3.1. Prevalence of Cannabis Use Disorders
8.4% of the participants met criteria for a lifetime history of DSM-IV cannabis abuse and 1.3% for a lifetime history of DSM-IV cannabis dependence (Grant et al., 1998
; Stinson et al., 2006
). Men were more likely than women to meet criteria for cannabis abuse (11.1% vs 4.9%) or cannabis dependence (1.8% vs. 0.8%). We elected to perform all analyses on the sub-sample of 8172 individuals who reported lifetime cannabis use and were therefore queried about individual symptoms of cannabis abuse and dependence. In cannabis users, rates of cannabis abuse and dependence were 39.9% and 6.5% respectively. shows the increased prevalence of individual abuse/dependence criteria in men versus women. With the exception of Quit, which was frequently, and equally endorsed by both genders, all other criteria were endorsed more often by men than women.
Prevalence of individual cannabis abuse/dependence criteria in men and women who report lifetime cannabis use (N=8172, in NESARC).
3.2. Factor Analysis
The EFA showed high loadings of abuse and dependence criteria on a single underlying factor (abuse/dependence), with high loadings for all criteria except Quit, Legal and Hazard. A two-factor model, which also fit the data well, did separate an abuse factor from a dependence factor but when the 2-factor CFA was fitted using the robust maximum likelihood estimator for complex survey designs, the abuse and dependence factors were found to be very highly correlated (inter-factor correlation of 0.92), and the addition of a second factor did not significantly alter the variance explained by each variable nor the factor loadings. A CFA with 2 uncorrelated factors fit poorly. Therefore, we opted to pursue the single-factor model for further analyses (). This single factor showed high loadings of all items except Legal, Hazard and Quit, of which Hazard and Quit also had lower thresholds.
Factor loadings (standardized parameter estimates) and thresholds from a confirmatory factor analysis of cannabis abuse and dependence.
We first fit the one factor model with 4 abuse and 7 dependence criteria (i.e. including Withdrawal) and subsequently fit a sub-model where Withdrawal was dropped from the CFA. The highly significant change in model-fit (Δχ2=662.79) coupled with the relatively high factor loading of Withdrawal () suggested that while the general factor structure for cannabis abuse/dependence replicates in the absence of Withdrawal, the addition of Withdrawal as a criterion of cannabis dependence appears to improve model fit.
3.3. MIMIC models
The regression of abuse/dependence on gender was significant. Males were more likely to have higher factor scores (or endorse more criteria on average) on the latent abuse and dependence construct. To this model, we individually added a regression path from each criterion to gender. As seen in , after controlling for its effect on the underlying latent construct of abuse and dependence, we found gender differences in Quit (β=0.22), Problems (β=0.41), Legal (β=−1.16) and Hazard (β=−0.66), with a trend-level significant gender difference for Give-up (β=0.42) and Intend (β=0.28). Interestingly, parameter estimates for all four dependence criteria were positive while the estimates for the abuse criteria for Legal and Hazard were negative suggesting that women were more likely than men to endorse the dependence criteria while men were more likely to endorse abuse criteria, independently of gender differences in the underlying abuse/dependence construct.
Table 3 Fit statistics and parameter estimates for MIMIC models testing whether gender explained additional variance in individual criteria for cannabis abuse and dependence even after accounting for its effect on the latent abuse/dependence construct (N=8,172 (more ...)
3.4. Two-group models
A model in which the factor loading and threshold of Quit (Δχ2 = 10.99, df=2), Problems (Δχ2 = 8.12, df=2), Legal (Δχ2 = 49.06, df=2) and Hazard (Δχ2 = 52.51, df=2) were freely estimated in men and women fit significantly better than the constrained model. For the two dependence criteria that showed trend-level significance in the MIMIC models, Give-up and Intend, no significant gender-variance was found in the two-group model. We noted that while women had lower thresholds for endorsement of Quit and Problems, and men had lower thresholds for Hazard and Legal, the factor loadings for these criteria did not vary considerably. Subsequently, we constrained the factor loadings to be equal across genders but allowed thresholds to differ, and this model provided a suitable fit to the data (Δχ2 = 4.14, df=4), results for which are presented in . Thus, results from the two-group models showed that abuse criteria for Legal and Hazard were more likely to be endorsed by women relative to men with high scores on the latent dimension while the dependence criteria for Quit and Problems were endorsed by men relative to women with higher scores on the latent abuse/dependence dimension.
Factor loadings (constrained across genders) and thresholds, with some cannabis abuse/dependence criteria showing gender variance: results from a two-group gender heterogeneity model.
Both MIMIC and two-group analyses were repeated using 6 dependence (i.e. excluding Withdrawal) and 4 abuse criteria as currently in DSM-IV. Dropping Withdrawal did not substantively change the findings reported above.