Mediation Analysis with ordinal mediators

[zeynep genç]

Hi, I am running a mediation analysis involving two ordinal mediators and an ordinal dependent variable. Lavaan uses DWLS estimator as default for ordinal mediators. I am wondering if I still need to use bootstrapping, or even it is possible?

Many thanks,

Zeynep

[Terrence Jorgensen]

I am wondering if I still need to use bootstrapping, or even it is possible?

I would worry about the threshold distribution varying across bootstrap samples, especially if some categories are sparse.

But you don't even "need" to use bootstrapping with continuous outcomes.  

• For asymptotically large samples, the delta method works fine for testing.  Those are the default output for user-defined parameters in lavaan's summary() output
• In modest sample sizes, I usually recommend Monte Carlo CIs, which have similar advantages but are much more computationally efficient.  To learn about this method (and how it compares to others), you can read the References paper cited on the ?semTools::monteCarloCI help page.  

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