Accepted Manuscript

Do children benefit from internet access? Experimental evidence from a developing

country

Ofer Malamud, Santiago Cueto, Julian Cristia, Diether W. Beuermann

PII: S0304-3878(18)30125-1
DOI: https://doi.org/10.1016/j.jdeveco.2018.11.005
Reference: DEVEC 2307

To appear in: Journal of Development Economics

Received Date: 23 February 2018

Revised Date: 20 November 2018
Accepted Date: 22 November 2018

Please cite this article as: Malamud, O., Cueto, S., Cristia, J., Beuermann, D.W., Do children benefit
from internet access? Experimental evidence from a developing country, Journal of Development
Economics (2019), doi: https://doi.org/10.1016/j.jdeveco.2018.11.005.

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT

Do Children Benefit from Internet Access? Experimental Evidence from a Developing

Country

Ofer Malamuda,*
Northwestern University, NBER and CESifo

Santiago Cuetob

PT
GRADE and Pontificia Universidad Catolica del Peru

Julian Cristiac

RI
Inter-American Development Bank

Diether W. Beuermannd

SC
Inter-American Development Bank

November 2018

U
AN
Abstract

This paper provides experimental evidence for the impact of home internet access on
M

a broad range of child outcomes in Peru. We compare children who were randomly
chosen to receive laptops with high-speed internet access to (i) those who did not
receive laptops and (ii) those who only received laptops without internet. We find
D

that providing free internet access led to improved computer and internet proficiency
relative to those without laptops and improved internet proficiency compared to
TE

those with laptops only. However, there were no significant effects of internet access
on math and reading achievement, cognitive skills, self-esteem, teacher perceptions,
or school grades when compared to either group. We explore reasons for the absence
EP

of impacts on these key outcomes with survey questions, time-diaries, and computer
logs.

JEL classifications: C93, I21, I28

C

Keywords: Internet access; Education; Academic achievement; Digital skills;

Cognitive skills; Technology; Experimental
AC

a
School of Education and Social Policy, Northwestern University, 2120 Campus Drive, Evanston, IL 60208
*
Corresponding author: ofer.malamud@northwestern.edu.
b
GRADE, Av. Almte. Miguel Grau 915, Lima 15063, Peru
c
Inter-American Development Bank, 1300 New York Ave NW, Washington, DC 20577
d
Inter-American Development Bank, 1300 New York Ave NW, Washington, DC 20577

1
 ACCEPTED MANUSCRIPT

1. Introduction

Despite the rapid worldwide expansion of the internet, large disparities in children’s internet

access remain. Internet access is practically universal for children in developed countries: over

95% of 15-year old students in OECD member countries report having a link to the internet at

PT
home (OECD, 2017). In contrast, access to the internet continues to lag for children in

RI
developing countries. For example, less than half of 15-year-old students in Algeria, Peru, and

Vietnam report having internet access at home (OECD, 2017). In an effort to alleviate this

SC
“digital divide”, many government and non-governmental organizations are investing substantial

resources to expand internet access to children in developing countries.1 However, rigorous

U
evidence for the impact of home internet access on children’s outcomes is currently limited to
AN
developed countries and may not generalize to settings where fewer resources can complement

or substitute for technology.2 Accordingly, this paper provides the first experimental evidence for
M

the impact of home internet access on a broad range of child outcomes in a developing country
D

context.
TE

Internet access can potentially affect a range of skills including academic achievement

and cognitive skills. If children lack educational materials, internet access may improve the
EP

development of academic skills by providing access to educational websites with subject-specific

content and exercises (e.g. Khan Academy). Moreover, children can access e-books and other
C

reading materials such as newspapers, blogs, and online encyclopedias (e.g. Wikipedia). On the
AC

other hand, internet access may diminish learning if children spend more time on activities that

are not conducive to developing academic skills, such as playing online games, and less time

1
For example, the “Yo Elijo Mi PC” program in Chile provides free internet along with laptops for eligible students.
Numerous national programs subsidize internet access more broadly, e.g. Brazil’s Plano Nacional de Banda Larga.
2
See Fairlie and Robinson (2013) and Vigdor et al. (2014) for evidence from the United States, and Faber et al.
(2016) for evidence from the United Kingdom. We discuss the findings from these studies in detail shortly.

1
 ACCEPTED MANUSCRIPT

reading and doing homework. Finally, internet access may affect cognitive skills by exposing

children to online activities that alter cognitive processes (Johnson 2006; Mills 2014).

To measure the effects of providing internet access, we implemented a randomized

experiment in Lima, Peru, between 2011 and 2013. We first provided access to XO laptops for

PT
home use to a random sample of 540 children enrolled in grades 3 to 5 in low-achieving public

RI
primary schools (in June/July 2011).3 Then, among children who received these laptops, we

randomly selected about 350 children to receive free high-speed internet access (in July/August

SC
2012). The laptops included 32 applications selected by the Ministry of Education of Peru for its

national program, and we offered training and manuals on how to use them. We also offered

U
tutorials and manuals to children who received internet access in which we showed them how to
AN
take advantage of freely available educational websites created by Peru’s Ministry of Education

and other online resources, such as Khan Academy and Wikipedia.

M

To evaluate the impacts of our interventions, we conducted a follow up survey in

D

November 2012, approximately 17 months after the laptops were initially distributed and 5
TE

months after the provision of internet access. We also conducted an additional follow-up survey

in March 2013 to check for longer-run impacts after the summer vacation. In prior work, we
EP

examined only the short-term impact of XO laptops without internet access (Beuermann et al.,

2015). In the current study, we compare (i) children who were randomly chosen to receive
C

laptops with internet access to (ii) those who did not receive laptops and to (iii) those who only
AC

received laptops without internet. Thus, we are able to estimate the impact of internet access both

separately from, and in conjunction with, the impact of the laptops themselves.

3
The XO laptops were developed by the One Laptop per Child (OLPC) program with an emphasis on self-
empowered learning and with specialized software intended to encourage such learning.

2
 ACCEPTED MANUSCRIPT

Our interventions were successful in increasing children’s access to technology at home

and led to substantial improvements in digital skills after just 5 months. Children who were

offered internet access were 30 percentage points more likely to have internet at home as

compared to those who were not offered internet, whether they had laptops or not. Furthermore,

PT
children who were offered internet access scored 0.3 standard deviations higher on a test of

RI
internet literacy than those who were not offered internet access, whether they had laptops or not.

They also scored 1 standard deviation higher on a test that measured proficiency on the XO

SC
laptop compared to those who were not offered laptops, but not significantly different from those

children who were offered laptops without internet. In addition, children who were offered

U
laptops (with or without internet) had significant improvements on a Windows-based computer
AN
test, suggesting that gains in computer literacy were not only limited to the specific XO platform

but transferred to skills for using other types of computers.

M

Despite the increase in access to technology and the improvements in digital skills, there
D

were no significant effects of internet access on academic achievement. We can rule out impacts
TE

larger than 0.08 standard deviations in math and 0.13 standard deviations in reading with 95%

confidence when comparing children who were offered internet access to those who did not get
EP

laptops. Nor were there any significant effects on a broad set of cognitive skills, as measured by

the Raven’s Progressive Matrices test, a verbal fluency test, a test of executive functioning, a
C

coding test, a working memory test and a test of spatial reasoning. Similarly, we did not find
AC

significant effects on a self-esteem index measured by a self-reported questionnaire. Based on

teacher reports, children in the treatment groups were equally likely to exert effort at school

when compared with their counterparts in the control group, and there were no differences on

grades obtained from administrative school records or in teacher perceptions of children’s

3
 ACCEPTED MANUSCRIPT

sociability. There was also no evidence of improvements when we resurveyed children 8 to 9

months after internet provision following the summer vacation, despite the potential benefits of

engaging children with the internet to counteract summer learning loss.4

Why were there no significant impacts on academic achievement and cognitive skills

PT
from providing children with internet access? Though we cannot provide a definitive answer to

RI
this question, we consider a number of possible explanations. First, while the intervention itself

was not directly linked with pedagogical activities at school, we did provide children with

SC
tutorials and manuals to make more effective use of their computers and the internet for

educational purposes. Second, while we do not have long-term outcomes, previous research has

U
shown that new technology can have short-term impacts within a year (Malamud and Pop-
AN
Eleches, 2011; Banerjee et al. 2007) or even just several months (Muralidharan, et al., 2016).

Third, it is possible that the impact on internet use was not sufficiently large. The provision of
M

internet led to initial increases in use but our computer and internet logs reveal a decline over
D

time. Approximately 5 months after being offered internet access, the fraction of children who
TE

report using the internet anywhere in the previous week remained significantly higher among

those who were offered internet access, but the difference was only 6 percentage points.
EP

Nevertheless, children who were offered laptops with internet access remained 22 percentage

points more likely to use internet at home during the previous week compared to those who were
C

offered only laptops, with 8 percentage points in lower use at internet cafés.
AC

So how did children use their home computers and what was the effect on time spent on

other activities? We use data from computer and internet logs to show that computer and internet

use was focused more on entertainment than on learning. This happened in spite of our efforts to

4
Many studies find that students score lower on the same standardized tests at the end of the summer than at the
beginning of the summer, often referred to as summer learning loss. See Alexander et al. (2016) for a collection of
recent contributions to the literature on summer learning loss.

4
 ACCEPTED MANUSCRIPT

promote internet use for educational purposes through the provision of training, tutorials and

manuals. Furthermore, although the time diaries do not reveal large and consistent changes in the

time spent on other activities, there is some evidence of a reduction in time spent watching TV

and doing homework. Overall, our analysis underscores the null effects of increasing internet

PT
access on the development of academic and cognitive skills, and suggests the need for parents to

RI
ensure that technological resources are used in ways that foster better educational outcomes.

There are a few recent studies that estimate the causal impact of home internet access on

SC
children’s outcomes, but all are based in developed countries.5 Fairlie and Robinson (2013) find

no impacts of home computers with subsidized dial-up internet access on standardized tests or

U
grades using a randomized experiment in California.6 Vigdor et al. (2014) exploit local variations
AN
in broadband internet penetration in North Carolina to show that children who live in areas that

receive additional internet providers experience a modest but significant decline in mathematics
M

test scores (and insignificant decline in reading). Lastly, Faber et al. (2016) exploit differences in
D

broadband connection speeds across neighboring residences in England and find no significant
TE

impacts on test scores or time spent studying. However, to our knowledge, no previous study has

separately identified the added effect of home internet access in an experimental setting.
EP

Accordingly, this paper makes several contributions to the existing literature on

technology in education. First, this study represents the first randomized experiment exploring
C

the effects of home internet access implemented in a developing country. Focusing on this
AC

setting is especially policy relevant given that governments and households in the developing

5
Papers that examine the causal impact of home computers without internet access on children’s outcomes in
developing countries include Malamud and Pop-Eleches (2011), Mo et al. (2013) and Beuermann et al. (2015).
Papers that evaluate the causal impact of school-based internet access in developing countries include Kho,
Lakdawala, and Nakasone (2018) and Sprietsma (2007).
6
Using this same experiment, Fairlie and Kalil (2016) find positive impacts on the likelihood of having a social
networking site and time spent communicating with friends, but no effects on school participation and engagement.

5
 ACCEPTED MANUSCRIPT

world are making significant investments to expand internet connectivity.7 Second, because our

individual-level randomization includes almost 2,000 children with follow-up data, we can

provide relatively precise estimates of impacts on a variety of short and medium-term outcomes.

Third, our study includes a broad range of outcomes, including not only academic achievement

PT
measures but also a full set of cognitive skills tests, teachers’ assessments, time diaries and

RI
school grades records. Finally, we use detailed information from computer logs of applications

and internet sites that provide objective measures to help us better understand the “black box” of

SC
computer use.

The paper is structured as follows. Section 2 describes the experimental design and

U
implementation of the interventions. Section 3 explains our data collection efforts and the
AN
empirical strategy we use to analyze the data. Section 4 presents the main impacts of our

interventions. Section 5 attempts to open the “black box” of our interventions using survey
M

questions, time-diaries, and computer logs. Section 6 provides a summary of our findings and
D

concludes.
TE

2. Experimental Design
EP

Our experimental study was implemented in several steps. We began by randomly selecting 14

schools from a sample of low-achieving public primary schools.8 Within these schools, we
C

provided laptops for home use to a random sample of children who were in third to fifth grade in
AC

2011. Then, among children that won laptops, we randomly provided high-speed internet access

to a sub-sample of them. The specific timeline of the study is as follows (and shown in Online

7
The yearly cost of providing (high-speed) internet access is now higher than the cost of most lower-end laptop and
desktop computers in many countries: see https://www.forbes.com/sites/niallmccarthy/2017/11/22/the-most-and-
least-expensive-countries-for-broadband-infographic/.
8
These schools all had morning shifts which enrolled between 400 and 800 students, 4 classes per grade or fewer, a
ratio of school computers to students lower than 0.15, and a classroom available for a computer lab in the afternoon.

6
 ACCEPTED MANUSCRIPT

Appendix Figure A1). We collected baseline data in April/May 2011 and conducted the lottery

(and delivered laptops) in June/July 2011. Training for the laptop and a manual for children was

offered in August/September 2011. We provided internet access in July/August 2012. Training

on internet use and a manual for children was offered in September 2012. Finally, we collected

PT
follow up data in November 2012 and March 2013. Note that Peru’s academic school year runs

RI
from March to December. It is worth noting that the Ministry of Education had automatically

blocked the XO laptops during the first summer of the project (January to March 2012).

SC
However, in order to measure the longer-run effects through March 2013, we obtained a waiver,

and the laptops were not blocked during the second summer (January to March 2013). The

U
remainder of this section describes the interventions in more detail.
AN
2.1 Laptops
M

We first offered XO laptops for home use to 540 randomly selected children in grades 3 to 5.9 In
D

June/July 2011, we conducted public lotteries for 4 laptops within each class among children
TE

whose parents provided written consent. The lotteries were conducted in class and parents were

invited to attend in order to assure transparency. These procedures were developed in close
EP

coordination with schools, principals, and teachers. The XO laptops were provided by the

Ministry of Education of Peru and specifically designed to be used by primary-aged students in

C

developing countries. The laptops had 512 MB of RAM, 2 GB of flash storage, and a Linux
AC

9
In total, 1048 laptops were distributed to children in grades 1 to 6. However, we focus on children in 3rd to 5th
grade because of the difficulties in collecting reliable baseline information from 1st and 2nd grades, and the challenge
of tracking 6th graders who moved on to secondary school by the time of the follow-up survey.

7
 ACCEPTED MANUSCRIPT

operating system. The graphical interface, known as Sugar, was designed to be used by

children.10

Thirty-two applications, selected by the Ministry of Education for nationwide use, were

installed in the distributed laptops. The applications included standard applications such as word

PT
processor, drawing software, and calculator; educational games including Tetris, Sudoku and a

RI
variety of puzzles; applications to create, edit and play music; two programming environments;

and other applications including sound and video recording and certain sections of Wikipedia.

SC
The laptops were also pre-loaded with age-appropriate e-books selected by the Ministry of

Education.

U
We provided all beneficiary children with an instruction manual and the opportunity to
AN
attend training sessions. The manual was designed for primary school children, with graphical

illustrations about how to use the laptop and in-depth practical instruction for 10 government-
M

prioritized applications. Weekly training sessions took place in each school for a seven-week
D

period in August and September 2011. On each Saturday during the training period, there were
TE

three two-hour sessions for children arranged by grade. Children did not receive help with

schoolwork or any other instruction during these sessions. Average student attendance was about
EP

50 percent, and approximately 70 percent of children attended at least one session.

C

2.2 Internet
AC

We provided high-speed internet access to a subset of laptop lottery winners who were studying

in grades 3 to 5 during the 2011 school year. This intervention was conducted in July and August

10
The laptops do not run Windows and they are not compatible with software designed for that operating system.
However, most files (e.g., images, sound and text documents) are compatible with the XO environment.

8
 ACCEPTED MANUSCRIPT

of 2012, approximately 13 months after the start of the first intervention.11 We randomly chose

two among the children that were offered laptops in each class and offered them a high-speed

internet connection for the XO laptop.12 Specifically, we randomly sorted the children who were

initially offered laptops and offered internet access to children ranked first and second. If either

PT
of these children ended up not receiving internet access (e.g. because the parents did not attend

RI
the session when the equipment was provided), we offered internet access to the child ranked in

the third position. If two of the children ranked first to third did not receive internet access, then

SC
we offered this service to the child ranked fourth. Approximately 80 percent of the internet

connections were provided to children ranked first or second––see Online Appendix Table A4.

U
All children who were offered internet access are considered as treated for the purposes of our
AN
ITT estimates (as described later).

In total, 354 children were offered high-speed internet access at home for eight months
M

until March 2013. We provided 8 additional training sessions and a manual to help children take
D

full advantage of the internet access. The manuals contained guidelines on using the internet
TE

safely, tutorials for educational websites produced by Peru’s Ministry of Education, instructions

on how to search for information using Wikipedia and other virtual libraries, as well as links to
EP

Khan Academy and other educational resources. However, children did not receive any help with

schoolwork during these sessions. We also attempted to minimize the possibility of exposure to
C

adult content by blocking certain websites and providing guidelines for use to children and
AC

parents.

3. Data and Empirical Strategy

11
Note that this intervention was only announced in June 2012, so children and parents were not aware of this
intervention until immediately prior to the start of the intervention itself.
12
Internet access was provided through Claro, a major telecommunications company in Peru.

9
 ACCEPTED MANUSCRIPT

3.1 Data

The primary data used in this evaluation were collected directly in baseline and follow-up

surveys.13 We conducted a baseline survey among children in targeted schools during April/May

2011 and collected information on basic demographics, computer literacy, computer and internet

PT
use, time use, and detailed information about social networks, as well as a test of cognitive skills

RI
(Raven’s Progressive Matrices) and standardized tests in math and language ability. We also

surveyed teachers on their perceptions of student sociability, effort devoted in class, and

SC
expected educational attainment.

A first follow-up survey was conducted in November 2012, approximately 5 months after

U
the provision of internet access. It covered most of the topics examined in the baseline survey
AN
plus a battery of cognitive tests including (i) a general cognitive test based on the Colored

Raven’s Progressive Matrices test together with an additional set of more difficult pattern-
M

recognition questions, (ii) a test of verbal fluency that requires children to list all the words they
D

can think of starting with a particular letter (“P”) in three minutes, (iii) a coding test similar to
TE

the one in the Wechsler test, (iv) a test of spatial reasoning based on mental rotation exercises,

(v) a working memory test, and (vi) a short test of executive functioning: the Stroop test
EP

measuring cognitive interference. We also tested children in three domains of digital skills: (i)

proficiency in the use of the XO laptops, (ii) competence in operating a Windows-based

C

computer, and (iii) knowledge and use of the internet. In addition, we administered a detailed
AC

time diary to a subsample of 838 children in November 2012 who had attended 4th grade at

baseline. These children were interviewed individually and were asked to mention each activity

performed during a 24-hour period starting at 7 am the prior day. We supplemented this follow-

13
In addition, we used administrative data at the school level from the Peruvian Education Statistics Unit and
individual-level standardized tests from Student Census Evaluation (ECE) at the Ministry of Education of Peru for
the initial selection of schools.

10
 ACCEPTED MANUSCRIPT

up survey by collecting official school records on grades corresponding to the 2012 academic

year.

A second follow up survey was conducted in March 2013, approximately 8 to 9 months

after the provision of internet access and shortly before families were required to return their

PT
laptops to the school.14 Children were administered an abridged questionnaire similar to the one

RI
in the previous follow-up, and tested on their digital skills, math and reading ability and one of

the cognitive tests (Raven’s progressive matrices).

SC
Finally, we extracted log files from the laptops for approximately two-thirds of the XO

beneficiaries (with parental consent). This enables us to examine detailed patterns of use in an

U
objective manner without relying on subjective reports from children. The logs recorded the date
AN
and time when each session started as well as roughly when every application was opened. They

also provide information about every internet site visited by children on these computers.
M
D

3.2 Empirical Strategy

TE

The empirical framework for our main analysis involves two different comparisons. First, we

compare children who were randomly offered laptops with internet access (“Laptop+Internet”
EP

treatment) to those who did not win laptops (“No Laptop” control group). Second, we compare

children who were randomly offered laptops with internet access (“Laptop+Internet” treatment)
C

to children who won laptops but were not offered internet access (“Laptop Only” control group).
AC

We can estimate both of these impacts jointly using the following regression model:


 =    +   +   +   +  +  (1)

14
This survey was only administered to children who attended 3rd and 4th grade in 2011. This because students who
were in 5th grade in 2011 graduated from primary school in December 2012 and we were no longer able to track
them.

11
 ACCEPTED MANUSCRIPT

where
 denotes an outcome of interest for child i, in class j and school k observed at follow-

up.   is an indicator which takes the value of 1 if child i in class j and school k

won a laptop but was not offered home internet access, and 0 otherwise.  +   is

PT
a treatment indicator which takes the value of 1 if child i in class j and school k won a laptop and

was offered internet access, and 0 otherwise. The omitted category is the “No Laptop” control

RI
group. Thus,  captures the impact of providing internet access for children who neither have

access to the internet or to a laptop; and  =  −  captures the impact of providing internet

SC
access for children who have access to a laptop without internet access (i.e. the added effect of

internet over and above the effect of having a laptop only).15

U
AN
In our preferred specifications,  only includes a constant and the baseline value of the

corresponding outcome variable (when available) to help improve precision. However, our
M

results are unchanged by the inclusion of additional baseline covariates or by excluding the

baseline value of the outcome variables. We always include a class lottery fixed effect,  , since
D

the individual randomization was carried out class by class. Finally,  is an error term that
TE

allows for heteroscedasticity across observations.

EP

3.3 Analytical Sample

C

The main sample used to analyze the effects measured in November 2012 includes 2,126
AC

children attending grades 3 to 5 at baseline whose parents provided consents for participating in

15 
This second comparison can be estimated directly by
 =    +  !   +   +  +
 ; where the omitted category is the group of children in the “Laptop+Internet” treatment. Then, reversing the
signs of the main coefficients in this regression equation, we get that  = −  and  =  −  = −  .

12
 ACCEPTED MANUSCRIPT

the study.16,17 Of these, 1,653 are in the “No Laptop” group, 163 are in the “Laptop Only” group

and 310 are in the “Laptop+Internet” group. The attrition rate in this sample from the

corresponding baseline sample is 13 percent, and not statistically different across treatment

arms.18,19 Figure 1 describes how we arrived at this final sample from our original sample in

PT
more detail and shows the level of compliance within each treatment arm (Online Appendix

RI
Figure A2 shows the analogous sample composition for the longer-run follow-up).

SC
[Figure 1 here]

U
Table 1 provides evidence that the within-school randomization was successful in
AN
generating balance between treatment and control groups for our analytical sample.20 Columns 1,

2, and 3 present the means of the baseline characteristics for the “No Laptop”, “Laptop Only”,
M

and “Laptop+Internet” groups respectively. Column 4 shows the estimated difference between
D

the “Laptop+Internet” group and the “No Laptop” group while Column 5 shows the estimated
TE

difference between the “Laptop+Internet” group and the “Laptop Only” group. Among the 42

estimated differences between treatment and control groups, only two were significant at the five
EP

percent level or lower. Thus, the randomization of computers and internet was successful in
C

16
Online Appendix Table A3 shows that children whose parents consented to the study had significantly lower
probability of having phone, electricity, computer and internet access but higher scores in reading, cognitive skills,
AC

and teacher assessments than those whose parents did not consent.
17
This analytical sample is restricted to observations with non-missing values for the corresponding variables at
baseline and follow-up so that we have consistent samples in our balance and outcome tables. Our results are
essentially unchanged when we relax this restriction, as shown in Online Appendix Table A5.
18
The likelihood of attrition in the November 2012 survey was slightly higher for males, those without siblings, and
lower math scores at baseline. However, these characteristics predict attrition similarly across treatment arms.
19
The attrition rate in the March 2013 follow-up survey was closer to 25 percent, but also not significantly different
across treatment arms. See Online Appendix Table A1 for the differential likelihood of attrition across treatment
arms.
20
Note that Table 1 shows balance for children that were observed at both baseline and follow-up. Online Appendix
Table A2 shows that baseline characteristics are also balanced for the full sample of randomized children at baseline
(including those who were not observed at follow-up).

13
 ACCEPTED MANUSCRIPT

creating balanced treatment and control groups to consistently identify the effects of the

intervention through equation (1).

[Table 1 here]

PT
RI
All our results are based on intent-to-treat (ITT) estimates; that is, we consider children

who were randomly assigned to the “Laptop Only” or “Laptop+Internet” groups whether or not

SC
they actually received a laptop or internet access. Given the 95 percent take-up rate of laptops

and the 81 percent take-up rate of internet in our analytical sample, the ITT estimates are

U
relatively similar in magnitude to the treatment on the treated (TOT) estimates that scale up our
AN
estimates by the rates of take-up.21 However, scaling up the estimates by the rates of computer

and internet access and use reported by children, as discussed in sections 4.1 and 4.2, would
M

yield substantially larger impacts.

D
TE

4. Main Impacts

This section presents the estimated impacts of our interventions on computer access and use,
EP

internet access and use, computer and internet skills, academic achievement, cognitive skills,

self-esteem, school grades, and other teacher assessments. Most of the tables are structured in a
C

similar fashion: the means of outcomes for the “No Laptop,” “Laptop Only,” and
AC

“Laptop+Internet” groups are shown in columns 1, 2, and 3 respectively; the difference between

the “Laptop+Internet” group and the “No Laptop” group in column 4; and the difference between

the “Laptop+Internet” group and the “Laptop Only” group in column 5.

21
We have estimated equation (1) instrumenting actual take-up with the indicators for random assignment and
results remain mostly unchanged. These results are available upon request.

14
 ACCEPTED MANUSCRIPT

4.1 Computer Access and Use

Panel A of Table 2 presents our findings related to self-reported computer access and use.

Column 4 shows that children randomly assigned to receive laptops with internet access were

PT
approximately 40 percentage points more likely to report having a computer at home compared

RI
to those in the “No Laptop” group. This represents a large and significant effect on access to

home computers but it is noteworthy that 54 percent of children in the “No Laptop” control

SC
group already own a computer.22 Unsurprisingly, column 5 shows no significant impact of

simply providing internet access on computer ownership relative to children in the “Laptop

U
Only” group who were provided laptops without internet access.
AN
[Table 2 here]
M
D

There were also significant differences in computer use when comparing children in the
TE

“Laptop+Internet” group to those in the “No Laptop” group in column 4. Consistent with the

nature of our intervention, children who were randomly assigned to receive laptops with internet
EP

were 33 percentage points more likely to use computers at home during the previous week

compared to those in the “No Laptop” group. However, these same children in the
C

“Laptop+Internet” group were 13, and 7 percentage points less likely to use computers in
AC

internet cafes, and friend’s houses respectively than those in the “No Laptop” group. Therefore,

it seems that children who won a laptop and internet access substituted computer use at home for

22
Note that only 43 percent of children in the control group report having a home computer at baseline, suggesting
that some of them acquired a computer after (and, perhaps, because of) our intervention. Although we did not collect
information about the nature of other computers in the household, these are unlikely to be XO laptops since such
computers were not commercially available.

15
 ACCEPTED MANUSCRIPT

use outside the home. Again, except for a 9 percentage points reduction in the likelihood of using

computers in internet cafes, column 5 shows no significant impacts of providing internet access

on computer use relative to children in the “Laptop Only” group.

PT
4.2 Internet Access and Use

RI
Panel B of Table 2 presents our findings on self-reported internet access and use. Compared to

children in either the “Laptop Only” or “No Laptop” group in columns 4 and 5 respectively,

SC
those randomly assigned to receive laptops with internet were 30 to 33 percentage points more

likely to report having internet at home. Thus, the internet intervention was successful in

U
increasing connectivity. Nevertheless, around 40 to 44 percent of children in the control groups
AN
also report access to the internet (this fraction was around 31 to 34 percent at baseline).23

We also observe that the internet intervention led to increased internet utilization at
M

home. Children in the “Laptop+Internet” group were 27 and 22 percentage points more likely to
D

report using internet at home during the previous week compared to those in the “No Laptop”
TE

and “Laptop Only” group respectively. However, those children who were randomly assigned

laptops with internet access did report a decrease in the likelihood of using internet in internet
EP

cafés (a significant decrease of 10 percentage points when compared to the “No Laptop” group).

This suggests that there was some substitution away from internet use outside the home and
C

towards internet use on children’s personal computers.

AC

4.3 Skills

23
These rates of internet access are similar to home internet penetration of 36 percent among households in
Metropolitan Lima in 2012 based on the Peruvian National Household Survey. This penetration is higher than in
other regions of Peru (especially the poorer highlands and Amazon regions where rates are below 10 percent).

16
 ACCEPTED MANUSCRIPT

Table 3 presents the impacts of internet access on a broad set of skills. Compared to children in

the “No Laptop” group, those who were randomly assigned to receive laptops with internet

access scored 1 standard deviation higher on an “XO test” measuring XO-specific laptop

knowledge (column 4). Similarly, we observe a positive impact on a “PC test” which measured

PT
skills related to using a Windows-based computer. Children in the “Laptop+Internet” group

RI
scored 0.21 standard deviations higher than those in the “No Laptop” group. It thus appears that

XO skills are transferable (at least to some degree) to Window-based platforms. However, there

SC
is no additional impact of internet access over and above the effect of receiving a laptop without

internet access for either the XO test or the PC test (column 5). This confirms that the provision

U
of internet access itself did not improve those digital skills that were not specifically related to
AN
internet use.
M

[Table 3 here]
D
TE

In our “Internet test” of internet knowledge and skills, we observe that children in the

“Laptop+Internet” group scored 0.33 and 0.26 standard deviations higher than those in the “No
EP

Laptop” and “Laptop Only” groups respectively. This confirms that the internet intervention

served to improve children’s internet skills. In other words, the impact on internet skills is almost
C

wholly captured by the additional impact of internet access over and above the effect of
AC

receiving a laptop without internet access. The combined impact of our XO test, PC test, and

internet test is captured by a digital skills index that is also standardized accordingly. We

attempted to gauge the magnitude of our impacts on digital skills by comparing them to

differences in digital skills between children with and without home computers and internet in

17
 ACCEPTED MANUSCRIPT

the control group. We find that providing children with access to computers and internet at home

effectively closes the gap in digital skills between those with and without home computers.

We administered standardized math and reading tests, but we did not find significant

impacts of providing internet access relative to either control group. These estimated impacts of

PT
zero are also quite precise. None of the estimates are larger than 0.05 standard deviations in

RI
magnitude. Using our academic achievement index which combines the standardized scores in

math and reading, we can rule out positive impacts larger than 0.08 standard deviations with

SC
95% confidence when comparing children with internet access to those who did not get laptops.

This is consistent with previous evidence provided by Beuermann et al. (2015) and Cristia et al.

U
(2017). We also administered a battery of cognitive tests which included the Raven’s Progressive
AN
Matrices, verbal fluency, executive functioning, coding, working memory, and spatial reasoning.

Again, our results reveal no impacts of providing internet access compared to either of the
M

control groups. Using our cognitive skills index which combines the standardized scores across
D

all of these tests, we can rule out positive impacts larger than 0.09 standard deviations with 95%
TE

confidence when comparing children with internet access to those who did not get laptops. This

contrasts with previous findings in Malamud and Pop-Eleches (2011) and Cristia et al. (2017)
EP

who found significant effects of exposure to computers on the Raven’s Progressive Matrices test.

Finally, we applied a survey instrument that yielded a self-esteem index but also found no
C

evidence of significant effects.24

AC

4.4 Grades and Teachers’ Perceptions

24
There is an extensive literature in psychology examining the effect of the internet on social involvement and
psychological well-being, starting with a seminal paper by Kraut et al. (1998).

18
 ACCEPTED MANUSCRIPT

Table 4 presents results based on teacher evaluations. We collected administrative data regarding

official school grades for the 2012 academic year running from March to December. We

computed indicators for the likelihood of being promoted to the next grade (Pass grade), for

being assigned to summer school because of poor performance during the academic year (Need

PT
summer school), and for failing the academic year (Fail grade). We also computed the

RI
percentage of courses in which students obtained the highest possible letter grade awarded by

their teacher (Percentage of top grades). Columns 4 and 5 in the top panel of Table 4 shows that

SC
there are no significant impacts of providing internet access on any of these measures.

U
[Table 4 here]
AN
We also surveyed teachers on their perceptions of their students’ social popularity, effort
M

at school and expected educational attainment.25 For the first two dimensions, we asked teachers
D

how they evaluate each pupil in their classroom: below average, average, or above average.
TE

Regarding educational expectations, the options reported by teachers consisted of whether the

child was expected to attain primary, secondary or post-secondary education. We then

EP

constructed summary indicators that take the value of one if the teacher reported the highest

possible outcome for the pupil and zero otherwise. As shown in the bottom panel of Table 4, we
C

did not observe any significant impact of providing internet access on teacher’s perceptions of
AC

social popularity or effort at school (although there was a marginally significant difference of 5

25
Teachers’ perceptions could have been affected if they knew their students’ treatment status. However, this is not
likely because of several reasons. First, internet randomization was conducted privately by the research team and
only internet winners were contacted. Second, laptops were provided for home use and students were not allowed to
take them to school. Third, most of the teachers surveyed did not witness the initial provision of computers in 2012
because this took place in the prior year and teachers typically do not follow the same cohort of students over time.

19
 ACCEPTED MANUSCRIPT

percentage points between the “Laptop+Internet” and the “No Laptop” groups in the likelihood

of expecting the child to complete university education).

4.5 Heterogeneous effects

PT
In addition to showing the average impacts of the intervention on the full sample, we also looked

RI
for heterogeneous effects by individual characteristics. We focused on differences by gender and

baseline academic achievement, as shown in Table 5.26 The impact of our interventions on access

SC
and use of home computers and internet by gender is similar. Girls show higher impacts on

digital skills as compared to boys, although these differences are not significant at the 5 percent

U
level. There are no statistically significant impacts on academic achievement, cognitive skills,
AN
and socio-emotional skills for either girls or boys with the exception of the effects of internet

provision on socio-emotional skills for girls. We also looked whether there were statistically
M

significant differential effects between boys and girls for the outcomes presented in Table 5 and
D

could not reject that the effects are similar across genders at the 5 percent level with the sole
TE

exception of the effect of internet provision on socio-emotional skills (larger for girls).
EP

[Table 5 here]
C

The impact of our interventions on access and use of home computers by baseline
AC

academic achievement is also similar, with slightly higher rates for children with higher baseline

achievement. The impacts on digital skills are larger for children with higher baseline

achievement, and significantly different from those with lower baseline achievement in the

26
Previous research by Banerjee et al. (2007), Bai et al. (2016), and Mo et al. (2015) suggests there may be
heterogeneous impacts of technology by baseline ability and gender. On the other hand, Linden (2008) and
Muralidharan et al. (2016) did not find significantly different impacts by these characteristics.

20
 ACCEPTED MANUSCRIPT

Laptop+Internet group. Again, there are no significant impacts on academic achievement,

cognitive skills, and socio-emotional skills for either high or low achievers with the exception of

positive effects of internet provision on academic achievement for high achievers. Moreover, we

checked whether there are differential statistically significant effects across low and high

PT
achievers and found that we cannot reject that the effects are the same with the exception of a

RI
larger effect of internet provision on internet use for high achievers.

Online Appendix Table A6 presents the differential impacts of our intervention by prior

SC
computer and internet access. Not surprisingly, the impacts on access and use of home computers

are larger in magnitude for children that reported no baseline availability of a computer or

U
internet at home. However, positive effects on digital skills are similar regardless of baseline
AN
access and there are no significant impacts on academic achievement, cognitive skills, or socio-

emotional skills for any of these subgroups. Finally, Online Appendix Table A7 presents the
M

differential impacts of our intervention by children’s grade at baseline. The estimated impacts are
D

mostly similar by grade, although the effects on use and digital skills appear to be larger for the
TE

youngest cohorts in a few cases.

EP

4.6 Longer-run effects

The results from the second follow-up survey in March 2013 are presented in Table 6, and
C

mostly mirror the main findings from the earlier follow-up survey conducted in November 2012.
AC

We continue to observe a pronounced impact of our interventions on measures of digital skills.

Children in the “Laptop+Internet” group scored over 1 standard deviations higher than those in

the “No Laptop” group on a test of XO proficiency, and 0.30 standard deviations higher on a test

of internet proficiency. However, as in our earlier survey, there were no significant effects on

21
 ACCEPTED MANUSCRIPT

academic achievement in math and reading, scores on the Raven’s Progressive Matrices test, or

on our measure of self-esteem.

[Table 6 here]

PT
RI
4.7 Spillovers

We also checked for the possibility of spillover effects by taking advantage of social network

SC
data reported by all children at baseline. In particular, we focused on children who did not win

the XO lottery and split them into three subgroups: (i) those reported as close friends of children

U
who won the XO lottery and were assigned to receive internet access, (ii) those reported as close
AN
friends only of children who won the XO lottery but were not assigned to receive internet access,

and (iii) those not mentioned as a close friend of any child who won the lottery.27 Under the
M

assumption that children who were not close friends with the lottery winners experienced little or
D

no spillovers, we can interpret the differences between groups (i) and (ii) and between groups (i)
TE

and (iii) as alternative measures of the spillover effect of internet access. However, Online

Appendix Table A8 indicates no significant spillover effects.28

EP

5. Opening the Computer’s “Black Box”

C

In this section, we explore some of the possible mechanisms that may explain why we do not
AC

observe significant impacts of internet access on the key child outcomes in our study. To do this,

27
Specifically, we estimate the following OLS regression equation:


 = " #$%&'()*(  + " #$%+,)-.,-)  +   +  + /! +  where we control for the number of
participating children who report child i as a close friend, Nijk, because children with more participating friends are
also more likely to have a lottery winner among their friends
28
We cannot rule out the possibility that internet-connected laptops generated positive externalities to everyone in
the school (regardless of how closely they were connected to the lottery winners). However, in previous work,
Beuermann et al. (2015) showed relatively little evidence of such spillovers for laptops without internet access.

22
 ACCEPTED MANUSCRIPT

we use traditional survey questions, as well as detailed time-diaries, and computer logs that

capture information about which laptop applications were used and which internet websites were

visited by the children who received XO laptops.

PT
5.1 Time Use

RI
Table 7 shows the impact of internet access on a broad set of activities measured through time

diaries applied to a random sub-sample of 837 children who were in 4th grade at baseline. These

SC
measures represent the number of minutes that the child reported being engaged in each activity

during the previous day. Columns 1, 2, and 3 show that time spent on computer or laptop is

U
substantially lower than the time spent watching TV, doing homework, playing without a
AN
computer or even doing domestic chores. Across all three groups, the reported time spent on a

computer or laptop ranges from 20 to 34 minutes per day.

M

Nevertheless, the provision of internet access does lead to significantly more time spent
D

using a computer or laptop based on the children reports. Relative to the children in the “No
TE

Laptop” group, those who were randomly assigned to receive laptops with internet used a

computer an additional 13 minutes per day, or 1.5 hours per week. This represents an increase of
EP

over 60 percent, and very similar to the impact on the reported number of minutes spent using a

computer specifically at home. Relative to the children in the “Laptop Only” group, the impact is
C

smaller and insignificant, at about 5 minutes per day, or half an hour per week. Still, despite the
AC

lack of significance, this represents an increase of over 15 percent.

[Table 7 here]

23
 ACCEPTED MANUSCRIPT

There is some evidence that providing internet access leads to substitution away from

other activities. We see that children in the “Laptop+Internet” group spend almost 30 minutes

less time watching TV as compared to those in the “Laptop Only” group. Meanwhile, children in

the “Laptop+Internet” group spend about 8 minutes less time doing homework as compared to

PT
those in the “No Laptop” group. However, there are also instances where exposure to internet

RI
appears to increase time spent on certain activities (e.g. domestic chores, working outside the

home), albeit not significantly. It is possible that internet access makes children more efficient at

SC
completing homework assignments such that they spend less time on homework and frees up

time for other activities. In terms of the type of computer use that children are engaged in, we

U
observe increases of 5 and 8 minutes per day on computer games relative to the “No Laptop” and
AN
“Laptop Only” groups, and an increase of about 7 minutes per day using computers for

homework as compared to the “No Laptop” group.

M
D

5.2 Type of computer and internet use

TE

We also examine how children use their computers and the internet in more detail in Table 8.

Compared to those in the “No Laptop” group, children who were randomly assigned to receive
EP

laptops with internet were significantly more likely to use their computers to do homework (10

percentage points) and to play games (11 percentage points). There were also positive impacts of
C

internet access over and above the effect of having a laptop itself in terms of using the computer
AC

to watch videos (11 percentage points) but a negative effect on using it to listen to music (10

percentage points).

[Table 8 here]

24
 ACCEPTED MANUSCRIPT

Regarding how children use the internet, we observe that those in the “Laptop+Internet”

group are significantly more likely to use the internet to search for information (13 percentage

points) and watch videos (11 percentage points) relative to those in the “Laptop Only” group.

PT
There is also evidence that those in the “Laptop+Internet” group were more prone to play online

RI
games, search information, and use educational programs than those in the “No Laptop” group.

SC
5.3 Activity Logs

In addition to time-use diaries and survey questions about time-use, we gathered more objective

U
assessments of computer use through log files which record the date and time when each
AN
application is opened. We focus on a measure of extensive use based on the fraction of days in

which a particular application, or set of applications, is opened.29 Furthermore, we classify

M

applications into four broad categories to facilitate the analysis: entertainment, learning,
D

information, and communication. However, because these log files were only available for 290
TE

children who received free laptops, this might not be a completely representative sample. Indeed,

Online Appendix Table A9 shows that children whose logs were obtained appear to have
EP

somewhat lower reading scores and higher computer access at baseline compared to their

counterparts without logs, although these differences are only marginally significant.30
C

Figure 2 uses these logs to show how computer use evolved over the course of our study
AC

for children who received laptops with and without internet access. During July/August 2011,

29
These logs also recorded the date and time when every application is closed so it is possible to estimate the
intensive margin of use in terms of minutes, although this is an upper bound because we cannot be certain that
children were actually using the computer throughout the time that an application remained open. Using this
alternative measure yields results that are broadly similar to those from extensive margin.
30
Online Appendix Table A10 also reveals some baseline differences between the “Laptop Only” and the
“Laptop+internet” treatments among children for whom we have computer logs, although only two (out of 19
estimated coefficients) are significant at the 10 percent level.

25
 ACCEPTED MANUSCRIPT

immediately after children received their laptops, laptop use was relatively high. Specifically,

laptops were used on average 40 percent of days. However, there was a steady decline over the

subsequent months so that, by December 2011, laptop use was only 16 percent. Following the

distribution of internet access to a subsample of children in July/August 2012, their levels of use

PT
increased sharply while those who did not receive internet access reduced use even further.

RI
These patterns are consistent with strong novelty effects, especially considering that laptop use

for those that received internet access also decreased over time. 31

SC
[Figure 2 here]

U
AN
The specific types of use are disaggregated in Table 9 which presents data based on logs

collected from July to November 2012 (i.e. after internet was provided). Children who received a
M

laptop without internet access opened computer applications on 17 percent of days. Among
D

children who also received internet access, this level of use was over 11 percentage points, or 65
TE

percent, higher. And unsurprisingly, the increase in use was dominated by internet applications.

Among the non-internet applications, those classified as entertainment represented the largest
EP

category of use, though closely followed by learning applications. Of course, even these learning

applications were educational games that were available on the XO platform chosen by the
C

Ministry of Education. The use of applications associated with either information or

AC

communication was substantially lower. These patterns are also disaggregated by gender and

baseline academic achievement in Online Appendix Table A11, and by prior computer and

internet access in Online Appendix Table A12.

31
We cannot rule out the possibility that the log files capture use by other family members. However, we do observe
that use during school days is concentrated after school hours (i.e. between 2pm-10pm).

26
 ACCEPTED MANUSCRIPT

[Table 9 here]

5.4 Internet Logs

PT
Our activity logs were supplemented by internet logs which recorded the date and time when

RI
each website was accessed. We constructed an analogous measure of use based on the fraction of

days on which a particular website, or set of websites, is opened. We also classified the websites

SC
into the same broad categories of entertainment, learning, information, and communication.

Table 10 describes the patterns of internet use among the 119 children who received internet

U
access, logs were recovered, and had effective internet activity within the period in which
AN
internet was provided. Table 10 also disaggregates internet use by gender, baseline academic

achievement, and baseline internet access.

M
D

[Table 10 here]
TE

As with the evidence from the activity logs, the largest category of internet websites is
EP

classified as entertainment, although communication is almost as large (and, indeed, identical

when rounding). The websites classified within the information and learning categories were
C

visited much less often. This is also reflected in the specific websites visited, with Facebook,
AC

Youtube, and Twitter being the most popular sites. The use of communication-related websites,

and Facebook in particular, was especially high among girls in our sample. Furthermore, children

who did not have prior access to the internet showed higher use in every category as compared to

those that already had access prior to our interventions.

27
 ACCEPTED MANUSCRIPT

6. Conclusion

This paper examines the effect of internet access on the development of children’s academic,

cognitive, and digital skills. We present findings from a randomized experiment in which free

PT
laptops and internet were provided for home use to children in Lima, Peru. These interventions

RI
were successful in increasing children’s exposure to technology at home and led to substantial

improvements in digital skills. We find that children who were randomly chosen to receive

SC
laptops with internet access showed higher computer and internet proficiency relative to those

who did not receive laptops. They also had higher internet proficiency compared to those who

U
received laptops without internet. On the other hand, we did not observe any significant impacts
AN
on academic achievement and on a large battery of cognitive skills. There were also no impacts

on children’s grades or on teacher’s assessments of their sociability, academic effort and

M

expectations of their eventual educational attainment.

D

We explore the reasons for the lack of impacts, showing that while computer and internet
TE

use do increase significantly following our respective interventions, there is a pronounced drop

in use over time. Moreover, computer use remains substantially lower than reported time spent
EP

watching TV, playing without a computer and doing domestic chores, and we do not find much

evidence of substitution away from these activities. We also find that the largest category of
C

computer use is entertainment. Thus, providing children with computers and internet at home
AC

appears to engage children in digital activities that are focused less on information or learning

and more on entertainment that do not translate into improved academic achievement, cognitive

or socio-emotional skills. It is important to recognize that our results are context-dependent.

Providing home internet access in other settings may produce positive (or negative) effects on

28
 ACCEPTED MANUSCRIPT

academic achievement and cognitive skills depending on how internet is used, whether there is

access to other learning platforms, and when children’s internet access is supervised more

closely.

Our results do indicate that providing children with access to computers and internet at

PT
home (together with some training) effectively closes the gap in digital skills between those with

RI
and without home computers and internet. Therefore, to the extent that improving children’s

digital skills is a relevant goal for an educational system, providing access to computers and

SC
internet at home may be one way to achieve this. However, it may also be possible to achieve

these gains at a lower cost. For example, Bet et al. (2014) show sizeable increases in digital skills

U
from relatively minor increases in access to shared computers at schools in Peru. There is also
AN
some evidence that the provision of school-based internet can generate gains in student learning

(Kho, Lakdawala, and Nakasone, 2018; Sprietsma, 2007). Perhaps the utilization of school-based
M

internet is monitored more closely than internet at home. In contrast, increased access to such
D

technology at home in circumstances similar to those analyzed in this paper, does not appear to
TE

improve academic achievement, cognitive or socio-emotional skills, which are arguably the more

important outcomes of such interventions.

EP

Acknowledgements
C

We want to especially thank Elena Arias Ortiz (IDB), Mariana Alfonso (IDB), and Jennelle
AC

Thompson (IDB) for their significant contributions. This project would not have been

materialized without the collaboration and commitment shown by the Dirección General de

Tecnologías Educativas in the Ministry of Education of Peru (DIGETE). We thank its former

directors Oscar Becerra and Sandro Marcone, respectively, and their excellent team including

29
 ACCEPTED MANUSCRIPT

among others Roberto Bustamente, Carmen Alvarez and Victor Castillo. We also thank Giuliana

Avila, Minoru Higa, Olga Namen, Elizabeth Rosales, Claudia Sugimaru, Diego Vera, and

Micaela Wensjoe for outstanding research assistance. Financial support from the Inter-American

Development Bank, USAID (Grant AID-527-F-12-00002) and the Spencer Foundation (Grant

PT
#201300081) is gratefully acknowledged. The views expressed in this paper are those of the

RI
authors and should not be attributed to the Inter-American Development Bank or any other

institution.

U SC
AN
M
D
TE
C EP
AC

30
 ACCEPTED MANUSCRIPT

References

Alexander, K., Pitcock, S. and M. Boulay (2016) The Summer Slide: What We Know and Can
Do About Summer Learning Loss. Teachers College Press, New York

Bai, Y., D. Mo, L. Zhang, M. Boswell, and S. Rozelle (2016). “The impact of integrating ICT
with teaching: Evidence from a randomized controlled trial in rural schools in China”.

PT
Computers & Education 96, 1-14.

Banerjee, A. et al. 2007. “Remedying Education: Evidence from Two Randomized Experiments

RI
in India.” Quarterly Journal of Economics 122: 1235–1264.

Bet, G., J. Cristia, and P. Ibarrarán. 2014. “The Effects of Shared School Technology Access on
Students’ Digital Skills in Peru.” IDB Working Paper WP-476. Washington, D.C., Inter-

SC
American Development Bank.

Beuermann, D., J. Cristia, S. Cueto, O. Malamud, and Y. Cruz-Aguayo. 2015. “One Laptop per

U
Child at Home: Short-Term Impacts from a Randomized Experiment in Peru.” American
Economic Journal: Applied Economics 7(2): 1-29.
AN
Cristia, J., P. Ibarrarán, S. Cueto, A. Santiago, and E. Severín. 2017. “Technology and Child
Development: Evidence from the One Laptop per Child Program.” American Economic Journal:
Applied Economics 9(3): 295 - 320.
M

Faber, B., R. Sanchis-Guarner, and F. Weinhardt. 2016. “ICT and Education: Evidence from
Student Home Addresses”. NBER Working Paper No. 21306
D

Fairlie, R. and J. Robinson. 2013. “Experimental Evidence on the Effects of Home Computers on
Academic Achievement among Schoolchildren.” American Economic Journal: Applied
TE

Economics 5(3): 211–240.

Fairlie, R. and A. Kalil. 2016. “The Effects of Computers on Children’s Social Development and
EP

School Participation: Evidence from a Randomized Control Experiment.” NBER Working Paper
22907. Cambridge, United States: National Bureau of Economic Research.

Johnson, G. (2006) “Internet Use and Cognitive Development: A Theoretical Framework” E-

C

Learning 3(4): 565-573

AC

Kho, K., L.K. Lakdawala, and E. Nakasone (2018). “Impact of Internet Access on Student
Learning in Peruvian Schools.” Unpublished working paper.

Kraut R., Patterson M, Lundmark V, Kiesler S, Mukopadhyay T, Scherlis W. 1998. “Internet

paradox: a social technology that reduces social involvement and psychological well-being.”
American Psychologist 53:1017–1031

Linden, L. L. (2008). “Complement or substitute?: The effect of technology of student

achievement in India.” InfoDev Working Paper 22928, World Bank, Washington, DC.

31
 ACCEPTED MANUSCRIPT

Malamud, O. and C. Pop-Eleches, C. 2011. “Home Computer Use and the Development of
Human Capital.” Quarterly Journal of Economics 126: 987–1027.

Mills, K.L. (2014) “Effects of Internet use on the adolescent brain: despite popular claims,
experimental evidence remains scarce” Science & Society 18(8): 385-387.

Mo, D., J. Swinnen, L. Zhang, H. Yi, Q. Qu, M. Boswell, and S. Rozelle. 2013. “Can One-to-

PT
One Computing Reduce the Digital Divide and Educational Gap? The Case of Migrant Schools
in Beijing.” World Development 46: 14–29.

Mo, D., L. Zhang, J. Wang, W. Huang, Y. Shi, M. Boswell, and S. Rozelle (2015). “Persistence

RI
of learning gains from computer assisted learning: Experimental evidence from China.” Journal
of Computer Assisted Learning 31 (6), 562-581.

SC
Muralidharan, K., A. Singh, and A. J. Ganimian (2016). “Disrupting Education? Experimental
Evidence on Technology-Aided Instruction in India.” NBER Working Paper (22923).

OECD (2017), PISA 2015 Results (Volume III): Students’ Well-Being, PISA, OECD

U
Publishing, Paris. http://dx.doi.org/10.1787/9789264273856-en
AN
Sprietsma, M. (2007). “Computers as Pedagogical Tools in Brazil: A Pseudo-Panel Analysis.”
ZEW Discussion Paper 07-040, Centre for European Economic Research, Mannheim, Germany.
M

Vigdor, J., H. Ladd, and E. Martinez. 2014. “Scaling the Digital Divide: Home Computer
Technology and Student Achievement.” Economic Inquiry 52: 1103–1119.
D
TE
C EP
AC

32
 PT
Figure 1. Sample Composition

RI
Final sample: students who at least completed
one instrument in November 2012

C
Received laptop +
internet

US
250

Won laptop + Won laptop + Received laptop

AN
internet internet only
354 310 50

Did not receive

M
laptop
Won laptop 10
540

D
3rd - 5th grade Participated in Received laptop +
at baseline laptop lottery internet
4,847 2,457
TE Won laptop Won laptop
0

Received laptop
EP
only only only
186 163 154
C

Did not receive

laptop
AC

Did not win Did not win Did not receive

laptop laptop laptop
1,917 1,653 1,653
AC
C EP
TE
D
M
AN
US
C RI
PT
 ACCEPTED MANUSCRIPT

Highlights
• Providing home internet access to children improves their digital proficiency
• Home internet access does not affect children’s math and reading achievement
• Home internet access does not affect children’s cognitive skills or self-esteem
• Children with home internet focused more on entertainment than on learning

PT
RI
U SC
AN
M
D
TE
C EP
AC