Abstract
The debate following Chater & Loewenstein's (2022) critique of individual-level (i-frame) behavioural interventions has reinvigorated interest in system-level (s-frame) approaches such as regulation, taxation, and network-based seeding. However, the comparative risk profiles of these intervention families remain underexplored. This paper addresses a critical gap: while s-frame interventions may offer greater transformative potential, they are also more vulnerable to policy misspecification — errors in design, structural assumptions, or implementation that cause policies to fail.
Using an agent-based model, I formalise and test three policy-relevant risks: (A) structural misspecification (errors in network knowledge used for targeting), (B) heterogeneity risk (adverse or weakly responsive subpopulations), and (C) shock risk (temporary dampening versus backsliding of behaviour). For each dimension, I track adoption levels, diffusion speed, cost-efficiency, and downside risk.
Two key findings emerge. First, when structural assumptions are approximately correct, targeted s-frame seeding ignites cascades and proves markedly more transformative and cost-efficient than i-frame benchmarks. Second, the same reliance on social spillovers creates fragility: network mis-targeting and backsliding shocks substantially reduce s-frame performance, whereas i-frame designs deliver smaller but steadier gains and degrade more gracefully under adversity.
These results imply a portfolio approach to behavioural policy: invest in structural measurement to earn the right to scale s-frame interventions, while maintaining i-frame backstops and shock absorbers to keep systems in a safe-to-fail corridor. The framework provides actionable guidance for choosing and sequencing policy levers under uncertainty, shifting attention from average effects to the risk profile of alternative intervention families.