The cross-lagged path analysis adjusted for sex. In addition, to diminish potential confounding by familial factors, parental education, smoking and alcohol use were used as covariates in all models. Parental education was assessed for mothers and fathers as an ordinal variable with four categories, ranging from compulsory schooling only to tertiary education, and the highest education of either parent was used as a covariate. Similarly, alcohol use frequency was assessed separately for mothers and fathers as an ordinal variable, ranging from never use to a few times a week or more often, and the highest frequency of parental drinking was used. Parental smoking denoted whether either of the parents reported any current smoking at study baseline. In addition to the parental covariates, externalizing behavior problems at age 12 were adjusted for in further analyses. Externalizing behaviors were rated by the twins’ teachers using a Finnish scale, the Multidimensional Peer Nomination Inventory (MPNI), which has sub-scales for hyperactivity-impulsivity, aggression, and inattention (Pulkkinen et al., 1999). A sum score of these sub-scales, previously shown to predict substance use (Korhonen et al., 2010, 2012), was used to assess externalizing behavior problems.
