6.2 Rationale for a Large Scale Study of the XO Laptop

Proposal for a Randomized Evaluation of
Laptops in Nigerian Schools
Faisal Anwar and Lauren Velazquez
A-165 (Economics of Education) Final Paper
May 9, 2008
1
Contents
EXECUTIVE SUMMARY .................................................................................................................................. 4
1.
Introduction and Problem Statement ................................................................................................... 7
2.
The Nigerian Context ............................................................................................................................ 7
2.1 Education in Nigeria ............................................................................................................................ 7
2.2 Education Finance and Governance ................................................................................................... 8
2.3 Nigerian Access to ICT ......................................................................................................................... 8
3.
Our Proposed Intervention ................................................................................................................... 9
3.1 About the XO Laptop from OLPC ........................................................................................................ 9
3.2 Organizational Model for the Intervention....................................................................................... 10
4.
The Promise of ICTs in Nigerian Education and Society...................................................................... 11
5.
Literature Review of Other Educational Interventions ....................................................................... 12
5.1 Technology-Based Interventions ...................................................................................................... 12
5.2 Other Interventions that Demonstrate Best Practices for Evaluation .............................................. 14
5.3 Key Results from Impact Evaluation ................................................................................................. 15
6.
Why a Large-Scale Evaluation of Laptops Makes Sense ..................................................................... 16
6.1
Setting the Context for Cost-Benefit Analysis of the Laptop ...................................................... 16
6.2 Rationale for a Large Scale Study of the XO Laptop ......................................................................... 17
7.
Experimental Design ........................................................................................................................... 18
7.1 Fundamental Research Questions .................................................................................................... 18
2
7.2 Outcomes of Interest ........................................................................................................................ 19
7.3 Defining Our Target and Sample Populations ................................................................................... 20
7.4 The Different Treatment Groups ...................................................................................................... 22
7.5 Randomization Strategy .................................................................................................................... 23
7.6 Important Operations Details ........................................................................................................... 24
7.7 Data Analysis ..................................................................................................................................... 27
7.8 Extensions ......................................................................................................................................... 28
8.
Conclusion ........................................................................................................................................... 30
Appendix A: Evaluation Timeline for Single-Year Study ............................................................................. 31
Summer 2008 (Prior to School Year)....................................................................................................... 31
School Year 2008-2009 (9th Graders) ...................................................................................................... 31
Summer 2009 .......................................................................................................................................... 32
Bibliography ................................................................................................................................................ 33
3
EXECUTIVE SUMMARY
The following proposal outlines an evaluation model to assess the impact of giving laptops and internet
access to secondary school children in Nigeria.
Background
Nigeria is currently one of the poorest nations in the world. According to the non-profit data
organization “Gapminder,”1 the average per capita income in Nigeria is only $1.33 a day. As part of a
global effort to eradicate poverty in nations like Nigeria, the UN has created targeted priorities for
government intervention. These “millennium development goals” ask nations to meet a broad range of
goals that relate to poverty and economic development. The millennium development project
specifically discusses the importance of enhancing educational attainment and informational
communication technology access in developing nations.2 Currently Nigeria is struggling with both of
these issues. Education in Nigeria today is woefully inadequate: less than 70% of the adult population
literate and less than 40% of the population finishes the 9th grade.
In addition to having a weak education system, Nigeria also has been unable to provide widespread
access to ICT: less than 0.6% of the population has computers. Our proposal evaluates the potential
benefits of government-funded procurement of XO $100 laptops in secondary schools to address these
concerns. We have created an evaluation model that will assess the impacts on student achievement,
school attendance, teacher attendance and ICT literacy from this intervention. In addition to looking at
these outcomes we can track future wages of program participants to understand the larger economic
ramifications of this intervention.
Proposed Intervention
To achieve gains in schooling and ICT access, we propose a program that will place laptops in secondary
schools throughout Nigeria. Some laptops will come with internet access, others will come with
educational software while other laptops will be provided to students as is. Students are free to use
laptops as they see fit and teachers can choose to integrate the laptops into their teaching if they desire.
1
www.gapminder.org
2
http://www.un.org/millenniumgoals/#
4
Promise of ICT
Employment of computers and other ICT in schools is based on the premise that access to technology
and global communication facilitate constructivist learning. Constructivist learning is based on the
premise that student-directed learning is the best way to achieve comprehension. Students can use the
built-in software on the laptops to complete class assignments in innovative ways. They can also tap in
to the informational power of the Internet to access knowledge beyond their immediate localities. Thus,
laptops provide an opportunity to complement classroom instruction with innovative student use of a
new technology.
A Randomized Evaluation for Laptops in Schools
In order to understand the impact of laptops on schooling and ICT literacy, we propose establishing a
randomized trial that will track cohorts of students within schools. Given the high stakes associated with
any scalable laptop program, we believe that investing in a thorough and relatively costly evaluation
process is warranted. Furthermore, making such an investment in a select few countries (like Nigeria)
can be done in a way that provides external validity to the broader developing world.
We will choose a random sample of high schools from across Nigeria. From this random sample, 6/7 of
the schools will receive some form of treatment – laptops that are configured with varying capabilities
and different distribution ratios for laptops per student. Ultimately, we hope that our randomized
design will help policymakers to answer some fundamental questions about the laptop program:

Do laptops in schools improve learning outcomes for students? Which outcomes (attendance,
literacy levels, ICT literacy, project-based and self-directed learning capacity) do they help to
improve the most?

In terms of achieving desirable learning outcomes, can laptops be shared among a group of 5
students, or must each child receive an individual unit?

How essential is internet access to improved education through laptops? Should policymakers
ensure that any laptop program starts with internet access, or are the educational gains through
internet connectivity insignificant?
5
In addition to gathering and analyzing this quantitative data, we will survey members from all sample
groups to assess other measures like ICT literacy, internet usage, satisfaction, time spent on work etc. By
using a combination of test and survey data we can measure a broad range of education outcomes in
order to assess the true revolutionary potential of laptops in schools.
Finally, we will outline several additional extensions to our experimental design that can help get at
larger issues that may be of interest to policy makers. One important extension is a proposed 10 year
study that would try to link laptop usage to employment outcomes. In a sense, this expanded study
would measure the final outcome that any educational intervention seeks to identify – the additional
value of schooling with the intervention in labor market outcomes.
6
1. Introduction and Problem Statement
Our proposed evaluation is intended to determine the impact of laptop computers in improving
educational outcomes. There are two fundamental problems that laptop-driven intervention is meant to
address. First, we shall show that the education system in Nigeria is in serious need of improvement,
especially at the secondary school level. Secondly, there is an acute need to develop competencies in
information and communications technologies (ICTs) among Nigerian students since this may be a
critical path to opening up economic opportunity for many of this nation’s poor.
We shall review research that shows the promise of computers in improving educational outcomes,
including technological literacy. This background will set the stage for motivating our proposed
intervention. The latter half of this proposal is devoted to detailing a randomized evaluation that
measures educational outcomes as defined by test scores, attendance rates and other variables. This
study will allow policymakers to decide whether a laptop-based intervention is indeed worthy of
investment in Nigerian secondary schools.
2. The Nigerian Context
Nigeria is currently not on target to fully meet its millennium development goals. One of the UN’s
priorities for promoting global partnerships is to have nations work “in cooperation with the private
sector, make available the benefits of new technologies— especially information and communications
technologies”. Nigeria is one of the 25 poorest nations in the world today, and has the 3rd largest
population of people living in poverty (as defined by living below $1 a day), behind only India and China.
As of 2004, 54% of Nigeria’s population lived below this poverty threshold-the national per capita GDP is
only $393 USD. To address poverty and promote growth, the UN has advocated for integration of
technology into developing societies and for improved education.
2.1 Education in Nigeria
Historically, Nigeria has had three separate education systems: a system for indigenous education, a
system for religious Islamic education and a system for secular “European style” education.3 Rural
communities often relied on indigenous systems that stressed agrarian and trade education. As formal
3
http://www.onlinenigeria.com/education/index.asp
7
education increasingly became a necessity, Nigeria grew its investment in education. Eventually,
schooling became the largest social investment made by the government. By the 1990s, more than 17
million children attended primary or secondary school in Nigeria.4 With respect to teacher salaries,
Nigeria spends significantly less than its neighbors, but it has dramatically improved pay over the past
decade.
While Nigeria has stressed education in recent years, levels of school completion and school attendance
remain worrisome. According to the World Bank, less than 40% of the Nigerian population stayed in
school until grade 9.5 A strong indicator of Nigeria’s education troubles is the adult literacy rate: less
than 70% of the adult population are literate today in Nigeria.6
2.2 Education Finance and Governance
The funding (and hence governance) of public education in Nigeria varies according to level. Primary
education has received the bulk of its support
from local governments, especially in recent years
as the federal government has pulled back its
subsidies and the local governments have been
asked to bear their constitutional responsibilities
to basic education. Our intervention and
evaluation focuses on secondary education, which
is largely funded by state governments. The
federal government also provides some support,
but local governments, in general, do not
contribute sizeable funds to secondary schools in Nigeria (Hinchliffe, 2002). Figure 1 breaks down
education funding for secondary schools in Nigeria.
2.3 Nigerian Access to ICT
As a developing nation, Nigeria has had limited access to technology. As of 2004, slightly over 1% of
Nigerians even had access to the internet. As a developing nation, Nigeria ranks near the bottom quartile
4
ibid
http://www.worldbank.org/research/projects/edattain/edattain.htm
6
http://www.unicef.org/infobycountry/nigeria.html
5
8
with regards to access to information and communication technology. While the national population
exceeds 140 million, there is only an estimated 860,000 computers nationally. Nigeria is beginning to
distribute computers but there is no guarantee that once individuals have access to computer
technology that they will be able to utilize technology and integrate it into communities to improve the
efficiency of Nigerian society. Research beyond our proposed evaluation is needed to determine how
capable Nigerians are in utilizing ICTs and whether specific educational interventions can address this
need more specifically.
3. Our Proposed Intervention
The intervention we will evaluate is to supply schoolchildren in Nigeria with laptops that are preloaded
with educational software. A core purpose of the evaluation is to test several different hardware and
software features that will determine the final configuration for any intervention that is rolled out at a
large scale. The one sure thing about the final intervention (and all treatment groups for the evaluation)
is the XO Laptop that is supplied by the One Laptop Per Child (OLPC) foundation.
3.1 About the XO Laptop from OLPC
XO laptops have been designed specifically for schoolchildren in developing countries and therefore
have many features that will be useful in the Nigerian context. Each laptop is capable of networking with
other computers and the wider internet through a mesh network mechanism. Such connectivity is
unique because it allows users to communicate even in the absence of a wider internet infrastructure.
Such a capability will be essential in places like rural Nigeria. The fact that each laptop can be manually
charged in the absence of an electrical outlet is also important for underdeveloped regions of Nigeria
(OLPC, 2008b).
Beyond the specific features of the laptop, the philosophy behind it development ensures that the
laptop is optimized for the best educational outcomes. While the OLPC project is associated with
introducing computers in schools, an essential part of the organization’s mission is to facilitate student
learning beyond the school walls:
“We believe the emerging world must leverage this resource by tapping into the children's
innate capacities to learn, share, and create on their own. Our answer to that challenge is
the XO laptop, a children's machine designed for ‘learning learning.’ (OLPC, 2008a)”
9
It is clear that a mobile computing solution for ICT literacy allows children greater opportunity to learn
on their own and to disseminate the benefits of educational technology to their families and local
communities. Furthermore, such a technology allows students to test their creativity and intellectual
abilities by applying the tools in the laptop to real problems in their communities. Our intervention will
encourage schools to use the laptops beyond the school building, but we may limit students’ abilities to
take laptops home with them based on logistical factors (especially for the evaluation stage).
3.2 Organizational Model for the Intervention
The basic business model for our intervention is to work with OLPC to procure laptops, but then to
actually define the program vision and manage implementation through a specialized non-profit agency.
This is a model similar to the Open Learning Exchange project in Nepal, where a non-profit organization
has created a curriculum program for
Nepalese schools that is distributed
through laptops purchased from OLPC
(OLE, 2008). Figure 2 illustrates the
arrangement we envision. The job of
the evaluation team will be to work
with the Laptops for Nigeria non-profit
and the state and national
governments of Nigeria to execute an
evaluation that will test the viability of
such an intervention. Since local
governments provide the majority of
secondary school funding in the public
sector (Hinchliffe, 2002), it is essential
that any evaluation and large-scale
intervention involve them from the start in order to succeed.
Another important question about the intervention itself is what the actual laptop capabilities will be. As
mentioned earlier, the configuration for any final program involving the XO laptop will depend on what
our evaluation finds regarding the value of different configurations. Our evaluation is designed to test
10
research questions that pertain to the efficacy of the following features in a large-scale laptop
distribution program:

Individual laptops for each student.

Internet access for laptops in the school facility.

Learning software for different subject areas.
With respect to software, there are millions of options and configurations that are available. In our
evaluation, we will focus on a leading suite of math-enrichment software tools that will be bundled with
laptops and test whether these have a significant impact on math scores. All laptops will be equipped
with a basic suite of productivity software as well (word processing, calculators, web browsers, etc.).
4. The Promise of ICTs in Nigerian Education and Society
Logically, XO laptops clearly hold the promise of improving the quality of education in Nigeria by
widening access to education and learning, promoting retention and eliminating inequalities between
urban and rural localities. The goal of our evaluation is to determine whether this potential is a reality
for secondary school students in Nigeria. Several sources in the literature support our hypothesis that
new learning technologies, like the XO laptop provided by One Laptop Per Child (OLPC), will improve
educational outcomes. One of the most relevant lines of research for the XO laptop is with regards to
constructivist learning.
Constructivist learning is a pedagogy based on the idea that humans learn best through learner directed
exploration where teachers act as facilitators. This model of learning was championed by early
developmental psychologist Jean Piaget. Computers and laptops may hold unique potential to promote
constructivist learning since students can program things according to their interests and once online
they can seek out information that they desire when they desire it. XO laptops come equipped with
standard software (such as the Scratch programming environment) that may facilitate constructivist
learning. Several studies have found that constructivist learning approaches do promote higher
achievement in math and reading literacy (Law, Chanand, & Sachs, 2008; White-Clark, DiCarlo, &
Gilghriest, 2008; Zhang, 2008).
11
5. Literature Review of Other Educational Interventions
Beyond the research on constructivist learning, there is also significant research on a variety of other
educational interventions. This literature includes technological innovations (such as flip charts and
specially designed electronic devices) as well as policy innovations (such as contract teachers). Below,
we describe some of the most prominent studies and then list the key findings in a table. Ultimately,
policymakers will need to compare the relative costs and benefits found in the table at the end of this
section with the costs and benefits of an XO laptop program. Comparing both the magnitude and
breadth of improvement in educational outcome between these interventions should guide decision
makers as to what approach is best for their schools.
5.1 Technology-Based Interventions
Below are some studies that evaluate how technology interventions into classrooms correlate with
student achievement and other educational outcomes (school attendance etc). All of the studies below
evaluate programs in developing nations much like Nigeria. While numbers may not correlate exactly
with the expected outcomes and costs that would occur in Nigeria, these studies can provide insight into
which programs have proven to be the most effective and efficient in developing regions. When
evaluating OLPC, we will need to compare how much change we can anticipate in measured outcomes
over the cost per child to assess relative cost effectiveness.
5.1.1 Pic Talk
In India, students were given access to simple machines that pronounced words and displayed pictures to
match words. This tool was a supplement to language learning programs were teachers were often weak
in English. These machines were associated with positive gains in student test scores (.29 SD). When
calculated on a per student basis, this intervention cost approximately $1.31 per SD gain in test score
(Muralidharan, 2008). While this program is extremely cheap, its gains are significantly smaller than
other programs like Computer Assisted Learning (CAL) that is discussed below.
5.1.2 Flip Charts
In an evaluation of classroom inputs in Keya, researchers assessed the impact that the addition of simple
flipcharts into classrooms had on student achievement. Flipcharts, like textbooks, combine visual aids
and information as a teaching tool. Flipcharts were a seen as a desirable intervention because they are
12
cheaper than textbooks (Glewwe, Kremer, Moulin, & Zitzewitz, 2000). In fact, providing wall charts to
4th grade classrooms would cost approximately $80 while providing textbooks to each child in an
average 4th grade school would cost $800. In this study, 89 schools were randomly chosen to receive flip
charts that had drawings and explanations for specific subjects. 89 other schools where used as control.
The researchers compared test score data of schools that did and did not receive the flipcharts and
teacher training. The effects of these tools as an input for student learning range from .20-.05 SD
(depending on whether a retrospective OLS or a DID model is used). When a prospective analysis was
used (to control for omitted variable bias), the findings were not statistically significant. Even if the
intervention only cost $80 per school, it would not be cost effective to spend money on a program that
had no significant effect.
5.1.3 Laptop Usage in Israel
Unlike the other studies referenced here, Israel’s “Tomorrow 98 Programme” is the only program that
occurred in a developed nation. While Nigeria is not a developing nation, it, like Israel, has invested
heavily in computer access in schools already. The study done in Israel provides useful insight to
different types of computer interventions and their relative effectiveness.
In this program, treatment schools received funding for computer training programs, specific software
and additional hardware and computer upgrades (Angrist & Lavy, 2002). In order to receive funding,
schools had to apply for funding and demonstrate a preexisting computer program and a need to
expand. Researchers sampled 200 schools from a stratified sample. 122 schools were chosen from the
sample to receive treatment. In treatment schools, one fourth grade class and one 8th grade class were
chosen to take a math and language test. In addition to test score data, researchers gave a survey to
teachers to assess their use of other technology like overhead projectors in the classroom. Each
computer cost $3000 (computer, training, hardware and software). This study found a negative
correlation between the intervention and gains in math scores. Other subjects had no statistically
significant gains and no spillovers were detected. Given the high cost of this intervention and the
negative effects, the researchers concluded that the intervention was not cost effective.
5.1.4 Computer Assisted Learning in Prantham India
In the Indian city of Vodadora, researchers evaluated the impact that a computer assisted learning
program, run by the NGO Pratham, had on student achievement. In this study, treatment students were
given 2 hours of shared computer time weekly where they played math computer games. 100 schools in
13
the city received 4 computers as part of a government program. Two years after schools received these
computers, Pratham randomly chose half of the primary schools that had received computers to
participate in the Balsakhi Computer assisted learning (CAL) program. CAL schools were chosen from a
stratified sample that accounted for a schools grade level, gender, instruction language and Balsakhi
treatment status (Banerjee, Cole, Duflo, & Linden, 2005). CAL treatment schools received specific
teacher training on computers and were given the specially designed math software. Control schools
had computers but received no training or specialized software. In year 2 of the study, schools switched
so that those that did not receive treatment in year 1 did. These games adjusted to the individual
student’s ability and progress.
Evaluations of this program reveal that when the treatment of computer games were introduced, math
scores increased by .35 SD in year 1 and .47 SD in year 2 (Banerjee et al., 2005). This intervention was
shown to have no impact on school attendance or dropout rates. The CAL program cost approximately
$15.18 per student (when divided by the first year gain of .35 SD this cost is $43 per SD).
5.2 Other Interventions that Demonstrate Best Practices for Evaluation
5.2.1 De-worming
Hookworm infections effect more than 740 million people worldwide and can lead to anemia,
gastrointestinal diseases and difficulties in childbirth (Hotez et al., 2004). School aged children are most
vulnerable to hookworm infection. Researchers have long hypothesized that children do not attend
school or do not perform well in school because they are suffering from the symptoms of hookworm
infection.
One of the most notable studies evaluating this question occurred in Kenya in 1998. In this study, the
school related effects of deworming medication was studied for more than 30,000 school aged children.
Treatment was assigned at the school level to minimize externalities (since hookworm is so easily
transmitted if one student is dewormed in a school it is likely that the rate of transmission for other
children in the school will also fall). In this study, absenteeism fell by 25% but there were not notable
gains in test score achievement (Miguel & Kremer, 2004).
5.2.2 Contract Teachers
Several studies have been conducted to assess the impact of using contract teachers in classrooms
(either college educated or not) on student achievement. The rationale behind these evaluations is that
14
teacher salaries are extremely costly and oftentimes schools suffer from overcrowded classrooms and
teacher absenteeism. By hiring non-certified teachers, districts hope to reduce class size, provide
additional help to children and motivate teachers to attend school. The results for these studies vary
and are summarized in table 1.
5.3 Key Results from Impact Evaluation
Table 1 below provides a condensed summary of many of the prominent evaluations discussed above.
Table 1: Summary of prominent educational impact evaluations.
INTERVENTION
REGION
COST
EFFECT
EFFECT SIZE
Hookworm
Kenya
$3.50 per
Increased health and
.076 gain(females) and
child per
school attendance
.088 (mal
Gains in Math scores in
.284 mean increase in
both evaluation years
scores
increased
year
schooling
Contract Teacher
India
wth smaller gains in
literacy
Pic Talk
India
$1.31 per sd
Flip charts
Kenya
$81 per
No notable gains in test
school
scores
Subsidize
Israel
Computer software
.29 SD
.05-.2 SD
$3000 per
computer
and hardware
upgrades
Computer Assisted
Learning
India
$43 per SD
Increased math scores,
.35
no increases in
attendance
15
This table provides some context in which to assess the impact of our own intervention in Nigeria. We
do acknowledge that results may not be directly comparable because of different national contexts and
research designs. But, we feel it is important to use this data as a starting point for any policy discussion.
Up to this point, we have stated the basic educational problem that our XO laptop intervention seeks to
address. We have also established why there is a strong reason to believe that this intervention will have
a positive effect on many different educational outcomes. We have reviewed related literature and
provided a context within which cost-benefit analysis for the laptop intervention can occur. We can now
turn our attention to describing the evaluation that will help determine the actual value of our proposed
intervention. In the next section, we provide an argument as to why a sizable and potentially costly
evaluation is warranted for the type of intervention proposed. We then detail the key features of our
impact evaluation.
6. Why a Large-Scale Evaluation of Laptops Makes Sense
Before describing in detail the implementation of our evaluation, it is important to step back and
explore the high level goals that such an evaluation would serve. In particular, we would like to describe
the balance we are setting between the costs of an actual rollout of the XO Laptop and the costs of an
evaluation that studies the benefits of a laptop-based program.
6.1 Setting the Context for Cost-Benefit Analysis of the Laptop
In the developing nation context, the cost of any program involving computers will be quite significant.
In sub-saharan Africa, it is estimated that the per-pupil spending on education is less that $40 per year
(Brown, 2005). Limited data is available on the education spending in Nigeria, partly because reporting
of costs and enrollment figures are unreliable across many of the states. However, we can make some
basic estimates of spending using data provided by Hinchliffe (2002). The range in annual per pupil
spending at the secondary level in Nigeria varies from about $65 for public schools to approximately
$210 in private schools7. Households in Nigeria spend around $30 yearly to supply materials for children
attending secondary school in government institutions.
7
Per pupil spending at public secondary schools was reported to be N 3080, although there was more than a
doubling of spending across the board in education because of higher teacher salaries by 2002. We thus used a
conservative estimate of spending at N 7500 in 2002 and then converted to U.S. dollars. Per pupil spending at
private schools also varied by state, but was roughly N 24,000. converted to U.S. dollars.
16
Looking at the current levels of expenditure on secondary education, we can see why any laptop
initiative will need to be justified by exceptional results. Even an arrangement that provides one laptop
for every 5 children and no internet connection will cost at least $20 per child, not including
maintenance and administrative costs. This would nearly double household spending on materials for
education if families are asked to bear the costs of the laptop. If, instead, the government must bear the
cost as part of its larger funding for public schools, then this will represent at least a 30% increase in per
pupil spending at the secondary level. Of course, the $20 per pupil figure is a very optimistic scenario,
since maintenance and administrative costs have yet to be accounted for. Not only are the per pupil
costs of the laptop very high, but many current arrangements for introducing the XO laptop have
involved extremely large orders to bring the cost of each individual laptop in to the realm of
affordability. A recent scheme in Nigeria called for the purchase of 1 million XO laptops, but that has
stalled due to politics and questions about whether such a large investment is warranted (IRIN, 2007).
6.2 Rationale for a Large Scale Study of the XO Laptop
Policymakers have a responsibility to identify which interventions will be the best use of very limited
resources in education. The previous section makes it clear than any meaningful rollout of a laptop
program at the national level will be extremely expensive. It is also clear that for such an initiative to
make reasonable policy there must be substantial and broad gains in educational outcomes (especially
in comparison to the costs and benefits of other educational interventions, such as those listed earlier in
table 1). Our evaluation is intended to help policymakers make this essential cost-benefit analysis before
they invest precious resources in a laptop-based program.
Given the high stakes associated with any scalable laptop program, we believe that investing in a
thorough and relatively costly evaluation process is warranted. Furthermore, making such an investment
in a select few countries (like Nigeria) can be done in a way that provides external validity to the broader
developing world. Such a model is similar to the literature on contract teachers, where a few high
quality studies in developing countries have helped to shed light on the efficacy of contract teachers
across the developing world (Banerjee et al., 2005; Duflo, Dupas, & Kremer, 2007; Duthilleul, 2005;
Muralidharan & Sundararaman, 2006). Our vision is therefore to have a fairly comprehensive and large
study in Nigeria that will not only shed light on whether purchasing millions of laptops in that country is
reasonable, but also whether such a policy makes sense in other developing nations. If funding is
available, we also suggest extending the evaluation to a ten year longitudinal analysis of the laptop
17
program so that long term benefits in labor market employment and wages can be accounted for along
with shorter term educational outcomes.
7. Experimental Design
Given the background and intervention described earlier, we can now detail the plans for our
evaluation. The goal of this evaluation is to determine whether Nigeria, the Laptops for Nigeria NGO and
state governments in Nigeria should go ahead with a laptop program in their schools. This evaluation
also seeks to inform these entities about what a final intervention should look like by determining which
features are essential to meaningful learning outcomes through the laptop and which features are not
worth the cost.
As argued in the previous section, our plan is ambitious since it is meant to inform extremely large and
disruptive investments in Nigeria and in the broader developing world. Our basic evaluation framework
is a randomized control trial, since this technique is currently one of the most robust methods for
identifying causal impacts on educational outcomes. We considered alternative approaches, such as
difference in difference analysis or regression discontinuity designs, but felt that randomized trials are
the most robust mechanism to determine causality. Given our willingness to spend a sizable amount of
money on the evaluation itself, we believe that the potentially higher cost of randomized trials is not an
issue. Under the randomized trial framework, we will need to test our research questions by randomly
assigning units to different treatment and control groups and then measuring our outcomes across our
sample. The rest of this section provides details on our randomized study.
We describe an ambitious study, but leave out some of the most ambitious and complex extensions
from the main part of the experimental design. In the last part of this section, we list s additional
research questions that can be answered through some of these extensions and how our study would
need to be modified to answer the additional questions.
7.1 Fundamental Research Questions
Our evaluation is designed to answer several research questions that are all intended to help
policymakers decide whether laptops are a cost-effective mechanism for improving learning outcomes,
especially in comparison to other educational interventions that have been studied in the developing
country context (de-worming, contract teachers, voucher programs, etc.):
18

Do laptops in schools improve learning outcomes for students? Which outcomes (attendance,
literacy levels, ICT literacy, project-based and self-directed learning capacity) do they help to
improve the most?

In terms of achieving desirable learning outcomes, can laptops be shared among a group of 5
students, or must each child receive an individual unit?

How essential is internet access to improved education through laptops? Should policymakers
ensure that any laptop program starts with internet access, or are the educational gains through
internet connectivity insignificant?
Through these research questions, we not only tackle the broader issue of laptop viability in developing
nations’ schools, but we also seek to understand what components of computing technology are most
vital to improving student gains.
7.2 Outcomes of Interest
Table 2 below lists the different outcomes that our evaluation will attempt to measure. The central
outcome for our study will be the score on a composite math and reading examination designed to
measure basic and higher level skills. We can then split this score across the reading and math sections
to get our outcome values for the individual math and reading categories. We intend to hire consultants
to design an appropriate test that will measure these outcomes, similar to the strategy employed by
Muralidharan (2006).
Despite the focus on a single test for our core outcome, this evaluation seeks to examine the impact of
laptops across a broad range of possible changes in learning. Potentially expensive and revolutionary
interventions like computers in the classroom should be evaluated based on their impact across a wide
spectrum of learning outcomes, not just a single test score. This is why we intend to closely track all of
the outcomes listed in table 2 and to present the whole breadth of our findings to policymakers at the
conclusion of the evaluation.
19
Table 2: Outcomes of interest for OLPC evaluation.
OUTCOME
Attendance
MEASUREMENT TOOL
School Records
Reading Scores
Examination
Math Scores
Examination
ICT Literacy
Team-Oriented Learning
Examination
Random classroom visits
and survey data.
Project-Based Learning
Random classroom visits
and survey data.
Self-Directed Learning
Disciplinary Incidents
Homework Assignment
Rate
Homework Completion
Rate
Teacher Attendance
Classroom visit and
observations.
Random classroom visits
and survey data.
Survey data
Survey data
Survey data
DESCRIPTION
Obtained by disaggregating a composite
score on a comprehensive examination.
Obtained by disaggregating a composite
score on a comprehensive examination.
Test will ask students to complete basic
tasks (writing a letter, copying and moving
files, etc.) on computers other than the
types they have been provided.
Visits and surveys will help to reinforce final
conclusion regarding the level of teamwork.
Visits and surveys will help to reinforce final
conclusion regarding the amount of projectbased learning
Field workers will sit in on a class and then
ask students if they can think of ways to
extend concepts being learned (how would
they try to learn about something they
haven’t yet been exposed to).
Sample process variable that is also
measured by field workers.
Sample process variable that is also
measured by field workers.
Sample process variable that is also
measured by field workers.
Sample process variable that is also
measured by field workers.
To help us ensure that our sample is balanced prior to running our evaluation and to increase the power
of our experimental design, we intend to collect baseline data for test scores and attendance rates. If
the funds are available, then we will collect baseline data for the survey-measured outcome variables as
well. Appendix A gives a rough timeline of how we anticipate carrying out our measurements.
7.3 Defining Our Target and Sample Populations
We are primarily concerned with answering our research questions for individual students. However, we
are choosing to cluster at the school level for several important reasons. First, evaluating our outcomes
at the school level helps to control one common threat to experimental validity: spillover effects from
students who do receive the laptop to those who do not. Secondly, we may not even be sure who is
20
using the laptop in a student distribution scheme since students may actually share laptops with their
peers in school. If there were treatment and control students in the same school, then it would be quite
likely that students in the treatment group would use their laptops less than measured and more likely
that students in the control group managed to receive some of the treatment (access to the laptops).
Furthermore, distributing each laptop individually would be logistically difficult and expensive. Classes
would have some students who do have the laptop and others who do not, so instruction could not be
re-optimized to make best use of the new technology. Some students and teachers may just perceive a
random assignment of laptops to some subset of the students in school as being unfair.
Thus, there is ample justification to distribute laptops at the school level. Our evaluation will focus on
public secondary schools in Nigeria. Secondary schools across the developing world (and in Nigeria in
particular) have much lower rates of attendance than primary schools while costing the government
more money per pupil to operate. It is thus more likely that a laptop intervention will be identified as
cost-effective at the secondary school level.
Furthermore, we also believe that the laptop may benefit secondary students more than primary school
students. The educational goals of secondary school go beyond just introducing basic math and reading
literacy. Such students should begin to develop higher order thinking skills while also sharpening the
basic skills that they learned in high school. Thus, we can test this population for its proficiency in basic
skills continued since primary schools and additional skills (teamwork, project-based learning, selfdirected learning, etc.) that are developed at the secondary level.
Our evaluation sample will be randomly drawn from the available public secondary schools in Nigeria.
This sample will then be split randomly in to one control group and several treatment groups. To
determine the optimal size for the control and treatment groups, we conducted a power analysis using
Optimal Design software.

We estimated the average number of pupils per 9th grade cohort in each school to be 150
students.

Our key outcome variable is the composite examination score, although we would also like to
detect differences in our other outcome variables as well. For simplicity, we set our target effect
size to be 0.2 standard deviations for our key outcome and for the other additional outcomes.
21

We would like to achieve a power of 80% at a significance level of 0.05.

We estimate the intra-cluster correlation (within each school) to be 0.1.
Given these parameters, we would require a minimum of 86 sample points for a one treatment study
(43 samples in the control and 43 in the treatment). We expect to have 6 treatment groups in addition
to 1 control group, so this would require 7*43 = 301 participating schools. Since our effect size is fairly
conservative (at 0.2 SD instead of something much larger like 0.5 SD), we believe that this sample size
will give us more than enough power to identify differential impacts to all our outcome variables,
including the composite test scores.
7.4 The Different Treatment Groups
To help answer our research questions, we will break up our final sample in to one control and several
treatment groups. Our final breakdown will look like what is shown in table 3.
Table 3: Control and treatment groups for randomized evaluation.
No Laptop
No Internet or Addt'l
Software
One Laptop per Child
CONTROL GROUP
(43 Schools)
One Laptop per 5 Children
43 Schools
43 Schools
Internet
43 Schools
43 Schools
Additional Learning
Software
43 Schools
43 Schools
7.4.1 Individual or Shared Laptop Distribution
Having separate treatments for one laptop per child and one for every 5 children will allow us to
determine whether individual and shared laptop arrangements result in different student outcomes. We
can also compare these two treatment groups to the comparison group to test whether there is indeed
a positive impact of laptops at all.
7.4.2 Internet Access
We also want to test whether Internet access is an essential component for computing technology in the
classroom and whether an investment in computers must be coupled with an investment in Internet
22
capacity. The Internet has the potential to expose students to information from across the world, both
supplementing and complementing the learning opportunities they have through their instructors and
curricular materials. We can determine whether students do indeed take advantage of the information
available on the Internet for learning by comparing students who receive internet access with those who
have laptops without internet.
7.4.3 Additional Learning Software
There are literally millions of software configurations that we can test on any computing machine. For
the sake of cost and simplicity, we will choose one leading software package intended to support high
school mathematics. We will test the effect of this software on the various learning outcomes, focusing
specifically on math literacy. Our goal is not to provide an exhaustive set of conclusions about the value
of all software, but to use this evaluation as a starting point for further study of specific software
packages and their value to education.
For Internet access and additional learning software, we could have created a design that tests all
possible combinations of the two features. For the purposes of simplicity and cost, we chose to test
each individual feature alone and the control status of having neither feature. This leaves out the case
where a computer is equipped with both Internet and additional learning software. We feel that just
testing each feature alone will provide a good sense of its individual value and that there is little reason
right now to believe that learning software will be greatly enhanced by Internet access or vice versa. If
there is software that requires internet access to be effective (such as software that downloads
questions or materials from the web) or if there is extra funding to test the combination, then we can
easily extend the design above to also include cells for both additional software and Internet
connectivity.
7.5 Randomization Strategy
Figure 3 below outlines our basic randomization strategy. As discussed earlier, our target population is
9th graders in all Nigerian public high schools. In order to conduct the evaluation, we will need the
cooperation of the national and state governments as well as the individual schools themselves. We
presume that we will get permission from the national government since otherwise this study will
simply not be possible. For state governments and individual schools, we intend to seek out the
permission of as many of them as possible. This is to help us ensure that our pool of potential
participants is close to the larger target population.
23
By having most or all schools willing to
participate, we can then randomize across
schools and ensure that our final evaluation will
have external validity (at least for Nigerian
public high schools). In such a situation, our
evaluation sample of size 301 would come
through random selection from the participants
who have agreed to participate. Once we have
our evaluation sample, we will then randomize
the assignment of each school to either the
control or one of the 6 different treatment
categories listed in table 3.
We will randomize and assign right after we measure our baseline data (refer again to our timeline in
Appendix A). In this way, we can use the baseline data to ensure that our randomized samples are
balanced across baseline outcome characteristics and any other variables of interest. We would check
for this balance before assigning each randomly selected group to treatment or control status. If we are
comfortable with the sample balance across the groups, then we will assign each group to one of the 7
cells in our study. If not, then we will re-randomize until we do have 7 balanced samples.
7.6 Important Operations Details
At this point, we have laid out the basic research questions for our experiment and the broad theoretical
framework (randomized experiments) under which we will evaluate the impact of laptops in schools. To
ensure that our conclusions are valid, we must now address essential details about operations and
anticipated threats to validity.
7.6.1 Data Collection
Our evaluation timeline provides a clear description of our data collection strategy. We will be using
three key instruments to measure our outcome variables: examinations, class observations by field
researchers, and survey data distributed to participants. We will focus on the cohort of 9th graders at
each high school. Since many students change schools once they jump to the secondary level, it is not
feasible to collect student data prior to their beginning the 9th grade. We will therefore distribute
baseline exams and conduct any additional baseline analysis within the first month of the school year.
24
Once we have collected our baseline data, we will randomize our evaluation sample into 7 cells. Then
we will check for sample balance across all the cells and re-randomize if it is unbalanced on one of the
variables we are tracking. Once we are sure that our cells are balanced, we will randomly assign them to
treatment and control groups and distribute the treatment. We will then give schools some time to get
accustomed to the use of the laptops before we conduct class observations and distribute posttreatment surveys and exams.
Our key outcome measure will be a composite score on a math and readings skills exam that we design
with the help of education researchers and local assessment firms. As is the case with exams like the SAT
in the United States, the scores can be disaggregated across subject as well. Unlike the SAT, however,
our exam will not be a norm-referenced test; rather, it will report a raw score that measures proficiency
in math and verbal skills. We intend to also categorize each question based on the content that it tests
and the type of skill that it measures (basic knowledge or higher order thinking). In our final analysis, we
can then drill down and see how different control and treatment groups did on different types of
questions. Later in this proposal, we describe several extensions that can be done to the study to answer
additional research questions and how these extensions may affect the operations of our evaluation.
The next parts in this section address specific threats to validity that further inform our data collection
and operations plans.
7.6.2 Piloting our Study in a Select Few Secondary Schools
If time and money allow, we would like to delay the actual evaluation outlined in appendix A by 1 year
and use the 2008-2009 cohort of 9th graders to ensure the quality of our operations and data collection
procedures. In such a scenario, we would select 7 schools that will not be in the evaluation sample and
run a test experiment on them for 1 year. Such a pilot allow us to ensure exams are fairly distributed,
laptops are correctly working, support for educators can be made available and any non-compliance is
identified, prevented and accounted for in the real study. Furthermore, we can use this first year to help
us get better estimates for parameters like the intra-cluster correlation that will go into our power
analysis and sample size selection.
7.6.3 Preventing Spillover Effects of Treatments
Spillover effects where students in the control schools are exposed to students in the treatment schools
may lead to underestimation of treatment effects if the laptops do indeed improve performance. We
have tried to prevent such a scenario by clustering at the school level, so students within a school all
25
have the same treatment status. We hope that a pilot will help us to determine whether there are still
problems with students sharing laptops when they take them home or if other issues of lost or stolen
laptops arise that may present attrition biases. Based on what we see during our pilot study, we will
determine whether students should be allowed to take laptops home with them or if the computers
must be kept within the school building (in secured cases) and used only during class.
7.6.4 Managing Attrition in Participating Schools
Several scenarios can lead to students in a treatment group not receiving exposure to laptops. One issue
will arise when students lose their laptops. In such a case, our policy will be to ask the student to look on
with others in class. In our final analysis, we will also keep track of how many student lost their laptops
as an outcome. This will allow us to calculate additional costs for the laptop program and also to adjust
our findings based on whether many students in treatment groups did not have access to their
computers.
Lost or stolen laptops that are not replaced are not the only risks for attrition or non-compliance. The
biggest source of attrition will be students dropping out of school. This can threaten the validity of our
experimental conclusions, especially if there is some systematic relationship between dropouts and the
various cells in our study. Typically, the weakest or most disadvantaged students can be expected to
drop out at the highest rates and struggle the most on assessments. If, however, the presence of laptops
keeps more of these students in school, then we risk underestimating our treatment effects as weaker
students drop out from the control schools in greater numbers as compared to treatment schools.
We will use incentives to minimize the number of students who participate in the study but do not
complete post-treatment assessments and surveys. All students (whether they are still in school or not)
will receive a relatively sizeable monetary reward (in the range of $5, which is substantial in Nigeria).
This reward will be communicated to students only after everyone has taken baseline exams and been
assigned to treatment or control groups. We believe this incentive structure is optimal for measuring the
impact of the laptop without substantially altering the dropout dynamics at participating schools.
Students may still dropout under this arrangement, but we ensure that close to everyone is measured
whether or not the student has remained in school.
7.6.5 Preventing Unauthorized Use of Computers
Another possible issue to consider is whether students in some groups will end up playing games or
using the computers for other unauthorized activity. We could potentially lock computers from installing
26
new programs and only have a specific predefined suite of software on each laptop. However, the
majority of games and distractions that students can engage in are online, especially with more
advanced Web 2.0 technologies that allow for software to run within web browsers. So, cells that have
internet access will have access to a limitless supply of sites that cannot all be tracked with a reasonable
cost.
We believe the best approach is to acknowledge this as a potential phenomenon among our treatment
groups, especially those that are given internet access, and to include it as part of our research inquiry.
We can still lock computers that don’t have internet access so that students do not install unauthorized
software. However, we must expect that students with internet access will be able to preoccupy
themselves with both educational and non-educational websites. Our results for the group of schools
that have internet access will guide further research on the impact of the internet on student outcomes.
If their learning outcomes are lower than the other groups in the study (including the control), then we
can at least conclude that internet usage is a distraction from school work. In this situation, further
measures need to be devised to determine why students are not performing as well and how they can
be motivated to use the internet for positive gains.
7.7 Data Analysis
Once we have all completed our evaluation, we can conduct data analysis to determine the effect of the
treatment. We have already discussed how we will use our baseline scores as well as any additional
variables (particularly any that are chosen to stratify our sample) to determine whether our sample is
balanced. A simple approach to quantifying effect size is to simply measure the means of different
outcome variables in each cell and then perform t-tests to determine statistical difference from other
cells. We can also implement this approach through a more sophisticated regression framework where
we create dummy variables for the different treatment options and then calculate parameter estimates
that represent the average treatment effect for each scenario (Duflo, Dupas et al., 2007; Muralidharan &
Sundararaman, 2006).
It will also be helpful to drill down in our data across relevant subgroups. One possible avenue for
exploration would be to group students according to baseline characteristics and see if there are
heterogeneous treatment effects based on initial levels of performance and attendance. For example,
we could follow those students who scored among the top 10% in baseline exams and those students
who were among the bottom 10% and determine how their scores changed after the treatment. This
27
would allow us to answer whether any of the treatments disproportionately benefit weaker students
over stronger students (or vice versa). As our next section on extensions discuss, we can define several
different variables on which to stratify our initial evaluation sample so that we can specifically track
differential effects across different subpopulations.
7.8 Extensions
We have presented a detailed outline of a randomized trial to test the effect of laptop computers in
schools. We posed a specific set of research questions that have important policy implications, and then
described a design that allows us to test each of the questions in our study. Due to budget constraints
and for the sake of simplicity, we did not include several additional research questions that can also be
answered with a few extensions to the study. In this section, we outline some additional research
questions that may be of interest to policymakers and provide a brief synopsis of how our study would
need to be modified to answer these additional questions.
7.8.1 Evaluating How Laptops Impact the Long Term Benefits of Education
It can be argued that all the outcome variables we have defined for our study are indeed process
variables, with the ultimate outcome being wage and quality of life measures that are not included in
typical evaluations. This evaluation can be extended over a period of 10 years to determine the impact
of laptop computers on life outcomes for a cohort of students. At the most fundamental level, we are
asking here whether laptops in school improve the employment and wage prospects of students.
Answering this question would allow policymakers to determine net present value of returns to
additional years of secondary education increases when schools have laptops. They can then perform a
more accurate cost-benefit analysis for procuring laptops that accounts for future returns.
To answer such a question, the control and treatment cohorts could be assigned and managed in the
same way as we have described already. The primary difference from our current study would be the
length of time that we would measure outcomes for students and the precise outcomes that we would
measure. The core outcomes would no longer be test scores, but would be student incomes. We could
measure these incomes for a specific set of years (perhaps 4, 6, 8 and 10 years after the study). Our
analysis would focus on determining whether the average incomes for students with and without
laptops differ significantly (through a regression framework). Since we have 6 different treatment
groups, we could also determine whether any of the specific configurations that were set up during high
school had a particularly strong impact on incomes.
28
7.8.2 Evaluating the Short and Long Term Impacts of Laptop Access
Another research question is whether exposure to laptops, internet and learning software has long term
as well as short term educational benefits. Do any gains in student achievement persist after students
have not used the laptop for some time? To answer this question, we could continue the laptop
program for half of the treated schools in the 10th grade. In such a scenario, we would have three groups
to compare as the initial cohort of students in the study completes the 10th grade:
1. Students in the comparison group who never received a laptop.
2. Students who received laptops as 9th graders, but did not have any during the 10th grade.
3. Students who received laptops in both the 9th and 10th grades.
In such a setup, we could compare groups (1) and (2) to determine if students who have not used
laptops recently have any persistent educational gains over students who never received them to begin
with. We could also compare the outcome means for groups (2) and (3) to determine how much more
benefit continuous use of laptops provides as compared to use during a single year. We could continue
to track students among the 6 different treatment cells to determine if there is any differential across
treatment types as well. Finally, we could compare groups (1) and (3) to determine whether laptops
have significant benefits after students have had them for longer than one year (just during the 9th
grade).
7.8.2 Stratifying on Other Relevant Variables
We can use stratification to ensure that certain underrepresented schools are included in our study.
Another benefit of stratification, however, is to help us compare the effects of the treatment on
different types of schools (Duflo, Glennerster, & Kremer, 2007). Policymakers may want to know
whether the gains of laptops are the same in rural and urban schools. To test such a question, we would
have two evaluation samples: one for urban schools and one for rural schools. We would randomize as
described earlier, except over both of these samples. Then we would conduct our experiment and use
our final results to help us determine the impact on rural and urban schools separately and Nigerian
public high schools as a whole (the latter would require combining the weighted means from our
individual strata). There are a multitude of other variables along which we can stratify, depending upon
29
the specific wishes of the state and local governments and the OLPC community: income levels for
students in schools, computer access outside of school, etc.
8. Conclusion
Computing technology for education, as promoted by OLPC, holds great promise for improving short
and long term educational outcomes. However, such technology remains extremely expensive in the
developing world, even if the cost has been reduced to $100 per unit. We have proposed a randomized
evaluation that can help policymakers determine whether laptops are truly a worthy investment for
secondary schools. Upon completion of this evaluation, we hope that policymakers will be able to have a
clear sense of the breadth and magnitude of benefits from computers in school and will be able to
compare these benefits to other educational interventions that have been tried in the developing world.
It’s important at this point to step back and remember that computers are not meant to simply improve
a single targeted educational outcome. Indeed, they have revolutionized nearly every other sector
where they have been employed. There is reason to believe, therefore, that a similar redefinition will
occur for schooling. We have tried to design our evaluation in a way that accounts for these systematic
changes by providing a diverse set of outcome measurements that will be tracked. We have also
proposed an extension to our evaluation that would track long-term student outcomes in terms of
employment outcomes. After all, the best way to determine whether a revolutionary approach to
education is actually working is to allow the labor market to decide between the quality of graduates.
30
Appendix A: Evaluation Timeline for Single-Year Study
This timeline is specific to an experiment that tests one cohort of students starting in the 9th grade. It
does not apply to the extension suggested where there are several grade-based cohorts to test for the
effect of the laptop on short-term and long-term learning outcomes. We are also assuming that a
standard school year runs from September thru June.
Summer 2008 (Prior to School Year)

Potential participants list compiled by gathering permission contracts from all relevant parties.

Randomized selection of evaluation sample from list of potential participants. Schools in
evaluation sample notified of their involvement in the study.
School Year 2008-2009 (9th Graders)
September, 2008

Baseline exams given to 9th graders. Other baseline data also collected.

Initial randomization of evaluation sample in to 7 cells. No assignment to treatment or control
made.

Baseline data compiled and entered into database by the end of the month.
October, 2008

Ensure sample balance across baseline data and any additional relevant variables. Once sample
is balanced, then randomly assign each cell to the control group or one of the 6 treatment
groups.

Distribute laptops and any additional equipment or software to all the treatment groups.
November, 2008 to March, 2009

Allow schools time to integrate technology and become accustomed to its use in the classroom.
Provide basic support for treatment schools to ensure that they know how to use the hardware
and software available.
April, 2009 to May, 2009

Send field teams to random schools unannounced. Have them record outcome measurements
that depend on classroom observation (project-based learning, discipline issues, etc.).
June, 2009

Conduct endline examinations for math and reading skills as well as for ICT literacy.

Distribute and collect surveys for teachers and students to measure additional outcomes.
31
Summer 2009

Upload all exam, observation and survey data in to database and begin analysis.

Depending on available funds, determine whether to continue evaluation for another year and
whether to also plan for a longer longitudinal analysis to follow wages and life outcomes (10
year study).
32
Bibliography
Angrist, J., & Lavy, V. (2002). New Evidence on Classroom Computers and Student Learning. The
Economic Journal, 112, 735-765.
Banerjee, A., Cole, S., Duflo, E., & Linden, L. (2005). Remedying Education: Evidence from Two
Randomized Experiments in India.
Brown, G. (2005). Remarks by the Rt Hon Gordon Brown MP, Chancellor of the Exchequer on Education
in Africa. Retrieved May 1, 2008, from http://www.hmtreasury.gov.uk/newsroom_and_speeches/speeches/chancellorexchequer/speech_chx_140105
_education.cfm
Duflo, E., Dupas, P., & Kremer, M. (2007). Peer Effects, Pupil-Teacher Ratios, and Teacher Incentives:
Evidence from a Randomized Evaluation in Kenya.
Duflo, E., Glennerster, R., & Kremer, M. (2007). Using Randomization in Development Economics
Research: A Toolkit. Centre for Economic Policy Research.
Duthilleul, Y. (2005). Lessons Learnt in the Use of Contract Teachers. Paris, France: UNESCO International
Institute for Educational Planning.
Glewwe, P., Kremer, M., Moulin, S., & Zitzewitz, E. (2000). Retrospective versus Prospective Analysis of
School Inputs: The Case of Flip Charts in Kenya. Cambridge, MA: National Bureau of Economic
Research.
Hinchliffe, K. (2002). Public Expenditures on Education in Nigeria: Issues, Estimates and Some
Implications. Washington, D.C.: The World Bank.
Hotez, P., Brooker, S., Bethony, J., Bottazzi, M., Loukas, A., & Xiao, S. (2004). Hookworm Infection. The
New England Journal of Medicine, 351(18), 799-807.
IRIN. (2007). Nigeria: Laptops-in-schools debate turns messy. Retrieved May 2, 2008, from
http://www.irinnews.org/Report.aspx?ReportId=76023
Law, Y.-K., Chanand, C., & Sachs, J. (2008). Beliefs about learning, self-regulated strategies and text
comprehension among Chinese children. Psychological British Journal of Educational Psychology,
78, 51-73.
Miguel, E., & Kremer, M. (2004). Worms: Identifying Impacts on Education and Health in the Presence of
Treatment Externalities. Econimetrica, 72(1), 159-217.
Muralidharan, K. (2008). A-165 Course Lecture for April 25, 2008.
Muralidharan, K., & Sundararaman, V. (2006). Teacher Incentives in Developing Countries: Experimental
Evidence from India.
OLE. (2008). Open Learning Exchange Nepal Website. Retrieved April 2, 2008, from http://olenepal.org/
OLPC. (2008a). One Laptop Per Child Mission Statement. Retrieved March 31, 2008, from
http://laptop.org/vision/mission/
OLPC. (2008b). One Laptop Per Child Wiki. Retrieved March 31, 2008, from
http://wiki.laptop.org/go/The_OLPC_Wiki
White-Clark, R., DiCarlo, M., & Gilghriest, N. (2008). Guide on the Side. The High School Journal(AprilMay 2008).
Zhang, L. (2008). Constructivist pedagogy in strategic reading instruction: exploring pathways to learner
development in the English as a Second Language (ESL) classroom. Intructional Science, 36, 89116.
33