Message/Author 


I've got two quick (and easy) questions that I can't find the answer to in Klien & Moosebrugger or the Mplus manual. First, is it necessary to center your variables when using the Mplus/QML approach for intearctions? Second, are you supposed to allow for the covariation between the latent variables comprising your intearcation term? For example, would I say: FCOM  FORMAL XWITH COMM; FORMAL WITH COMM*.375; OR WOULD I CONSTRAIN THE COVARIANCE TO BE ZERO AS FOLLOWS: FORMAL WITH COMM@0; Thanks. 


It is not necessary to center your variables when using the XWITH command. You should not mention the covariance between the two variables that are part of the interaction. Because these are covariates, the model is estimated conditioned on these variables. Their covariance is not estimated as part of the model. If you want to know the value, you can obtain it using TYPE=BASIC;. 


Hi there. I am conducting an analysis where I model an interaction between two latent continuous variables (A and B) in the prediction of an observed continuous variable. I noticed that I am unable to specify a correlation between the main effects and the interaction (e.g., A with interaction and B with interaction). Is there someway that I can find out what the correlation would be. I appreciate any help that you can provide. Thanks. 


You cannot have this parameter in the model. You can however derive this value. See the Aiken and West book Multiple Regression: Testing and Interpreting Interactions starting on page 178. 

Study Hard posted on Thursday, November 23, 2017  5:20 pm



Dear Dr. Muthen I have 2 questions regarding the QuasiML software developed by you and Dr. Andreas Klein. (This is a program designed to run latent interaction test.) Will this program be incorporated in Mplus in the near future? Is this program publicly available now? I contacted Dr. Klein, but haven't got any response. Thank you Ken 


Q1: No Q2: I don't think so. 

Study Hard posted on Monday, September 17, 2018  2:40 am



Dear Dr. Muthen I've seen in the FAQ page that Mplus runs a simple slope test for 2 observed variables using the code below (copied & pasted from the website). This givesthe simple slopes at z values of 1 and 1. MODEL: y ON m x z zx; m ON x (b1) z (b2) zx (b3); MODEL CONSTRAINT: NEW(modlow modhigh); modlow = b1+b3*(1); modhigh = b1+b3*1; I was wondering if the above code can be extended to test a simple slope for latent variables too. So, a code would be... analysis: estimator=MLR; type=random; MODEL: y by y1y3; m by m1m3; x by x1x3; z by z1z3; zxz xwith x; y ON m x z zx; m ON x (b1) z (b2) zx (b3); MODEL CONSTRAINT: NEW(modlow modhigh); modlow = b1+b3*(1); modhigh = b1+b3*1; thank you Ken 


Yes, this can handle simple slope for latent variables using the approach you list. Note that you don't need to restrict yourself to 1 and 1 for the moderator but can consider a range of values using the LOOP and PLOT options in Model Constraint as shown on our Mediation page: http://www.statmodel.com/Mediation.shtml Then you also get a graph as seen at http://www.statmodel.com/download/Loop%20plot%203.18.pdf 

Back to top 