I have 4 exogenous variables, 3 mediators, and 3 endogenous variables. AMOS has calculated direct and indirect effects, and I have relied on standardized indirect effects to report in my dissertation. The reason for not conducting a bootstrap was that when I tried to calculate it, even with 200 samples, my PC hanged for 1 hour, and then it gave me an irrelevant error result. How can I defend my findings?!is it a big problem?
1 Answer
$\begingroup$ $\endgroup$
2 If Amos supplies the asymptotic matrix of sampling (co)variances of the estimated parameters, then you could use the much more computationally efficient Monte Carlo method, which does not require fitting the model because it samples parameters rather than data:
https://quantpsy.org/medmc/medmc.htm
If you fit your model using the modern, free, open-source R package lavaan to fit your SEM, you could just define your indirect effects in the model syntax:
https://lavaan.ugent.be/tutorial/mediation.html
Then you could use the semTools package's monteCarloCI() function to easily obtain a Monte Carlo CI for any user-defined parameters.
- $\begingroup$ Interesting! Does the monteCarloCI() function provide asymmetric confidence intervals for indirect effects? $\endgroup$Christian Geiser– Christian Geiser2025-09-17 11:21:11 +00:00Commented Sep 17 at 11:21
- 1$\begingroup$ Yes, you can read the reference listed on the help page for details. It is a resampling technique, and confidence limits are calculated from an empirical distribution (not analytically, so symmetry is not required), similar to bootstrap CIs. $\endgroup$Terrence– Terrence2025-09-18 13:07:16 +00:00Commented Sep 18 at 13:07