Mediation model with complex data and... PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
Message/Author
 Allison Pack posted on Thursday, October 04, 2018 - 5:23 am
Hi,

I am working on a proposal and would like to know if it's possible to do a path analysis to examine a fully longitudinal indirect effect with the data I will have.

The data include stratification (3 countries), survey weights, and clustering - and I expect I will need to use multiple imputation with them. There are about 500 individuals in the sample, all of whom have data for the outcome variable.

At time 1, I have a dichotomous predictor variable and I can control for the mediator at this time as well.

At time 2, I have a sum score mediator and I can control for the dichotomous outcome.

At time 3, I have a dichotomous outcome variable.

In reading through the message boards and several papers, it seems to me that Mplus can account for much, if not all, of these things individually using:

TYPE=COMPLEX
WEIGHTS=WeightVar
CLUSTER=StratVar
Estimator= WLSMV

But what I don't know is if Mplus can do all of these things together - and/ or if I am missing anything else that I should be aware of.

I am also not sure whether, with a sample of 500, it makes sense to use the Sobel test or bootstrapping - and which might work best in Mplus with this data and analysis.

I would very much appreciate some advice. Thank you!
 Allison Pack posted on Thursday, October 04, 2018 - 7:43 am
I maybe should have included that I think I am going to need to do a log-binomial model rather than a logistic model. This is because my outcome is greater than 10% (VanderWeele, 2016 Mediation Analysis, A Practitioner's Guide)
 Bengt O. Muthen posted on Friday, October 05, 2018 - 2:53 pm
WLSMV and ML can do these things together. Even though your binary outcome has prevalence > 10%, you can use the Mplus counterfactually-defined indirect/direct effects for logistic link - they don't assume a rare event for the outcome as the VanderWeele approximate odds ratio approach does. See chapter 8 of our RMA book.
 Allison Pack posted on Saturday, October 06, 2018 - 7:15 am
Thank you so much. This is very encouraging to hear, and I will check out your book as well.
 Allison Pack posted on Tuesday, April 09, 2019 - 10:43 am
Hi again, I wanted to ask about this whether it is possible to obtain a 95% confidence interval with imputed data. It is possible I may need to impute this data. Thank you!
 Tihomir Asparouhov posted on Tuesday, April 09, 2019 - 8:00 pm
Yes. You can use the formula

Estimate +-1.96*SE

or you can have Mplus do that for you with the option output:cint;
 Allison Pack posted on Wednesday, April 10, 2019 - 6:32 am
Thank you. Would that be a bias-corrected bootstrap 95% CI? Is that possible to use with imputed data?
 Tihomir Asparouhov posted on Wednesday, April 10, 2019 - 2:43 pm
No and it is not possible at this point in Mplus. It would be the above symmetric CI.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: