PREADOLESCENT PREDICTORS OF SUBSTANCE INITIATION: A TEST OF BOTH THE DIRECT AND MEDIATED EFFECT OF FAMILY SOCIAL CONTROL FACTORS ON DEVIANT PEER ASSOCIATIONS AND SUBSTANCE INITIATION

Abstract

A review of the literature suggests that early substance initiation is related to a variety of negative outcomes, including substance misuse or abuse in adolescence and adulthood. This study examines potentially modifiable predictors of early substance initiation, including both family and peer factors known to influence early initiation. A theoretically derived model of substance initiation was tested using structural equation modeling. Results indicate that both family and peer factors have an impact on early substance initiation when children in this sample were 11 and 12 years old. The model explained 60% of the variance in substance initiation. Prosocial family processes (rules, monitoring, and attachment) had a significant impact on child peer association, decreasing involvement with antisocial peers. These prosocial family processes had a significant negative effect on substance initiation even while modeling the influence of antisocial peers. Implications for drug use prevention practice are discussed.

• Family management
• Peer association
• Substance initiation

Notes

^*Some of the students in this sample are part of a prevention program. This raises the question of whether the program affects the measurement model examined here, leading to different measurement models for experimental and control groups [[56]]. To examine the consistency of the measurement model, we compared the fit of an unconstrained multisample confirmatory factor analysis (CFA) and constrained CFA model between experimental (n = 491) and control (n = 414) groups [[55]], [[56]]. Specification of these models was identical to the final CFA for the main analysis. The unconstrained CFA multisample fit the data well, χ²(178 df) = 365.74, CFI (comparative fit index) of 0.947, and RMSEA (root mean square error of approximation) of 0.036. With equality constraints placed on all the factor loadings, the constrained model also fit the data well, χ²(179) = 393.00, CFI of 0.942, and RMSEA of 0. 036. The constrained and unconstrained solutions were statistically significantly different, suggesting that some factor loadings were not equivalent (Δχ²₍₁₁₎ = 27.26) [[57]]. The factor loading with the greatest difference in the two groups was V14, which is a youth report item about whether the youth has ever smoked marijuana. For the control group, the unstandardized factor loading is 0.428, while for the experimental group the unstandardized factor loading is 0. 303. Given that the factor loadings are in the same direction and different only in magnitude, we assumed partial measurement variance as specified by Muthen and Christoffersson [[58]] and Byrne [[57]] and reran the model freeing the equality constraint between groups on V14. This resulted in a nonsignificant chi-square difference test between groups (Δχ²₍₁₀₎ = 16.49) suggesting that with V14 free across groups, the remaining measurement model is invariant across groups [[57]]. Consequently, given the measurement model consistency across groups, we combined the data for the subsequent analysis.