www.cmm.bristol.ac.uk/forum

Afternoon,

I am new to MLwiN and to multiple imputation of missing data.

I have a longitudinal data set with missing data present in some categorical/binary predictor variables.

I wish to use MI to increase the working sample size for the modelling procedure.

What is the best way to do this? I get errors when using the MI macro saying that there is an outcome length mismatch - I guess this is because individuals may be measured between 1 and 9 times.

I am thinking that REALCOM can cope with this scenario. I am also currently looking at MICE in R.

Any help would be appreciated.

Thanks,
Justin Grace

additional:

I get a data length mismatch error when imputing using MLwiN and similar using the macro. I presume this is because the data are unbalanced and these methods cannot handle this?

You can ceratinly use realcom for this. The data should be presented as a 2-level structure, i.e. sorted by measurement occasion within individual. Refer to manual for details.