Remedial Education and Student Achievement: A Regression-Discontinuity Analysis

8,120 (grade 3, one-year) and 7,623 (grade 3, two-year); 5,018 and 4,552 for grade 6

Chicago Public Schools third and sixth graders near the ITBS reading promotion cutoff under the 1996 end-of-social-promotion policy

Review of Economics and Statistics 86(1), 226-244; full text read from NBER working paper 8918 (May 2002), so published figures may differ slightly. THE FOUNDING RD STUDY of this literature. Running variable is the August ITBS reading grade equivalent; cutoffs of 2.8 (grade 3), 5.3 (grade 6) and 7.0 (grade 8) grade equivalents, roughly the 20th national percentile. First-stage F of 236.0 (grade 3) and 149.7 (grade 6). The treatment is mandatory summer school PLUS retention, and the paper separates the two by exploiting the June-then-August exam sequence - a decomposition almost nobody else attempts. Outcomes are on the ITBS, the same instrument used to assign retention, and the authors flag the resulting incentive confound explicitly: one year out, retained students face high-stakes testing again while promoted students do not, and two years out the incentives reverse for sixth graders. Their own reading is that 'the two-year estimates provide the most accurate view of the retention effects for third graders' and that the grade-6 two-year estimates 'probably reflect an upper bound on any negative retention effects'. Comparison is SAME-AGE, on a common Rasch logit scale.

The age gradient appears here first and it is the durable finding of the whole literature. THIRD GRADE: retention raises reading by 0.174 logits (SE 0.050) and maths by 0.227 (0.044) after one year - 41% and 33% of an average annual learning gain - and both fall to statistical insignificance after two years (0.035 and 0.091). SIXTH GRADE: nothing positive at one year, and after two years reading is 0.154 logits LOWER (SE 0.065), which the authors describe as 27% of an annual learning gain. Same policy, same district, same instrument, opposite sign by grade. The fade in third grade is complete within two years even on the outcome measure most favourable to the treatment.

Low at the cutoff. Promotion is determined by a test score with a sharp threshold, so children just above and just below have the same expected ability and family background; the fuzzy design instruments actual retention with cutoff position. The residual problem is not genetic but instrumental - the outcome is the assignment test, and the incentive to perform on it differs between the arms in a way that flips direction between the one- and two-year horizons.

The early-grade short-run positive replicates in Florida, Indiana and Texas. The grade-6 result - that later retention is where the harm is - replicates in New York City (grades 7 and 8: dropout +10.8 percentage points, credits -7.8) and in Louisiana. But Eren, Depew and Barnes explicitly contradict the companion finding that early retention is harmless for dropout: in Louisiana, grade-4 retention raises dropout by 5.1 percentage points, and they tested and rejected the catch-up explanation, concluding that 'the effect of grade retention on dropping out may depend significantly on the setting and institutions of where it occurs'.