Social Correlates of Delinquency for Youth in Need of Mental Health Services: Examining the Scope Conditions of Criminological Theories

Abstract

While both traditional criminological inquiry and mental health research have identified internal and external constellations of risk factors associated with juvenile offending, interdisciplinary discourse has been limited. This paper takes a step in bridging the gap between criminological literature and work in the field of children’s mental health by evaluating the combined effects of social and mental health predictors on juvenile delinquency in a sample of youth with diagnosed clinical disorders. Results of multivariate analyses indicate that both traditional social risk factors as well as indicators of the nature and severity of youths’ mental health disorders contribute to delinquency. Moreover, the influence of one well-established risk factor, self-control, on delinquency is moderated by the presence of oppositional defiant disorder. The results of this study suggest that researchers and practitioners should consider the cumulative influence of social risk factors and psychological impairment in the etiology of delinquency.

Keywords:

• delinquency
• mental health
• system of care
• correlates of delinquency

Acknowledgments

This study was funded by Contracts #280-03-1603 and #280-03-1604 to Macro International Inc. (now ICF Macro, an ICF International company) from the Center for Mental Health Services at the Substance Abuse and Mental Health Services Administration, US Department of Health and Human Services. We are grateful to LuAnn McCormick, Rose Green, Phyllis Gyamfi, and the anonymous referees for comments on earlier drafts of this paper.

Notes

1. Due to research restrictions, self-report information was not collected on youth under the age of 11. This analysis includes youth aged 11–22 at the time of the baseline interview. SAMHSA allows participants to receive services up to the age of 21 (and in one case 22). Only 44 respondents were over the age of 18 at the time of the baseline interview. We report results for the entire SAMSHA population, even though the older segment extends beyond common understandings of “youth.” All subsequent analyses were also conducted without these 44 cases yielding no difference in substantive findings. These additional analyses are available upon request.

2. Although potentially masking variation in the frequency of offending, the variety scale’s distributional properties are preferred for modeling event-count data. We also estimated the models using an offending frequency scale, computed as the cumulative sum of the frequency in which a respondent engaged in each of the 21 delinquent acts (Appendix 1). The results of these models were consistent with those reported in Table 5 and indicate that these findings are robust across various operationalizations of delinquency.

3. Items are drawn from the Youth Information Questionnaire.

4. Items are drawn from the Behavioral and Emotional Rating Scale (2nd ed.).

5. Items are drawn from the Behavioral and Emotional Rating Scale (2nd ed.).

6. Items are drawn from the Behavioral and Emotional Rating Scale (2nd ed.).

7. Items are drawn from the Youth Information Questionnaire.

8. Item drawn from the Child Behavior Checklist.

9. The use of both DSM IV and ICD likely reflects the preference of the mental health professional making the initial diagnosis. As this information is collected from medical records at the intake interview, there is no information currently available to indicate why some youth have DSM IV diagnoses while others have ICD diagnoses.

10. We estimated model fits by multiplying the difference between likelihood functions of the full and restricted models by −2. The quotient approximates a chi-square distribution with degrees of freedom equal to the difference in parameters between the two models. A significant chi-square value indicates the full model is preferred over the restricted model.

11. A potential issue with comparing across multiple models is the increased risk of committing type 1 error. To adjust for this potential bias, we utilized the Bonferroni correction and divided the nominal probability of committing type one error (p < 0.05) by the number of tests. We therefore failed to reject the null hypothesis for coefficients in the interaction models with probabilities greater than 0.016 (0.05/3).