Results
The rate of daily smoking increased from 3% at age 13 (1990) to 31% at age 18 (1995), which is equivalent to an average increase of 5.6 percentage points per year (Table 2). The increase in smoking "at least once a week" and "smoking every week" was much weaker. There was a significant increase in smoking rates between each wave and the subsequent one. The reported rate for the best friend being a smoker also increased significantly, from 11% to 48%, corresponding to an annual growth exceeding 7 percentage points (Table 2). Furthermore, the proportion of participants who reported that none of their friends were smokers decreased significantly, from 73% when the participants were 13 to 11% when they were 18 (Table 2).
There was no gender difference in smoking status (no smoking versus smoking at all) at baseline (χ = 0.04, df = 1, p > 0.05) (Table 3). Both father's (χ = 5.74, df = 1, p < 0.01) and mother's smoking behaviours (χ = 7.86, df = 1, p < 0.01) were positively associated with adolescents' smoking behaviour (Table 3). The distribution of whether the adolescent was allowed by their parents to smoke was too skewed to permit chi-square testing (expected frequency < 5). Parents' high SES was significantly associated with lower smoking prevalence (χ = 5.93, df = 2, p < 0.05) (Table 3). However, the two strongest predictors for adolescents to smoke were whether older siblings were smokers (χ = 51.43, df = 1, p < 0.001) and, in particular, whether the best friends were smokers (χ = 305.36, df = 1, p < 0.001) (Table 3).
Smoking prevalence increased among both boys and girls from one data collection to the next over the entire 1990–1995 period (data not shown). The gender differences were mixed over the follow-up period; girls generally had a higher smoking prevalence, but the shape of the gender differences varied both as a function of how smoking is defined and depending on which of the waves were emphasised (data not shown).
Dropout
The attrition rate from the baseline to wave five was 36%. Neither the participant's own smoking behaviour nor the best friend's smoking behaviour was significantly associated with attrition. However, participants with a high SES had a significantly lower attrition rate than those with a low and mid SES. In a post-hoc analysis regressing the intercept and slope of the smoking growth curve on whether the participants had a valid SES value, a missing SES was positively associated with intercept but was not significantly associated with slope.
Univariate Unconditional Growth Model of Adolescents' Smoking Behaviour
The univariate model (Table 1, M1) had a fair to good fit (χ = 46.73(9), CFI = 0.99, RMSEA = 0.06). The intercept mean is set to 0 by Mplus when the outcome variables are categorical, and the significant slope mean of .32 (p < 0.05) represents an increasing developmental trend of smoking behaviour with increasing age. Furthermore, the intercept
and slope
were characterised by significant variances indicating individual variability of both starting point and slope of smoking behaviour. There was a significant correlation between intercept and slope (0.35, p < 0.05).
Univariate Conditional Growth Model, Best Friend's Smoking Behaviour, Parent's SES and Smoking Behaviour at Home (Figure 1)
The question whether smoking was allowed by parents was excluded because very few informants were allowed to smoke. The unconditional LCM, reflecting the adolescents' smoking trajectory, was regressed on the best friend's smoking behaviour, parental and older siblings' smoking behaviours and parents' SES ( Table 1 , M2). The insignificant loadings on both intercept and slope were omitted from the model (Figure 1, Table 1 , M3). All predictors had a significant impact on intercept; friends' smoking behaviour (.52, p < 0.05) was the strongest, followed by parents' high SES (−0.26, p < 0.05) and older siblings' smoking behaviour (0.18, p < 0.05). The best friend's smoking behaviour was the only predictor not associated with the slope. Parents' smoking behaviour was the strongest (0.33, p < 0.05), followed by parents' SES (0.29, p < 0.05) and older siblings' smoking behaviour (0.13, p < 0.05). Parents' SES changed from a negative association to a positive association.
(Enlarge Image)
Figure 1.
Own smoking by best friend's smoking, parents' and older siblings' smoking, and parents' SES. A conditional univariate latent curve model covering the age span from 13 to 18
Univariate Conditional Growth Model, Parent's SES Mediated Through Smoking Behaviour at Home (Figure 2)
The unconditional LCM was also regressed on parental and older siblings' smoking behaviours and parents' SES ( Table 1 , M4) without the best friend's smoking behaviour. The insignificant loadings on both the intercept and slope were omitted from the model, and the final model ( Table 1 , M4) also included the indirect associations. These indirect effects were associations between parents' high SES and the intercept and slope via both parents' smoking behaviour and older siblings' smoking behaviour (Figure 2). The final model showed good fit (χ = 68.43, CFI = .99, RMSEA = 0.05), and all associations shown in Figure 2 were significant. Parents' high SES was significantly associated with older siblings' smoking behaviour (−0.21, p < 0.05), parents' smoking behaviour (−0.27, p < 0.05), intercept (−0.19, p < 0.05) and slope (0.30, p < 0.05). Older siblings' smoking behaviour was associated with both the intercept (0.25, p < 0.05) and slope (0.18, p < 0.05). Parents' smoking behaviour was also associated with both intercept (0.15, p < 0.05) and slope (0.32, p < 0.05).
(Enlarge Image)
Figure 2.
Own smoking by parents' and older siblings' smoking and parents' socioeconomic status. A conditional univariate latent curve model covering the age span from 13 to 18, direct and indirect associations
There were also significant indirect associations. The total effect from parents' high-SES (−0.28, p < 0.05) on the intercept, the direct effect (−0.19, p < 0.05) and the indirect effects via siblings' smoking behaviour (−0.05, p < 0.05) and parents' smoking behaviour (−0.04, p < 0.05) were significant. The total effect from parents' high SES (0.18, p < 0.05) on the slope, the direct effect (0.30, p < 0.05) and the indirect effects via siblings' smoking behaviour (−0.04, p = 0.05) and parents' smoking behaviour (−0.08, p < 0.05) were also all significant.