Multiple imputation assuming missing at random: auxiliary imputation variables that only predict missingness can increase bias due to data missing not at random

Data Availability

Stata code to verify algebraic results, and also to generate and analyse the data as per the simulation studies is included in Supplementary Material, Section S6. Stata code to perform the real data analysis is included in Supplementary Material, Section S7. The real data are not publicly available due to privacy restrictions. Requests to access these datasets should be directed to alspac-data@bristol.ac.uk.