Patterns of agreement between children and parents in SDQ subscales: latent class and latent profile analyses

Abstract Background Multi-informant assessment is considered best practice in child and adolescent mental health research, yet agreement between child and parent reports is often only moderate. The present study examined patterns of agreement and discrepancy between children’s and parents’ ratings on the Strengths and Difficulties Questionnaire (SDQ) using both variable-centered and person-centered approaches in a nationally representative German sample. Method Data are used from Wave 2 of the German Health Interview and Examination Survey for Children and Adolescents (KiGGS), including 5,224 child–parent dyads of adolescents aged 11–17 years. To identify latent patterns of agreement and discrepancy, latent profile analyses (LPA) and latent class analyses (LCA) were conducted. Configural frequency analysis (CFA) was applied to test for patterns across the SDQ domains. Results Descriptive analyses showed moderate agreement across SDQ subscales, with children generally reporting higher levels of difficulties than their parents, particularly for the subscales Emotional Problems, Peer Problems, and Hyperactivity/Inattention. Agreement varied substantially across items, with higher concordance for more observable behaviors and lower agreement for internal or evaluative symptoms. The scale-level LPA identified six profiles differing in symptom severity and degree of child–parent agreement. Two profiles were characterized by systematic child-greater-than-parent discrepancies, whereas no clear parent-greater-than-child profile emerged. Item-level LCAs across all five SDQ subscales revealed both concordant and discordant classes, including distinct patterns in which either children or parents consistently reported more symptoms. Across domains, concordant low-symptom classes represented the largest groups, while concordant high-symptom classes were smaller but clearly identifiable. Cross-domain analyses with CFA showed only limited evidence for a generalized discrepancy type across all SDQ domains. However, a second-order LCA identified small but meaningful subgroups characterized by systematic child-greater-than-parent or parent-greater-than-child reporting patterns across multiple domains. Conclusions Although the results obtained using different analytical approaches did not fully converge and need to be replicated, the findings clearly suggest that the discrepancies between informants are structured and domain-specific, and cannot simply be explained by measurement error. Person-centered approaches provide important additional insights beyond conventional correlational analyses and may improve the interpretation of multi-informant SDQ assessments in both research and clinical contexts.