Triangulating Child Abuse Reports across Sources: Implications for Measurement

Background and Purpose. Accurate measurement of child physical abuse remains a longstanding methodological challenge in social work research. Discrepancies across data sources, including administrative records, parent reports, and self-reports, are well-documented, yet researchers frequently rely on a single source without clear empirical guidance on how to integrate multiple reporters. Triangulation has been proposed as a potential solution, but concrete quantitative frameworks for operationalizing this concept remain limited. This study systematically compares five analytic approaches to triangulating child abuse data and examines their relative utility for capturing a common underlying construct and predicting adult depression.

Methods. Data were drawn from the Lehigh Longitudinal Study, a prospective study of child maltreatment spanning over four decades. The analytic sample included 303 participants (mean age = 46 years) originally recruited in the 1970s from economically diverse families in eastern Pennsylvania, over 60% of whom lived below the federal poverty line at baseline. Physical abuse was assessed across six reporter sources: two prospective parent reports collected during early and middle childhood, three retrospective self-reports gathered during adolescence and across two adult waves, and administrative child welfare records. Adult depression was measured using the validated PHQ-8. Five analytic approaches were applied: network analysis, ordinary least squares regression (OLS), structural equation modeling (SEM), latent profile analysis (LPA), and cumulative index regression.

Results Across analytic approaches, adult retrospective self-reports of childhood physical abuse showed the strongest intercorrelations and the highest loadings on the latent abuse construct in SEM (β = 0.75–0.78). The SEM latent construct was also significantly associated with depression (β = 0.22, p < .01). In OLS models, only adult self-report (wave 2) significantly predicted adult depression (β = 4.41, p < .01). Network analysis indicated stronger clustering by reporter types, with self-reports and administrative records forming a distinct community cluster with depression, while parent reports formed a separate cluster with weaker associations with depression.

Conclusions and Implications. SEM and network analysis provided the most informative approaches for integrating multi-reporter child physical abuse data. Retrospective self-reports demonstrated strong intercorrelations and meaningful alignment with administrative records, supporting their value as a measurement source despite concerns about recall bias. In contrast, prospective parent reports showed weaker, non-significant associations with both the latent abuse construct and adult depression, suggesting that reliance on parent-reported data alone may limit predictive validity. Simpler aggregation methods, such as cumulative index models, were less effective in capturing the complexity of multi-source data.

These findings highlight the importance of how methodological choices shape the interpretation of multi-source maltreatment data. Decisions about which reporters to include and how to integrate them can influence study conclusions and their implications for research and practice. Future research using longitudinal designs and causal approaches is needed to clarify the mechanisms linking reporting sources to adult outcomes.