The findings need to be interpreted in light of several limitations. Firstly, responders and non-responders to the self harm questionnaire administered at age 16 years differed on a range of characteristics, showing that the data were not missing completely at random, and this non-random response may have biased our complete case analyses. However, we also performed analyses using imputed data and results were consistent with the complete case analyses. Under the missing at random assumption, multiple imputation should correct biases that may arise in the complete case analyses. Although we cannot say with certainty that the data are missing at random, our imputation models included a wealth of auxiliary information, which increases the plausibility of the missing at random assumption. Secondly, we were not able to distinguish those who had self harmed with suicidal intent from those who had self harmed without suicidal intent during the past year at age 21 years. Thirdly, we cannot rule out the possibility of reverse causation for associations with GCSE qualifications, as these school examinations are generally taken at age 16 years, when self
