Say I have an SEM model with 1 predictor variable (IV), 2 mediators (MV1, MV2) and 1 dependent variable (DV). Amos reports combined indirect effects for an IV on a DV. So the combined indirect effect would be the sum of the component indirect effects (i.e., the IV-MV1-DV indirect effect and the IV-MV2-DV) indirect effect.

Amos makes it easy to get bootstrap confidence intervals on combined indirect effects. However, hypotheses often concern component indirect effects (e.g., that MV1 mediates the effect of IV on DV).

How can I get confidence intervals on component indirect effects using Amos?

I'd be interested in both asymptotic and bootstrap confidence intervals. Could I just use the values and standard errors for the individual coefficients (i.e., MV1<-IV, and DV<-MV1)?

  • $\begingroup$ Did you solve that problem? I think you can use nested models for that, and you constrain to zero each path you are not interested in for each of the models. $\endgroup$ – YouTu Apr 24 '14 at 15:26
  • $\begingroup$ @Benny I'd be interested in more information about the nested model approach. On the face of it, constraining a parameter to zero or treating it as known when it is unknown seems like it would alter the standard errors of other parameters, but I'd be keen to read more about the approach. $\endgroup$ – Jeromy Anglim Apr 26 '14 at 0:43
  • $\begingroup$ You are right. Sometimes one wonders "what would happen if the indirect effect were only via w or v?" But this is not your question. I heard about "the phantom model approach", but I havent read it yet. $\endgroup$ – YouTu May 2 '14 at 8:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.