! Straightforward and easy to follow instructions. Have you made the follow-up video without the matching?

[@dawmi3140]

Can you do one with an interaction between an observed variable and a latent varaible!

[@banajitdebnath4991]

I am getting an error "If the match-paired approach is used, the number of variables in all sets must be equal." I have unequal no of items for those variables ! Also i wish to know how to handle multiple moderators acting on single IV

[@noorawegelius3240]

This video has been super helpful, thanks a lot! Would you be able to make a video about how to plot such latent interaction (moderation) with R? Plotting moderation seems easy enough, except when using latent variables.

[@benr1813]

This is a great suggestion. The semTools package has (had?) some functionality for this but I'm not sure it's supported any longer. I have played around with it to see if I can get a consistent result or a simple workflow but I'm still not satisfied enough to make a video. But I'll keep plugging away : )

[@yuanjiebao9739]

Thanks for the video. I see that the IV and the moderator all had 3 items, is it necessary for them to have same number of items? I if have IV with 3 items and moderator with 5 items, should I parcel? thanks a lot.

[@benr1813]

Hi Yuanjie, it is not necessary that the IV and M have the same number of items but it gets complicated fast so I do recommend parceling.

[@jessperry1703]

Please help! Can you run interactions in a SEM with second-order latent variables (i.e., a latent variable explained by other latent variables). I want to run an interaction between 2 predicting latent variables, but one of them is a second order LV, with 4 sub-constructs. Is is a case of running interactions between each sub-construct and the other predicting LV?

[@benr1813]

Hi Jess, I doubt I am replying to this with enough time to be helpful, but for whatever it's worth, I think this would be a good case for parceling. So what I would do is run a CFA with the second order LV, then fit a parceled model in which the four sub-constructs are used to create parcels for the second-order variable. Note: it is often tempting to create one parcel for each sub-dimension, but if you want the unique components of each sub-factor in the latent variable, it is better to distribute items from each sub-factor into each parcel so that the parcels share common information about each sub-factor. So you could have three parcels that contain items from each of the four sub-scales and use them to create the interaction variable. Hope that helps! (sorry it took me so long to reply)