Attachment | Size |
---|---|
FIML_example.R | 1.79 KB |
misdat_example.csv | 2.76 KB |
Dear Dr. Cheung and forum pannels,
This is Jihyun, a doctoral student.
I have a question about (currently) two approaches to use FIML to handle missing covariates in meta-regression.
At first, I had adopted the code in Dr. Cheung's book (2015) using RAM formulation and recently also tried to use metaFIML after it was introduced. However, I noticed that the two approaches showed pretty big differences in the results (parameter estimates and standard error estimates).
Could you see my code if I did something wrong? Or, since you noted that metaFIML maybe not stable, should we not use the function yet?
In the dataset, smd is the effect size estimate, v is variance, V1 is the covariate.
I enclosed a data misdat_example.rds
and code (FIML_example.R) with FIML approach 1 and approach 2.
I appreciate any of your comments!
Thank you so much.
And again, thank you for distributing metaSEM package!
Hi Jihyun,
The metaFIML works fine. There are two issues here.
1) The known v is incorrectly labelled as
v
, which should bedata.v
.2) The estimated Tau2 is negative, which is not allowed. A lower bound is required.
Mike
I appreciate your the time to take a look at this. I see what was the problem.
Thank you!
Regards,
Jihyun