Do students benefit from longer school days? Regression discontinuity evidence from Florida's additional hour of literacy instruction

Florida administrative records 2005-06 to 2012-13, grades 3-10, linked to birth records for children born 1992-2002; 25-accountability-point bandwidth

Elementary schools in the lowest 100 statewide on reading, and their students

Economics of Education Review 67, 171-183; full text read from CALDER working paper 201-0818-1 (August 2018). Sharp regression discontinuity on the accountability score that determined which 100 Florida elementary schools were required to add an hour of literacy instruction in 2012-13, with a triangular kernel and a linear fit. THE CLEANEST TEST IN THE ARCHIVE OF ADDING TIME TO A LOW-PERFORMING SCHOOL, and it includes the placebo one wants: the pseudo-effect in the year BEFORE the mandate is 0.001. The authors are candid that identification degrades after year one - schools could choose to continue voluntarily, so 'there are several reasons why this analysis is less likely to yield unbiased estimates of the causal effects of the ESD program after the first year' and 'long-run effects are difficult to assess'.

An extra hour of daily literacy instruction produced 'effects of 0.05 standard deviations of improvement in reading test scores for program assignment in the first year' - the student-level estimate is 0.045 - and THEN IT DECAYED TO NOTHING: school-average reading rose 0.20 SD in 2012-13, 'which decreased to an improvement of 0.07 SD in 2013-14, and further declined to -0.004 SD in 2014-15'. Three years from a 0.20 SD school-level gain to zero. THE DISTRIBUTIONAL RESULT IS THE ONE THAT SHOULD CHANGE POLICY: the students the mandate targeted benefited LEAST. Children at the lowest achievement level gained 0.027 SD and not significantly, while levels 2-3 gained 0.077 SD and levels 4-5 gained 0.062 SD, both at p < 0.01. By socio-economic proxies the pattern is mixed - persistently free-lunch-eligible students gained 0.060 SD versus 0.030 for others, but children of married mothers gained 0.066 versus 0.042 for others, and maternal education barely mattered (0.044 versus 0.049).

Low. Treatment is assigned by a school-level accountability score with a sharp administrative cutoff, so neither families nor children select into it, and the pre-period placebo confirms the schools on either side were on the same trajectory. The heterogeneity results are within-treatment comparisons and therefore descriptive rather than identified.

Unreplicated as a design, but the pattern - a real first-year effect from a large dose increase that decays toward zero - recurs across the extended-time literature. Massachusetts added 300 hours a year for five years and found no achievement effect; a Dutch randomised trial added five weekly hours for three months and found nothing; Chile added roughly 27% more annual instructional time for 0.05-0.07 SD. That the smallest, most targeted dose here produced the clearest short-run effect is consistent with Lavy's concavity and with Rivkin and Schiman's finding that time returns least in the schools with the worst classroom environments.