Before beginning the twin analyses we examined whether the categories created for smoking and alcohol consumption were within the same continuum of liability. This was done by estimating the tetrachoric correlations (and thresholds) separately for each of the zygosity groups for these variables. χ2 was not significant for smoking or alcohol consumption at wave 2 (χ2=8.4, p=0.21; χ2=7.0, p=0.32) or wave 4 (χ2=6.4, p=0.37; χ2=6.0, p=0.42), suggesting that these variables satisfy assumptions of multivariate normality. However, χ2 was significant for both substances at wave 3 (χ2=14.6, p<0.05; χ2=19.5, p<0.05). Analyses were repeated separating twins by age (>16.5 or ≤16.5 years) and χ2 was no longer significant. This suggests that wave 3 variables are approximately multivariate normal and the differences detected in our initial analysis are due to age.
