BriDGE the gap: Improving behavioral research by integrating DAGs and GAMs into experiments

Understanding the mechanisms through which behavioral interventions work remains a critical challenge in behavioral science. While randomized controlled trials (RCTs) provide reliable evidence for intervention efficacy, they are seldom designed to reveal the underlying causal pathways that drive observed outcomes. We introduce a comprehensive data-driven methodological protocol - BriDGE - that combines advanced causal inference techniques, such as directed acyclic graphs (DAGs), causal discovery algorithms, and generalized additive models (GAMs), to enhance mechanistic insights in behavioral applications. BriDGE modifies conventional experimental analysis with a stepwise approach including DAG-based hypothesis formulation, modeling of nonlinear relationships with GAMs, and detailed mediation analysis. Using bootstrapping and sensitivity checks, BriDGE ensures robust and reliable detection of both direct and indirect effects. We use a simulation study to validate BriDGE's ability to identify complex causal mechanisms, offering researchers a robust framework for deepening understanding of causal mechanisms and optimizing intervention design. To support adoption, we additionally provide practical guidance on mediator dimensionality, computational feasibility, and simulation-based power planning, including benchmarking templates implemented in the accompanying code. There are natural limitations of BriDGE - we discuss their implications when applied to public policy. We call for a greater integration of these methods in the toolkit of applied policy analysis to bridge the gap from "what works" to "why and how it works". We also release BriDGE, an open-source R package that implements the workflow to facilitate adoption and reproducibility.