You are here

Inquiry about handling missing data with moderators of more than three categories~

2 posts / 0 new
Last post
aexwtt's picture
Offline
Joined: 03/05/2020 - 23:21
Inquiry about handling missing data with moderators of more than three categories~

Dear Mike,

This is Wen-Ta from Department of Applied Foreign Languages, National Taiwan University of Science & Technology.

I am very interested in meta-analysis and now applying your metaSEM package to my research project.

May I possibly ask a question about handling missing data with moderators of more than tree categories:

After I created four indicator variables for the four categories of the moderator (Proficiency Level), and then I applied the 'meta3X' code and specified the level of analysis (x2), I failed to obtain the parameter estimates.

Adv <- ifelse(SA2$Level=="Advanced", yes=1, no=0)

Int <- ifelse(SA2$Level=="Intermediate", yes=1, no=0)

Base <- ifelse(SA2$Level=="Basic", yes=1, no=0)

Mix <- ifelse(SA2$Level=="Mixed", yes=1, no=0)

summary( Model2 <- meta3X(y=y, v=v, intercept.constraint=0, x2=cbind(Adv, Int, Base, Mix), cluster=studyID, data=SA2, model.name="Model 2") )

Where did I go wrong? I am wondering whether you can guide me the correct way of doing this?

I have attached the dataset, R codes, and output for your reference.

Best Regards,

Wenta

Mike Cheung's picture
Offline
Joined: 10/08/2009 - 22:37
Dear Wenta,

Dear Wenta,

meta3X() uses FIML to handle missing covariates. It means that the covariates are assumed multivariate normal. I don't think that it works for categorical variables.

One approach is to apply MI on the covariates first. Then you may use meta3() on the imputed datasets.

Best,
Mike