The Effect of Strain, Affect, and Personal/Social Resources on Problem Substance Use among Incarcerated and Non-Incarcerated Youth

ABSTRACT

We explore the applicability of General Strain Theory to juvenile substance use. In doing so, we offer three advances over prior work. First, we put forward the concept of “problem substance use” to move beyond frequency-based and more medicalized conceptions of the phenomenon toward a conceptualization that views substance use as problematic when it disrupts social relationships and expectations and when it reflects a loss of self-control. Second, we employ a dataset that permits us to explore how strains originating in different life domains influence problem substance use as well as how negative emotions and personal coping resources mediate the relationship between strain and problem substance use. Third, we move beyond prior work by comparing how strain and strain mediators operate differently in schools, alternative learning centers, and juvenile correctional facilities. We find strong effects of the strain variables and strain mediators indicated by General Strain Theory and evidence that alternative learning centers and juvenile correction facilities are associated with especially high odds of problem substance use even when strain and other variables are controlled. We conclude by discussing the limitations of current work and the implications for next steps in the strain-substance use research literature.

Acknowledgments

A special thanks to Zachary Psick, MA who helped procure the data used in this project.

Notes

¹ This point echoes Hagan and McCarthy’s (1997) critique of the “school criminology” approach to studying juvenile delinquency, which relies on samples drawn from schools rather than the non-school environments that many young people find themselves. Hagan and McCarthy use this argument to highlight the importance of studying the involvement of homeless youth in delinquency and deviance.

² The DSM V, adopted in 2013, changed the definition for “substance use disorder” to include 11 diagnostic criteria that span the factors discussed above from physiological indications to inability to control impulses to consequences for social relations (Fisher et al. 2017).

³ Although the 2010 dataset is not the most contemporary, it was selected for analysis because the relatively more racially diverse Minneapolis metropolitan public schools are included in the 2010 survey, but excluded from the 2013 survey, making the 2010 MSS survey the most representative of the broader population.

⁴ The variable measuring “school attachment” is the only variable where the ALC and JCF groups do not have a significantly lower mean, in comparison to the school group.

⁵ Since the Games-Howell test does not rely on equal variances and sample sizes, it is often recommended over other approaches such as Tukey’s test. However, the Games-Howell test will often report looser confidence intervals compared to Tukey’s test. In our analysis, the both the Games-Howell and Tukey test provided similar conclusions, but we report the results for the Games-Howell analysis to be conservative.

⁶ The LR test compares the log likelihoods of the two models and tests whether this difference is statistically significant. If the difference is statistically significant, then the less restrictive model (the one with more variables) is said to fit the data significantly better than the more restrictive model. This statistic is distributed chi-squared with degrees of freedom equal to the difference in the number of degrees of freedom between the two models (i.e., the number of variables added to the model).

⁷ Winship and Mare (1984) note that direct comparisons of coefficients between nested logistic models are potentially misleading because coefficient estimates can change, not just because the effect of a variable increases or decreases as other variables are controlled, but because the variance of y is changing as new variables are added. After extensive testing, our models and subsequent conclusions do not seem significantly affected by this issue. Adding new independent variables to an equation has the potential to alter the variance of y and thus the remaining coefficients in the model, even if the new independent variables are uncorrelated with the original independent variables. Winship and Mare refer to this issue as “rescaling.” They suggest y-standardization as an approach to appropriately rescale the estimated coefficients to a constant variance for the latent dependent variable across equations. We tested our models using the advised y-standardization, but the substantive conclusions did not change. Our supplementary analyses are available upon request from the authors.

Additional information

Notes on contributors

Konrad Franco

Konrad Franco is a PhD student within the Department of Sociology at the University of California, Davis. His research interests pertain to the sociology of law and punishment, crime, deviance, and health. He specializes in the application of quantitative research methods and casual analysis.

Ryken Grattet

Ryken Grattet is Professor and Chair of Sociology at UC Davis who specializes in the sociology of law, governance, and social control. His current research focuses on correctional reforms in California that aim at downsizing prison and jail incarceration.