Bootstrapping Indirect Effects in Two-Level SEM or Uni-Level Models with Cluster-Robust Standard Errors

[Anna Freitag]

I have a question regarding a multilevel SEM path model. I would like to test a mediation by bootstrapping the indirect effect in a two-level path model, just as I did in a uni-level path model. However, this does not seem to be directly possible for two-level models. I also tried doing it for a uni-level model with cluster-robust standard errors, which also does not seem to be possible in lavaan.

Does anyone know how to work around this issue? Or are there any alternatives to interpreting the indirect effets in a two-level path model with lavaan?

[Terrence Jorgensen]

Does anyone know how to work around this issue? Or are there any alternatives to interpreting the indirect effets in a two-level path model with lavaan?

Bootstrapping a CI is about inference from sample to population, not about interpreting the point estimate.  You can use parametric bootstrap instead, sometimes called a Monte Carlo CI in this context.  If your indirect effects are already defined (using the := operator) in your lavaan syntax, just load the semTools package and pass the fitted model to monteCarloCI().  There is a reference on the help page you can read for more info.

Terrence D. Jorgensen    (he, him, his)
Assistant Professor, Methods and Statistics
Research Institute for Child Development and Education, the University of Amsterdam
http://www.uva.nl/profile/t.d.jorgensen