Technology and Child Development: Evidence from the One Laptop per Child Program

318 rural primary schools (209 treatment, 110 control after randomization within strata; the working-paper version reports 319 and the published version 318) — roughly 4,100 students tested, drawn from second grade, a followed cohort (mostly fourth grade at follow-up) and sixth grade; about 2,620 students in the interviewed sub-sample

Students in small, poor rural public primary schools in Peru; second, fourth (followed cohort) and sixth graders, tested 15 months after laptops arrived

ACCESS intervention at national scale, cluster-randomized at the school level with stratification by region, share of over-age students, and school size; two-thirds of schools treated. This is the largest randomized evaluation of One Laptop per Child ever run and the design is sound: strata fixed effects, standard errors clustered at the school level, and pre-treatment administrative balance. The first stage is enormous — computers per student went from 0.12 to 1.18, computer use last week from 32% to 84%, home computer use from 4% to 42%. THE PAPER MEASURED DISPLACEMENT and found essentially none: time studied at home, reading a book last week, and enrolment were all unmoved. So this is a null produced by a delivered treatment with no measured crowding-out, which makes it a null about laptops rather than about implementation. THE COGNITIVE-SKILLS RESULT NEEDS SPECIFIC HANDLING and this archive should not let it pass as a far-transfer or g finding. Three reasons. (1) Magnitudes are marginal: Raven's +0.112 (SE 0.057, p<0.10), verbal fluency +0.134 (SE 0.090, ns), coding +0.086 (SE 0.097, ns), pooled index +0.110 (SE 0.060, p<0.10). Only the index and Raven's clear the 10% level and neither clears 5%. (2) The tests were MIS-ADMINISTERED in 40% of schools — the coding and verbal-fluency tests are timed at three minutes, and in 40% of schools at least some students got roughly ten. The larger estimates the paper reports (0.14-0.24 SD) come from restricting to the 60% of schools timed correctly, which is a post-hoc subsample selected on a protocol failure, not a robustness check. (3) The authors themselves published the hedge: the AEJ abstract reads "some evidence, though inconclusive, about positive effects on general cognitive skills". A four-year follow-up by overlapping authors (Cueto et al. 2025) then found nothing on the same battery. Under this archive's rules — far transfer and g claims are presumptively false and carry the burden of proof — a marginal, unreplicated, partly-mis-timed effect on Raven's does not discharge it. Recorded as outcome_class g so it can never be quietly reclassified as "learning". Run by the Inter-American Development Bank with GRADE (Peru) and the Peruvian Ministry of Education; three authors were at the IDB, which financed the evaluation but did not build or sell the laptops. Numbers below are from the full text of the IZA/IDB working-paper version (IZA DP 6401 / IDB-WP-304, March 2012); the published AEJ: Applied 2017 abstract matches on every qualitative claim.

Fifteen months after the largest randomized OLPC deployment ever run, math was +0.046 SD (SE 0.061) and language -0.039 SD (SE 0.057) — flat, with enough precision to rule out effects above about 0.11 SD. Nothing moved on enrolment, home study time, or reading. What did move was laptop access (computers per student 0.12 to 1.18) and, marginally, three general cognitive tests (pooled +0.110, p<0.10). That cognitive result is the paper's most-cited and least-durable finding: it is significant only at 10%, rests partly on tests that were mis-timed in 40% of schools, is hedged as "inconclusive" in the published abstract, and did not replicate in a four-year follow-up of the same programme. The defensible summary is that giving every rural child a laptop changed what they could do with a laptop and did not change what they knew.

Minimal (cluster-randomized at the school level, stratified, with documented baseline balance on administrative test scores).

The achievement null is replicated and extended. Beuermann et al. (2015) found no achievement effect from the same laptops taken home in Lima; Cueto et al. (2025) followed a non-overlapping sample of 531 schools from the SAME Peruvian randomization for ten years using national examinations and administrative enrolment records and found nothing, ruling out effects above 0.04-0.05 SD. The COGNITIVE-SKILLS result did NOT replicate: Beuermann et al. (2015) found +0.05 SD (ns) on Raven's in Lima, and Cueto et al. (2025), surveying children after four years of exposure, found +0.046 SD (ns) on Raven's, +0.034 (ns) on coding, +0.187* on verbal fluency, and no significant effect on a pooled cognitive index. Treat the 2017 cognitive finding as an isolated marginal result that a larger, longer, same-programme follow-up failed to confirm.