Abstract
This study examines the impact of changes in the influence of several individual-level constructs (e.g., risk-seeking, family, peers, attitudes) on involvement in substance use and delinquent behavior during mid to late adolescence. Data were drawn from the longitudinal evaluation of the Gang Resistance, Education, and Training (G.R.E.A.T.) program. The final pooled time series samples used in the analyses were n = 2,515 in the delinquency model (original sample: n = 847) and n = 2,250 in the substance use model (original sample: n = 849). Fixed effects negative binomial modeling indicated that: (1) changes in risk-seeking, parental and peer influences, and attitudes are significantly related to the frequency of self-reported delinquency and substance use and (2) similar risk factors, drawn from multiple domains, influence the likelihood of greater involvement in both delinquency and substance use during mid to late adolescence. Findings highlight the importance of considering a variety of dynamic processes during adolescence and their potential changing influence on deviant behaviors.
Notes
1The social environment may be influenced by the individual as well. This is apparent in explanations of delinquency such as Thornberry's (Citation1997) interactional theory.
2The decision to exclude the pre-test data was based on the need to ensure equally spaced time intervals between the data points. The pre-test data were collected in the Spring of 1995, the post-test data were collected in the Fall of 1995, and the annual follow up data were collected in the Fall of each consecutive year (i.e., 1996–1999) (Esbensen Citation2003).
3In a prior study testing the stability of self-control, Winfree et al. (Citation2006) used a similar portion of these data (n = 965). Attrition analyses suggested that the youths included in the study were somewhat “lower risk” than the youths in the full sample. Specifically, those included in the analysis were more likely to be older, white, live in two parent families, and have higher levels of self-control.
4Clearly, the lack of inclusion of all proposed aspects of self-control reduces our ability to thoroughly address the theory. Still, in assessing potential social influences on delinquency, it is important to capture individual factors, such as those that reflect self-control, to ensure proper specification of the model (see, e.g., Pratt and Cullen Citation2000).
5These measures revealed a bivariate correlation of .58 (p < .001).
6Importantly, these estimated effects do not account for variation in the influence of covariates over time (i.e., variation in coefficients). They merely assess the effect of change in the covariate, relative to its mean, on delinquency/drug use. Further, because this is a “pooled panel” analysis, there is only one estimated effect for each covariate in the model. In effect this estimate summarizes the overall picture across the years in question. The use of interaction terms with age helps to provide some further insight on the time varying effects of key variables.
7All but three of the respondents included in the sample were between the ages of 11and 14 at the initial age of observation (>99%). Specifically, one youth was ten years old, one youth was sixteen years old, and one youth was nineteen years old. All analyses reported in this study were re-analyzed excluding these three outliers. However, no significant differences in the results were revealed. Therefore, to preserve the age range of the original sample, these three cases were included in the results reported in the following section.
*p < .05; ∧Based on 50 bootstrap replications.
a Brame et al. (1999) suggest that the time-stable variables are included in the model to control for the stability in each measure.
*p < .05; ∧Based on 50 bootstrap replications.
8Diagnostics were run on the data to check for multicollinearity. For both models, multicollinearity does not seem to be a problem; the VIF value of each variable was under 3. Belsley (Citation1991) recommends a cut-off value of four, with values greater than four indicating multicollinearity in the data.
9The effect for delinquency on “parental attachment by age” had a z value of 1.73, but its magnitude was quite small (i.e., less than a one percent change in delinquency count for a unit increase in the interaction term).
10Only two of the six elements of self-control (risk-taking and impulsivity) were included in the G.R.E.A.T. survey. Furthermore, Winfree et al. (Citation2006) reported significant variation in both risk-seeking and impulsivity using this same data set.
11Attempts were made to contact survey participants who were absent, suspended, expelled, or transferred schools. However, the sample was originally based on adolescents who attended school (Esbensen et al. Citation2001).