Multiple imputation with mice of twin datasets

ISG Workshop 2022
1

Hi,

I am struggling with imputation of covariates in a twin dataset. I am using co-twin design (generalised estimating equations) to estimate the within- and between-family effects. For the outcome and exposure, I am using complete cases. I would like to adjust the estimates for covariates that are twin-specific, but they have missing data.

How is the data imputation typically done for twin data? Is it possible to do it using mice package in R?

I have checked mice guidelines for imputation on multilevel data, but I am not sure I understand how to best adapted for twin data as the size (n) of each cluster/family is small.

Is it wrong to impute the twin data in wide format ignoring the clustering?

With many thanks and my best wishes,
Sinziana

2

Hi Sinziana

Usually, we use full information maximum likelihood so imputation is unnecessary. If the data are continuous, covariates can be regressed out as a preliminary step, but if they are ordinal or binary, it’s typically necessary to do the covariate adjustment in the model.

HTH
Mike

3

Hi Mike,

Thank you very much for your answer.

With my best wishes,
Sinziana