The Medium-Run Effects of Florida’s Test-Based Promotion Policy

Five cohorts of Florida third graders, outcomes followed to five years after the retention decision

Florida third graders near the FCAT reading promotion cutoff

Education Finance and Policy 7(3), 305-330; full text read from the author-hosted published PDF. Fuzzy RD with instrumental variables on the same Florida policy as Schwerdt and colleagues. The paper's distinctive contributions are two. First, it reports results on the NRT, a norm-referenced test SEPARATE from the FCAT used for assignment, so at least part of the effect is verified on an independent instrument. Second, the authors defend the SAME-GRADE comparison on explicit policy grounds rather than adopting it by default: 'we follow the within-grade approach... if we think of schooling in the long-term context, the additional year of schooling itself is one of the most important interventions under consideration.' That is a coherent position and readers should know it is a position, because it is the choice that makes retention look best. Independence is recorded as developer-involved: both authors were Manhattan Institute fellows and the institute campaigned for test-based promotion; the precursor report was published by it.

Same-grade reading effects are large and decay slowly: about a third of a standard deviation in grade 4 (0.340, SE 0.022 and 0.292, SE 0.025 across two specifications), 0.236 and 0.168 in grade 5, and 0.223 and 0.190 in grade 6. Mathematics runs at about a fifth of a standard deviation. CRITICALLY, the effects survive on the independent norm-referenced test at roughly 55 to 70 percent of the FCAT magnitude - grade 4 NRT 0.268 and 0.190, grade 5 0.231 and 0.131 - which is the best single answer to the objection that the Florida results are practice on the assignment test. The authors also state the bundle problem in their own words: 'students subjected to the policy were retained in the third grade, were required to be assigned to a high-quality teacher during the retained year, and were required to attend summer school... though we cannot completely separate the differential effects of the treatments attached to the policy', although they provide evidence that teacher quality is not the primary driver.

Low at the cutoff. The larger interpretive risk is not genetic but definitional: a same-grade comparison credits retention with an extra year of maturation and an extra year of schooling, and Goos and colleagues quantify that choice as worth about 0.16 of a standard deviation across 84 design-gated studies.

The same-grade persistence replicates in Schwerdt, West and Winters, in Indiana and in Florida's English learners. What does not replicate is the implication commonly drawn from it: the same Florida data show the effect is largely gone on a same-age comparison within five years, and the Texas evidence finds long-run harm on graduation and earnings. Berne and colleagues name this paper among those whose instrumented estimates may be inflated by exclusion-restriction failure.