Assessing the Psychometric Properties of the Child Behavior Checklist in the ABCD Study

The Child Behavior Checklist (CBCL) is widely used in developmental research to assess core dimensions of psychopathology and has been used as the main such measure in the landmark Adolescent Brain Cognitive Development (ABCD) study, which comprises over 11,000 individuals receiving comprehensive, longitudinal assessments during the transition from childhood to adolescence. Despite rapid and widespread use of the CBCL in this sample, the hierarchical structure, subscale validity, item-level measurement properties, and measurement precision of the instrument in this sample remain unclear. Here, we use a multi- method framework integrating confirmatory factor analysis (CFA), Bayesian CFA (B-CFA), factor mixture modelling (FMM), item response theory (IRT), and exploratory structural equation modelling (ESEM) in an analysis of baseline ABCD data (N=11,861; 6,188 male; age range 8–11 years) to identify fundamental psychometric limitations of the CBCL. Specifically, we show that neither bifactor nor unidimensional models adequately capture the CBCL’s structure, as indexed by measures of poor global and local model fit and substandard amounts of explained variance. Subscale-level analyses further reveal insufficient construct validity for seven of eight subscales. IRT revealed poor measurement precision across the latent trait continuum, particularly at lower severity levels; a failure of item response options to discriminate between different levels of the latent traits; and a high item overlap in differentiating individual differences in the latent traits. IRT-based refinement of subscale item content failed to resolve these issues. Exploratory methods yielded unstable factors with weak discriminant validity. These findings challenge the CBCL’s utility as a dimensional measure of psychopathology in the ABCD study. We caution against uncritical use of CBCL sum scores or latent constructs within this dataset and advocate for measures explicitly validated for the assessment of dimensional psychopathology in population-scale research.