Message/Author 


Hi, I am running a model with two latent exogenous variables, one observed categorical endogenous, and one latent variable interaction between the two latent exogenous variables. The link is logit, and the output gives an odds ratio for the regression of the outcome on the interaction term, but I am unsure of how to interpret this. Could you tell me what an odds ratio of 0.75 for the interaction term indicates? Thank you! 


See our FAQ Latent variable interactions which tells you how to interpret interaction models with a continuous DV. Then just change that reasoning to apply to a logodds ratio (which is what we print). and then exponentiate that logodds ratio to get your answer. 

Franziska posted on Friday, August 14, 2015  11:02 am



Hi everyone, I'm trying to run a probit regression model with latent EFA factors (out of the ESEM based on categorical variables). The regression model is supposed to include interaction effects of one categorical variable with the latent factors. However, if I run the analysis, I get this message: *** ERROR in MODEL command The use of EFA factors (ESEM) is not allowed with TYPE=RANDOM. However, I need the TYPE=RANDOM statement to define the interactions. Is my only possible solution to calculate the interactions first (with the factor scores of the latent factors) and then use this new datatset to run the regression? Even though I would like to run the model completely simultaneously. I would really appreciate any hint or comment. 

Franziska posted on Friday, August 14, 2015  12:31 pm



This would be the respective output. Thanks for your help!! CATEGORICAL ARE ArrogantSocial local knowledge esteem relevance usage; USEVARIABLES = ArrogantSocial local knowledge esteem relevance usage; Missing are all (66); IDVARIABLE = id; AUXILIARY = brand categorySK categoryCD con; MODEL: !The label 1 following an asterisk in parentheses specifies that the factors are a set of EFA factors; f1f8 BY ArrogantSocial(*1); !latent variable interactions; Int1  local XWITH f1; Int2  local XWITH f2; Int3  local XWITH f3; Int4  local XWITH f4; Int5  local XWITH f5; Int6  local XWITH f6; Int7  local XWITH f7; Int8  local XWITH f8; !simultaneous regressions for all dependent variables; knowledge ON f1 f2 f3 f4 f5 f6 f7 f8 Int1 Int2 Int3 Int4 Int5 Int6 Int7 Int8; esteem ON f1 f2 f3 f4 f5 f6 f7 f8 Int1 Int2 Int3 Int4 Int5 Int6 Int7 Int8; relevance ON f1 f2 f3 f4 f5 f6 f7 f8 Int1 Int2 Int3 Int4 Int5 Int6 Int7 Int8; usage ON f1 f2 f3 f4 f5 f6 f7 f8 Int1 Int2 Int3 Int4 Int5 Int6 Int7 Int8; ANALYSIS: TYPE = RANDOM; ITERATIONS = 10000; SAVEDATA: FILE IS GERMANY05_FS8F,inv,aux.dat; SAVE = FSCORES; 


You cannot have interactions with ESEM factors. Please limit posts to one window. 

Franziska posted on Friday, August 14, 2015  2:44 pm



Hi Linda, thank you very much (and sorry for the two posts  I couldn't put the input into the first one). Do you think it could be an option to use the factor scores out of the ESEM then and manually calculate the interaction effect and then run the regression? I would greatly appreciate your opinion on this. Thanks again! 


Factor scores and factors are not the same unless factor score determinacy is extremely high. I don't this is a good option. If you cannot fit the entire post in one window, it is too long for Mplus Discussion. 


Dear prof. Muthen, If ESEM factor scores are not a good option to model the interaction effect, is there a better alternative then to model an interaction between latent ESEM factors? 


You can consider the ESEMwithinCFA (EwC) approach described here http://www.vanderbilt.edu/psychological_sciences/graduate/programs/quantitativemethods/quantitativecontent/marsh_morin_parker_kaur_2014.pdf 

Back to top 