Meta-analysis of Grade Retention Research: Implications for Practice in the 21st Century

20 studies, 21,229 participants; 1,161 retained and 1,651 comparison students for socio-emotional and behavioural outcomes; 175 achievement effect sizes

US school children, kindergarten through grade 8, from studies published 1990-1999

School Psychology Review 30(3), 420-437. THIS IS THE RECEIVED WISDOM, and it is recorded here as the position the causal literature had to argue against rather than as evidence for a verdict. Read at SECONDARY depth only: the article is not open access and could not be obtained, so every number below comes from sources read in full - the CRD/DARE critical abstract, Hughes and colleagues (2010), and Lorence (2006). Grade D is a design judgement, not a competence judgement: the pooled studies are matched-comparison and correlational work with no credible control for selection into retention, and a meta-analysis inherits the confounding of what it pools. The CRD abstract records three further defects: study validity was 'not formally assessed', there were 'no formal tests for assessing heterogeneity', and 'no formal test for publication bias' was reported. Effect sizes were averaged within study and each study weighted equally regardless of size. Lorence documents that Jimerson 'made no effort to differentiate which of the studies in his synthesis utilised better-matching or statistical controls' - only 9 of 18 met a reasonable control standard - that 47% of the negative findings came from a single study whose own authors concluded retention helped, and that several samples are double-counted. Citation hazard: this paper is cited as 'Jimerson 2001a' by Allen and colleagues and as '2001b' by Hughes and colleagues.

Achievement: the reported pooled effect of retention is -.39, and of 175 achievement effect sizes reported in 20 studies, 84 (52%) showed no statistically significant effect while of the remaining 91, 'only 9 favored the retained students, whereas 82 favored the comparison group'. Behaviour tells a different and much less quoted story: 'approximately 9% of the analyses favoured the comparison group, approximately 5% favoured the retained students, and 86% indicated no significant differences.' The socio-emotional case against retention, in other words, was mostly null even in the meta-analysis that established the case against retention. The number that would resolve the modern dispute - the same-age versus same-grade decomposition of the -.39 - is not obtainable and was NOT obtained here.

HIGH, and this is the whole problem. Children are retained because they are performing poorly, and school performance is substantially heritable; comparing them with promoted classmates without matching on prior achievement compares two different populations. Allen and colleagues later quantified the consequence directly: the studies at the low end of design quality average -0.30 and those at the medium and high end average +0.04. Jimerson's corpus sits mostly at the low end.

Superseded rather than refuted, and recorded as mixed for that reason. Two later syntheses that condition on design quality - Allen and colleagues (2009) and Goos, Pipa and Peixoto (2021) - both return effects indistinguishable from zero. The 1989 Holmes meta-analysis, which Jimerson's work extends, already contained the same signal in strata nobody quoted: -0.28 across 25 best-matched studies but -0.01 across the five matched on achievement AND socio-economic status. Note also that the -0.19 to -0.31 range routinely attributed to Holmes could not be verified in any primary text.