SEM R lavaan: Latent Interactions (Moderation) With Double Mean Centering

[@aissahaka]

Thank you. Would love to see a tutorial involving moderated mediation using latent interaction.

[@Y45HV1N]

thank you for your content! It's really helpful to have tutorials like this that are easy to follow.

[@gavinaustin4474]

Exactly what I needed. Thanks so much.

[@עינתיאור]

Thank you so much, that was a very clear explanation - very helpful for me

[@drryangagnon]

Excellent work!

[@reza8558]

Hi Arndt, thank you very much for the generous sharing. I wonder if you could share the academic reference indicating that model fit estimation could be evaluated without the latent interaction... thanks!

[@RegorzStatistik]

You could look at Schoemann and Jorgensen (2021), full reference information see the video description. "Traditional methods of assessing model fit do not take this dependence among observed variables into account and, thus, provide incorrect estimates of model fit. To assess model fit for latent interaction model we recommend first fitting a model without the latent interaction variable, or any product-indicators, and assessing model fit on this “main effects” model." (section 1.3, last paragraph)

[@reza8558]

@RegorzStatistik I am grateful for your guidance... Thank you, brother...

[@reza8558]

@@RegorzStatistik I wonder, if interaction terms are not included in estimating the model fit, what is the merit of conducting interaction analysis with latent variables? Wouldn’t ignoring the impact of latent interactions on model fit effectively overlook the influence of their error terms in the model estimation? In this case, why shouldn’t one fit the structural model using latent variables for the main effects, while evaluating moderation effects at the observed variable level through techniques such as Hayes' PROCESS model?

[@RegorzStatistik]

@@reza8558 I think you should try and ask this question to Schoemann and Jorgensen (2021) - maybe they can help you with that question.

[@reza8558]

@@RegorzStatistik Thanks for the attention, brother!

[@dailyrandomsmiles380]

Hello, thank you for the tutorial! In my case, I have 6 indicators for the IV and 5 indicators for the MOD. I was considering item parceling as an option. How would I integrate the item parceling code with the code provided in the video? I don't know how or if the random repeated item parceling code using parcelAllocation would fit. Thank you again for your guidance.

[@RegorzStatistik]

I am not sure if/how random parceling can be combined with double mean centering. I think I would choose another parceling technique here.

[@Sandy-m3v]

Hi, thank you for the video. My question is if the moderator a dichotomized variable classified as low and high based on the medain score and the predictor variables are latent variables with a binary outcome variable. how can one go about it when it involves conducting a moderated mediation analysis using lavaan?

[@RegorzStatistik]

I haven't come across a situation like that. I think you could use double mean centering to create the latent interaction. And lavaan supports binary outcome variables, see lavaan.ugent.be/tutorial/cat.html

[@Sandy-m3v]

Thank you so much for your response. If I understand corrrectly, you are suggesting that I mean center the moderator instead of treating it as a dichotomized variable using the median score as cut off? Also, for context, the moderator is an observed variable. Is the double mean centering still rcommended for observed and latent interaction?

[@catherineaudrin944]

Thank you ! Do you have any references, notably to assess the model fit without the interaction ?

[@RegorzStatistik]

One of the papers referenced in the video description: Schoemann, A. M., & Jorgensen, T. D. (2021). (last paragraph of section 1.3)

[@ahmadvalikhani6290]

Great! Can we do this approach for a moderated mediation and mediated moderation too? If yes, of course, the next question would be how?

[@RegorzStatistik]

You could build an SEM model to model a moderated mediation with latent variables, yes. You would have to draw a structural model modeling the moderated mediation and then the measurement model to measure the latent variables.

[@artisan_charcutier]

A very instructive video. Thank you very much. I have followed all the steps, but get a "subscript out of bounds" error when I run the probe2WayMC. I wonder what this is due to. Do you have an idea?

[@RegorzStatistik]

Unfortunately, I have no idea what the reason is for that.

[@ЮлияМузыченко-х5п]

Hello, thank you for the videos! I have a question, if we have a complex model with mediations and moderations, should we interpret their path coefficients step by step(that is how they do in AMOS), first run a model without interactions and interpret direct a indirect effects for mediations and then run model for moderation with interactions to interpret them, is it so?

[@RegorzStatistik]

Unfortunately, it is not possible to give a general answer to that, it all depends (primarily on your hypotheses).

[@tjwang3349]

Hi, IV and moderators in the measurement model, should I put the raw items in this case, or the centered items to those latent variables? Thanks!

[@RegorzStatistik]

I think it doesn't matter (but I am not a 100% sure).