General descriptions of the design and methods used in the 2007 Adult Psychiatric Morbidity Survey have been provided elsewhere , and the methods relating specifically to the analysis of sexual abuse are described in more detail by Bebbington et al. [5, 7]. The survey sought to establish a random sample of household residents aged 16 and above, representative of the adult population of England (5,689 of 7,461 participants provided sufficient data for these analyses). First phase interviewing was carried out by experienced non-clinical interviewers; while in a second phase, selected participants were interviewed clinically, in particular with the Schedules for Clinical Assessment in Neuropsychiatry (SCAN, version 2.1) .
Affect was assessed in relation to the preceding week during phase 1, using the Clinical Interview Schedule (revised) (CIS-R) . This can be used to create overall scores for anxiety (range 0–8) and depressive symptoms (range 0–9). A category of probable psychosis included people diagnosed as such in the second phase, plus participants who did not attend the second phase but met at least two of four screening criteria based on the Psychosis Screening Questionnaire , antipsychotic medication use and medical contact with a given diagnosis of psychosis or schizophrenia. A detailed history of sexual abuse was obtained in phase 1, using confidential computer-assisted self-completion to encourage disclosure. In the current analysis, we use information about sexual abuse occurring before the age of 16 years: three levels of abuse were recorded, sexual talk (reported by 10.3 % of the sample), physical molestation (reported by 8.2 %), and non-consensual sexual intercourse (reported by 1.9 %). The commonest form, sexual talk, is more subjective than the first two, and is generally less strongly associated with psychiatric outcomes . In the current analysis, we provide data separately for non-consensual sexual intercourse, and for “contact abuse”, a category derived by combining intercourse and unwanted sexual touching. Survey data were weighted to take account of survey design and non-response, to render the results representative of the English household population.
In the current paper, we used the ‘survey’ commands in STATA (Version 13 for MAC): these allow for the use of clustered data modified by probability weights, and provide robust estimates of variance. To test the role of affect as a mediator of the relationship linking CSA with psychosis, we applied the Karlson Holm Breen (-khb-) command in Stata. This method of mediation analysis decomposes the total effect of a variable into direct and indirect effects  and can be used in logit models. In these analyses, the confounding effects of concomitants on the decomposition were controlled for. These concomitants comprised sex, age, ethnicity, educational qualifications, equivalised household income, and whether the participant had been brought up by both biological parents until 16 years of age. Confidence limits were derived using the delta method of Sobell .