Helping or Hurting: The Effects of Retention in the Third Grade on Student Outcomes

8,376 students at grade 4 falling to 3,144 by grade 7, within a 25-point bandwidth; cohorts from 2011-12

Indiana third graders at the IREAD-3 reading cutoff of 446

Educational Evaluation and Policy Analysis; full text read from EdWorkingPaper 22-688 (December 2022), which carries the earlier title 'Holding Back to Move Forward'. Fuzzy RD on Indiana's third-grade reading gate, bandwidth chosen as the minimum of sixteen cross-validation-optimal bandwidths. THE FEATURE THAT MAKES THIS SOURCE VALUABLE is that the outcome test is the ISTEP+, a DIFFERENT instrument from the IREAD-3 used to assign retention - one of only three studies in the literature that breaks the shared-instrument problem. It also reports both same-grade and same-age estimates and both behavioural and achievement outcomes.

Same-grade effects are large and persistent: ELA +0.528 SD in grade 4, +0.274 in grade 5, +0.315 in grade 6 and +0.245 in grade 7; maths +0.522, +0.378, +0.188 and a non-significant +0.154. Same-age effects are 'somewhat smaller and statistically less certain' but are positive and significant for ELA up to five years out - a worked example gives about 0.18 SD three years on - while maths is NULL throughout on the same-age comparison. That split is the fadeout pattern in miniature: what survives is reading, on the outcome closest to the gate, and only when the child is compared with younger classmates or with a generous same-age model. Discipline and absence are null under BOTH framings at every grade, which the authors note is 'at odds with Özek (2015)'.

Low. Sharp test-score cutoff, and an outcome test independent of the assignment test, which removes the incentive confound that troubles the Chicago and Florida estimates.

Replicates the short-run positive achievement effect of early retention found in Chicago, Florida and Florida's English-learner population, and does so on an independent instrument - which matters, because the objection that these studies simply measure practice on the assignment test does not apply here. It fails to replicate Özek's Florida discipline harms. Its same-age ELA persistence is the strongest single piece of evidence against complete fadeout, and it is subject-specific.