The number of cannabis users increases with age as does the frequency of use. Cannabis users did not differ from non-users with respect to SES (t(1447) = −.9, p = .387), gender (χ2 (1) = 1.1, p = .289), familial vulnerability for internalizing (t(1447) = −.4, p = .705) and externalizing behaviour (t(1447) = −1.8, p = .071). Cannabis users and non-users differed significantly with respect to alcohol use at T2 (χ2 (1) = 90.3, p < .001), alcohol use at T3 (χ2 (1) = 95.0, p < .001), tobacco use at T2 (χ2 (1) = 137.3, p < .001)

and tobacco use EGFR inhibitor at T3 χ2 (1) = 346.8, p < .001), with cannabis users using alcohol and tobacco more often than non-users (57.8% vs. 31.2% reported monthly alcohol use at T2; percentages for T3: 94.0% vs. 70.7%; 19.8% vs. 2.2% reported weekly tobacco use at T2; percentages for T3: 57.4% vs. 11.1%). Tobacco and alcohol use were also related to both internalizing and externalizing behaviour and therefore included as covariates in subsequent find protocol path analysis (for detailed information, see Table 2). Factor loadings of the indicators of the latent variables of internalizing behaviour and externalizing behaviour of all three measurement waves are presented in Table 3. Table 4 shows

the correlations between all latent variables. The independence model testing the hypothesis that all cannabis scores and internalizing behaviour scores were uncorrelated was rejected: χ2 (30, N = 1,449) = 56.4, p < .003. The model provided an acceptable fit to

the data (CFI = .99, RMSEA = .03). However, as shown in Table 3, correlations between internalizing behaviour problems (T1-2-3) and cannabis use (T2-T3) ranged from .02 to .06 and thus are very small. Although these correlations were significant (probably due to the large sample size), they of were indicative of non-relationships between cannabis use and internalizing behaviour. This was confirmed by the Wald test. Dropping parameters indicative of associations between internalizing behaviour (T1, T2 and T3) and cannabis use (T2 and T3) resulted in a non-significant change of the model [χ2 (6, N = 1,449) = 11.2, p = .081]. Path-analysis revealed that although our model represented the data well [χ2 (66, N = 1,449) = 215.2, p < .001; RMSEA = .04, CFI = .97], all paths between internalizing (T1-2-3) and cannabis use (T2-T3) were non-significant. The independence model that tested the hypothesis that all cannabis scores and externalizing behaviour scores were uncorrelated, was rejected: χ2 (9, N = 1,449) = 64.4, p < .001. Also, although RMSEA was relatively high (.07), the CFI was .99 and therefore our model provided an acceptable fit to the data. Correlations between externalizing behaviour (T1-2-3) and cannabis use (T2-T3) ranged from .19 to .58 and thus were indicative of a relationship between externalizing behaviour problems and cannabis use (see Table 4).