The Effect of Retention Under Mississippi's Test-Based Promotion Policy

4,729 students in a single cohort (third grade 2014-15) followed to grade 6 in 2018-19

Mississippi third graders near the MKAS English language arts Level 2 threshold

Wheelock Educational Policy Center Working Paper 2023-1; no DOI. Fuzzy RD on the MKAS grade-3 reading gate, 20-point bandwidth, outcome measured on the MAAP - a different test from the assignment instrument, which is a genuine strength. THREE THINGS DOWNGRADE IT TO GRADE C. First, THE FIRST STAGE IS 5.8 PERCENTAGE POINTS (SE 0.016) against Florida's 28-37, because 'within this cohort half of the students who scored below the passing threshold passed on their second or third attempt'. Dividing a 0.067 SD reduced form by 0.058 is what produces the headline 1.15 SD, and Berne and colleagues name precisely this arithmetic as the exclusion-restriction failure mode - Mississippi is the most exposed study in the literature on their own ranking. Second, it is a SINGLE COHORT. Third, 'within-age comparisons of student test scores are not possible in Mississippi because scores on the state's standardized tests are comparable within grades over time but not across grades', so only the flattering same-grade comparison is available. Independence is recorded as developer-involved: the project 'was made possible by a grant from ExcelinEd', the advocacy organisation that promotes the Florida third-grade model nationally, and Winters co-authored the Florida anchor study.

The headline is that retention raises grade-6 ELA achievement by 1.153 SD (SE 0.657) - and that number should not be quoted without three qualifications printed beside it. It is significant at the 10% level only; it is a same-grade comparison, by data necessity; and it is the product of dividing a reduced-form 0.067 SD (SE 0.035) by a 5.8-point first stage. Mathematics is null in both forms (reduced form -0.025, instrumented -0.436). Absences and special-education placement are null. The subgroup claim that the effect is driven by Black and Hispanic students rests on REDUCED-FORM significance (0.106 and 0.379) while the corresponding instrumented estimates are not significant (2.842, SE 1.731; 2.042, SE 1.250). The right way to read this paper is as a reduced-form finding that being flagged under Mississippi's policy raised sixth-grade reading by roughly 0.07 SD.

Low at the cutoff, as in every RD here. The problem with this study is not selection on ability but instrument strength: with only 5.8% of the below-cutoff group actually retained, the estimate is dominated by whatever else the flag triggers, and the archive's own critique source shows those channels are real.

Unreplicated - one cohort, one state, one publication, no independent re-analysis. Directionally consistent with the Florida short-run reading gains, though at a magnitude no other study approaches. Berne, Jacob, Weiland and Strunk name it explicitly among the papers whose instrumented estimates 'might be biased upward', and their Michigan case (3.4-point first stage implying an implausible 1.3 SD) is the closest analogue to the arithmetic here.