Multi-informant latent difference score models for comparing theories of developmental psychopathology: common cause versus dynamic mutualism
Mental health difficulties rarely occur in isolation. Symptoms of attention-deficit/hyperactivity disorder (ADHD), conduct problems, and internalising difficulties co-occur at rates substantially above chance across childhood and adolescence, yet the mechanisms generating this comorbidity remain debated. Two influential theoretical frameworks offer competing explanations. Dynamic mutualism proposes that distinct psychopathology dimensions reinforce one another through reciprocal developmental processes, such that symptoms in one domain directly amplify symptoms in another. The common-cause model proposes instead that comorbidity reflects a shared latent liability underlying multiple symptom dimensions simultaneously. Prior longitudinal tests of these frameworks have relied on single-informant measurement, limiting the psychometric quality of the latent constructs under investigation. The present study demonstrates how multi-informant structural equation modelling can be combined with latent difference score methodology to provide a more rigorous evaluation of competing developmental theories.
We analysed data for parent, youth, and teacher reported indicators of ADHD, conduct problems, and internalising symptoms obtained through the Adolescent Brain Cognitive Development (ABCD) Study (N = 7,558) at Waves one (10-11 years of age) and two (11-12 years of age). Using structural equation modelling we estimated two competing latent difference score models, one representing hypotheses of dynamic mutualism and the other representing hypotheses of common cause theory of development. Both models represented ADHD, conduct problems, and internalising symptoms as latent factors combining parent, youth, and teacher information, along with orthogonal respondent-general factors to account for covariance arising from shared method effects. Under the dynamic mutualism model mechanisms of change were specified for each domain and predicted by both self-feedback and cross-domain regressive effects, the hypothesis being that dynamic mutualism would be evidenced by positive cross-domain regressive effects. Under the common-cause model mechanisms of change were specified at the level of a single higher-order general psychopathology factor that represents effects that act upon all three domain factors.
Both models fitted the data well by conventional criteria (CFI ≥ .974, RMSEA = .027, SRMR ≤ .048). Information criteria provided divergent guidance: AIC favoured the mutualism model (ΔAIC = 40.5) while BIC, which more heavily penalises model complexity, favoured the more parsimonious common-cause model (ΔBIC = −35.7). Inspection of the substantive parameters differentiated the models more clearly. In the dynamic mutualism model, significant self-feedback was observed for conduct problems (standardised β = −.48, p = .002) and ADHD (β = −.28, p < .001), but five of six cross-domain coupling paths were non-significant. The one significant coupling effect indicated that higher initial ADHD predicted less subsequent change in internalising (β = −.15, p = .001) — a negative direction contrary to the positive reciprocal reinforcement that mutualism theory predicts. In the common-cause model, all three domains loaded significantly on the general psychopathology factor, and the proportional self-feedback parameter was significant and negative (β = −.33, p < .001), indicating that higher initial general psychopathology predicted less subsequent increase in the common factor. Individual differences in general-factor change were also significant (standardised residual variance = .89, p < .001).
These findings highlight value of the multi-informant latent difference score approach. The pattern of results is more consistent with a common-cause account of developmental comorbidity than with dynamic mutualism during this developmental period. The study demonstrates that multi-informant measurement enriches the reliability of latent constructs in developmental theory testing, and that embedding competing theoretical models within a unified latent difference score framework allows for hypothesis-driven evaluation of their relative empirical support. Implications for the measurement of developmental psychopathology and the design of future longitudinal studies will be discussed.