Message/Author 

Fay Chen posted on Thursday, April 21, 2016  9:40 am



Dear Drs.Muthen, I am conducting latent interaction analysis in Mplus. Instead of mean centering all indicators in Mplus, I centered the indicators for latent independent variables in R first (x1 through x6) and formed the latent interaction product terms there (i.e., x1x4, x2x5, x3x6). Then in Mplus I centered the x1x4, x2x5 and x3x6 using the CENTERING command before modeling it. Do you think this is the correct way? Does Mplus automatically center the latent variables, or do I still need to center latent variables manually before modeling it? I appreciate your help! 


You don't need to do any centering when using XWITH. 

Fay Chen posted on Wednesday, April 27, 2016  1:28 pm



Thank you for your suggestion, Dr. Muthen! I am trying to use the unconstrained product indicators approach proposed by Marsh, Wen, & Hau (2004). If I am forming the latent interaction variable using the product indicators from the two latent independent variables, should I simply regress the latent dependent variable on 1)latent IV1, 2)latent IV2, and 3)latent interaction variable as formed using product indicators, or do I need to use the XWITH command? Below is part of my input dealing with the interaction using the product indicators approach. Do you think this is correct, or should I use XWITH somewhere for the interaction? Thank you so much! VARIABLE: NAMES ARE y1 y2 y3 x1 x2 x3 x4 x5 x6; usev = y1 y2 y3 x1 x2 x3 x4 x5 x6 x1x4 x2x5 x3x6; CENTER = grand mean (x1 x2 x3 x4 x5 x6); DEFINE: x1x4 = x1*x4; x2x5 = x2*x5; x3x6 = x3*x6; MODEL: ksi1 BY x1 (L1) x2 (L2) x3 (L3); ksi2 BY x4 (L4) x5 (L5) x6 (L6); ksi1ksi2 BY x1x4 (L7) x2x5 (L8) x3x6 (L9); eta BY y1 y2 y3; eta ON ksi1 ksi2 ksi1ksi2; 


Sorry, I have not used the unconstrained product indicators approach  you may want to ask on SEMNET. Why not use XWITH. 


Hi Dr. Muthen, I am also attempting to use Mplus for my moderation analysis using latent variables. I was trying to follow the discussion above. When you say "You don't need to do any centering when using XWITH," do you mean that this command "XWITH" centers the variables for you? Or do we center our variables prior to bringing the dataset to Mplus for analysis? I am sorry this is all very new to me as I have only used Mplus with latent variables for mediation analysis. 


XWITH does not do the centering for you and no centering is needed. 


Thank you Dr. Muthen for your quick reply. Do mean centering is not needed when you are using latent variables in moderation analysis? Thanks. 


Sorry, I meant to say: Do you mean that mean centering is not needed when you are using latent variables in moderation analysis? 


Hi Dr. Muthen, Sorry for all of the emails back and forth. If using latent variables in moderation analysis, do we not mean center because latent variable have mean zero in most models and therefore do not need be altered. I read this from one of your comments to another researcher who posted a similar question as mine. Also, do we just use the XWITH when we are doing moderation analysis with latent variables in Mplus? Thanks, Sara 


Answer to 6:46: Right. Answer t0 8:17: Centering is not needed even with nonzero factor means. XWITH is used whenever you want a predictor that is f1*f2, f*f, or f*y. 

Sara Namazi posted on Wednesday, May 30, 2018  8:44 am



Hi Dr. Muthen, I am examining whether a masculine organizational culture (MASCC) moderates the relationship between a masculine gender identity (BM_SM)and emotional suppression (EMOSUP). I wanted to verify with you whether my syntax looks correct before I report my results. All of my variables are continuous and treated as latent variables. As such, my interaction term is also latent. USEVARIABLES ARE MascC1 MascC2 MascC3 EmoSup1 EmoSup2 EmoSup3 EmoSup4 BEM_SM1 BEM_SM2 BEM_SM3 BEM_SM4 Analysis: type = general; Iterations = 10000; Type = Random; ALGORITHM=INTEGRATION; model: BM_SM by BEM_SM1 BEM_SM2 BEM_SM3 BEM_SM4 ; MascC by MascC1 MascC2 MascC3; EmSup by EmoSup1 EmoSup2 EmoSup3 EmoSup4 ; Int1  BM_SM XWITH MascC; EmSup on BM_SM MascC Int1 ; OUTPUT: STAND TECH1 tech4 sampstat PATTERNS RESIDUAL modindices ; 


Looks ok. 

Sara Namazi posted on Sunday, August 05, 2018  8:58 pm



Hi Dr. Muthen, I wanted your advice on a question I had about moderated mediation. I have two predictors in my model that both interact with a single moderator. My moderator points to both the a path and the c’ path. Do you recommend I run the interactions one at a time or keep both in a single model? Currently I have both interactions in a single model and it ran fine. Syntax is this: Emo (mediator) on x1 x2 Moderator interaction 1 ! Moderator x X1 interaction 2 ! Moderator x X2 Stress (main outcome) on Emo x1 x2 Moderator interaction 1 ! Moderator x X1 interaction 2 ! Moderator x X2 In another analysis, I have three moderators pointing to the b path. These moderators are Depression, anxiety and hostility. Should I also run those separately or keep them together in a single model? I currently ran them separately. Syntax is this: Emo (mediator) on X1 X2 Stress (main outcome) on Emo X1 X2 Depression (moderator) Interaction 3 ! Depression x Emo And I repeated this for anxiety and hostility. Your advice is greatly appreciated. Sara 


I have no strong feelings either way  although if your theory suggest these moderations, I would want to include them all at once. For analysis strategy questions you may instead want to consult SEMNET. 

Sara Namazi posted on Monday, August 06, 2018  6:14 pm



Thank you, Dr. Muthen. I also wanted to ask two additional questions: (1). whether one should grand mean center all continuous latent variables before doing moderatedmediation and (2). whether one should graphically plot the interactions even if there is no statistically significant conditional indirect effect Sara 


1) Latent variables typically have zero means already. 2) I think that can be useful. 

Back to top 