One Laptop per Child at Home: Short-Term Impacts from a Randomized Experiment in Peru

~1,000 XO laptops distributed (1,048 delivered) via in-class public lotteries; 14 treatment and 14 control schools in Lima; roughly 2,600-2,850 students in the analysis samples depending on outcome

Children in large, low-performing public primary schools in Lima, Peru (schools selected from the lowest-scoring on the national second-grade examination, with 400-800 pupils and fewer than 0.15 computers per student); laptops given for HOME use

ACCESS intervention, and the sharpest test in the cluster of "does the device do anything once it leaves the classroom". Two-stage randomization: 14 treatment / 14 control schools by pair-wise Mahalanobis matching on class size and computer-to-student ratio, then public in-class lotteries for four XO laptops per eligible section within treatment schools. Compliance was near ideal (93% of treatment students received laptops; zero control students did). The within-school design was chosen deliberately to measure spillovers, and the paper checks them two ways (social networks reported at baseline, plus the pure-control schools); spillovers appear only in XO-specific computing skill (+0.14 SD), not in time use or achievement. THE PAPER MEASURED DISPLACEMENT and found it — reading down 6 percentage points, unstructured play down 4, domestic chores UP 4, homework and TV unmoved. It also measured a teacher-reported behaviour outcome that is easy to miss and worth flagging: laptop winners were 5 percentage points less likely to be rated as showing high academic effort in class. That is the same sign as the Romanian grade decline, from a randomized design. Follow-up ran only six months (baseline April/May 2011, lottery June/July, follow-up November 2011), which is the main limitation and the likeliest reason the Raven's estimate is small; the authors say so. VERSION NOTE: the full text read here is NBER working paper w18818 (February 2013), then titled "Home Computers and Child Outcomes", in which math and reading tests were administered ONLY at baseline. The published AEJ: Applied 2015 version adds follow-up achievement and reports "no impacts on academic achievement or cognitive skills as measured by the Raven's Progressive Matrices" in its abstract; the achievement row below is therefore taken from the published abstract rather than from a table I read, and is marked accordingly. Run by the Inter-American Development Bank with GRADE (Peru); laptops supplied by the Peruvian Ministry of Education. No vendor authored or ran the evaluation.

Roughly a thousand children in Lima won an XO laptop for home use in a public lottery. Home computer use rose 34 percentage points and about 19 minutes a day, and part of that was substitution rather than addition — internet-café use fell 11 percentage points. What the children gained was proficiency with the specific machine they were given (+0.88 SD on an XO test) and nothing else: PC and internet skills were flat (+0.02, ns), Raven's was +0.05 (ns), and the published version reports no impact on academic achievement. What they lost was reading (-6pp) and unstructured play (-4pp), and their teachers rated them 5 percentage points less likely to show high academic effort in class. A laptop at home bought laptop skill, cost reading time, and produced no learning.

Minimal — individually randomized by public lottery within classrooms, on top of school-level pair-matched randomization.

This is itself close to a direct replication of Cristia et al. (2017) with the delivery channel changed from school to home, and of Malamud & Pop-Eleches (2011) with random rather than discontinuity-based assignment. It reproduces their computer-skills gain, reproduces the achievement null, reproduces the reading displacement found in Romania, and FAILS to reproduce the Raven's gain reported in both of those papers (+0.05 SD, ns, here). Fairlie & Robinson (2013) is the matching US null. Cueto et al. (2025) extends the Peruvian null to ten years.