Internalizing Problems as a Predictor of Change in Externalizing Problems in At-Risk Youth

Abstract

Intervention and prevention programs for children with externalizing problems frequently involve children with co-occurring internalizing problems. Little is known about how these co-occurring internalizing problems predict outcomes, particularly for programs involving cognitive-behavioral strategies. The current study examined how a set of child-related risk factors (including anxiety and depressive symptoms) predicted change in parent- and teacher-reported externalizing problems following a school-based preventative intervention for children at risk for externalizing problems. Participants included 112 preadolescent children (ages 9–12) who participated in a study designed to evaluate the efficacy of the Coping Power Program (Lochman & Wells, 2004). Participants included 81 boys (68%) who were primarily African American (69%) or Caucasian (30%). Regression analyses were conducted to examine predictors of change in parent- and teacher-reported externalizing problems on the Behavior Assessment System for Children (Reynolds & Kamphaus, 1992). Results indicated that greater child depression symptoms (as reported by parent or teacher) were associated with a larger reduction in externalizing behavior problems based on parent or teacher report. This effect was found in both the parent and teacher models and held after controlling for a number of child-oriented baseline variables including baseline aggression. Future research studies should examine whether co-occurring symptoms of depression relate to enhanced changes in externalizing problems following intervention for externalizing problems, particularly when cognitive-behavioral interventions are utilized. In addition, it will be important for studies to examine such effects relative to a control group and/or alternative treatment conditions and to further explore possible mechanisms of change.

Notes

Note: Change Scores = T1 – T2. Gender: 1 = male, 0 = female. Race: 1 = African-American, 0 = Other. Repeat Grade: Child repeated grade = 1, child did not repeat grade = 0. SES = socioeconomic status; BASC = Behavior Assessment System for Children.

*p < .05.

Note: Change scores = T1 – T2. Gender: 1 = male, 0 = female. Race: 1 = African-American, 0 = Other. Repeat Grade: Child repeated grade = 1, child did not repeat grade = 0. Positive change scores reflect improvement in the Externalizing Problems Composite, whereas negative change scores reflect an increase in the Externalizing Problems Composite. BASC = Behavior Assessment System for Children; SES = socioeconomic status.

*Significant coefficients at p < .05.

¹Although we originally planned to include Time 1 BASC Externalizing Problems as a predictor, we encountered problems with multicollinearity. When regression models were ran with pretreatment BASC Externalizing Problems and BASC Depression as predictors, BASC Depression had elevated multicollinearity statistics that are considered problematic based on expert recommendations (Allison, 1999; recommendation of requiring all predictors to have a VIF ≤ 2.5 and tolerance > .40). These statistics were elevated in both the parent and teacher models. Given this multicollinearity issue, we chose to pursue gain scores rather than repeated measures analysis of covariance, because repeated measures analysis of covariance would require that pretreatment BASC Externalizing Problems be included in the model. Using gain scores allowed us to use a substitute measure for pretreatment externalizing problems as a predictor of the average change in BASC Externalizing Problems. It should be noted that the aggression screener measure was significantly associated with BASC Externalizing Problems (.44 for teachers, .28 for parents) but not significantly associated or as strongly associated with parent or teacher depression (.09 for parents and .27 for teachers; this latter correlation is significant). When models were ran using the gain score approach with the screener measure for Time 1 aggression, multicollinearity diagnostic statistics were not in a problematic range.