Publications

Behaviorally-informed “nudges” are widely used in government outreach, but face criticism for being too modest to address poverty at scale. Indeed, when used to increase the take-up of social safety net programs, results are often mixed. In this study, we test adjustments to behavioral interventions that are hypothesized to increase their effectiveness. In four field experiments over two years (n = 542,804), we examine whether more proactive communication, variations in message framing, and more precise targeting increase take-up of critical anti-poverty benefits in California, above and beyond traditional light-touch approaches. Our interventions targeted extremely vulnerable low-income households, many of whom have no prior-year income and so were at risk of missing out on large cash transfers during the Covid-19 pandemic, namely the Child Tax Credit and economic stimulus payments. We find that light-touch interventions significantly and consistently increase take-up by 0.14 to 2 percentage points – a 150% to over 500% relative increase – regardless of message, sample, timing, or modality, resulting in over $4 million disbursed to low-income families. Importantly, we show that light-touch outreach was extremely cost-effective in this context: every $1 spent yielded from $50 to over $8,000 in benefits disbursed. However, higher-touch proactive outreach, refined messaging, and precise targeting yielded minimal additional benefits, with proactive outreach delivering a negative return on investment. These findings suggest an urgent policy need to rethink what outreach strategies – if any – can do better than a “nudge” if we are to close the take-up gap in anti-poverty programs.

Governments increasingly use randomized controlled trials (RCTs) to test innovations, yet we know little about how they incorporate results into policymaking. We study 30 US cities that ran 73 RCTs with a national nudge unit. Cities adopt a nudge treatment into their communications in 27% of the cases. We find that the strength of the evidence and key city features do not strongly predict adoption; instead, the largest predictor is whether the RCT was implemented using preexisting communication, as opposed to new communication. We identify organizational inertia as a leading explanation: changes to preexisting infrastructure are more naturally folded into subsequent processes.