Effects of Abolishing Grade Retention on Educational Achievement, Mental Health and Adult Crime

4,463 individuals (CFPS 2020) for attainment; crime panel aggregated to 313 province-birth-cohort cells covering ~202,000 property and ~213,000 violent criminal convictions (China Judgements Online, 2014-2018)

Chinese adults surveyed in CFPS 2020 who passed through compulsory schooling as provinces successively banned repetition from the early 2000s; crime sample restricted to offenders aged 18-24

Staggered provincial adoption of a ban on grade repetition during compulsory education, used as an exogenous shock in a flexible difference-in-differences model with cohort-varying exposure (grades 1-3, 4-6, 7-9 at reform), province, birth-year, birth-month and region-by-birth-year fixed effects. This is the mirror image of the US test-based-promotion RD literature: it identifies the effect of REMOVING retention rather than of applying it, on the whole affected population rather than at a test-score cutoff. The authors run the Goodman-Bacon (2021) decomposition and re-estimate with the Callaway-Sant'Anna (2021) estimator, which matters because naive TWFE is biased under staggered adoption — the robust estimate is roughly half the TWFE one and still significant. Two real limits: it is an SSRN preprint that has not been peer reviewed, and the crime analysis is ECOLOGICAL (province-cohort crime rates, not individual convictions linked to individual retention), with only 6.8% of the crime sample exposed to the reform in grades 1-3, so those estimates are marginal (p<0.10) and rest on 313 cells.

Abolishing retention made attainment WORSE, and the damage is concentrated in the earliest grades. Exposure to automatic promotion in grades 1-3 cut the probability of completing junior high school by 6.42 percentage points (p<0.01) against a 94.8% completion base — 6.77% of the mean; exposure in grades 4-6 cut it by 4.42pp and in grades 7-9 by 2.79pp, a monotone gradient the authors read as early primary being the formative window. Under the Callaway-Sant'Anna estimator the grades 1-3 effect halves to -3.15pp but stays significant, so the sign survives the staggered-adoption correction while the magnitude does not. Effects are a rural phenomenon: -0.170 in the rural subsample versus -0.009 and insignificant in the urban one. Two prominent NULLS run against the standard case for social promotion: depression (CES-D) is unmoved (0.475, n.s.) — the psychological-harm argument for abolishing retention does not show up on the outcome it predicts — and college enrolment is unmoved, which the authors attribute correctly to the marginal student being nowhere near the college margin. Adult crime rises: property-crime rate +4.858 per 10,000 for grades 1-3 exposure (t=1.92, p<0.10; 32.7% of the 14.841 mean) and rape +0.175 (p<0.05; 53% of the 0.328 mean); affray +1.403 is imprecise (t=1.07) and aggravated assault is small and null.

Low for the identified contrast. Provincial reform timing is not chosen by families, so exposure is not an expression of child or parental traits; the DiD nets out province and cohort levels. The residual worry is the usual one for policy DiD — provinces that reformed earlier may have been on different educational trajectories — which the event-study and leave-one-province-out sensitivity checks address rather than eliminate.

Directionally consistent with Koppensteiner's Brazilian evidence that automatic promotion lowered fourth-grade maths scores, and directionally opposed to Cabrera-Hernández's Mexican finding that removing the retention threat for third graders left test scores unchanged while REDUCING dropout one year on. Three developing-country abolition studies, three partly different answers, which is why this is recorded as mixed rather than replicated. No independent replication of the Chinese reform exists, and the paper is a preprint.