Robust nudge-capacity computation under distribution shift

Develop a method for making historical-data-based nudge-capacity calculations robust to year-to-year shifts in applicant behavior and the market environment when scaling personalized recommendations in capacity-constrained school matching markets.

Background

The recommendation system computes program-level nudge capacities from historical applicant data and a match simulator, then uses those capacities to preserve acceptance rates for students with high offer likelihoods. The authors compare capacities computed from the 2024–25 and 2025–26 applicant populations and find substantial variation at some programs, including cases where the capacity falls to zero because the targeted acceptance rate cannot be maintained under the newer data.

These results show that historical calculations remain informative but do not provide exact guarantees when applicant behavior, program conditions, or other features of the matching environment change across years. The authors therefore identify robustness to distribution shift at scale as an unresolved methodological problem.

References

Together, these results imply that computing nudge capacities on historical data is informative even under the observed distribution shift, but that they do not induce exact guarantees. It is thus important to deploy conservative interventions, as we did---an open methodological question remains on how to best be robust to such shifts at scale.

— Personalized Recommendations Without Inducing Congestion: Mitigating Disparities in the NYC High School Match  (2610.08275 - Chiang et al., 6 Oct 2026) in Section 7, “Scale-up and robustness: What are the limits of student-side interventions?”, paragraph “Distribution shift”