Estimating causal effects on psychological networks using item response theory

This methodological study introduces a novel approach for estimating causal effects on psychological networks by leveraging the mathematical equivalence between network models (e.g., the Ising model) and item response theory (IRT). The authors show that when direct estimation of large or complex networks is computationally infeasible, researchers can instead use explanatory IRT models to recover causal effects on both network state (overall level of a construct) and network strength (the degree of interconnection among items). Through simulations, a detailed case study of vocabulary learning from a content literacy intervention, and applications to 72 randomized controlled trials across education, health, and social science domains, the study demonstrates that interventions often affect network strength independently of mean-level outcomes. These findings reveal that many interventions may meaningfully restructure how skills, symptoms, or beliefs are interconnected—even when traditional analyses show weak or null average effects—offering a powerful new lens for understanding mechanisms of change in complex psychological systems.