Data come from the first wave (1994–1995) of the National Longitudinal Study of Adolescent Health, a nationally representative school-based sample of 20,745 adolescents in grades 7–11. After excluding individuals without sampling weights (n=1,821) and individuals missing information on key variables (outcome variable, n=43; basic demographics, n=34), a primary analytic sample of 18,847 adolescents was derived. Missing values on other variables were handled using multiple imputation.
Two dichotomous outcome measures were created for adolescents who reported receiving any psychological or emotional counseling in a clinical setting (private doctor’s office, community health clinic, and/or hospital) or a school setting within the previous 12 months. Race/ethnicity is measured with five mutually exclusive categories (white, Hispanic, black, Asian American and Pacific Islander [AAPI], and other). Control measures for predisposing, enabling, and need characteristics are presented in [7
]. The measures of mental health need include several dichotomous indicators as well as two scales: an 18-item version of the Center for Epidemiologic Studies Depression Scale (CES-D; α=0.86) [8
], and a delinquent behavior index created by summing the response categories for involvement in 15 different behaviors such as stealing, getting into a physical fight, or selling drugs (α=0.84).
Comparison of Weighted Descriptive Statistics for Adolescents in Three High-Need Samples to Full Sample
Weighted logistic regressions are estimated for the three following high-need subsamples: (1) adolescents whose CES-D scores meet the cutoff established by Roberts et al. (22+ for males, 24+ for females) to identify likely cases of major depressive disorder and dysthymia (N=2,004) [9
], (2) adolescents who reported seriously thinking about committing suicide within the past 12 months (N=2,498), and (3) adolescents whose scores on the delinquent behavior index exceed one standard deviation above the mean (N=2,378). To assess whether language accounts for any observed racial/ethnic differences in service use across settings, two model specifications are used: (1) a model that does not control for language, and (2) a model that controls for whether a language other than English is spoken at home.