Message/Author 


Hello: I am running a path model that is essentially an APIM model of infantmother dyads. In the model I want to test the interaction of an exogenous variable (mother attachment status) on infant affect in predicting mother behavior. I am able to create the continuous interactions using the DEFINE command but I have a few questions: 1) When I grand mean center the interaction variables either in Mplus by: DEFINE: Eu2FFAff=(Eu22.150478)*(FFaff4.149212); DEFINE: Eu2SFAff=(Eu22.150478)*(SFaff4.822054); or by creating them externally in SPSS before saving my input file, the overall test of model fit and the main effects estimates for the interaction variables are different than what they were in the raw (uncentered) model. Interestingly, the interaction effects are identical. Why is this? 2)I have searched high and low but have not found an adequate explanation of how to probe the interaction. Is there a resource I can be pointed to? My greatest appreciation. J 


You can take a look at the article by Preacher, Rucker, Hayes (2007) in MBR referred to in the Mplus V7 UG. I don't know the answer to 1). 


Thank you, Dr. Muthen. Interestingly re: #1 for anyone interested the issue is discussed in Aiken & West chap 3. I had passed over this chapter before, but it explains entirely question #1 above. 


In running my continuous path interaction analyses (all variables involved in the interaction centered) the model converges but I get the warning message re: the SE's being untrustworthy due to....a possible linear dependency. Tech 1 gives me the problematic parameter and it indeed is one of the interaction terms. Given that such a term is likely to be a linear dependency on its constituents, how problematic is this warning in the context of a model with interaction terms? 


I would subtract the mean from each of the variables involved in the interaction before creating the interaction term. 


Thank you, Linda. Am I correct in thinking that the below syntax accomplishes subtracting the mean from each of the involved interaction variables before it creates the interaction term itself? DEFINE: Eu2Cen=(Eu22.150478); DEFINE: Ds2Cen=(Ds23.020335); DEFINE: FFAffCen=(FFaff4.144492); DEFINE: SFAffCen= (SFaff4.825626); DEFINE: Eu2FFAff= Eu2Cen*FFAffCen; DEFINE: Eu2SFAff= Eu2Cen*SFAffCen; DEFINE: Ds2FFAff=Ds2Cen*FFAffCen; DEFINE: Ds2SFAff=Ds2Cen*SFAffCen; 


It looks correct. 

Tracy Witte posted on Wednesday, March 20, 2013  11:19 am



I have a question about centered variables in Mplus. I have created the centered terms in SPSS and want to use them in a path analysis with an interaction term. Since there is missing data, and Mplus uses FIML to handle it, I've noticed that the means for the centered variables differ from 0. Given this, I'm concerned that this will affect interpretation, should I find a significant interaction. Is anyone aware of guidelines regarding how to deal with centered variables when using FIML to handle missing data? 


You should center the variables in Mplus. 

Tracy Witte posted on Wednesday, March 20, 2013  4:41 pm



Even when I center in Mplus, my centered variables end up with small decimal values for the sample statistics, rather than being 0. This is an excerpt of my syntax: DEFINE: CENTER ESRBVA HOURF (GRANDMEAN); ANALYSIS: type=basic; When I look at the descriptives, the mean for centered ESRBVA is .209, and the value for centered HOURF is .007. Any other suggestions for what I can do? (Thanks!) 


Please send TYPE=BASIC outputs with and without centering and your license number to support@statmodel.com. 

Back to top 