Multiple imputation is becoming increasingly established as the leading practical approach to modelling partially observed data, under the assumption that the data are missing at random. However, many medical and social datasets are multilevel, and this structure should be reflected not only in...