Disentangling symptoms of externalizing disorders in children using multiple measures and informants.

Abstract

The trait impulsivity theory suggests that a single, highly heritable externalizing liability factor, expressed as temperamental trait impulsivity, represents the core vulnerability for externalizing disorders. The present study sought to test the application of latent factor models derived from this theory to a clinical sample of children. Participants were 474 German children (age 6-12 years, 81% male) with symptoms of attention-deficit/hyperactivity disorder and externalizing behavior problems participating in an ongoing multicenter intervention study. Using confirmatory factor analyses (CFA) and exploratory structural equation modeling (ESEM), we evaluated several factor models of externalizing spectrum disorders (unidimensional; first-order correlated factors; higher-order factor; fully symmetrical bifactor; bifactor S-1 model). Furthermore, we assessed our prevailing factor models for measurement invariance across raters (clinicians, parents, teachers) and assessment modes (interview, questionnaires). While both CFA and ESEM approaches provided valuable insights into the multidimensionality, ESEM solutions were generally superior since they showed a substantially better model fit and less biased factor loadings. Among the models tested, the bifactor S-1 CFA/ESEM models, with a general hyperactivity-impulsivity reference factor, displayed a statistically sound factor structure and allowed for straightforward interpretability. Furthermore, these models showed the same organization of factors and loading patterns, but not equivalent item thresholds across raters and assessment modes, highlighting cross-situational variability in child behavior. Our findings are consistent with the assumption of the trait impulsivity theory that a common trait, presented as hyperactivity-impulsivity symptoms, underlies all externalizing disorders.

Empirically based dimensions of externalizing symptoms in children and adolescents: a multitrait-multisource approach

“…As the bifactor model approach is based on the assumption that the g-factor causes all correlations between the s-factors, the reference facet, the authors found the IN domain of ADHD and ODD to remain a stable component, which is strongly associated with, but still distinct from, ADHD HI. Junghänel et al, (2020) and Thöne et al, (2021) reported similar findings regarding the association of these domains and the stability of the ADHD IN and the ODD factor in clinical samples of German children, thus strengthening the results of Burns et al, (2020b) with respect to the latent factor structure of ADHD and ODD. Regarding ADHD and CD, around 20% of children diagnosed with ADHD also met criteria for a CD diagnosis (Willcutt et al, 2012).…”

“…These very small or even negative factor loadings do not concern the ADHD dimension, but only the ODD/CD dimension in all samples. Such statistical anomalies related to traditional bifactor models are consistent with methodological concerns (Eid et al, 2017) and empirical studies demonstrating a variety of anomalous results associated with the application of traditional bifactor models to externalizing symptoms (Arias et al, 2018;Burns et al, 2020a;Rodenacker et al, 2018;Thöne et al, 2021). Anomalous loadings might be a result of the generally mistaken assumption of interchangeability of domains, which is a statistical prerequisite for accurately applying a bifactor model (Heinrich et al, 2021).…”

“…Specifying HI as the general rated their child's behavior in the same context (Burns et al, 2014). However, other research noted informant-specific variance across parent and teacher ratings, highlighting meaningful cross-situational variability in child behavior (Burns et al, 2020a;Thöne et al, 2021;Vitoratou et al, 2019). Studies including a self-report and systematically comparing this to teacher and parent reports are still rare.…”

“…Some of the major issues inherent in traditional bifactor models cannot be identified when solely examining model fit, but rather require the additional assessment of bifactorspecific indices. Such issues pertain to frequently observed anomalous factor loadings (Burns et al, 2020a;Eid et al, 2017;Junghänel et al, 2020;Rodenacker et al, 2018;Thöne et al, 2021) and vanishing s-factors (Burns et al, 2020a). Anomalous results are any results that are not in line with the general structure of the model, and include negative or non-significant factor variances and/or factor loadings > 1 (Eid et al, 2017).…”

A Developmental Perspective on Facets of Impulsivity and Brain Activity Correlates From Adolescence to Adulthood

“…In line with Dalley and Robbins ( 6 ), for the multidimensional concept of impulsivity, we distinguished trait impulsivity assessed via self-report, questionnaire-based methods from more objective neurocognitive measures. Thereby, trait impulsivity is defined as a predisposition for rapid, but often premature actions without appropriate foresight ( 6 ), reflecting preferences for impulsive responding and immediate rewards ( 10 , 11 ). It tends to explain impulsive behaviors such as impaired response inhibition as a consequence of dysfunctional processing of delayed rewards ( 3 ).…”