Homicides and community care: the evidence.

gender ratio between cases and controls was unequal, albeit non-significantly so, this could not account for the marked differences observed in IQ scores. On the one hand, the difference between the two groups was around one standard deviation, which is much more than the commonly observed difference of 0 to 1 points between genders (Hedges & Nowell, 1995). This was the case both for verbal (difference=12.5 points, P=0.01, Student's t-test) and performance IQ (difference= 19 points, P=0.002, Student's t-test). On the other hand, males had a higher mean IQ (WISC or WISC-R) score, and this was observed both for cases and controls, and there were more males among the cases but more females among the controls. The mean IQ scores in the sample were 97.1 With respect to the second main issue, it has been routine practice at the Maudsley to defer IQ testing of psychotic or very depressed adolescents and those on high doses of antipsychotic medication. The performance IQ may yet not adequately reflect premorbid IQ. However, the verbal IQ was also significantly lower for cases than controls (cases: 92.6, s.d. 15.9; controls: 105.1, s.d. 13.1, P=0.01, Student's t-test). Scores that were regarded as unreliable by the assessing psychologist were not included in the study. On a few occasions more than one IQ test had been performed on the same subject in adolescence, in which case only the highest scores were entered into the database for statistical analysis. Subjects for whom psychometric results were not available had IQ estimates im-puted by clinicians contemporaneously, as this was and remains an item in the 'item sheets' which teams are expected to complete for every patient. This is usually estimated in light of school reports and grades, history given by informants and interviews with the subjects, the main purpose being to identify those with very low IQ. We only used these estimates to create a binary variable with a cutoff of 85 for logistic regression modelling. In the total sample, eight subjects (all cases) had an estimated IQ below 85. The validity of this categorisation had been confirmed by formal psychometric testing. Lastly, 14 out of those 15 patients in the sample who had experienced a neuro-developmental delay (1 1/12 cases, 313 controls) had their IQ tested. The mean IQ scores differed significantly between these subjects and those who had not experienced neurodevelopmental antecedents (88.6 v. 104.3, P=O.OOS, Student's t-test). This …