Roger Brown posted on Thursday, November 20, 2014 - 10:58 am
While I am aware that Mplus currently doesn't allow bootstrapping of two-level models with indirect effects. Is there a solution to this? I have discovered a pieced together approach by Lau and Cheung (2012) using Mplus in conjuction with LISREL/PRELIS, but wow lots of work here. Is Mplus planning this in the near future? Thanks.
Lau, R. S., and Cheung, G. W. (2012) Estimating and comparing specific mediation effects in complex latent variable models. Organizational Research Methods, 15(1), 3-16.
It's high on our list. In the meanwhile, I recommend Bayesian analysis which like bootstrap allows non-symmetric confidence intervals.
Dan R. posted on Thursday, June 11, 2015 - 7:36 pm
Hello, Dr. Muthen. Are there any other means of obtaining asymmetrical confidence intervals in Mplus with two-level models with indirect effects? Do you know of any literature (or resources) describing how this is accomplished in Mplus, using either Bayesian analysis or other means? Thanks!
Bayes is a good choice here. I would use the 2-level mediation literature (using ML) and use Bayes together with Model Constraint to express the effects. I can't think of 2-level Bayes mediation literature.
Hi, I have a 2 level moderated mediation model for testing. I have opted to assess the indirect effect using Confidence Interval, but face major limitations with the inability to bootstrap. I tried Estimator = Bayes, only to realize that this feature cannot be used in conjunction with Model Indirect.
I followed your advice and used the Estimator=BAYES option with MODEL CONSTRAINT to bootstrap the CI of the indirect effect in my multilevel mediation model. However, the estimates of most parameters differ greatly between the MLR and BAYES estimation methods (several even switch signs). Why could this be? and is there way to resolve this?
Too few Bayes iterations were used. Check Tech8 to see that you have a long string of iterations with PSR close to 1. Also, Bayes gives non-symmetric CIs that may be needed particularly for the indirect effect.
Hello, I'm currently running a two-level path model with a continuous dependet variable, a continous mediator variable and various independet variables on both levels (1 and 2). Additionally, I have used a multiple imputation to imputate the missing data before the path model. I reported direct and indirect effects (mediation), but now I should control the indirect effects with a bootstrap analysis. I read that I should use a Bayesian analysis, but I can`t find comparable examples. Can you tell me, where I can find them or how I can work this out?
I would recommend using Bayesian estimation without first doing an imputation step. If you have missing data on covariates, they can be brought into the model. The analysis setup is the same as for ML - but you get the non-symmetric CIs that you want for the indirect effects.
Is there an example for this analysis in the User`s guide or somewhere else? I have 9 indepente variables on level 1 and 11 on level 2. With the Model Constraint Option, this will be pretty complex, right?