Phi Adajar

High-stakes final exams play two roles in university admissions: they give students an incentive to exert effort, and they give universities information with which to screen applicants, shaping the equilibrium match. We study these forces using the staggered rollout of unconditional offers in the UK, which let some students matriculate without requirements on their age-18 exams (A-levels). Students who accept an unconditional offer score 0.14 SD below those who decline one, conditional on observables. However, using a two-way fixed effects specification, we find that exposure to unconditional offers has precisely zero impact on students' age-18 exam performance, indicating that the observed gap is driven by private information. Exposure to unconditional offers increases university matriculation by 1.1 percentage points, mostly at middle-tercile universities. A structural model that incorporates the value students place on certainty of admission shows that, in equilibrium, recipients' gains come at the expense of non-recipients. Relative to a ban, unconditional offers lower total student welfare by 0.7%, hurting the three highest-achieving quartiles of students. Post-qualification admissions, under which students apply after receiving their scores, raise welfare by 12.5% through better sorting. Our results highlight how the distributional effects of admissions policies depend on student information, informing current policy debates in the UK.

This paper presents evidence from a randomized controlled trial designed to test how to best identify and support gifted students from disadvantaged backgrounds. To do so, we offered access to an advanced online prealgebra course and an internet-enabled tablet to 7th-grade public-school students from Tamil Nadu, whom we identified as having high potential using several alternative strategies. Our primary finding is that we were able to successfully identify students who succeeded in challenging coursework originally designed as enrichment material for students with much stronger math preparation.

We randomly assigned identification methods, which allowed us to investigate their relative effectiveness. We find that the combination of existing administrative scores and engagement with an extract of the textbook distributed as a physical booklet best predicts engagement and success with the course material. We additionally cross-randomized treatments to evaluate program design (including group and individual tutoring, incentives for students, access to computer labs), and find that the simplest-to-scale version without tutoring led to the highest engagement.

Phi Adajar

Centralized school assignment with lottery tie-breaking generates quasi-experiments that identify school effects, but existing methods require applicants’ rank-order lists and lottery numbers, data that is not always available. I characterize which data from a match identify a propensity score under which offers are ignorable. When each school draws its own lottery, offers among applicants who reach a school are randomly assigned, so comparing applicants offered and rejected there requires only waitlists. Under a single lottery, this comparison is biased, and the order of each applicant’s rejections is also needed. Rank-order lists identify the score without lottery numbers, while the assignment and published aggregates do not in general. I show how to compute each score in finite markets, how waitlist order reveals which schools share a lottery, and how to handle priorities and screened schools. In simulations, all valid designs recover the true effect, and richer data yields more precise estimates.