Hold Back To Move Forward? Early Grade Retention And Student Misbehavior

1,298,460 students in the full sample; regression-discontinuity samples of 43,793 (5-point bandwidth) and 178,248 (20-point); long-term subsamples 21,712 and 87,924

Florida third graders near the FCAT reading Level 2 threshold, seven cohorts 2003-04 to 2009-10

Education Finance and Policy 10(3), 350-377; full text read from CALDER working paper 100 (updated September 2014). Note the DOI: 10.1162/EDFP_a_00166, not _00174, which is a different paper on teacher transfer incentives. Fuzzy RD on the same Florida cutoff as Schwerdt and colleagues, first stage 0.313-0.322, but the outcomes here are disciplinary records rather than tests. The paper reports both same-age and same-grade framings and they agree, which is rare enough in this literature to be worth noting. It also tests a mechanism most of the field ignores: instrumenting relative age by birth month.

Third-grade retention makes behaviour worse for exactly two years and then stops. Any disciplinary incident rises 3.1 to 6.4 percentage points in the first year, '30 to 50 percent of the control mean of 0.107', driven by out-of-school suspension (+3.7 to +6.2 points) with in-school suspension null. At two years the incident effect is +4.8 to +5.5 points, '40 to 50 percent of the control mean of 0.132', now driven by in-school suspension. Years four and five are all null. The same-grade framing agrees exactly - grade 4 +4.5 to +5.0, grade 5 +5.5 to +6.5, grades 7 and 8 null - and the apparent negative year-three estimates turn out to be an artefact of promoted students entering middle school first, vanishing once peer-school controls are added (-0.092 becomes +0.012, not significant). THE MECHANISM MATTERS FOR INTERPRETATION: instrumenting relative age gives 0.0008 per month (p < 0.01), so a twelve-month increase in relative age raises the probability of a disciplinary incident by one percentage point on its own - between a fifth and a third of the first-year effect may be age-in-grade rather than the stigma of being held back. Effects are concentrated in economically disadvantaged students (a 7 to 9 percent increase, versus no significant effect for more affluent students), though the author notes the subgroup effects are not statistically different from each other.

Low. Sharp cutoff on a state test, administrative discipline outcomes, and the relative-age instrument provides an independent check on the mechanism. Externalising behaviour is heritable and correlates with low achievement, which is precisely why the pre-RD literature found retained children behaving worse; identifying at the cutoff removes that.

Contradicted by two later studies with the same design in other states. New York City finds nothing surviving multiple-comparison correction except a beneficial attendance effect, and Indiana finds discipline and absence null under both same-grade and same-age framings, with Hwang and Koedel noting explicitly that this is 'at odds with Özek (2015)'. Goos and colleagues' meta-analysis puts psychosocial outcomes at g = +0.08 (95% CI -0.01 to 0.17). The most likely reconciliation is that the effect is real, transient, and state-specific, and that a substantial share of it is relative age rather than retention.