Home Computer Use and the Development of Human Capital

3,354 households interviewed (52% response from a list of 6,418 applicant families in seven Romanian counties); parent reports cover roughly 5,900 children; the cognitive, computer-skills and child-report measures cover the subset of children present at home on survey day

Children aged 7-22 (only 3% over 19) in low-income Romanian families who applied to the 2008 round of the "Euro 200" home-computer voucher programme. Predominantly rural, mean monthly income about 48 RON (~USD 20) per household member; applicants scored roughly 0.3 SD below the national average on school performance. Counties sampled — Arad, Bistrita-Nasaud, Braila, Cluj, Maramures, Mures, Sibiu.

ACCESS intervention, not an instructional one: a 200-euro voucher toward a home computer, with no software mandate, no training, and no curricular integration. Identification is a regression discontinuity on the administrative income cutoff (62.58 RON per household member) at which the oversubscribed 2008 round stopped awarding vouchers; results are reported across six nonparametric bandwidths and three polynomial splines, and the paper is honest that magnitudes swing with bandwidth. First stage is large and clean: winning moved home-computer ownership by over 50 percentage points and computer use by 3-4 hours a week. Crucially the paper measured WHAT THE COMPUTER REPLACED, which almost nothing else in this literature does — daily homework, daily TV, and daily reading for pleasure, from both child and parent questionnaires. Outcome measures are mostly self- and parent-reported school grades rather than administered tests, which is the main measurement weakness; the computer-skills test and Raven's were administered by the interviewers. Survey response was 52% and non-representative of the applicant pool, but response rates did not differ between winners and losers. Only 14% of households had internet, and only ~9% had educational software installed despite the Ministry of Education distributing it free — so this is a clean test of "a computer in the house", not "a computer with a learning programme on it". Funded by the Spencer Foundation, NORC/Chicago PRC, ISERP at Columbia and CHPPP; no vendor involvement of any kind. NUMBERS BELOW ARE FROM THE FULL TEXT OF THE NBER WORKING PAPER (w15814, March 2010); the published QJE 2011 version carries the same design and the same qualitative claims, with the Raven's result softened from "increased cognitive ability" to "some evidence... increased cognitive skills".

Winning a home-computer voucher lowered school grades by roughly one-third of a standard deviation in Math, Romanian and English, raised measured computer skills by about the same amount, and left behaviour grades and non-cognitive outcomes untouched. The reason is legible in the time-use data: the extra computer time went into games (+14pp daily), not homework (+0.001, a precise zero) or educational software (+0.005), and it came out of homework, TV and reading for pleasure. The paper also reports a positive Raven's Progressive Matrices coefficient, which is the single most over-read number in this literature — it is significant at some bandwidths, near zero (0.013, 0.027) at the two smallest, and has since failed to replicate in two later randomized Peruvian trials by overlapping authors. Read this paper for the displacement evidence, not for the g claim.

Minimal. Regression discontinuity at an administrative income threshold — heritable family traits do not jump discontinuously at 62.58 RON per household member, and the paper shows covariate balance across the cutoff. Residual risk is survey non-response (52%), which did not differ by voucher status.

Split record, and the split matters. The NEGATIVE academic finding replicates in direction: Beuermann et al. (2015, Lima) found teacher-rated academic effort 5pp lower among laptop winners, and Cristia et al. (2017) found a small negative effect on self-perceived school competence. Fairlie & Robinson (2013, California) is a precise null rather than a negative — the same intervention, a richer country, and no detectable displacement of homework, which is the most likely reason the sign differs. The POSITIVE Raven's finding has since failed: Beuermann et al. (2015) found +0.05 SD (ns) on Raven's for the full sample, and Cueto et al. (2025), surveying rural Peruvian children four years into laptop exposure, found +0.046 SD (ns) on Raven's and no effect on a pooled cognitive index. The computer-skills gain replicates everywhere it has been looked for.