Seminar Measuring and predicting long-run impacts of vaccination incentives

Abstract

When researchers test interventions for behavior change, they typically only have access to short-run data on behavior or survey data, like stated intentions, which they use as proxies or surrogates to predict long-run behavior change. We overcome typical data limitations using data from two large-scale field experiments (N=5,324 and N=8,286) with a unique combination of data on vaccination intentions, clicks on a link to make an appointment, and long-run administrative data containing each vaccination date up to two years after the incentives. We first document that guaranteed incentives of \$20 increase booster dose uptake by 13 percentage points in the short run and 9 in the long run. Guaranteed incentives are more effective at accelerating booster doses than lottery-based, prosocial, or participant-chosen incentives and also accelerate first-dose vaccination. Second, we study whether commonly used surrogates highly correlated with long-run uptake, including intentions, link clicks, and short-run uptake actually predict long-run impacts. Relying on surrogacy methods (Athey et al., 2025), we find that using intentions and link clicks as surrogates in our context would mislead decision-makers about long-run impacts. Using vaccination uptake within 30 days as a surrogate, however, predicts long-run behavior change. Our findings highlight the potential of incentives for long-run health behavior change and the risks of using common behavioral surrogates.