Moderation Analysis with SEM

Comments

• patc3

  May 2023

  I don't know for sure what the source of your problem is, but just bouncing ideas here: I don't think interaction terms with latent variables are implemented in lavaan (the SEM library JASP uses) yet
• caro96

  May 2023

  Thank you so much! I think the source of the problem is the variable AGE. As long as I don't include it into the regression, everything works fine (see code). As soon as I add it, I get the error. Do I have to specify age first after doing the cfa? It's a normal scale variable in my dataset so I don't see the problem...

  #cfa

  BI =~ lambda_1_1*BI_1 + lambda_1_2*BI_2 + lambda_1_3*BI_3

  T =~ lambda_2_1*T_1 + lambda_2_2*T_2 + lambda_2_3*T_3 + lambda_2_4*T_4

  PE =~ lambda_3_1*PE_1 + lambda_3_2*PE_2 + lambda_3_3*PE_3 + lambda_3_4*PE_4

  EE =~ lambda_4_1*EE_1 + lambda_4_2*EE_2 + lambda_4_3*EE_3 + lambda_4_4*EE_4

  SI =~ lambda_5_1*SI_1 + lambda_5_2*SI_2 + lambda_5_3*SI_3

  FC =~ lambda_6_1*FC_1 + lambda_6_2*FC_2 + lambda_6_3*FC_3

  SA =~ lambda_7_1*SA_1 + lambda_7_2*SA_2 + lambda_7_3*SA_3 + lambda_7_4*SA_4

  PR =~ lambda_8_1*PR_1 + lambda_8_2*PR_3

  TS =~ lambda_9_1*TS_1 + lambda_9_2*TS_2 + lambda_9_3*TS_3

  #regression

  T ~ PE + SA + TS + PR

  BI ~ PE + EE + SI + FC + T
• caro96

  May 2023

  I thought about calculating the interaction term as a new variable in my dataset before adding it to the regression function.
• patc3

  May 2023

  Again, this could still be due to the interaction terms. Have you tried including age without interaction terms? My guess is still that the problem is due to interaction terms with latent variables.

  I have personally never dealt with this problem, but a cursory google search yields interesting results:

  • https://m-clark.github.io/sem/intro2R.html which points to an example: https://rpubs.com/njaalf/workshop2014 (I think they use the parameter estimates of the CFA to generate factor scores from a normal distribution)
  • https://groups.google.com/g/lavaan/c/ZnRn9O6yWrA In particular, one comment points to this reference: Foldnes, N., & Hagtvet, K. A. (2014). The choice of product indicators in latent variable interaction models: Post hoc analyses. Psychological Methods, 3, 447-457.