Childhood sexual abuse (CSA) is found to have occurred to a substantial minority of women with bulimia nervosa. Its clinical significance is unclear.

We studied 87 bulimic women in a clinical trial. Structured interviews determined the presence of CSA, DSM–III–R disorders, global functioning, and depressive and bulimic symptoms.

Forty-four per cent reported a history of CSA. Bulimic women with CSA reported earlier onset of bulimia, greater depressive symptoms, worse global functioning and more suicide attempts, and were more likely to meet criteria for bipolar II disorder, alcohol and drug dependence, conduct disorder and avoidant personality disorder.

Although those with CSA had greater comorbidity, it was not an important modifier of bulimic symptoms.

Statistical error rates in the medical literature are generally high.

All papers published in the British Journal of Psychiatry in 1993 which presented numerical results were reviewed by the author for statistical errors.

A total of 248 papers were published, of which 164 (66%) presented numerical results.

Sixty-five (40% of 164) papers contained statistical errors. Many errors were not serious in nature, but some were serious enough to cast doubt on conclusions. The error rates are similar to those found in an earlier study.

The statistical error rate is unacceptably high. There is no evidence of a change in the statistical error rate over time.

The paper by McGuigan in this issue of the BJP examines the frequency of errors in statistical methods presented in papers published in the BJP in 1993. In that year, 248 papers appeared, 164 contained numerical results, and 65 of these contained statistical errors. So, one might infer, some 26% of the papers in the BJP draw potentially doubtful conclusions. Moreover, the level of misunderstanding implied by some of the examples McGuigan cites is frightening.