The Cost of Grade Retention

99,729 observations on 73,621 individuals, school years 1996 and 1997

Uruguayan public non-vocational junior high (Liceo) students in grades 7-9

Review of Economics and Statistics 94(2), 596-606; full text read from the October 2010 accepted manuscript. Fuzzy RD where the running variable is the NUMBER OF FAILED SUBJECTS (0 to 9) with a sharp cutoff at more than three, quartic polynomial interacted with the above-cutoff dummy plus school, grade and year fixed effects. First stage 0.426 (SE 0.085). VERSION TRAP WORTH RECORDING: CEP Discussion Paper 878 (2008) is a DIFFERENT study using a 25-days-absence cutoff, and the file advertised on some repositories as the open-access copy of this article is that earlier version. The author reports a real weakness honestly - there is bunching at exactly three failed subjects, indicating some manipulation, and a bounding exercise that reassigns the excess mass shrinks the estimate by roughly 75% while leaving it negative and significant. Covariate balance also fails on 2 of 15 covariates. This paper has NO achievement outcomes at all; it is an attainment study.

Grade failure in junior high costs about 0.76 to 0.91 years of eventual schooling, with the IV estimate ranging from -0.760 to -1.773 across five specifications - a spec sensitivity the paper does not hide. The mechanism is immediate and permanent rather than cumulative: 'grade failers are on average 50 percentage points less likely to be in school after one year', the gap narrows to -0.280 at two years and is gone by three (+0.064), yet additional grades attended stays at about -0.8 to -0.96 at every horizon. The authors' own summary is the important one for this topic: 'early drop-out rather than the compounded effect of grade failure... explains why grade failers end up with lower educational attainment.' THE ATTAINMENT GAP DOES NOT FADE even though the enrolment gap does - a pattern that recurs across the archive, where test-score effects fade and attainment effects persist, here running in the harmful direction.

Low at the cutoff in principle, weakened in practice by the bunching. Students are compared with others who failed a similar number of subjects, so measured ability is balanced; but if teachers or students can push a marginal fourth failure back to three, the manipulation is selective on exactly the family and school characteristics that predict attainment. The 75% shrinkage under bounding is the honest measure of that exposure.

The finding that retention in the lower-secondary years durably reduces attainment replicates in Chicago (grade-8 dropout +8 points), Louisiana (grade-8 dropout +7.2 points), New York City (grades 7-8 credits -7.8, dropout +10.8 points) and Belgium (Cockx and colleagues: first-time grade-8 repeaters 14 points less likely to graduate). Four countries, four designs, one direction. Valbuena and colleagues' systematic review treats it as one of the field's more secure results.