Self-selection into social programs can lead to socially excessive or insufficient participation. We propose a framework to detect and address these inefficiencies, applying it to diabetes care, where individuals above a biomarker threshold receive nudges to seek care. Crossing the threshold increases healthcare utilization and improves health outcomes. However, those who opt into care—both compliers and beyond-compliers—are generally healthier and benefit less, indicating reverse selection on gains. Targeting based on observable characteristics reduces excessive participation and improves welfare, while outreach to individuals reluctant to seek care despite high potential benefits mitigates insufficient participation and may further raise welfare.
