The Effects of Grade Retention on High School Outcomes: Evidence from New York City Schools

66,742 eligible students; 56,254 within the main bandwidth (33,468 in grades 3-5, 27,715 in grades 6-8); cohorts 2004-2012 followed to 2016

New York City public school students in grades 3-8 near the promotion cutoff, about 86% eligible for free or reduced-price lunch and about 90% Black or Hispanic

Journal of Research on Educational Effectiveness 18(1), 1-31; full text read from RAND working paper WR-1259-DEIES (July 2018), where the third author appears as Tiffany Tsai. Fuzzy RD on the end-of-summer-school NYC assessment score at the Level 1/Level 2 cutoff. THE FIRST STAGE IS 0.641 (SE 0.017), by a wide margin the strongest in this literature, which matters because it makes the instrumented estimates roughly 1.6 times the reduced form rather than 3 to 30 times, and therefore largely immune to the exclusion-restriction inflation Berne and colleagues identify. The authors pre-empt that critique directly: 'since our research design centers on comparing students close to the Level 1 cutoff, both treatment and control students in our analyses would have been eligible for and received the supplemental services in the following year.' Funded by IES, not by the NYC Department of Education.

The cleanest statement of the age gradient on attainment. DROPOUT: retention in grades 3 and 5 is null (-0.017, SE 0.026, and similar across specifications); retention in grades 7 and 8 raises dropout by 10.8 percentage points (SE 0.047), 'an increase of more than 30 percent of the control complier mean'. The timing is diagnostic - dropout by on-time grade 10 is null (+0.004), so 'retention in middle school does not cause students to quit high school in earlier grades, but it does make them less likely to persist after their promoted peers begin completing high school'. ATTAINMENT: highest grade attended -0.413 (SE 0.108) overall; credits earned -5.173 overall and -7.834 (SE 1.981) for grades 7 and 8, about 25% of the control complier mean. GRADUATION: on-time graduation collapses by 27.7 points, which is largely mechanical, but graduating WITHIN TWO YEARS is null overall (-0.036, SE 0.025) and null or positive for grades 3 and 5 - the harm is concentrated in grades 7 and 8 (-0.093, SE 0.052, about 21% of the complier mean, imprecise). ACHIEVEMENT: Regents exam scores rise about 0.2 SD (Algebra I +2.776, ELA +3.947) on a test independent of the assignment instrument - but retention also cuts the probability of TAKING those exams by 10-12 points, so the authors call it 'suggestive evidence at best' and note it 'could also reflect an age effect in that retained students are older when they take the tests'. Retention also cuts future retention by 19.4 points and raises special-education placement by about 4 points.

Low. A sharp cutoff on a district-administered assessment, with the strongest first stage in the literature, so compliers on either side are balanced on ability by construction. The Regents result carries a selection problem of its own - differential exam-taking - which the authors flag rather than bury.

The late-retention harm on attainment replicates Jacob and Lefgren's Chicago grade-8 dropout result, Eren and colleagues' Louisiana results, and Manacorda's Uruguayan finding of 0.76-0.91 fewer years of schooling from failure in grades 7-9. The early-grade null on attainment replicates Florida. This is the study that ties the two halves of the literature together on one population with one design.