Learning in One-to-One Laptop Environments: A Meta-Analysis and Research Synthesis

96 sources screened (65 journal articles and 31 doctoral dissertations, January 2001 - May 2015); 10 studies pooled in the quantitative meta-analysis; the remaining 85 were synthesized narratively because they lacked an experimental design and/or quantitative results

K-12 students in one-to-one laptop programmes, overwhelmingly in the United States

READ-DEPTH DEBT: the full text is paywalled (Review of Educational Research; no open repository copy exists per OpenAlex/Unpaywall/Semantic Scholar) and this record was built from the verbatim published abstract plus press coverage of the paper, principally Education Week (May 2016), which reported the subject-level effect sizes. The individual pooled studies, their designs, their outcome measures and any publication-bias analysis WERE NOT INSPECTED. Every judgement about the pool below is therefore an inference from what is verifiable, and is flagged as such. What is verifiable and already damning for the pooled estimate: of 96 sources screened over fourteen years, only 10 contained a quantitative comparison usable in a meta-analysis at all. The authors themselves note that "a disproportionate amount of the research to date on this topic consists of small case studies in one or a handful of schools". A literature that yields ten comparable effect sizes in fourteen years is not one in which the pooled average carries much information about causation, and the base rate for what fills that gap in ed-tech is school- or district-level comparisons of places that CHOSE to go one-to-one against places that did not — the exact selection problem this archive's premise predicts, since districts that adopt one-to-one differ in funding, leadership, parental composition and everything else that also predicts test scores. ALSO: this meta pools four different subjects, mixed grade levels, and — on the evidence of the base literature — a mix of state-test and locally-administered outcomes, which is the incommensurability problem METHODOLOGY.md rejects in Hattie-style aggregation. It is not a like-for-like average. GRADE: C is the METHODOLOGY.md row for "meta-analyses of mixed-quality designs" and is where this sits on what can be confirmed. It should be revisited on a full read, and the plausible direction is down to D, not up: if the ten pooled studies turn out to be uncontrolled or adopter-versus- non-adopter comparisons, the meta inherits that confounding and belongs in the D row. It should never be cited alongside the randomized nulls as though the two were commensurable evidence. INDEPENDENCE: recorded as `unclear` rather than `independent`. The authors are university researchers with no product to sell, but funding and any vendor or district-programme relationships could not be checked from the abstract, and the pooled primary studies (several of them dissertations evaluating their own districts' programmes) may not be independent of the programmes they evaluate.

Reviews 65 journal articles and 31 dissertations on K-12 one-to-one laptop programmes (2001-2015) and pools 10 of them, reporting a small positive overall effect on achievement — about +0.16 SD, with significant positive averages in English/language arts, writing, mathematics and science, and a non-significant effect on reading. The number is small on its own terms (well under the archive's benchmark for a year of schooling) and the pool is thin: only 10 of 96 sources contained a usable quantitative comparison, and the authors describe the base literature as dominated by small case studies of one or a handful of schools. Set against four randomized trials of the same intervention class that all return nulls, the most economical reading of +0.16 is not that one-to-one laptops work, but that schools which adopt them differ from schools which do not.

High for the pooled estimate. Where the comparison is between schools or districts that adopted one-to-one and those that did not, the families in adopting districts differ in every heritable and selected respect that also predicts achievement, and the meta-analysis inherits that confounding regardless of how carefully the pooling was done. Note the contrast with the four randomized studies of the same intervention, where genetic confounding is minimal by construction and the effect is zero.

The pooled positive is contradicted by every well-identified randomized trial of the same intervention class in this archive. Fairlie & Robinson (2013) randomly gave 1,123 US schoolchildren home computers and found nothing, with 95% CIs tight enough to exclude 0.16 SD. Cristia et al. (2017) randomized one-to-one laptops across 318 rural Peruvian schools and found +0.046 SD in math and -0.039 in language. Beuermann et al. (2015) randomized laptops to ~1,000 Lima children and found no achievement effect. Cueto et al. (2025) followed 531 randomized Peruvian schools for ten years on national examinations and ruled out effects above 0.04-0.05 SD. Four randomized designs, three countries, all null; one meta-analysis of ten mostly non-randomized US school comparisons, +0.16. When a pooled non-randomized estimate and a set of RCTs on the same intervention disagree by that margin, this archive sides with the RCTs and treats the gap as the size of the selection effect.