Accurate Identification of Boys at-Risk for Serious Aggression and Violence – Challenges in Classification Methods

Introduction: In the context of decreasing budgets for prevention and increasing numbers of children growing up in multiple risk families and communities, selective or targeted interventions have become a preferred choice. However, the public health impact of targeted interventions hinges on the accurate identification of individuals with whom to intervene. Classification attempts typically involve two types of misclassification: False negatives, i.e., individuals at-risk do not receive the intervention and false positives, i.e., individuals not at-risk receive the intervention. This study will present findings on variation in classification accuracy due to the timing of the assessment (i.e., grades 1 to 5), the informant (i.e., parent versus teacher), as well as the race/ethnicity of the student (i.e., African-American [AA] versus Non African-American[NAA]).

Methods: The proposed study will include 503 males from the Pittsburgh Youth Study who were originally recruited from 1^st graders in the Pittsburgh Public Schools. In this sample, 55.7% identified as African-American and 45.5% engaged in serious violent behavior between 6^th grade and two years past high school using self-report and official arrest data. Parent (CBCL) and teacher (TRF) ratings of externalizing behavior are used as predictors. Receiver Operating Characteristics analyses are carried out and in addition to standard accuracy statistics (e.g., sensitivity, and specificity) the generalized the kappa coefficient will be presented.

Results: Preliminary results yielded important differences: Using teacher ratings, AUC for AA ranged from .596 to .674 peaking in 1^st grade. Sensitivity (SEN) ranged from .625 -.847 peaking in 3^rd grade. Specificity (SPEC) ranged from .287 to .605 peaking in 2^nd grade. AUC for NAA ranged from .697-.763 peaking in 3^rd grade. SEN ranged from .250 to .328 peaking in 3^rd grade. SPEC ranged from .911 to .929 peaking in 2^nd grade. Using parent ratings, AUC for AA ranged from .578 to .608 peaking in 2^nd grade. SEN ranged from .755 to .807 peaking in 4^th grade. SPEC ranged from .231 to .315 peaking in 5^th grade. AUC for NAA ranged from .623 to .668 peaking in 3^rd grade. SEN ranged from .086 to .229 peaking in 3^rd grade. SPEC ranged from .938 to .974 peaking in 1^st grade.

Conclusions: When implementing targeted program, it is imperative to employ accurate screening instruments. Preliminary analyses show important ethnic differences, i.e., higher levels of sensitivity for AA compared to NAA. In addition teacher ratings tend to be more accurate as compared to parent ratings with some variation around the timing of the screening.