Logistic regression was used to test the associations of the tobacco use variables with the SRB The odds ratios (OR) and adjusted odd ratios (AOR) with 95% confidence intervals (95%CI) were computed first adjusting for age and sex. Sex-by-tobacco use interactions were tested for all SRB. If such interaction was observed (p<. 10), we conducted the subsequent analyses separately by sex. Otherwise, females and males were pooled together. The final analyses were adjusted for age, sex (if no interaction), school achievement, DSM-IV MDD, impulsivity, and number of alcohol dependence symptoms at baseline, as well as for parental education. When assessing the association of early onset tobacco use with later suicide ideations, we conducted an incidence analysis where we removed those reporting any baseline suicidal thoughts. In these analyses we treated the twins as individuals using robust estimators of variance when estimating standard errors (Williams 2000). For all these individual-based analyses, we conducted sensitivity analyses, where we added the variable reflecting type of sampling (random versus non-random) into the final multiple adjusted models. None of these sensitivity analyses suggested confounding by the type of sampling; i.e. the estimates reflecting the association between tobacco use and SRB were not clearly attenuated.
