Does Gifted Education Work? For Which Students?

Three RD samples in one large district (anonymised in the paper as "the District"; Broward County FL) — Plan A n = 2,679, Plan B n = 4,472, high achievers n = 4,144, plus a school-level design on 255 grade-4 classes at 94 schools

Grade 4 students entering separate gifted classrooms in a large, diverse US district, followed into grade 5. Outcomes are state standardized reading, math, writing and (grade 5) science, standardized within district-year-grade.

Three separate RDs on the District's three admission pathways. (1) Plan A — non-disadvantaged students at the IQ 130 threshold. (2) Plan B — free/reduced-lunch and English-learner students at the lower IQ 116 threshold. (3) High achievers — non-gifted students who fill the remaining ~20 seats in the gifted classroom, ranked by the previous year's state test scores within their school-grade cohort; RD is on rank, not IQ. Each estimates a LATE at its own admission margin. The Plan A/Plan B nulls are strengthened by differenced and donut-hole specifications showing near-identical gains across the whole IQ range (i.e. not just a marginal-student artefact), and by a school-level design confirming the high-achiever result and showing no spillover onto classmates ranked 25-44. NBER working paper — no journal DOI; the DOI recorded is the NBER series DOI. Funded by IES grant R305E110019.

Two findings, and the archive previously carried only one. (1) The gifted LABEL is inert: at both IQ thresholds, admission to gifted education — label, separate classroom, abler classmates — produces essentially no reading or math gain, with point estimates mostly negative at the IQ 130 margin, and this holds infra-marginally too. The one exception is a real +0.12 SD (≈0.3 SD TOT) on WRITING for disadvantaged Plan B students. (2) The same classroom works for students selected on past achievement rather than IQ: +0.09 SD reduced form, ~0.3 SD TOT on grade-4 reading and math, rising to 0.4-0.6 SD for low-income and Black/Hispanic high achievers, with the math gain and a new science gain persisting into grade 5 and no harm to the classmates left behind. The authors' own reading: separate gifted classrooms are more effective for students targeted on past achievement than on cognitive ability, and a relative (per-school) admission rule creates a pathway for able children in low-performing schools.

The raw gifted-versus-non-gifted gap is selection on heritable prior ability; the discontinuities isolate causal effects at each margin, which is the whole point of the design. Note the result that most resists a hereditarian reading is the contrast between the IQ margins (null) and the achievement-rank margin (positive) — the same classroom, the same peers, different selection criterion.

The Plan A/Plan B null is matched by Bui, Craig & Imberman (2011/2014) in a different district, and by exam-school studies (Abdulkadiroglu et al. 2011/2014; Dobbie & Fryer 2011). The high-achiever positive is corroborated within this paper by an independent school-level design, and by the same authors' 2016 AER universal-screening result.