
Message/Author 


Dear Prof. Muthen, I am running an SEM (with both direct and indirect effects). I also specified an interaction term for two latent variables to see if the joint effect of two of my predictors have an effect on another latent variable. I understand that I have to specify: 1) Type is random and 2) ALGORITHM=INTEGRATION; Each time I ran my model, I get an error message: "MODEL INDIRECT is not available for TYPE=RANDOM." I ran the model without the indirect effect and it worked. Does this mean that I cannot examine mediation and moderation at the same time? Thank you very much. Here's the code that I used. Model: ident on nmedia desire real norm f1 f2; f1xf2  f1 XWITH f2; ident on f1xf2; expectat on SIMILAR f1 f2; Model indirect: Expectat ind f2 nmedia; Behav ind expect f2 nmedia; ANALYSIS: TYPE IS RANDOM; ALGORITHM=INTEGRATION; ESTIMATOR IS MLR; ITERATIONS = 1000; CONVERGENCE = 0.00005; 


In this case you replace Model Indirect with Model Constraint where you define your indirect effect using labels given to your slope parameters in the Model command. 


Thank you very much for your answer. Unfortunately, I am still getting an error message after I replaced Model Indirect with Model Constraint. " A parameter label or the constant 0 must appear on the lefthand side of a MODEL CONSTRAINT statement. Problem with the following: NMEDIA(0) =". I am not quite sure what I am doing wrong because this is the first time that I am testing both mediation and interaction effects. 1) Is there any such example in the Users' guide or on the website that I could use as a guide? 2) When I run the multiple group analysis to see the effect of gender on both the direct and indirect effects, do I just constrain the mediated paths as I would normally do with the direct effects? 3) I am using 6.12 but it won't give me a threeway interaction. Is this available in the 7.0? Thank you in anticipation 


Please send the output and your license number to support@statmodel.com. 


I am modeling a bifactor model.Its factors are outcome variables. I test a 3way interaction between continuous variables as predictors. I predict the 3 outcomes in separate models because I want the factors to remain orthogonal and do not want factors to correlate through predictor variables. Because I estimate missing data for X variables, I constrain the correlation between X's and the other outcome factors (not predicted) to 0. I am having trouble interpreting the 3way interaction, which is significant when all 3 predictors are centered at 0. But, when I center the moderator at high or low values, recalculate interaction terms, and reestimate the model, the coefficient and pvalue of the 3 way interaction term changes significantly. I know this 3way interaction term should NOT change. I include all lower order terms (x,z,w,xz,zw,xw,and xzw) I think this has to do with testing a 3 way interaction. Testing only a twoway interaction and centering at high or low values of the moderator doesn't change the 2way interaction term. I think this also has to do with estimation of missing X variables (specify variances). When I don't specify variances of X, the coefficient/pvalue of the 3way interaction term does not change when I center the moderator at a high value and recalculate interaction terms.Is there something I am overlooking or is this an issue that has a solution? Any advice would be very much appreciated! 


This question seems more appropriate for a general discussion forum like SEMNET. 

Back to top 

