Handling missingness in mediation model with binary outcome?

[Zinrc3 K]

Hi all,

I've referred to the syntax that Yves has kindly shared with the group previously for mediation with binary outcomes. I'm wondering if there is a way to carry out imputation (if this seems appropriate) with this code? 

For instance, I've used the following code for mediation with continuous outcome variable.

Med=lavaan::sem(INVMED, data=dat, meanstructure=TRUE, missing="fiml", se="boot",

                      bootstrap=1000) 

whereas the sample code for binary mediation reads:

Med=lavaan::sem(INVMED, data=dat, ordered=c("outcome")) 

(1) Is there an appropriate missing option I can use here and

(2) Can/should I specify the estimator here?

Thank you

[Terrence Jorgensen]

(1) Is there an appropriate missing option I can use here and

You could specify missing="pairwise" to use all available information to estimate each polychoric correlation matrix.  Like the default "listwise" deletion, it assumes MCAR data, which is more restrictive than FIML's MAR assumption.

To work with the less restrictive MAR assumption, you can use multiple imputation to obtain completed copies of your data set, then pass the list of imputations to the sem.mi() function in the semTools package.

(2) Can/should I specify the estimator here?

lavaan only supports DWLS estimation when there are categorical endogenous variables, so there's no point trying to override that default. 

Terrence D. Jorgensen

Assistant Professor, Methods and Statistics

Research Institute for Child Development and Education, the University of Amsterdam

http://www.uva.nl/profile/t.d.jorgensen