multiple imputation in a four-level setting

[yongjookim78]

Hello,

I am building four-level multilevel linear and logistic models with MLwiN by using MCMC approach. The dataset had missing information for dependent and independent variables. I am wondering where might be a good place that I can start to figure out how I can perform four-level multiple imputation. In particular, I am wondering whether there is any MLwiN macro or manual/guideline for the procedure? Or, any other software packages that you'd recommend than MLwiN? Any comments/tips/advice would be appreciated greatly!

Best wishes,
Yongjoo

[ChrisCharlton]

Are the variables with missing data at a higher level, and what kind of variable are they (e.g. continuous, binary, etc)?

[yongjookim78]

Dear Chris,

Thank you for your reply!

The missing values are at the lowest (individual) level. The primary outcome is continuous, primary exposure is categorical (with three categories), and other covariates include continuous (e.g., BMI), binary (e.g., dichotomized behavioral factors), and categorical (e.g., education, income).

I would appreciate your helps again!

Best wishes,
Yongjoo

[ChrisCharlton]

The imputation directly provided by MLwiN only handles missing variables in continuous and binary variables, so I would suggest looking at the software described in http://www.bristol.ac.uk/cmm/research/m ... jr_missing. Realcom-Impute handles missing data in level-1 and/or level-2 variables, has been used for a number of years but is relatively slow. Imputation via Stat-JR provides more options (as detailed on that page) and is considerably faster, but has not yet been as widely used.

[yongjookim78]

This is really helpful! Thank you so much Chris! Will try those approaches and keep you updated with more questions!
Best wishes,
Yongjoo