Transporting behavioral insights to low-income households: A field experiment on energy efficiency investments

Established behavioral insights may not generalize when applied to under-represented target populations. Iterative approaches that identify such external validity failures to guide policy deployment at scale and to inform subsequent design refinements gain importance. We implement this approach in a field experiment involving 1800 low-income households in one of the largest energy efficiency assistance programs. Despite extensive co-designed vetting, most behavioral insights proved ineffective or inefficient. Notably, loss-framed interventions backfired by reducing take-up. We rank insights by their Marginal Value of Public Funds (MVPF), identifying ‘unexpected winners’ that outperform alternatives rated as promising by experts and laypeople. Our analysis suggests that factors specific to low-income households, such as scarcity, fundamentally alter behavioral responses, necessitating guided deployment to optimize future public assistance design.