The relative risk of schizophrenia was estimated by log-linear Poisson regression14 with the GENMOD procedure, using SAS statistical software, version 6.12.15 All relative risks were adjusted for age and its interaction with sex, calendar year, and history of mental illness in a parent or sibling. Age, calendar year, and history of mental illness in siblings were treated as time-dependent variables,16 whereas history of mental illness in a parent was treated as a variable that was independent of time. Age was categorized using the following cut points: 15, 16, 17, 18, 19, 20, 22, 24, 26, 28, 30, 35, and 40 years. Calendar year was categorized using the following cut points: 1971, 1976, 1979, 1982, 1985, 1988, 1991, 1993, 1994, 1995, 1996, 1997, and 1998. P values were based on findings from the likelihood ratio tests, and 95% confidence limits were calculated by the Wald test.16 The adjusted-score test17 suggested that the regression models were not subject to overdispersion. Apart from the reduction in the size of the study population, omission of adjustment for seasonality, maternal and paternal age, and the inclusion of information on permanent address for all cohort members since 1971, the material described is identical to that used in our previous study,3 where we found that the effects of urbanicity at birth and mental illness in a family member were invariant to the diagnostic system and the inclusion of outpatient information.