Ill Communication: Technology, distraction & student performance

130,482 student-level observations across 91 schools, 2001-2011; 91 completed surveys = 21% of target high schools in the four cities

Students at the end of compulsory education (age 16, GCSE) in state high schools in four English cities — Birmingham, London, Leicester and Manchester. Sampled schools were more disadvantaged than average (more minority and free-school-meal pupils) but had higher value-added than other schools in the same cities.

Identification is a difference-in-differences on the STAGGERED, SCHOOL-CHOSEN adoption of mobile phone bans. The authors surveyed head teachers about when (if ever) their school restricted phone use on premises and linked that to England's National Pupil Database; schools that had not yet banned (or never banned) serve as controls for schools that had. Specifications add prior achievement at age 11, student characteristics, other contemporaneous policy changes, and prior peer achievement. The outcome is the nationally standardized sum of GCSE points at age 16 — a genuine high-stakes external exam, not a researcher instrument, which is the paper's real strength. THE WEAKNESS IS THE ONE THE AUTHORS CANNOT DESIGN AWAY: schools choose when to ban. If a head teacher bans phones as part of a wider disciplinary tightening, or in response to a bad year, the estimate absorbs that. The authors probe this (they check for compositional change in intake, control for other stated policy changes, and present an event study by years of exposure) and find no sorting on observables, but adoption timing remains endogenous, which is exactly why this does not reach grade B: METHODOLOGY.md reserves B for DiD with a CLEAN shock. Two further limits: the 21% survey response rate makes the sample a selected set of schools willing to answer a researcher's survey, and by 2011 all but one sampled school had banned, so late-period identification is thin. Note also the significance level actually achieved — the headline pooled estimate is 0.0641 with SE 0.0373, i.e. p is about .09, significant at 10% and NOT at 5%. The paper's robust result is the INTERACTION with prior achievement, not the main effect. Independent academic work (Louisiana State University, University of Texas at Austin); no funder with a stake in phones or phone bans. Displacement was NOT measured: nothing records what students did with the time the phone stopped taking, so the mechanism (attention in class? attendance? sleep?) is unidentified.

After a school bans mobile phones, its students' age-16 exam scores rise by about 6.4% of a standard deviation (0.0641, SE 0.0373 — p about .09; 0.0567 to 0.0669 across specifications). That average is not where the paper's claim lives. The gain is monotone in prior achievement and concentrated entirely at the bottom: students in the lowest quintile of age-11 achievement gain 14.2% of a standard deviation (SE 0.040, p < .01), the second quintile 9.9% (p < .05), and the top quintile gains nothing (-0.025, ns). The interaction of the ban with prior test scores is -0.060 (SE 0.013, p < .01), and bans also help free-school-meal (+0.066, p < .05) and special-educational-needs (+0.110, p < .01) students more. The policy reading the authors offer — a near-zero-cost way to compress the achievement distribution — is why this paper became the most-cited evidence for phone bans worldwide. The caveat that most changes how it should be read is that the design cannot separate "the school banned phones" from "the school got stricter", and that the closest replication attempt, in Sweden, rejects an effect this large.

Medium. Within-school before/after comparison with age-11 prior achievement controls means pupil genotype is not what changes at the ban date; but the low-achiever heterogeneity is estimated on a prior-achievement split that is itself substantially heritable, so the subgroup finding describes which children respond, not an environmental cause of their being in that quintile.

Kessel, Hardardottir & Tyrefors (2020, Economics of Education Review; see `edt-kessel-2020-phone-ban-sweden`) ran the closest thing to a direct replication this literature has: the same difference-in-differences design on staggered school-chosen phone bans, but on the UNIVERSE of Swedish grade-9 schools with a 76% survey response rate instead of 21% of schools in four English cities. They find -0.034 (SE 0.029), 95% CI [-0.092, +0.024] on merit points — a null whose upper bound EXCLUDES this paper's +0.064, so the failure is not merely "not significant", it is a quantitative rejection of the effect size. Two later papers sit on either side and neither rescues the specific claim here. Abrahamsson (2024/2026, Norway; `edt-abrahamsson-2024-smartphone-bans-norway`) finds NO average effect on GPA or grades — matching Kessel — with gains confined to girls, which is a different heterogeneity story from this paper's (low prior achievers, of either sex; the male interaction here was insignificant). Figlio & Ozek (2025, Florida; `edt-figlio-ozek-2025-cellphone-ban-florida`) do find test-score gains, but only in year two, larger for BOYS, and mediated by attendance rather than by classroom attention. So: the general proposition that removing phones can raise attainment survives; this paper's particular estimate and its particular subgroup story do not have a clean corroboration anywhere. Per METHODOLOGY.md's replication rule any verdict resting on this source is capped at `mixed`. The disagreement is real and is left standing rather than resolved.