EFFECTS OF GRADE RETENTION POLICIES: A LITERATURE REVIEW OF EMPIRICAL STUDIES APPLYING CAUSAL INFERENCE

42 studies published 2001-2020 that credibly address the endogeneity of the retention decision, screened from 150+ candidates across ERIC, Web of Science, Scopus, EconLit, SocINDEX and Google Scholar

Kindergarten (2 studies), primary (27), secondary (9), mixed levels (4), across the US, Latin America, Europe and China

Journal of Economic Surveys 35(2), 408-451. A systematic review with an explicit inclusion rule that is the point of the paper: studies using ordinary least squares or other methods whose estimates are "highly unlikely to reveal causal implications" were excluded, which throws out the entire pre-2001 meta-analytic literature (Holmes 1989; Jimerson 2001) as a body that "did not address endogeneity or selection problems in a credible way". Design mix of the 42 survivors: RDD 7, IV 6, DiD 6, matching 4, hierarchical models 4, structural models 2, growth-curve 3, other 10. This is a vote-count review, not a pooled meta-analysis — it reports signs and study characteristics rather than a summary effect size, which is the right call given the heterogeneity but means it carries no precision estimate.

Two structural findings that reorganise the topic. FIRST, the same-age/same-grade distinction is the single largest source of disagreement in the literature, and the field is overwhelmingly on the flattering side of it: 34 of 42 studies use SAME-GRADE comparisons, which credit retention with an extra year of maturation and an extra year of schooling, while only 9 use same-age comparisons. The review's verdict is blunt — studies using same-age comparisons "identify effects which are considerably more negative", so "results associated to same-grade studies in this review could be read as the upper-bound of the true effect of grade retention." Schwerdt et al. (2017) is singled out as the one paper doing both. SECOND, the causal literature does NOT uniformly rehabilitate retention. Positive effects are short-run, institution-specific and concentrated in early primary; negative findings on later outcomes are common (Manacorda: failure causes dropout and 0.8 fewer years of schooling four to five years on; Cockx et al.: first-time grade-8 repeaters are 14pp less likely to graduate; Andrew: early-primary retention cuts the odds of high school completion by ~60% under sibling fixed effects; Fruehwirth et al.: kindergarten retention lowers achievement 9%, early retention 14%, late retention 4%, with the lowest-ability students hurt most). Against these, Eide & Showalter and Mahjoub find retention lowering dropout or raising junior-high graduation, and Cabrera-Hernández finds abolishing retention lowers dropout. The authors' overall conclusion is that "grade retention is unlikely to be an efficient public policy: its impact on student performance, when positive, is weak, and the negative consequences can easily outweigh its potential benefits."

Directly addressed and it is the review's organising principle. Retention is assigned on low measured achievement, which is substantially heritable, so any comparison of retained to promoted children without a design is confounded by exactly the trait the intervention targets. The review's inclusion rule exists to purge that confound, and the papers surviving it are RD, IV, DiD, sibling fixed effects and structural models. What remains is that most surviving designs identify only at a cutoff or on compliers — informative about marginal students, silent about the child a teacher is actually deciding about.

A review rather than a replicable estimate, so replication is not-applicable. Its central methodological claim — that measured retention effects get less negative as design quality and comparison group change — is itself a replication of Allen et al. (2009), which the authors cite in exactly that role. Two independent syntheses reaching the same structural conclusion by different routes is the strongest thing this literature has.