School Choice International

School Choice International
School Choice International
Exploring Public-Private Partnerships
edited by Rajashri Chakrabarti and Paul E. Peterson
The MIT Press

Cambridge, Massachusetts
London, England
( 2009 Massachusetts Institute of Technology
All rights reserved. No part of this book may be reproduced in any form by any elec-
tronic or mechanical means (including photocopying, recording, or information storage
and retrieval) without permission in writing from the publisher.
For information about special quantity discounts, please email special_sales@mitpress

.mit.edu.
This book was set in Palatino on 3B2 by Asco Typesetters, Hong Kong.
Printed and bound in the United States of America.
Library of Congress Cataloging-in-Publication Data
School choice international : exploring public-private partnerships / edited by Rajashri

Chakrabarti and Paul E. Peterson.
p. cm.
Includes bibliographical references and index.
ISBN 978-0-262-03376-3 (hardcover : alk. paper)
1. School choice—Cross-cultural studies. 2. Public schools—Cross-cultural studies.
3. Charter schools—Cross-cultural studies. 4. Education and state—Cross-cultural
studies. I. Chakrabarti, Rajashri. II. Peterson, Paul E.
LB1027.9.S333 2009
379.1 0 11—dc22 2008007297
10 9 8 7 6 5 4 3 2 1
Contents
Preface vii
I School Choice: Worldwide Perspectives
1 Perspectives in Public-Private Partnerships in Education 3

Rajashri Chakrabarti and Paul E. Peterson
2 Public-Private Partnerships and Student Achievement:

A Cross-Country Analysis 13
Ludger Wößmann
3 Mobilizing the Private Sector in the United States:

A Theoretical Overview 47
Thomas J. Nechyba
4 The Practice of Public-Private Partnerships 71

Norman LaRocque
II Traditional Forms of School Choice
5 Public-Private Schools in Rural India 91

Karthik Muralidharan and Michael Kremer
6 School-Sector Effects on Student Achievement in India 111

Geeta G. Kingdon
III Public Funding of Privately Managed Schools
7 School Vouchers in Colombia 143

Eric Bettinger
vi Contents
8 The Public-Private School Controversy in Chile 165

Cristian Bellei
9 The Concession Schools of Bogotá, Colombia 193

Felipe Barrera-Osorio
10 Public and Private Schooling Initiatives in England 219

Stephen Machin and Joan Wilson
IV Conclusions
11 Education Contracting: Scope of Future Research 245

Harry A. Patrinos
List of Contributors 259

Index 261
Preface
The importance of school choice in education cannot be overempha-

sized and public-private partnerships play an important role in pro-
viding such choice all over the world. While they have received
considerable attention in the United States, research on such initiatives
in other parts of the world has been very limited. This book brings
together a set of essays that provide a comprehensive overview and
analysis of international initiatives of this kind—on the one hand, it
presents essays that conduct rigorous impact evaluations of these ini-
tiatives in other countries, and on the other hand, it presents essays
that provide comprehensive descriptions of the initiatives and discuss
their theoretical underpinnings.
The increased role of such initiatives in today’s education policy
arena, and the lack of adequate understanding of the effects of such ini-
tiatives, prompted Harvard’s Program on Education Policy and Gover-
nance and the World Bank to come together to organize a conference
on public-private partnerships in education. The event, held at Har-
vard University on October 5–6, 2005, attracted many of the foremost
economists, political scientists, policy makers, and practitioners with
an interest in education policy. Each paper presented at the conference
was critically discussed by practitioners and researchers specializing in
education policy. A selected set of papers presented at this conference
was subsequently revised in the light of these comments and editorial
suggestions; they form this volume.
We are indebted to many people who helped us in the preparation
of this volume. In addition to the authors, we would like to thank the
following conference participants, presenters, and discussants for their
valuable comments, suggestions, and critical discussions of the papers:
Joshua Angrist, Carl Bistany, Xiaonon Cao, Jorge Cela, Christopher
Cerf, John Chubb, Paul Dovre, Susan Dynarski, Guy Elena, Ronald
viii Preface
Ferguson, Chester Finn, Ariel Fiszbein, Roland Fryer, Edward Glaeser,

Charles Glenn, Stephen Goldsmith, Isabel Guerrero, Frederick Hess,
William Howell, Caroline Hoxby, Brian Jacob, Pablo Jaramillo, Thomas
Kane, Elizabeth King, Michael Latham, Bridget Terry Long, Frank
Lysy, José Mora, Neil McIntosh, Richard Murnane, Ronald Perkinson,
George Psacharopoulos, Richard Romano, Robert Taylor, Carlos Vélez,
Martin West, and Richard Zeckhauser. We would also like to thank the
minister of education of Colombia, Cecilia Maria Vélez White, for her
insightful speech at the conference.
We gratefully acknowledge financial support for the conference
from the Center for British Teachers (CfBT), International Finance Cor-
poration, John M. Olin Foundation, Harvard’s Program on Education
Policy and Governance, Taubman Center for State and Local Gov-
ernment at the John F. Kennedy School of Government of Harvard
University, the World Bank, and the World Bank Institute. We are
thankful to Antonio Wendland, associate director of the Program on
Education Policy and Governance, for his invaluable help in organiz-
ing the conference, and also to Mark Linnen of the Program on Educa-
tion Policy and Governance, and Suzanne Roddis of International
Finance Corporation for their able assistance.
I School Choice: Worldwide Perspectives
1 Perspectives in Public-Private Partnerships in
Education
Rajashri Chakrabarti and Paul E. Peterson
The undeniable importance of human capital for economic growth has

made its accumulation a top priority in developing and developed
countries alike. To further that goal, many have looked ever more
closely at the contributions that can be made by public-private partner-
ships in the field of education. Indeed, there has been a burgeoning of
such initiatives across the world—from Latin America to Asia, New
Zealand, and Australia to Europe and North America. They vary
importantly in their forms and structures as well as in the extent of
public and private participation. Understanding the different forms
of such partnerships and the role they play in the production and de-
livery of education is crucial, if broader policy implications are to be
teased out. Yet apart from initiatives in the United States, careful docu-
mentation and systematic evaluation of their impact has been scat-
tered, with very little attention paid to bringing together in a coherent
fashion the work that is being done. The essays in this volume are
designed to take some initial steps in this direction as well as to insti-
gate further exploration and research.
Resources and management are two of the main factors that give
shape to a school, Ludger Wößmann points out in chapter 2. Consider-
ing just these two critical factors, public-private partnerships can be
divided into four broad categories, as shown in figure 1.1—publicly
funded resources that are publicly managed, privately funded re-
sources that are privately managed, publicly funded resources that are
privately managed, and privately funded resources that are publicly
managed. Schools in the first group constitute the most conventional
form of public schools, while the second group constitutes the private
schools. Unlike these two categories, the third category of initiatives
requires both public and private participation. Some examples of such
partnerships discussed in this book are vouchers in the United States,
4 Rajashri Chakrabarti and Paul E. Peterson
Figure 1.1
A Typology of the Interconnection between Public and Private Sectors in Education
Chile, and Colombia; charter schools in the United States; concession

schools in Colombia; and city academies in the United Kingdom (see
figure 1.1). Publicly funded and privately managed schools are also
often considered to be public. For example, publicly funded schools
in Britain managed by churches or synagogues are regarded as pub-
lic schools. Charter schools in the United States are also considered
public schools, though the first group of schools (publicly funded, pub-
licly managed) is regarded as the traditional form. Unless other-
wise stated, ‘‘public schools’’ in this volume will refer to this first
group of schools. Publicly funded privately managed schools can in
turn be divided into two subcategories: schools where the nongovern-
mental authority approached the state for funding (example, many
schools in France are managed by the Catholic Church) and schools
where private authorities have been invited to manage. The second
subcategory includes charter schools, concession schools, and city aca-
demies. The final category of initiatives (privately funded, publicly
managed) is less common, and consists of government tuition schools
where the government manages the schools and families pay tuition
to attend them. As Wößmann points out, these are mainly found in
Mexico, but also to a lesser extent in Italy, New Zealand, Brazil, and
Greece.
Part I of this volume provides a series of broad, overarching assess-
ments of the promise that public-private partnerships provide. Parts
II and III focus on specific partnerships in particular countries. The
chapters in part II examine the most conventional division between
the public and private sector—publicly financed and managed schools
Perspectives in Public-Private Partnerships in Education 5
versus privately financed and managed schools. Chapters in part III

evaluate initiatives that are publicly funded but privately managed.1
Following this introductory chapter, part I consists of three chapters.
Chapter 2 provides a big-picture scenario of the efficacies of these
different forms of public-private partnerships. In an instructive cross-
country analysis using student-level data from the Program for Inter-
national Student Assessment (PISA) international test, Wößmann finds
that public school operation is associated with lower student out-
comes, but public school funding with higher student outcomes. Thus
public-private partnership systems that combine private operation
with public funding are the most efficient, while systems that combine
public operation with private funding are less so.
Chapter 3 by Thomas Nechyba provides a penetrating analysis of
the arguments behind mobilizing the private sector for public educa-
tion. He starts with the public school choice and private school choice
systems, and their efficiency and equity aspects of these systems. Then
he moves on to vouchers, the most common form of publicly funded
and privately managed initiative in education. He shows that the form
of the voucher and the underlying conditions in the economy are key
determinants of how vouchers affect schools and students—differences
in these factors can have vastly different effects on these agents. He
argues that effective design of voucher proposals can go a long way
to enhance both the efficiency and equity aspects of an educational
system.
Chapter 4 by Norman LaRocque provides an excellent overview of
international examples of different types of public-private partner-
ships. It is especially valuable for any policy maker or researcher inter-
ested in examples of public-private initiatives in education. These
initiatives differ in forms, structure, and scope, and hence are likely
to have different effects on schools, students, and teachers. LaRocque
considers experiences in Australia, Brazil, Colombia, Côte d’Ivoire,
Germany, Lebanon, the Netherlands, New Zealand, Philippines, Vene-
zuela, the United Kingdom, and the United States.
Part II of this volume brings together country-specific studies that
focus on the first two types of schools: public and private. Using data
from an elaborate nationally representative survey of rural private
schools in India, in chapter 5 Karthik Muralidharan and Michael
Kremer find that private schools are much less likely to be plagued
by teacher absenteeism than public schools, although private school
teacher salaries are considerably lower than those in public schools. A
possible answer to this apparent puzzle is that head teachers in private

schools are much more likely to take disciplinary actions against shirk-
ing teachers, while such actions are virtually absent in public schools.
The authors go on to compare school facilities, teacher characteristics,
student and parent characteristics, and student achievement in these
two types of schools, to consider relative efficacies of these two sectors
in providing education in India.
Chapter 6 in part II also focuses on India. Geeta G. Kingdon com-
pares the achievement and cost effectiveness of the different types of
schools in India: public schools, private schools, and aided (publicly
funded, privately managed) schools. She finds that private schools are
considerably better than public schools on both counts. She also argues
that the passage of several acts in the early 1970s led to a draining of
accountability from the aided schools so that they became virtually in-
distinguishable from the public schools.
Part III focuses on publicly funded, privately operated initiatives.
The most well known of such initiatives are publicly funded voucher
programs, and the most hotly debated ones are found in the United
States. The first such program in the United States, the Milwaukee Pa-
rental Choice Program (MPCP), was implemented in 1990. The pro-
gram made low-income families residing in the city of Milwaukee,
Wisconsin, eligible for vouchers that enabled them to move to private
schools. While the program started small, with less than 1 percent of
the city’s students, it has grown steadily in size. In the 2005–06 school
year around 15 percent of the students availed themselves of vouchers
under the program.
The Milwaukee program was closely followed by the Cleveland
Scholarship and Tutoring program in 1996 that made low-income stu-
dents in Cleveland, Ohio, eligible for vouchers to move to private
schools. Moreover, Ohio recently enacted a statewide voucher pro-
gram that allows students in failing schools to enroll in private schools
using publicly funded vouchers from the 2006–07 school year.
The third publicly funded voucher program in the United States was
established in the state of Florida in 1999. Unlike the two previously
enacted programs, the Florida Opportunity Scholarship program
made students in failing schools eligible for vouchers to move to pri-
vate and higher performing public schools. (A January 2006 ruling of
the Florida Supreme Court found the movement to private schools with
vouchers unconstitutional; however, students in failing schools can still
move to higher performing public schools.) The first federally funded
voucher program in the United States was established in Washington,

D.C., in 2004. Like the Milwaukee and Cleveland programs, this pro-
gram made low-income students residing in D.C. eligible for vouchers
to move to private schools. There is a considerable body of literature
that studies publicly funded voucher programs in the United States.
This literature mainly focuses on two issues: the effect on students
who utilize vouchers2 and the competitive effect of vouchers on public
schools.3
Publicly funded vouchers are also found in Colombia and Chile.
Chapters 7 and 8 in part III consider the effect of vouchers in these two
countries on students and schools. While the U.S. experience with
vouchers has been mostly in terms of small experiments and pilot pro-
grams restricted to specific cities or states, in both Colombia and Chile,
vouchers have been implemented on a large-scale, country-wide basis,
which makes these two chapters all the more instructive.
In chapter 7, Eric Bettinger analyzes the effect of school vouchers in
Colombia on student outcomes. In 1991, Colombia implemented a
voucher program that made private school vouchers available to low-
income students entering sixth grade, the start of Colombian second-
ary school. The program also mandated the use of lotteries whenever
the demand for vouchers exceeded supply. Since the demand always
exceeded supply, the program generated two groups of students—one
group was randomly selected to receive vouchers (the treatment
group), the other randomly rejected (the control group). Taking advan-
tage of this random design, Bettinger reports that voucher lottery win-
ners scored higher on standardized exams, were more likely to have
attended private school, were less likely to have repeated a grade, and
in the longer run were more likely to have taken a college entrance
exam and score higher in this exam.
While results from Colombia show considerable promise, the effects
in Chile are much less clear. In chapter 8, Cristian Bellei looks at the
effect of vouchers in Chile on student outcomes. Bellei starts with a
careful review of the Chilean literature on vouchers, pointing out the
discrepancies among the various studies and their merits and limita-
tions. He points out that different studies on vouchers in Chile have
reached very different conclusions. He argues that this can be attrib-
uted to the studies’ differences in methodologies; data limitations in
some of the studies; and differences in researchers’ abilities to include
appropriate controls and interaction effects, and to controll for vari-
ous biases such as selection bias. Chapter 8 highlights the immense
importance of using correct empirical methodology and high-quality

data in impact evaluation analyses. Given the continuing methodologi-
cal debate, it remains a matter of interpretation whether the Chilean
experience with vouchers has been positive or negative.
While vouchers publicly fund private schools with but minimal
oversight and regulation, another form of public-private partnership
involves a more direct supervisory role for the government. Once
again, the most well known of such initiatives, the charter school, are
to be found in the United States. Charter schools are privately man-
aged entities that must receive a governmental charter in order to oper-
ate. Their charter typically runs for five years, at which point the
authority that granted the charter may renew it. The first charter school
opened its doors in Minneapolis-St. Paul in 1992; today there are more
than 3,500 charter schools in the United States, enrolling more than
one million students. A number of studies have looked at the effect of
charter schools in the United States on students who move to these
schools.4 In contrast, research on such initiatives in other parts of the
world has been very limited. In part IV, chapters 9 and 10 bring to-
gether evidence on such initiatives in two countries, Colombia and the
United Kingdom.
The concession schools program was implemented in Bogotá, Co-
lombia, in 1999. In chapter 9, Felipe Barrera-Osorio studies the effect of
this initiative in Colombia on student outcomes in concession schools
compared to nearby public schools. Using propensity score matching,
Barrera-Osorio finds that dropout rates were lower and test scores
higher in concession schools in comparison to similar public schools.
Further, he presents evidence that competition from concession schools
has led to a decline in dropout rates in the public schools.
In chapter 10, Stephen Machin and Joan Wilson present evidence
from a high-profile public-private initiative in the United Kingdom.
The city academies program, implemented in 2000, authorized the es-
tablishment of academies that represented partnerships between the
central government and private sector sponsors. These academies
mainly serve the socially and academically disadvantaged, and the ob-
jective of the policy was to raise the educational standards of these
groups of students. In their study, Machin and Wilson analyze the ef-
fect of these city academies on student achievement. Matching the
academy schools to an appropriate group of public schools and taking
into account pre-policy time trends, the authors find that there is
no short-run positive-impact effect of academy status. However, they
point out that the program is still in its very early phase and a more
conclusive picture is likely to be obtained a few more years into the
program.
In the concluding chapter of this book, Harry A. Patrinos reviews the
evidence on the impact of educational contracting in both developing
and developed countries, including vouchers, charter schools and their
variants in different countries, and private finance of school infrastruc-
tural arrangements. He suggests that research on these initiatives in
developing countries is still very limited, and calls for rigorous impact
evaluations.
We believe that the chapters that follow give us important insights
relating to the different kinds of educational choice prevailing in other
countries. Combining informative descriptions of such programs along
with rigorous policy evaluations, the essays provide a balanced and
comprehensive look at this very important strand of educational initia-
tives in other parts of the world.
Overall, the results seem very promising. While Nechyba in chapter
3 shows that private school vouchers have the potential to raise both
equity and efficiency, Wößmann’s study (chapter 2) shows that pub-
licly funded, privately managed initiatives provide an effective means
of improving cognitive outcomes of students, and Bettinger (chapter
7) and Barrera-Osorio (chapter 9) respectively provide evidence that
two forms of public-private partnerships in Colombia have led to
improvements of educational outcomes for the students involved. The
chapters by Muralidharan and Kremer (chapter 5) and Kingdon (chap-
ter 6) show similar promising findings for private schools in India.
Although the establishment of explicit public-private partnerships
(under the ‘‘Right to Education’’ Bill) is still being vigorously debated
in India, the comparative efficiency of the private sector implies that
such initiatives hold considerable promise there.
The results from Chile and the United Kingdom are, however, much
less definitive. The English city academies program is still in its very
early stages and it is too early to say anything conclusive as to its
impacts. The Chilean program, however, has been in operation for a
long time. In the absence of random assignments of students between
public and voucher schools (as is the case in Chile), it is very difficult
to accurately determine whether or not voucher schools are more effec-
tive than public schools. While Bellei’s essay (chapter 8) constitutes an
improvement over some of the existing literature on Chile, even his
most preferred specifications are likely to suffer from endogeneity
bias.5 As a result, the results from the Chilean study cannot be consid-
ered as conclusive. However, this study does an excellent job of dem-
onstrating that results are often sensitive to the methodology and data
used by researchers, which in turn points to the importance of choos-
ing the correct methodology and data.
The findings of the chapters that follow suggest that the effects of a
program depend crucially on the conditions under which it is imple-
mented, the design of the specific policy, as well as the design of the
empirical methodology. While the chapters in this volume provide im-
portant and valuable insight into public-private partnerships in educa-
tion under various situations and in various countries, we need much
more research before we can understand in a conclusive manner the
effects of different kinds of initiatives.
Notes
Chakrabarti was a postdoctoral fellow at Harvard University in the John F. Kennedy

School of Government’s Program on Education Policy and Governance when work on
this chapter and book was initiated.
1. While Barrera-Osorio’s chapter on concession schools also considers the effect on
dropout rates, the outcome measure mainly considered in this volume is student test
scores. This is not to say that other measures are not important, but this is an area where
research is very limited and further research on other outcome measures would certainly
help our understanding of the effect of such partnerships. Also, ‘‘choice’’ in this volume
mainly refers to choice between schools. However, there are other kinds of choices
involved in the delivery of education, for example, choice of text books, choice of test
preparation service and materials, choice of the consulting firm or the contracting firm
for various services, etc. Again, these are beyond the scope of this volume. Finally, the
chapters are limited to primary and secondary education, though choice pervades not
only these two sectors, but also other sectors of education such as higher education, voca-
tional education, and technical education. These caveats should be kept in mind while
reading the chapters that follow.
2. See John F. Witte, Troy D. Sterr, and Christopher A. Thorn, Fifth year report: Milwau-
kee Parental Choice Program, mimeo, University of Wisconsin, 1995; John F. Witte,
Christopher A. Thorn, Kim M. Pritchard, and Michele Claibourn, Fourth year report: Mil-
waukee Parental Choice Program, mimeo, University of Wisconsin, 1994; Jay P. Greene,
Paul E. Peterson, and Jiangtao Du, The effectiveness of school choice: The Milwaukee ex-
periment, Harvard University, Program on Education Policy and Governance working
paper, PEPG No. 97-1, 1997; Cecilia E. Rouse, Private school vouchers and student
achievement: An evaluation of the Milwaukee voucher program, Quarterly Journal of Eco-
nomics 113 (2): 553–602; Kim K. Metcalf, Natalie A. Legan, Kelli M. Paul, and William J.
Boone, Evaluation of the Cleveland scholarship and tutoring program, 1998–2003, Indi-
ana University School of Education, 2004; Paul E. Peterson, William G. Howell, and Jay
P. Greene, Evaluation of the Cleveland voucher program after two years, Harvard Uni-
versity, Program on Education Policy and Governance working paper, PEPG No. 99-02,
1999; Jay P. Greene, William G. Howell, and Paul E. Peterson, Lessons from the Cleve-
land scholarship program, in Paul E. Peterson and Bryan C. Hassel (eds.), Learning from
School Choice (Washington, D.C.: Brookings Institution, 1998); Patrick J. Wolf, Babette
Gutmann, Nada Eissa, Michael Puma, and Marsha Silverberg, Evaluation of the DC Oppor-
tunity Scholarship Program: First Year Report on Participation, U.S. Department of Educa-
tion, National Center for Education Evaluation and Regional Assistance (Washington,
D.C.: U.S. Government Printing Office, 2005). For a comprehensive analysis and lessons
from school choice in New Zealand, see Edward B. Fiske and Helen F. Ladd, When
Schools Compete: A Cautionary Tale (Washington, D.C.: Brookings Institution, 2000).
3. See Caroline M. Hoxby, School choice and school productivity (Or, could school
choice be the tide that lifts all boats?), in Caroline M. Hoxby (ed.), The Economics of School
Choice (Chicago: University of Chicago Press, 2003); Caroline M. Hoxby, School choice
and school competition: Evidence from the United States, Swedish Economic Policy Review
10 (2003); Rajashri Chakrabarti, Can increasing private school participation and mone-
tary loss in a voucher program affect public school performance? Evidence from the Mil-
waukee voucher experiment, forthcoming, Journal of Public Economics; Jay P. Greene and
Marcus Winters, When schools compete: The effects of vouchers on Florida public school
achievement, Education Working Paper 2, Manhattan Institute, 2003; Jay P. Greene, An
evaluation of the Florida A-Plus Accountability and School Choice Program (New York: Man-
hattan Institute for Policy Research, 2001); Rajashri Chakrabarti, Impact of voucher de-
sign on public school performance: Evidence from the Florida and Milwaukee voucher
programs, mimeo, Harvard University, Program on Education Policy and Governance,
2004; David Figlio and Cecilia Rouse, Do accountability and voucher threats improve
low-performing schools?, Journal of Public Economics 90 (2006): 239–255; Martin R. West
and Paul E. Peterson, The efficacy of choice threats within school accountability systems:
Results from legislatively induced experiments, The Economic Journal 116 (2006): C46–
C62; Rajashri Chakrabarti, Vouchers, public school response and the role of incentives:
Evidence from Florida, mimeo, Harvard University, Program on Education Policy and
Governance, 2006. For a comprehensive review of this literature, and the literature on
privately funded voucher programs in the United States, see William G. Howell and
Paul E. Peterson with Patrick J. Wolf and David E. Campbell, The Education Gap: Vouchers
and Urban Schools, revised edition (Washington, D.C.: Brookings Institution, 2006); and
see William G. Howell and Elena Llaudet, School vouchers, International Encyclopedia
of the Social Sciences, second edition (Framington Hills, MI: MacMillan Reference USA/
Thomas Gale, 2007), for a review of the U.S. and international literature.
4. Eric P. Bettinger, The effect of charter schools on charter students and public schools,
Economics of Education Review 24 (2005): 133–147; Robert Bifulco and Helen F. Ladd, The
effects of charter schools on student achievement: Evidence from North Carolina, Educa-
tion Finance and Policy 1 (2006): 50–90; Eric A. Hanushek, John F. Kain, and Steven G.
Rivkin, The impact of charter schools on academic achievement, Hoover Institution, Stan-
ford University, 2002; Bryan Hassel, Charter school achievement: What we know, Char-
ter School Leadership Council, 2005; Robin J. Lake and Paul T. Hill (eds.), Hopes, fears,
and reality: A balanced look at American charter schools in 2005, Center on Reinventing
Public Education at the University of Washington, 2005; Caroline M. Hoxby and Jonah E.
Rockoff, The impact of charter schools on student achievement, hhttp://post.economics
.harvard.edu/faculty/hoxby/papers/hoxbyrockoff.pdfi, 2004; Caroline M. Hoxby and
Jonah E. Rockoff, The truths about charter schools: Findings from the City of Big Should-
ers, Education Next (Fall 2005); Robert Bifulco and Helen F. Ladd, The truths about char-
ter schools: Results from the Tar Heel State, Education Next (Fall 2005).
5. Some of the explanatory variables, such as books at home, school mean books at home, %
selected students in school, and selected student are likely to be correlated with unobserved
characteristics of families and schools and hence endogenous.
2 Public-Private Partnerships and Student
Achievement: A Cross-Country Analysis
Ludger Wößmann
The issue of public-private partnership (PPP) is a much-debated topic,

and increasingly so in the education sector (for examples, see Human
Development Network 2001; Peterson 2003). However, given that
PPPs are mostly either project-based endeavors or systemic features of
whole education systems, evidence usually comes only in the case-
study form (e.g., World Bank 2004, chapter 7; Patrinos 2000, 2002).
This chapter, by contrast, uses the opportunities of internationally
comparable data to provide cross-country evidence on the association
between student achievement and PPPs across different countries. The
Program for International Student Assessment (PISA) tested students’
basic skills in math, reading, and science in an internationally compa-
rable way. The PISA microdatabase is unique among recent interna-
tional tests in containing information for each tested school both about
whether it is publicly or privately operated and about what share of its
funding comes from public and private sources. These data provide
the opportunity of presenting ‘‘big picture’’ cross-country evidence on
PPPs in schooling.
Such an international perspective carries two particular advantages
relative to analyses within a country. First, comparisons across coun-
tries allow for the recognition of systemic effects, in that the existence
of private schools may affect the behavior and performance of nearby
public schools. If public schools behave differently because there are
private schools nearby, then there may be effects of private involve-
ment even though the performance between individual private and
public schools may not differ. Cross-country evidence can detect such
systemic effects where both private and public schools may perform at
a higher level because of the existence of private competition. The sec-
ond advantage of cross-country evidence is that it allows analyzing
14 Ludger Wößmann
possible differences in the effects of PPPs when they exist in different

situations.
In basically all countries, the ultimate responsibility and supervision
of the school system remain with the state—whether the system makes
use of PPPs or not. But beneath this state supervision, both the opera-
tion and the funding of schools may show differing shares of public
versus private involvement. If we think of school operation and school
funding as the two broad tasks under consideration, and if we under-
stand PPPs as any collaboration between public and private entities,
then conceptually there are two specific ways in which PPPs can exist
in the school system. In the first case, schools are operated (managed)
by a public entity, but draw heavily on private funding—that is,
parents have to pay tuition fees. In the second case, schools are oper-
ated by a private entity—be it a business, the church, or other—but
get most of their funding from a public entity—be it through base
funding or vouchers.
As figure 2.1 shows, both forms of PPPs exist in a system-wide man-
ner.1 In the first type of PPP, prevalent in the systems in the top left
quadrant of figure 2.1, the majority of schools are operated by private
entities, but all schools receive the vast majority of their funding from
public sources. This combination is given in the Netherlands, Belgium,
Ireland, and, to a lesser degree, in Denmark. The second type of
system-wide PPP combines a high share of public operation with a
relatively low share of public funding. This combination of private
financing of publicly managed schools, depicted in the bottom right
quadrant of figure 2.1, exists particularly in Mexico, but to a lesser ex-
tent it can also be observed in Italy, New Zealand, Brazil, and Greece.
Figure 2.1 depicts two more groups of countries that do not consti-
tute partnerships between the public and the private, but rather are
mostly private or purely public. The systems in the bottom left quad-
rant combine relatively low shares of public operation and public
funding. This is true in Korea, where about half of both operation and
funding is private, and to a lesser extent in Japan, France, and Spain.
Finally, in the systems in the top right quadrant, the vast majority of
schools is both publicly operated and publicly funded. This is particu-
larly true in Norway, Iceland, Finland, Sweden, Latvia, and Germany.
But actually, most countries have the vast majority of their schools
both publicly operated and publicly funded. Both in terms of the share
of publicly operated schools and in terms of the average share of pub-
Public-Private Partnerships and Student Achievement 15
Figure 2.1
Public Funding and Public Operation of Schools
Note: Average share of public funding and share of publicly operated schools in the
country, respectively. The acronyms stand for: AUT: Austria, BEL: Belgium, BRA: Brazil,
CZE: Switzerland, CHE: Czech Republic, DEU: Germany, DEN: Denmark, ESP: Spain,
FIN: Finland, FRA: France, GBR: United Kingdom, GRC: Greece, HUN: Hungary, IRL:
Ireland, ICE: Iceland, ITA: Italy, JPN: Japan, KOR: Korea, LUX: Luxembourg, LVA: Lat-
via, MEX: Mexico, NLD: Netherlands, NOR: Norway, NZL: New Zealand, POL: Poland,
PRT: Portugal, RUS: Russian Federation, SWE: Sweden, USA: United States.
Source: Own calculations based on PISA microdatabase.
lic funding of schools, 20 out of 29 countries have more than 87%

public involvement.
This chapter will analyze the efficacy of the four types of systems—
private operation with public funding, public operation with private
funding, substantial private operation and funding, and purely public
operation and funding—in terms of student outcomes. While it will
detect substantial performance differences between the different forms
of systems, a quick glance at figure 2.1 already reveals that a simple
division between public operation and funding on the one side and
private operation and funding on the other side does not seem to be
fundamentally decisive for student performance. For example, two
well-known top performers in PISA, Finland and Korea, characterize
the opposite systems of sole public responsibility (100% public fund-
ing and 97% publicly operated schools in Finland) and large private
involvement (51% private funding and 49% privately operated schools
in Korea). But we will see that the more intricate combination of public
and private involvement in the two forms of PPPs seems to have im-
portant consequences for students’ educational performance.
While there are the discussed advantages of cross-country evidence,
it also has shortcomings. Despite the extensive information on family
and school background that allows accounting for other observable in-
fluence factors, thereby allowing the comparing of students who are
equal in terms of other observable characteristics, the international stu-
dent achievement test still provides observational data. In this data,
private involvement is not randomly divided between a treatment
group that has private involvement and a control group that does not.
Therefore, in contrast to randomized experimental evidence, the evi-
dence presented here has to be interpreted cautiously in terms of de-
scriptive conditional correlations, which do not necessarily allow for
causal inferences because they may also reflect effects of other, un-
observable characteristics. Still, the multivariate analysis goes substan-
tially beyond bivariate correlations in terms of detecting underlying
relationships by disentangling these relationships from other observ-
able influences at the student level. Furthermore, as will be discussed
in section 2.3.1, theory offers some guidance as to which direction
some of the main sources of potentially remaining bias in the presented
‘‘higher-level descriptives’’ point, which can help in the interpretation
of results. But ultimately, remaining bias due to selection on unobserv-
ables cannot be ruled out.
While this chapter goes into the question of effects of different forms
of private involvement in some detail, it should also be clarified right
from the start what the chapter is not about. The topic here is explicitly
the effectiveness of PPPs in providing cognitive skills for students. There-
fore, this chapter does not deal with questions of efficiency (for which
relative costs would have to be taken into account), or with questions
of equity of school systems with differing private involvement, or with
questions of the provision of noncognitive skills. While it goes without
saying that all these issues are of tremendous importance, they go be-
yond the scope of this chapter.
2.1 Public and Private Involvement in Schooling around the World
Before we describe the models and econometric evidence, this section

provides some background on the international data as well as some
more thorough descriptive patterns of public and private involvement
in schooling around the world.
2.1.1 The PISA Database

The database of the Program for International Student Assessment
(PISA) distinguishes itself from previous international tests by provid-
ing data both on whether individual schools are publicly or privately
operated and on what shares of schools’ funding stems from public
and private sources. PISA is an international student achievement test
of representative random samples of 15-year-old students conducted
in 2000 by the Organisation for Economic Co-Operation and Develop-
ment (OECD). The study tested student performance in math, reading,
and science in 32 developed and emerging countries, 29 of which can
be used in this chapter.2 The OECD ensured a consistent and coherent
study design and as much comparability as possible among the partic-
ipating countries.3
The PISA 2000 study had a special focus on the reading literacy of
students, with a sample size of the database used here of 130,242 stu-
dents. The sample sizes in the other two academic subjects are 72,493
students in math and 72,388 students in science. In this chapter, the
main focus will be on reading performance—because of the larger
sample size—and on math performance, which is generally viewed as
being most readily comparable across countries. Also, math perfor-
mance has often been found to be most strongly related to productivity
(e.g., Bishop 1992). The specific student-level database used in this

chapter was constructed by Fuchs and Wößmann (2007), who provide
more detailed information and notes on the specific database. They
combine the test results with rich background information on students
and schools from PISA background questionnaires answered by the
specific students and heads of schools tested in PISA and with addi-
tional country-level data.4
As general background on the data, table 2.1 provides descriptive
statistics for each country on the level and variation of test perfor-
mance in the three subjects. The test results were scaled on a score
with an OECD mean of 500 points and an OECD standard deviation
of 100 points. In all data descriptions and analyses presented in this
chapter, only students in schools that had both data on private versus
public operation and on the share of private versus public funding
were included. The sample size stands at an average of 4,500 students
in 168 schools per country.
2.1.2 Differing Patterns of Public and Private Funding and

Operation
In the PISA database, data on public versus private operation of
schools comes in the form of a dummy that classifies each school as ei-
ther public or private. In the school background questionnaire, a public
school was defined as ‘‘a school managed directly or indirectly by a
public education authority, government agency, or governing board
appointed by government or elected by public franchise.’’ A private
school, by contrast, was defined as ‘‘a school managed directly or indi-
rectly by a non-government organization; e.g., a church, trade union,
businesses, other private institutions.’’
Data on public versus private funding of schools is recorded as the
respective share of total funding coming from public and private
sources. Heads of school were asked, ‘‘About what percentage of your
total funding for a typical school year comes from the following
sources?’’ with the answer categories ‘‘Government (includes depart-
ments, local, regional, state and national),’’ ‘‘Student fees or school
charges paid by parents,’’ ‘‘Benefactors, donations, bequests, sponsor-
ships, parent fund raising’’ and ‘‘Other.’’ For our purposes here, only
the first category was classified as public funding, and the remaining
sources as private funding.
Descriptive statistics on the share of publicly operated schools and
the average share of public funding in each country, already visualized
in figure 2.1, are provided in table 2.2. On average across the 29 coun-
tries, 83% of schools are publicly operated, and the remaining 17% are
managed by a private entity. But the share of publicly operated schools
varies substantially across countries, with Belgium (25%) and the Neth-
erlands (26%) at the bottom end and Russia (100%), Latvia (99%), and
Iceland (99%) at the top end. The average share of public funding of
schools across the countries is 87%, with Mexico (37%) and Korea
(49%) at the lower end and Luxembourg, Sweden, Finland, Norway,
and Iceland all with a public share above 99%.
From the perspective of PPPs, it is particularly interesting to note
that the shares of public involvement in operation and funding can dif-
fer substantially within a country (column M in table 2.2). At the one
extreme, the public share is substantially larger in funding than in op-
eration in the Netherlands (a difference of 69 percentage points), Bel-
gium (63), and Ireland (52). At the other extreme, the public share can
also be substantially larger in operation than in funding, as for exam-
ple in Mexico (difference of 48 percentage points) and Italy (19).
It is also revealing to look at the relative shares of public funding in
publicly versus privately operated schools in each country (columns N
and O in table 2.2). Thus, while in most countries publicly operated
schools receive nearly all of their funding from public sources, 56% of
the funding of publicly operated schools in Mexico stems from private
sources, and 45% in Korea. On average across all countries, the share of
public funding in publicly operated schools (at 91%) is 25 percentage
points higher than in privately operated schools (at 66%). The share
of public funding in privately operated schools differs particularly
strongly across countries, with the privately operated schools tested in
Brazil, Greece, and Mexico receiving no funding at all from public
sources (and those in the United Kingdom only 2%), while privately
operated schools in the Netherlands, Finland, Luxembourg, and Swe-
den receive more than 95% of their funding from public sources.
Even more striking is the difference in public funding that publicly
and privately operated schools in a country receive (column P in table
2.2). At the one extreme, the share of public funding in U.K. public
schools is 96 percentage points higher than in U.K. private schools.
This difference is also quite large in other countries, such as Greece
(87 percentage points), Brazil (87), Poland (76), the United States (74),
and Switzerland (73). At the other extreme, there is no difference at
all in the share of public funding between publicly and privately oper-
ated schools in the Netherlands and Luxembourg, and the differences
Table 2.1
20
Test Performance and Sample Size

Number of
privately
Sample Sample operated Missing
Math Reading Science
size size schools in in original
Score S.D. Score S.D. Score S.D. (students) (schools) sample sample1
(A) (B) (C) (D) (E) (F) (G) (H) (I) ( J)
AUT 515 92 507 93 519 91 4,501 203 20 5.1
BEL 520 106 507 107 496 111 6,345 203 149 4.9
BRA 334 97 396 86 375 90 3,956 256 27 19.1
CHE 529 100 494 102 496 100 5,830 269 26 4.4
CZE 498 96 492 96 511 94 5,343 228 23 0.4
DEU 490 103 484 111 487 102 4,442 192 8 12.4
DNK 514 87 497 98 481 103 4,009 208 51 5.3
ESP 476 91 493 85 491 95 5,818 174 65 6.4
FIN 536 80 546 89 538 86 4,864 155 6 0.0
FRA 517 89 505 92 500 102 4,080 153 36 12.7
GBR 529 92 523 100 532 98 8,676 337 16 7.1
GRC 447 108 474 97 461 97 4,305 144 5 7.9
HUN 488 98 480 94 496 103 4,766 183 8 2.5
IRL 503 84 527 94 513 92 3,793 137 83 1.6
ISL 514 85 507 92 496 88 3,236 125 2 4.0
ITA 457 90 487 91 478 98 4,704 161 9 5.6
JPN 557 87 522 86 550 90 5,217 134 39 0.7
KOR 547 84 525 70 552 81 4,676 137 64 6.1
Ludger Wößmann
LUX 446 93 441 100 443 96 3,251 22 3 7.9
LVA 463 103 458 102 460 98 3,038 123 1 22.0
MEX 387 83 422 86 422 77 3,894 156 26 15.3
NLD 556 89 530 91 527 96 2,246 89 66 10.3
NOR 499 92 505 104 500 96 3,972 165 3 4.2
NZL 537 99 529 108 528 101 3,456 144 7 5.8
POL 470 103 479 100 483 97 3,586 125 5 1.9
PRT 454 91 470 97 459 89 4,554 148 11 0.7
RUS 478 104 462 92 460 99 6,566 242 0 2.0
SWE 510 93 516 92 512 93 4,416 154 6 0.0
USA 493 98 504 105 499 101 2,947 116 6 23.4
Mean 492 94 492 95 492 95 130,4872 4,8832 7712 6.6
Notes: See note to figure 2.1 for a list of country acronyms.
1. Percentage of observations without information on public operation or funding. These observations are not counted in the columns on sample
size.
2. Sum.
Public-Private Partnerships and Student Achievement
21
Table 2.2
International Differences in Public Funding and Public Operation of Schools
Average Average
share of share of
public public
Average Share of funding in funding in
share of publicly publicly privately
public operated operated operated
funding schools Difference schools schools Difference
(K) (L) (M) (N) (O) (P)
AUT 90.4 88.8 1.6 96.2 44.2 52.1
BEL 87.5 24.6 63.0 95.1 85.0 10.1
BRA 78.0 89.4 11.4 87.3 0.0 87.2
CHE 93.8 93.5 0.3 98.5 25.5 73.0
CZE 94.9 93.9 1.0 96.2 74.5 21.7
DEU 97.3 95.9 1.4 98.2 77.1 21.1
DNK 94.3 75.5 18.8 99.9 76.9 23.0
ESP 82.9 61.9 20.9 95.3 62.8 32.5
FIN 99.8 97.2 2.6 99.9 98.3 1.5
FRA 75.5 77.8 2.4 77.3 69.1 8.2
GBR 89.8 90.8 1.0 98.7 2.2 96.4
GRC 83.7 95.8 12.0 87.4 0.0 87.4
HUN 87.4 95.3 7.8 87.9 78.3 9.5
IRL 91.1 39.5 51.6 98.0 86.6 11.4
ISL 99.4 99.2 0.2 99.9 40.4 59.5
ITA 75.2 94.2 19.0 79.2 10.1 69.1
JPN 72.5 69.6 2.9 88.4 36.0 52.4
KOR 49.1 50.6 1.5 54.9 43.3 11.6
LUX 100.0 87.9 12.1 100.0 100.0 0.0
LVA 95.6 99.2 3.6 95.7 89.0 6.7
MEX 36.8 84.5 47.6 43.6 0.0 43.6
NLD 94.7 26.0 68.7 94.7 94.7 0.0
NOR 99.5 98.5 1.0 99.8 82.3 17.5
NZL 80.2 95.1 14.9 83.4 18.4 65.0
POL 92.2 97.1 4.9 94.4 18.1 76.4
PRT 87.9 92.7 4.8 88.5 80.1 8.4
RUS 93.5 100.0 6.5 93.5 — —
SWE 99.9 96.6 3.3 99.9 99.3 0.6
USA 91.6 94.6 2.9 95.6 22.1 73.6
Mean 86.9 83.0 3.9 91.2 65.9 25.4
Notes: In percent. See note to figure 2.1 for a list of country acronyms.
are also very small in Sweden (0.6 percentage points) and Finland
(1.5).
2.2 Why Should the Public-Private Division Matter?
Do these cross-country differences in public versus private involve-

ment in the operation and funding of schools matter for student
achievement? From a theoretical point of view, positive and negative
aspects of both operation and funding of schools by the state versus
the private sector have been advanced in the literature. The main case
usually made in terms of the operation of schools is that private opera-
tion is more efficient than public operation because market forces
create incentives for cost containment and performance-conducive
qualitative innovation in private school management (e.g., Chubb and
Moe 1990; Hanushek et al. 1994; Shleifer 1998; Bishop and Wößmann
2004). In accordance with this reasoning, empirical evidence tends to
find that performance in privately managed schools is superior to per-
formance in publicly managed schools.5 Some of the empirical contri-
butions also show that the existence of private schools improves the
performance of nearby public schools that face their competition, an
issue that complicates the empirical identification of the effects of pri-
vate school operation, to which we will return.
The case in favor of public provision of schools is less clear, if we
separate it from the conceptually different case of public versus private
funding, in particular because the government could always write spe-
cific contracts with private providers to ensure that certain require-
ments are observed (cf. Shleifer 1998). However, one point sometimes
advanced in favor of public provision is that only direct public provi-
sion of schooling could yield an inculcation of students with ideologi-
cal and cultural goals and beliefs pursued by the government, a task
that might not be easily contracted out to the private sector (e.g., Grad-
stein and Justman 2002; Pritchett 2003). Ideological inculcation may be
an issue hard to pin down empirically, and it is certainly beyond the
scope of a paper focusing on cognitive skills. But in a similar vein, di-
rect public school operation might allow a closer monitoring of imple-
mented curricula—although the incentives for monitoring may again
be stronger in the private sector.
In terms of the relative merits of public and private funding (as
opposed to operation) of schools, it is sometimes argued that pri-
vate or parent-based funding can increase accountability and provide
incentives for efficient behavior from the demand side (cf. Jimenez and
Paqueo 1996; Chubb and Moe 1990). It is not obvious, though, to what
extent this benefit of private involvement would go beyond the benefit
of private provision and the choice that parents can make between dif-
ferent private providers, which would already induce performance-
conducive incentives from the demand side.
This latter point can even be turned into the opposite case favoring
public funding, if combined with the idea that some families may be
too poor to choose privately operated schools if they have to be funded
privately. As long as there are credit constraints that prevent poor fam-
ilies from borrowing against possible future income gains of their chil-
dren due to improved educational performance (cf. Loury 1981; Galor
and Zeira 1993; Gradstein, Justman, and Meier 2004), poor families’
choices of better schools that require higher private funding will be
constrained. Public funding can relax the credit constraints, which can
allow greater choice for all families and therefore increase schools’
incentives to behave efficiently. The empirical evidence of positive
performance effects of (mostly publicly provided) school vouchers to
finance the attendance of privately operated schools (see preceding
references) can be viewed as one aspect of this possible positive effect
of public funding.6
Given the different theoretical arguments as summarized in table 2.3,
we might expect the public versus private nature of the operation and
funding of schools to have an impact on student achievement. The di-
rection and size of this impact remains an empirical question, though.
2.3 The Empirical Models
This section describes the different specifications of the empirical

model of the association between PPPs and student performance ana-
Table 2.3
Aspects of Public and Private Involvement in School Provision and Funding
Positive aspects of involvement of:
Public sector Private sector
Operation Inculcation of beliefs and Incentives for cost containment
cultural values and qualitative innovation
Funding Enabling choice for credit- Increased accountability
constrained families
lyzed in this chapter, and how they may or may not be affected by se-
lection bias.
2.3.1 Observables, Unobservables, and the Possibility of Selection

Bias
Given the separate arguments for and against public involvement in
the operation and funding of schools, assume that the true achieve-
ment model includes separate effects of operation and funding and
can be represented by:
Ti ¼ a þ b1 Os þ b2 Fs þ Bi b3 þ Ui b4 þ ei ; ð2:1Þ
where Ti is the achievement test score of student i, Os is a dummy

showing whether the student’s school s is publicly (as opposed to pri-
vately) operated, and Fs is the share of the school’s funding stemming
from public (as opposed to private) sources. B are additional back-
ground features that can be observed, like parents’ level of education,
U are additional features affecting performance that remain un-
observed, for example parents’ valuation of their children’s education,
and ei is an error term.
What are the consequences of B und U for our estimates of the asso-
ciation between public involvement and student performance? Two
observations can help us understand the specific nature of one of the
main sources of possible bias in our empirical models. First, given that
we look at the effects of operation and funding separately, the argu-
ments on credit constraints discussed in section 2 bear on the associa-
tion between family background and funding, but not on the private
versus public operation of schools. Second, families who can afford to
send their children to schools that require large shares of private fund-
ing may tend to have other, often unobserved features conducive to
the children’s learning. For example, they may show a greater valua-
tion of educational outcomes per se, or they may simply use their
larger income also to buy private afternoon lessons for their children if
they are underperforming in a given subject. Therefore, the selection
bias due to credit constraints would generally make privately funded
schools look better than they really are. Thus, we would expect this
particular source of bias to affect estimates of funding effects rather
than operation effects, and we would expect this bias to point in the di-
rection that publicly funded schools look worse than they are. We will
depict these ideas more formally before we go on to discuss other pos-
sible sources of remaining bias.
Consider first the observable background features B. Following the

above-mentioned theories of credit constraints, it seems reasonable to
assume that B will be positively related with the share of private fund-
ing in the school, that is, negatively related with F. Therefore, there will
be nonzero elements in the covariance matrix between the B features
and F. But if F 0 B 0 0, estimating the effect of F under disregard of con-
trols for B will yield a biased estimate of the true effect b2 of equation
(2.1). However, given that we assume B to be observable, we can easily
control for B in our regression, so that the estimated effect is no longer
biased by the observables. This is the reason why we include an un-
usually extensive set of controls for background factors B in all our em-
pirical specifications. Specifically, the control vector B of background
data encompasses 60 variables, including 8 variables on student char-
acteristics, 28 variables on students’ family background, 14 variables
on resource inputs at home and at school, and 10 variables on institu-
tional features of the school system.7
Now, consider the unobserved background features U. If the co-
variance matrix F 0 U 0 0, estimating the funding effect under disregard
of controls for U yields the following biased estimate g, using the stan-
dard formula for omitted-variable bias (cf., e.g., Greene 2000, p. 334):
EðgÞ ¼ b 2 þ ½ðFs Fs Þ1 Fs0 Ui b4 : ð2:2Þ
Assuming for the moment only one unobserved variable Ui , we can

also write:
covðFs ; Ui Þ
EðgÞ ¼ b 2 þ b : ð2:3Þ
varðFs Þ 4
This derivation allows us to pin down the likely direction of the bias
that emanates from credit constraints. The estimate g will be lower than
the true effect b2 of equation (2.1) if b 4 is positive and covðFs ; Ui Þ is neg-
ative. And this is likely to be the case in this particular application.
Let’s say that Ui is (unmeasured) parents’ valuation of education.
Then b 4 is positive, that is, parents’ valuation of education has a
positive impact on their children’s educational performance. And
covðFs ; Ui Þ is negative, because parents’ valuation of education is posi-
tively associated with their willingness (and probably ability) to pro-
vide private funding, and thus negatively related to the share of public
funding in their school. Therefore, the estimate g will be a lower bound
for the true effect b 2 . The larger the covariance between the unobserved
features U and the share of private funding, and the larger the effect b 4
of the unobserved features on student performance, the larger will be
the underestimation of the effect of public funding. However, once we
allow for multiple unobserved variables, the uniqueness of this result
is no longer given.
Note also that there may be other possible sources of remaining bias
than credit constraints. Although credit constraints are often viewed as
the main cause of concern for selectivity in education, other unobserv-
able features may give rise to different kinds of bias. As one example,
the selection may not only originate on the student/parent side, but
also on the school side. If privately operated schools have more free-
dom to choose their students, and if the selected students have features
unobserved by the researcher that differ from those of students in
publicly operated schools, then this may give rise to bias also on the
operation side. For example, privately operated schools may have a
preference to admit students who are particularly smart for their ob-
servable features, which would bias the coefficient estimate on O
downwards. Moreover, in empirical applications that use variation
across countries, there may be unobserved country features that are
associated with the share of public operation and funding in a nonran-
dom way. Therefore, the extent and direction of any remaining bias
must ultimately remain an open issue.
2.3.2 Alternative Empirical Specifications

Altogether, in this paper we will estimate six different versions of the
basic empirical model. In specification (2.4), the variables on public
versus private involvement are measured at the country level and
entered in a cross-country regression performed at the student level:
Ti ¼ a þ b1 Oc þ b 2 Fc þ Bi b3 þ nc þ hs þ ei ; ð2:4Þ
where Oc and Fc are the share of publicly operated schools and the av-
erage share of public funding of schools in country c, respectively. Both
student test scores Ti and background features Bi are measured at the
student level (the school characteristics in B at the school level), so as
to yield as clean a control for other influence factors as possible. Note
that in the empirical application, the error term will have higher-level
components at the school and country level, which is implemented
by clustering the standard errors at the higher level, and that students
are weighted by their sampling probabilities within each country (see
section 4.3 in Wößmann 2006 for details on the specifics of the micro-
econometric model).
The reason we start with a cross-country specification with country-
level public-private data is the possibility discussed in section 2.2: that
the mere existence of private schools in a city may have systemic
effects due to their effect on how the public schools in the city perform,
because these public schools are now faced with competition from
private schools. Thus, looking at the simple relative performance of
privately and publicly operated schools may well fail to observe the
effects of the existence of the privately operated school. By contrast,
such systemic effects will be captured in the specification that measures
private involvement at the level of the country. The specification
shows whether countries with a larger sector of publicly operated
schools and with a larger share of public funding fare differently on
average on the PISA test than countries with larger shares of private
involvement.
A second advantage of the specification is that it evades the problem
of selection bias just discussed. While it may be the case that students
whose performance differs for other reasons may select (or be selected)
into private schools in nonrandom ways, such selection effects will can-
cel out at the country level. Under the quite confident assumption that
there is no school selection across country borders of an order of mag-
nitude that might affect the presented results, any nonrandom selection
would occur within the observation level of the public-private mea-
sures in specification (2.4) and would therefore not affect the estimates
of b1 and b2 in this specification. Note, though, that one can never
perfectly rule out remaining endogeneity due to unobservables at the
country level.
Specification (2.5) simplifies the picture even further, by classifying
the countries into the four quadrants of figure 2.1 established by the
shares of public operation and public funding. That is, public-private
involvement will be measured just by attributing dummies to the
countries whether they belong to the top left ðTLÞ, the bottom left ðBLÞ
or the bottom right ðBRÞ quadrant of figure 2.1, where the reference
category is the top right quadrant:
Ti ¼ a þ b1 TLc þ b2 BLc þ b3 BRc þ Bi b4 þ nc þ hs þ ei : ð2:5Þ
This quadrant-dummy specification has the same advantages as the

specification with country-level public-private data, except that it pro-
vides the results in an even simpler (but also coarser) way: Do the per-
formances of the countries in the four quadrants of figure 2.1 differ sys-
tematically from each other?
Specification (2.6) adds an interaction term between public operation
and public funding to specification (2.4):
Ti ¼ a þ b1 Oc þ b 2 Fc þ Bi b3 þ ðOc Fc Þb4 þ nc þ hs þ ei : ð2:6Þ
The coefficient on the interaction term ðOc Fc Þ depicts whether any ef-
fect of public funding F differs between countries with lower or higher
shares of publicly operated schools.
The next specification makes use of the individual-level data of
public-private involvement in schools. That is, both public operation
Os and public funding Fs are now measured at the level of each school
s, still in a regression encompassing all countries:
Ti ¼ a þ b1 Os þ b2 Fs þ Bi b3 þ ðOs Fs Þb 4 þ hs þ ei : ð2:7Þ
Note that Os is now a dummy representing whether the student’s

school is publicly (as opposed to privately) operated, while Fs is the
share of public funding of the school.
In this specification, the considerations on possible selection bias
now come into play, which suggested that one particularly relevant
source of bias, due to credit constraints, points in the direction that the
estimate b 2 on the effect of public funding may be biased downwards,
that is, biased in favor of schools with larger shares of private funding.
The specification can again be estimated with and without an interac-
tion effect. In the specification using school-level data on public-private
involvement, the coefficient on the interaction term depicts whether
any association between student performance and public funding F
differs between publicly and privately operated schools.
Specification (2.8) adds country fixed effects to specification (2.7):
Ti ¼ ac þ b1 Os þ b2 Fs þ Bi b3 þ ðOs Fs Þb 4 þ hs þ ei ; ð2:8Þ
where ac is a country-specific intercept implemented by adding a full

set of controls for country dummies. By disregarding any variation
that exists between countries, this specification in effect estimates the
average effect of public operation and funding within the countries in
the pooled dataset. That is, the specification shows whether on aver-
age, publicly operated and publicly funded schools in a country fare
differently from privately operated and privately funded schools in the
same country. The previous considerations suggest that the relative

importance of selection bias may get ever more severe in this specifica-
tion, because the selection-free variation that exists between countries
is now no longer considered.
Finally, specification (2.9) estimates the same specification separately
within each country c:
Ec: Ti ¼ a þ b1 Os þ b 2 Fs þ Bi b 3 þ hs þ ei if i A c: ð2:9Þ
By doing so, this specification can depict whether the within-country

associations between public-private involvement and student perfor-
mance are heterogeneous across the countries. The presented results
on this specification will not consider an interaction term ðOs Fs Þ, be-
cause initial experimentation showed that samples seem to get too
small in most countries to properly identify the interaction effect with-
in individual countries.
2.4 Empirical Results on Public-Private Division and Student

Achievement
This section presents the empirical results on the association between

PPPs and student achievement. It shows how student performance is
associated with public versus private involvement in the operation
and funding of schools both across and within countries.
2.4.1 Public vs. Private Funding and Operation Measured at the

Country Level
The specifications (2.4) and (2.5) that use only the variation between
countries and disregard variation within countries by aggregating the
measures of public-private involvement at the country level have
the advantages of capturing system-level effects and evading within-
country selection biases. Column Q in table 2.4 presents the results of
the quadrant-dummy specification (2.5) for math performance. This
specification compares student performance in the four basic system
types of figure 2.1, with the type representing the largest number of
countries—the top right quadrant of systems with large public shares
in both operation and funding—serving as the reference category.8
The results show that there are large and statistically significant sys-
tematic performance differences between systems that make strong use
of PPPs and systems that do not. Systems with a relatively low share of
public operation, but a large share of public funding perform best.
Table 2.4
Public-Private Involvement and Math Performance across Countries
Country-level public-private measures School-level public-private measures

(Q) (R) (S) (T) (U) (V) (W) (X) (Y)
Top left quadrant 37.93***
(12.52)
Bottom left quad. 0.02
(7.73)
Bottom right quad. 36.64***
(8.17)
Public operation 74.55*** 93.80*** 193.15** 19.68*** 24.69*** 9.05
(14.78) (13.81) (94.26) (2.40) (2.69) (5.98)
Public funding 24.51 91.05*** 3.73 1.64 18.56*** 30.18***
(26.69) (27.03) (77.00) (3.53) (3.96) (6.40)
Interaction 113.45 20.37***
(109.86) (7.63)
Observations 72,493 72,493 72,493 72,493 72,493 72,493 72,493 72,493 72,493
Strata 29 29 29 29
Public-Private Partnerships and Student Achievement
PSUs 29 29 29 29 29 4,870 4,870 4,870 4,870

R2 0.314 0.309 0.293 0.315 0.316 0.297 0.293 0.298 0.298
Notes: Dependent variable: PISA international math test score. Least squares regressions weighted by students’ sampling probabilities. Regres-
sions include 60 control variables for student, family, and school characteristics. Robust standard errors adjusted for clustering in parentheses
(columns Q to U: clustering by country; columns V to Y: clustering by school). Quadrant dummies refer to country’s position in figure 2.1.
Interaction ¼ interaction term between public operation and public funding.
Significance level (based on clustering-robust standard errors): *** 1 percent; ** 5 percent; * 10 percent.
31
Their average performance is 37.9 PISA test-score points higher than

the average performance of systems that are mainly publicly operated
and funded. Given that the test scores are scaled to have an interna-
tional standard deviation among the OECD countries of 100, the effect
size can be interpreted as percentage points of an international stan-
dard deviation. That is, PPP systems that combine public funding with
private operation perform more than one third of an international stan-
dard deviation better than pure public systems. To provide an alterna-
tive benchmark for the size of this performance difference, we can also
compare it to the unconditional performance difference between ninth-
grade and tenth-grade students (the two largest grades in PISA), which
is 30.3 PISA test-score points in math.9 That is, 15-year-old students
in public-funding private-operation PPP systems on average perform
more than the equivalent of a whole grade level better than same-aged
students in mainly publicly funded and operated systems.
By contrast, students in systems that combine large shares of public
operation with relatively low shares of public funding, that is, the sec-
ond type of PPP, perform 36.6 test-score points worse than students in
purely public systems. Interestingly, there is no difference at all in the
average performance of students in systems that combine large shares
of private operation and funding and students in systems that combine
large shares of public operation and funding.
Figure 2.2 depicts the result pattern graphically. It shows that the
two forms of PPPs—public funding with private operation and public
operation with private funding—have diametrically opposite conse-
quences relative to all-public or mainly private systems. The perfor-
mance difference between the two forms of PPPs adds up to 74.6
points. These results suggest that it makes a fundamental difference
how the partnership between public and private in PPPs is conceived:
reserving funding for the public side but contracting the operation to
the private sector brings huge gains in performance, but transferring
funding to the private side and leaving the operation of schools in pub-
lic hands brings huge losses. The picture also suggests that there are no
significant interactions between operation and funding at this level:
public operation has a negative effect, independent of the mode of
funding, and public funding has a positive effect, independent of the
mode of operation. Therefore, in a mainly privately operated and
funded system, the two effects cancel out and average performance is
similar to a mainly publicly operated and funded system.
Figure 2.2
Student Achievement in the Four Quadrants of Public-Private Involvement
Note: The distinction into countries with relatively low and relatively high shares of pub-
lic operation and funding follows the quadrants in figure 2.1.
Source: table 2.4, column Q.
The basic pattern of results is exactly the same in the other two sub-
jects, reading and science, as the results reported in columns Z and AE
in table 2.5 show. The size of the performance differences in these two
subjects is somewhat smaller, though, and there are slight but statisti-
cally insignificant differences in the size of the estimate on the coeffi-
cient on the bottom left quadrant.
The very same pattern of negative effects of public operation and
positive effects of public funding emerges in specification (2.4) with
country-level data on public-private operation and funding. As the
results reported in columns T, AA, and AF in tables 2.4 and 2.5 reveal,
the coefficient on public operation is statistically significantly negative
in all three subjects, and the coefficient on public funding is positive in
all three subjects and statistically significantly so in math and read-
ing.10 Note that the positive coefficient on public funding in math is
Table 2.5
34
Public-Private Involvement and Performance in Reading and Science across Countries

Reading Science
Country-level public-private School-level public- Country-level public-private School-level public-
measures private measures measures private measures
(Z) (AA) (AB) (AC) (AD) (AE) (AF) (AG) (AH) (AI)
Top left quadrant 28.28** 17.00
(10.45) (12.45)
Bottom left quad. 9.56 5.86
(6.39) (6.61)
Bottom right quad. 13.06** 18.36***
(5.76) (6.53)
Public operation 56.95*** 35.15 19.27*** 7.04 55.61*** 183.89** 17.94*** 6.31
(10.71) (70.13) (2.30) (4.88) (11.86) (80.54) (2.42) (5.01)
Public funding 59.06** 78.07 8.35** 17.42*** 22.07 90.88 0.79 9.39*
(23.62) (59.05) (3.29) (5.37) (20.49) (66.11) (3.36) (5.24)
Interaction 24.80 15.91** 146.01 15.11**
(84.64) (6.30) (94.11) (6.41)
Observations 130,242 130,242 130,242 130,242 130,242 72,388 72,388 72,388 72,388 72,388
Strata 29 29 29 29
PSUs 29 29 29 4,882 4,882 29 29 29 4,870 4,870
R2 0.310 0.311 0.311 0.306 0.306 0.254 0.256 0.257 0.252 0.252
Notes: Dependent variable: PISA international reading/science test score. Least squares regressions weighted by students’ sampling probabilities.
Regressions include 60 control variables for student, family, and school characteristics. Robust standard errors adjusted for clustering in parentheses
(columns Z to AB and AE to AG: clustering by country; columns AC, AD, AH, and AI: clustering by school). Quadrant dummies refer to country’s
position in figure 2.1. Interaction ¼ interaction term between public operation and public funding.
Significance level (based on clustering-robust standard errors): *** 1 percent; ** 5 percent; * 10 percent.
Ludger Wößmann

significant only once the mode of operation is controlled for (compare

columns R, S, and T).
2.4.2 Cross-Country Regressions with School-Level Public-Private

Measures
Columns X, AC, and AH in tables 2.4 and 2.5 report the results of spec-
ification (2.7) that uses individual-level data on public-private involve-
ment, first without an interaction term.11 Note that the difference
between this and the previous specifications is not in the level of esti-
mation, because all specifications use student-level data on test scores
and background features. The difference is in the level at which the
measures of public versus private operation and funding of schools
are measured.
The qualitative results of the specification measuring public-private
involvement at the school level are the same as those of the specifica-
tion with country-level public-private measures in all three subjects:
Public operation is negatively associated with student achievement and
public funding is positively associated with student achievement.
As argued in section 2.3.1 above, the selection bias due to credit con-
straints is likely to bias downward the estimate on public funding in
the specification with school-level public-private measures. However,
we still get a statistically significant positive estimate. Therefore, the ef-
fect of public funding on student achievement seems indeed to be pos-
itive, and likely to be even larger than the reported coefficient estimates
of this specification suggest. This pattern is also consistent with the rel-
ative size of the coefficient estimate in the specifications with country-
level and school-level public-private measures. The lower size in the
specification with school-level public-private measures could be attrib-
uted to the selection bias due to credit constraints, which is operative
within countries but not across countries. But obviously, selection
biases stemming from other sources may also be at play.
The coefficient estimate on public operation is also smaller in abso-
lute terms in the specification with school-level public-private mea-
sures. As previously argued, theories of credit constraints would not
predict a bias of the coefficient on public versus private operation. By
contrast, such a bias could, for example, be attributed to nonrandom
selection of students on the part of the privately operated schools.
However, note that most standard versions of selection bias would
predict a difference in the coefficient estimates of the two specifi-
cations that would go the other way, pushing the coefficient in the
specification with school-level public-private measures even more into

the negative. Therefore, a more convincing explanation for the fact that
the estimated effect of private operation is larger in the specifications
where it is measured at the country level than at the school level is
that there are strong systemic effects of competition from privately
operated schools also on the publicly operated schools in a system, as
discussed in sections 2.2 and 2.3.2.
In sum, all specifications that use the cross-country variation, using
country-level and school-level measures of public-private involvement,
yield the result that public operation has a negative effect and public
funding a positive effect on student performance.
2.4.3 Interactions between Funding and Operation

To see whether the effect of public funding differs under public versus
private operation of schools, the next specification adds to the model
an interaction term between operation and funding. Columns U, AB,
and AG in tables 2.4 and 2.5 report results of specification (2.6) that
adds the interaction term to the country-level measure specification
(2.4). As already apparent in the quadrant-dummy specification de-
picted in figure 2.2, there is no significant interaction effect in any of
the subjects in this specification with country-level public-private data,
which in this specification is mainly due to the fact that the estimates
have large standard errors and thus lack statistical power.
But the interaction results are different in specification (2.7) that esti-
mates the interaction term on individual-level data on public-private
involvement, reported in columns Y, AD, and AI. In all three subjects,
the interaction term is statistically significantly negative. This means
that public operation has a slightly negative effect already in schools
with low public funding, but this effects gets ever more negative with
increasing public funding. This pattern is depicted graphically in panel
(a) of figure 2.3. The pattern might either be driven by increasing lack
of accountability in schools that do not receive any private funding, or
by the selection bias of higher-performing students into schools with
larger shares of private funding.
At the same time, the negative interaction term means that the posi-
tive effect of public funding is strongly concentrated in schools that are
privately operated, while the effect of public funding decreases to
about zero in publicly operated schools (cf. panel (b) of figure 2.3). This
pattern might suggest that it is particularly the role of public funding
in allowing everyone—including low-income families—to opt for pri-
Figure 2.3
The Interaction of Public Operation and Public Funding
(a) Effect of public operation depending on type of funding; (b) Effect of public funding
depending on type of operation
Source: table 2.5, column AD.
vately operated schools that drives the positive association between

public funding and student performance. In this sense, public funding
can enable more choice, in that families that would otherwise be credit
constrained can choose privately operated schools if funding comes
from public sources.
2.4.4 Regressions Using Only Variation within Each Country

Results of specification (2.8), which adds country fixed effects to the
previous specification (2.7), are reported in the top panel of table 2.6.
An interaction term between public funding and operation was never
statistically significant in the three subjects and was thus dropped
from the specification. In this specification that disregards the
between-country variation and exploits only the within-country varia-
tion, the coefficients on both public operation and public funding are
negative, and the coefficient on public funding is the larger one in ab-
solute terms. This might suggest that the association between public
funding and student performance within countries is mostly driven
by the selection bias of higher-performing children into schools with
larger shares of private funding. However, this specification using the
within-country variation in the pooled cross-country dataset also fore-
shadows systematic differences in the associations between countries.
This becomes apparent in the lower panels of table 2.6, which report
results of estimating the model separately for each country, as in speci-
fication (2.9).12 We restrict our attention to reading performance here
because the reading samples are substantially larger than the samples
in the other two subjects, which becomes particularly relevant when
restricting the estimations to within each country. It should still be rec-
ognized that in some countries, cell sizes get worryingly small, as is ev-
ident from the small number of privately operated schools in some
country samples, reported in column I in table 2.1.
A broad systematic pattern emerges when we look at the countries
separately by their affiliation to the four quadrants of figure 2.1 again.
Running the pooled within-country regressions separately for each
quadrant (again including country fixed effects), it becomes apparent
that the negative coefficient on public operation in the specification
pooling all countries is mainly driven by the countries in the top left
panel, while the negative coefficient on public funding is mainly driven
by the countries in the bottom right panel. All other panel-wise coeffi-
cients are statistically insignificant, which may be partly due to a sub-
stantially smaller statistical power relative to the pooled cross-country
Table 2.6
Estimations within Countries
Observa-
Public operation Public funding tions PSUs R2
Pooled
Math 1.63 (2.45) 10.63*** (3.98) 72,493 4,870 0.346
Science 4.01* (2.32) 10.51*** (3.51) 72,388 4,870 0.281
Reading 5.99*** (2.14) 10.43*** (3.52) 130,242 4,882 0.341
Reading
Top left 13.66*** (3.62) 12.40 (11.02) 16,388 637 0.348
BEL 38.51*** (7.07) 0.17 (14.27) 6,345 203 0.506
DNK 5.94 (11.36) 44.56 (42.58) 4,006 208 0.238
IRL 28.17*** (6.18) 16.86 (13.95) 3,791 137 0.330
NLD 9.91 (6.81) 17.80 (35.79) 2,246 89 0.471
Bottom left 7.51** (3.40) 5.75 (5.99) 19,773 598 0.331
ESP 13.05* (7.34) 11.58 (9.52) 5,805 174 0.383
FRA 6.99 (4.75) 3.18 (8.73) 4,077 153 0.494
JPN 36.29* (21.56) 28.67 (18.40) 5,215 134 0.241
KOR 6.10 (5.67) 12.04 (11.57) 4,676 137 0.261
Bottom right 4.41 (7.78) 13.88** (5.65) 20,284 861 0.418
BRA 15.11 (22.75) 20.29** (10.01) 3,951 256 0.342
GRC 23.64 (17.23) 2.86 (16.66) 4,301 144 0.364
ITA 26.79 (19.39) 56.49*** (21.08) 4,704 161 0.356
MEX 32.26*** (11.01) 0.88 (6.12) 3,873 156 0.433
NZL 20.21 (14.80) 68.81*** (19.70) 3,455 144 0.294
Top right 2.79 (4.73) 11.12 (8.24) 74,022 2,795 0.319
AUT 14.60 (12.70) 50.38*** (17.32) 4,501 203 0.400
CHE 34.69* (19.14) 69.35** (28.05) 5,822 268 0.429
CZE 9.14 (12.19) 59.03* (30.20) 5,343 228 0.384
DEU 25.25 (20.40) 6.03 (54.93) 4,430 192 0.486
FIN 2.12 (11.11) 238.14 (254.08) 4,863 155 0.221
GBR 83.86*** (26.85) 5.76 (27.73) 8,658 337 0.331
HUN 3.99 (12.55) 28.81 (24.81) 4,761 183 0.425
ISL 48.75*** (17.90) 32.96 (24.56) 3,229 125 0.175
LUX 21.89* (11.29) — 3,132 22 0.459
LVA 40.72*** (13.41) 20.17 (44.35) 3,037 123 0.315
NOR 35.83 (24.07) 276.67** (115.33) 3,967 165 0.182
POL 23.41 (52.29) 24.23 (66.03) 3,585 125 0.290
PRT 4.83 (20.68) 3.01 (10.43) 4,549 148 0.527
RUS — 5.81 (25.34) 6,563 242 0.279
SWE 10.32 (7.85) 275.32 (403.53) 4,415 154 0.209
USA 47.06 (15.07) 11.81 (19.04) 2,942 116 0.355
Table 2.6
(continued)
Notes: Dependent variable: PISA international test score. Least squares regressions
weighted by students’ sampling probabilities. Regressions include 60 control variables
for student, family, and school characteristics. Robust standard errors adjusted for clus-
tering in parentheses (clustering by school). Regressions that pool countries control for
country fixed effects. The organization of countries follows the four quadrants of figure
2.1. See note to figure 2.1 for a list of country acronyms.
Significance level (based on clustering-robust standard errors): *** 1 percent; ** 5 percent;
* 10 percent.
analyses because of the smaller number of observations.13 Note that

the lower performance of publicly operated schools in the top left
quadrant is unlikely to be caused by credit-constraint-based selection
patterns, as all schools in the systems of this quadrant—be they pub-
licly or privately operated—receive the vast majority of their funding
from public sources.
By contrast, the lower performance of schools with larger shares of
public funding in the countries of the bottom right quadrant may well
be driven by credit-constraint-based selection bias. Nearly all schools
in the systems of this quadrant are publicly operated, but rich families,
whose children may perform better for reasons other than the public-
private division of school operation and funding, can provide addi-
tional private funding for their schools. Finally, note that in the top
right quadrant, there is a statistically significant negative association of
student achievement with public school operation and a statistically
significant positive association with the share of public school funding
in several countries, as is the pattern in the specifications using the
cross-country variation.
Still, it should be borne in mind that the specifications using the vari-
ation within each country face the fundamental problem of selection
bias. In particular, credit constraints might be expected to bias the esti-
mates on public funding in the negative direction.
2.5 Conclusion
This paper has presented cross-country evidence on the effectiveness

of public-private partnerships (PPPs) in providing cognitive skills to
students. The main result is that across countries, public operation of
schools is negatively associated with student performance in math,
reading, and science, while public funding of schools is positively asso-

ciated with student performance in the three subjects. This suggests
that school systems based on PPPs in the sense that the state finances
schools but contracts their operation out to the private sector are the
most effective school systems. By contrast, school systems based on
PPPs in the sense that they require a lot of private funding but keep
the operation of schools in the public sector fare even worse than sys-
tems where operation and funding is either both public or both pri-
vate. Thus, the results favor the particular form of educational PPPs
where the state does the funding and the private sector runs the
schools.
While this paper has looked at the relative effectiveness of PPPs in
providing cognitive skills, aspects of efficiency, equity, and noncogni-
tive skills have been left for future research. To look at the relative effi-
ciency of PPPs, defined in terms of output per input, one would have
to add a comparison of the costs at which PPPs operate relative to
mere-public or mere-private schools. If the different systems show sys-
tematic differences in their spending levels, then effectiveness is only
one side of efficiency (note, though, that the reported results condition
on several measures of resource inputs). It would also be interesting to
analyze whether the effect of PPPs differs for students from different
parts in the performance distribution, in other words, whether the
effects are heterogeneous between elite and disadvantaged students.
Specifications that interact the public-private measures with family-
background measures could also provide evidence on equity aspects
of private involvement. It also remains an open issue whether and
how PPPs affect the noncognitive skills, behaviors, and beliefs of
students.
Another road for future research would be to look at the different
channels through which the effects of PPPs may come about. For exam-
ple, PPPs may differ from other schools in the education level of the
teachers that they hire or in the autonomy that they are granted in dif-
ferent areas, and one may get glimpses of the importance of these
different channels by comparing the presented results to regressions
that do not control for these other aspects of educational production.
Finally, countries differ in the extent to which they impose government
restrictions and graduation requirements on private schools, and some
of these regulations may limit the extent to which PPPs are allowed to
differ from public schools.
Notes
Paper prepared for the PEPG–World Bank conference, ‘‘Mobilizing the Private Sector for
Public Education,’’ Kennedy School of Government, Harvard University, October 2005. I
would like to thank the participants of the conference—in particular the two conference
organizers, Rajashri Chakrabarti and Paul Peterson—and my discussant Marty West, as
well as Giorgio Brunello and the referees for very useful comments and discussions. A
first draft of this chapter was circulated under the title ‘‘Public-Private Partnerships in
Schooling: Cross-Country Evidence on their Effectiveness in Providing Cognitive Skills’’
as Harvard University’s Program on Education Policy and Governance Research Paper
PEPG 05-09.
1. Details on the data underlying figure 2.1 will be provided in section 2.1.2. Note that
the specific subdivision into the four quadrants in the figure is arbitrary, chosen so that
each quadrant has at least four countries, and is undertaken for purposes of visualization
only.
2. Among the PISA participants, Australia and Canada did not provide data on the pub-
lic vs. private operation of schools. Liechtenstein was not included in the analysis be-
cause it features only 11 schools (with 314 students), 2 of which are private, and because
it lacks several internationally comparable country-level data. Data for the Netherlands
are provided, although there is a caveat in that the response rate in the Netherlands was
relatively low.
3. Adams and Wu (2002), OECD (2001), and the PISA Web page at hwww.pisa.oecd
.orgi provide detailed information on the PISA study.
4. Fuchs and Wößmann (2007) also provide imputed data for missing observations,
which are used in this chapter with the exception of the data on private vs. public opera-
tion and funding, for which only original data are used.
5. Important contributions to the empirical literature include Howell et al. (2002), Hoxby
(2003a, 2003b), and Neal (1997) for the United States; Bradley and Taylor (2002) and
Levaĉić (2004) for England; Sandström and Bergström (2005) for Sweden; Angrist et al.
(2002) for Colombia; Cox and Jimenez (1991) for Colombia and Tanzania; James, King,
and Suryadi (1996) and Bedi and Garg (2000) for Indonesia; and Mizala and Romaguera
(2000), Mizala, Romaguera, and Farren (2002), Sapelli and Vial (2002), Vegas (2002), and
Hsieh and Urquiola (2006) for Chile.
6. A lot of the discussion of the relative merits of public vs. private involvement in
schooling, in particular on the funding side, also surrounds equity rather than effective-
ness outcomes (cf., e.g., Glomm and Ravikumar 1992; Epple and Romano 1998; Nechyba
2000; Ladd 2002). For reasons of scope, in this chapter we do not deal with the issue that
public funding might serve to redistribute income or to raise opportunities for specific
disadvantaged subgroups of the population, leaving this important issue for future
research.
7. See the working-paper version of this paper (Wößmann 2006) for a list of the control
variables.
8. Note that given the somewhat arbitrary subdivision into the four quadrants of figure
2.1, this quadrant-dummy specification is meant as a depiction of a broad pattern only.
It receives its validity only from the fact that the depicted pattern is vindicated by the
richer specification that follows, where public operation and public funding are entered
as linear variables.
9. The values in reading and science are 33.2 and 32.4, respectively.
10. The results on public operation are consistent with evidence from a previous interna-
tional student achievement test, TIMSS, showing that student performance was superior
in countries with a larger share of private enrollment and with a larger share of public
funding going to privately operated schools (Wößmann 2001, 2003).
11. This specification and the results are very similar to those reported in Fuchs and
Wößmann (2007).
12. Previous studies have estimated similar models of the effects of private school opera-
tion within several countries from international achievement tests, but without account-
ing for differences in the source of school funding. Thus, Toma (1996; cf. also 2005)
estimates the effect of private school operation in five countries using the 1981 second
international mathematics test. She notes that the positive effect of private provision is in-
dependent of whether the countries tend to finance the schools publicly or not. Vanden-
berghe and Robin (2004) estimate the effect of private school operation in eight countries
in PISA, comparing several estimation methods that try to address selection bias, but dis-
regarding the funding side of schools.
13. The positive coefficient on public operation in the bottom left quadrant is driven by
the highly imprecise estimate in Japan.
References
Adams, Ray, and Margaret Wu, eds. 2002. PISA 2000 Technical Report. Paris: Organisation
for Economic Co-operation and Development (OECD).
Angrist, Joshua, Eric Bettinger, Erik Bloom, Elizabeth King, and Michael Kremer. 2002.
Vouchers for private schooling in Colombia: Evidence from a randomized natural exper-
iment. American Economic Review 92, no. 5: 1535–1558.
Bedi, Arjun S., and Ashish Garg. 2000. The effectiveness of private versus public schools:
The case of Indonesia. Journal of Development Economics 61, no. 2: 463–494.
Bishop, John H. 1992. The impact of academic competencies on wages, unemployment,
and job performance. Carnegie-Rochester Conference Series on Public Policy 37: 127–194.
Bishop, John H., and Ludger Wößmann. 2004. Institutional effects in a simple model of
educational production. Education Economics 12, no. 1: 17–38.
Bradley, Steve, and Jim Taylor. 2002. The effect of the quasi-market on the efficiency-
equity trade-off in the secondary school sector. Bulletin of Economic Research 54, no. 3:
295–314.
Chubb, John E., and Terry M. Moe. 1990. Politics, Markets, and America’s Schools. Washing-
ton, D.C.: Brookings Institution.
Cox, Donald, and Emmanuel Jimenez. 1991. The relative effectiveness of private and pub-
lic schools. Journal of Development Economics 34, no. 1: 99–121.
Epple, Dennis, and Richard E. Romano. 1998. Competition between private and public
schools, vouchers, and peer-group effects. American Economic Review 88, no. 1: 33–62.
Fuchs, Thomas, and Ludger Wößmann. 2007. What accounts for international differences
in student performance? A re-examination using PISA data. Empirical Economics 32, no.
2–3: 433–464.
Galor, Oded, and Joseph Zeira. 1993. Income distribution and macroeconomics. Review of
Economic Studies 60, no. 1: 35–52.
Glomm, Gerhard, and B. Ravikumar. 1992. Public versus private investment in human
capital: Endogenous growth and income inequality. Journal of Political Economy 100, no.
4: 818–834.
Gradstein, Mark, and Moshe Justman. 2002. Education, social cohesion, and economic
growth. American Economic Review 92, no. 4: 1192–1204.
Gradstein, Mark, Moshe Justman, and Volker Meier. 2004. The Political Economy of Educa-
tion: Implications for Growth and Inequality. Cambridge, MA: The MIT Press.
Greene, William H. 2000. Econometric Analysis. Fourth edition. Upper Saddle River, NJ:
Prentice Hall.
Hanushek, Eric A. et al. 1994. Making Schools Work: Improving Performance and Controlling
Costs. Washington, D.C.: Brookings Institution.
Howell, William G., Patrick J. Wolf, David E. Campbell, and Paul E. Peterson. 2002.
School vouchers and academic performance: Results from three randomized field trials.
Journal of Policy Analysis and Management 21, no. 2: 191–217.
Hoxby, Caroline M., ed. 2003a. The Economics of School Choice. A National Bureau of Eco-
nomic Research Conference Report. Chicago: University of Chicago Press.
Hoxby, Caroline M. 2003b. School choice and school competition: Evidence from the
United States. Swedish Economic Policy Review 10, no. 3: 9–65.
Hsieh, Chang-Tai, and Miguel Urquiola. 2006. The effects of generalized school choice on
achievement and stratification: Evidence from Chile’s voucher program. Journal of Public
Economics 90, no. 8–9: 1477–1503.
Human Development Network. 2001. Handbook on Public Private Partnerships in Educa-
tion. Washington, D.C.: World Bank. (http://www.ifc.org/ifcext/edinvest.nsf/Content/
PublicPrivatePartnerships)
James, Estelle, Elizabeth M. King, and Ace Suryadi. 1996. Finance, management, and
costs of public and private schools in Indonesia. Economics of Education Review 15, no. 4:
387–398.
Jimenez, Emmanuel, and Vicente Paqueo. 1996. Do local contributions affect the effi-
ciency of public primary schools? Economics of Education Review 15, no. 4: 377–386.
Ladd, Helen F. 2002. School vouchers: A critical view. Journal of Economic Perspectives 16,
no. 4: 3–24.
Levaĉić, Rosalind. 2004. Competition and the performance of English secondary schools:
Further evidence. Education Economics 12, no. 2: 177–193.
Loury, Glenn C. 1981. Intergenerational transfers and the distribution of earnings. Econo-
metrica 49, no. 4: 843–867.
Mizala, Alejandra, and Pilar Romaguera. 2000. School performance and choice: The Chil-
ean experience. Journal of Human Resources 35, no. 2: 392–417.
Mizala, Alejandra, Pilar Romaguera, and Darı́o Farren. 2002. The technical efficiency of
schools in Chile. Applied Economics 34, no. 12: 1533–1552.
Neal, Derek. 1997. The effects of Catholic secondary schooling on aecondary achieve-
ment. Journal of Labor Economics 15, no. 1: 98–123.
Nechyba, Thomas J. 2000. Mobility, targeting, and private-school vouchers. American Eco-
nomic Review 90, no. 1: 130–146.
Organisation for Economic Co-operation and Development (OECD). 2001. Knowledge and
Skills for Life: First Results from the OECD Programme for International Student Assessment
(PISA) 2000. Paris: OECD.
Patrinos, Harry A. 2000. Market forces in education. European Journal of Education 35, no.
1: 61–80.
Patrinos, Harry A. 2002. Private education provision and public finance: The Netherlands
as a possible model. NCSPE Occasional Paper No. 59. New York, NY: National Center
for the Study of Privatization in Education, Teachers College, Columbia University.
Peterson, Paul E., ed. 2003. The Future of School Choice. Stanford, CA: Hoover Institution
Press.
Pritchett, Lant. 2003. ‘‘When Will They Ever Learn’’? Why all governments produce
schooling. Bread Working Paper 31. Cambridge, MA: Harvard University, Bureau of Re-
search in Economic Analysis of Development.
Sandström, F. Mikael, and Fredrik Bergström. 2005. School vouchers in practice: Compe-
tition will not hurt you. Journal of Public Economics 89, no. 2–3: 351–380.
Sapelli, Claudio, and Bernardita Vial. 2002. The performance of private and public
schools in the Chilean voucher system. Cuadernos de Economı́a—Latin American Journal of
Economics 39, no. 118: 423–454.
Shleifer, Andrei. 1998. State versus private ownership. Journal of Economic Perspectives 12,
no. 4: 133–150.
Toma, Eugenia F. 1996. Public funding and private schooling across countries. Journal of
Law and Economics 39, no. 1: 121–148.
Toma, Eugenia F. 2005. Private schools in a global world: 2004 presidential address.
Southern Economic Journal 71, no. 4: 693–704.
Vandenberghe, Vincent, and Stephane Robin. 2004. Evaluating the effectiveness of pri-
vate education across countries: A comparison of methods. Labour Economics 11, no. 4:
487–506.
Vegas, Emiliana. 2002. School choice, student performance, and teacher and school char-
acteristics: The Chilean case. World Bank Policy Research Working Paper 2833. Washing-
ton, D.C.: World Bank.
World Bank. 2004. World Development Report 2004: Making Services Work for Poor People.
Washington, D.C.: World Bank.
Wößmann, Ludger. 2001. Why students in some countries do better: International evi-
dence on the importance of education policy. Education Matters 1, no. 2: 67–74.
Wößmann, Ludger. 2003. Schooling resources, educational institutions and student per-
formance: The international evidence. Oxford Bulletin of Economics and Statistics 65, no. 2:
117–170.
Wößmann, Ludger. 2006. Public-private partnerships and schooling outcomes across
countries. CESifo Working Paper 1662. Munich: CESifo.
3 Mobilizing the Private Sector in the United States:
A Theoretical Overview
Thomas J. Nechyba
Even in the absence of any government efforts to support private

schools, we observe a substantial number of private schools operating
successfully alongside nominally ‘‘free’’ public schools. When evaluat-
ing possible channels through which the private sector can be mobi-
lized, a first step is therefore to identify the existing channels through
which private schools manage to compete in public school environ-
ments despite charging nontrivial levels of tuition to parents. Put dif-
ferently, private schools must have some competitive advantages that
they currently leverage to attract parents, and these advantages deter-
mine the ways in which private schools form and the types of families
they attract. Public policy can then alter how the private sector evolves
in a public school environment, which families it attracts, and how the
nature of public schools changes as a result. While ‘‘mobilizing the pri-
vate sector’’ can be accomplished through a variety of policy tools, we
will use private school vouchers as a focusing device in this chapter.1
Many of the conceptual lessons, however, are more broadly applicable.
The overview in this chapter is organized around four types of com-
petitive advantages that can potentially explain the existence of private
schools. First, private schools may enjoy greater resource efficiency—
that is, they may be more effective at translating dollars into outcomes
valued by parents. Second, private schools may be able to serve niche
markets by more effectively targeting their pedagogical approach to
the needs of particular subsets of families. Third, private schools may
derive an advantage from positive peer effects by skimming the cream
of students. Finally, private schools may allow parents to unbundle
schooling and housing choices within a metropolitan area.
There is no firm consensus how these advantages combine to explain
the existence of private schools in different settings. Private schools
in Brazil are likely to exist for different reasons than private schools in
48 Thomas J. Nechyba
the United States, and even within the United States, private schools in
the more Catholic northeast may exist for different reasons than private
schools in the more Protestant south. The challenges for mobilizing the
private sector in positive ways may thus take different forms in differ-
ent regions, and the optimal design of public policies may therefore
differ depending on the context.
Our analysis begins in section 3.1 with an overview of the complex-
ity faced by education policy makers—and the resulting complexity of
an economic approach to analyzing the policy makers’ problem. I then
proceed in sections 3.2 through 3.5 to a discussion of how each of the
private school advantages relate to the benefits and costs of mobilizing
the private sector. In section 3.6 I present results from a model in
which different private school advantages are modeled and illustrate
how different assumptions we make about private school advantages
(and public school responses to competition) can alter our conclusions.
Finally, section 3.7 concludes with some final comments.
3.1 The Complex Challenges for Education Policy
A focus on different subsets of the many features of a local economy

that might be relevant for education policy may lead different policy
makers with identical policy preferences to different conclusion regard-
ing the tradeoffs involved in policy design. Policy makers may further
differ in their preferences over what outcomes to emphasize. Before
discussing the conceptual issues involved in mobilizing the private sec-
tor, I therefore discuss briefly the relevant features of the underlying
economic environment.
3.1.1 The Underlying Economic Environment

A typical economic model begins with an outline of the types of eco-
nomic agents that make decisions and respond to incentives, the endow-
ments, tastes, and production possibilities that these agents bring to
their economic decisions, and the mechanism by which individual
choices lead to an equilibrium in which every agent is doing the best
she can given the circumstances she faces. Policy plays a role in de-
fining the constraints that individuals face—and thus impacts the
decisions individuals make which then aggregate to the equilibrium
outcomes we observe.
To consider relevant economic agents, we can then begin by asking:
who are the decision makers most relevant for predicting how educa-
Mobilizing the Private Sector in the United States 49
tion policy translates into real world outcomes? First, parents decide
where to send their children and how much they are involved in mon-
itoring what happens at home and at school. They have some notion of
what they consider important (Hoxby 1999), and they face resource
constraints given the financial circumstances. Second, children may be
viewed as decision-making agents in the sense that they may exercise
some choice over which peer groups within schools and neighbor-
hoods to join and how much effort toward advancing their education
to expend (Harris 1998; Cooley 2005). Their relevant endowments are
the characteristics (such as innate ability) they bring to the school, and
their tastes may be shaped by a combination of parental, school,
and peer influences.
Third, school officials—both teachers and administrators, may re-
spond to changes in incentives. While their tastes may align to some
extent with those of parents, they may also care about personal ‘‘rents’’
(Chubb and Moe 1990; McMillan 2004; Hoxby 1996) that may be con-
strained in part by the degree of competition they face (Hoxby 2000)
or the types of accountability measures that are in place. Fourth, school
entrepreneurs operate in both the public and the private sector. As
with school officials, their tastes may take a variety of forms and may
include a desire to serve children of particular types as well as to
achieve personal rents of some form. Both school entrepreneurs and
school officials face technological constraints—the mapping of inputs
to outputs that is handed to them by the nature of education produc-
tion (Hanushek 1999; Krueger 1999).
Finally, homeowners and developers may play an important role
when public school access is determined primarily by where parents
live. Parental decisions regarding schools are then inherently bundled
with decisions regarding housing (Oates 1969; Bayer, Ferreira, and
McMillan forthcoming). This, as we will see, impacts both the evolu-
tion of housing prices within established housing markets and the
incentives by suppliers of new housing.
3.1.2 Equilibrium
In equilibrium, some mechanism must then exist to ration parents and
students into neighborhoods and schools and to determine how many
and what kinds of private schools operate. Since public schools typi-
cally do not charge tuition, other prices—in particular those associated
with the cost of housing, adjust to equilibrate supply and demand for
such schools. Alternatively, part of the ‘‘price’’ of choosing a public
school may involve monitoring schools or supplementing school pro-

duction with home production of education. In private schools, on the
other hand, explicit pricing governs much of the rationing of children
into schools. And an equilibrium typically has a political dimension as
well, with resources provided to public schools resulting from voting
by residents. School policies—including those seeking to mobilize the
private sector—thus result in observable outcomes as individuals do
the best they can given their circumstances, which arise in part from
equilibrium in housing, private school and political markets.2
3.1.3 Outcomes and Policy Preferences

Once a policy is implemented and a new equilibrium is reached, we
can observe a number of outcomes that depend on the particulars of
the policy, the way in which individuals respond to the changed incen-
tives from that policy, and the mechanisms by which equilibrium is
reached. New schools might emerge, households might move, public
school population and school resources might change, and school offi-
cials might behave differently—all leading to different outcomes for
children.
Even if agreement is reached on how a policy translates into changes
in observable equilibrium outcomes, however, policy makers may dis-
agree on the desirability of different policies because they disagree
about the appropriate objective for education policy. Are we attempt-
ing to maximize average achievement, or achievement of those with
the least initial opportunities, or the achievement of the ‘‘cream’’ that
might become the engine of economic growth? Is achievement as mea-
sured through standardized exams the most important variable of
interest? Is one of the aims of the education system to internalize exter-
nalities not taken into account by parents? It is one thing to mobilize
the private sector in education, but quite another to do so in such a
way that it is consistent with potentially competing social aims.
3.1.4 Beginning Simple and Introducing Complexity

Education policy therefore faces a number of complex challenges, and
there is little hope of unraveling all of these at one time. It is then most
productive to begin with simple and admittedly unrealistic settings
and to investigate the most salient forces that may be unleashed by a
greater fostering of private school markets. In the following sections,
I will therefore discuss particular issues raised by particular private
school advantages—and how this translates into different ways of

thinking about how private school markets can be mobilized most ef-
fectively given particular social aims.
3.2 A Simple Economic Environment: The Role of Resource

Efficiency
Suppose we begin by restricting ourselves to a world in which parents

and children are identical, in which there are no geographical con-
straints that limit access to schools (and thus no role for housing mar-
kets), and in which we can think of public schools as a single entity.
Parents in this world agree on what makes a good school, and peer
effects—to the extent they exist—are the same in all schools. Of our
four possible private school advantages, we have thus eliminated
three—with no private school advantage resulting from grouping
(identical) peers either by ability or through pedagogical targeting,
and no advantage to parents of private-school attending students from
unbundling housing and school decisions. All that remains is the pos-
sibility that private schools may arise as a result of inefficient use of
resources in the public school.
In this scenario, private schools emerge to the extent private school
entrepreneurs find sufficiently more innovative and efficient uses of
resources to produce sufficiently higher school quality that parents are
willing to pay for despite their access to the free public school. This
may, for instance, involve less rent-seeking behavior or less wasteful
use of resources. Alternatively, even with identical peers, private
school entrepreneurs may find ways to harness peer pressures and
thus coordinate students (who pay attention to what their peers think)
on ‘‘higher effort’’ peer equilibria.
3.2.1 Voice versus Exit and the Free Rider Problem

Parents can now either exercise pressure (‘‘voice’’) at the public school
in an attempt to extract greater effort from school officials while sup-
plementing their children’s education at home, or they can seek out
private school alternatives. The former, however, gives rise to a free
rider problem as each parent would prefer to have other parents
undertake the costly task of monitoring the public school, and in the
absence of parents solving the resulting coordination problem, eco-
nomic theory suggests that a suboptimal level of overall pressure will
be exerted within the public school. In the presence of a competitive

private school market, on the other hand, competition by private
schools with one another reduces (or, under perfect competition, elimi-
nates) inefficiencies within that sector, leaving parents free to sub-
stitute costly monitoring efforts (and home production) with tuition
payments.
3.2.2 Equilibrium
An equilibrium with a coexistence of private and public schools can
then emerge in a number of ways. As private schools enter, the public
school shrinks in size—thus reducing the free rider problem and in-
creasing the pressure felt by public school officials as parents can more
effectively coordinate. If efficient private schools can be easily repli-
cated, the resulting equilibrium would be one in which no rents are
possible (due to competition) in the private sector, and all parents are
indifferent between the lower-quality public school and the higher-
quality private schools (that charge tuition). Alternatively, if entrepre-
neurial talent in the private sector is scarce, each new private school
would be less efficient than existing private schools, and rents within
the private school sector would remain for more efficient entrepreneurs
who can charge higher tuitions for higher school quality. Any equilib-
rium, however, would be one in which private schools would provide
higher quality education than the public school, with the possibility
that private schools themselves can be hierarchically ordered based on
the quality they offer (due to the entrepreneurial skill differences in the
private sector). Parents would be uniformly happier, with some chil-
dren (whose parents pay higher tuition) experiencing higher school
quality than others.
3.2.3 Mobilizing the Private Sector in this Simple Economic Setting

Policy aimed at mobilizing the private sector in such a setting—either
through the removal of barriers to entry for private schools or through
explicit tuition subsidies in the form of vouchers, then gives rise to rel-
atively straightforward changes in outcomes that are almost certainly
positive from most policy perspectives. All students experience an im-
provement in school quality as a result of such mobilization as compe-
tition from the private sector serves to reduce resource inefficiencies.
Unless one places sufficiently high negative weight on the emergence
of variance in outcomes, fostering private school markets is unambigu-
ously positive.
3.2.4 Introducing Income Heterogeneity

Now suppose that households differ in one respect: their income. This
would result in a potentially more disturbing feature of the equilib-
rium in that educational quality for children could become correlated
with parental income as higher income parents are more willing to
pay private school tuition. At the same time, the introduction of in-
come heterogeneity raises the possibility of different ways of fostering
private school markets—with tuition subsidies potentially set in in-
verse proportion to household income. Thus policy could be designed
to empower the poor to exert greater competitive pressure through the
creation of a private school market that serves the poor.
3.2.5 Introducing Politics

We have implicitly assumed above that per-pupil public resources are
exogenously fixed as private school markets are mobilized. This means
school quality for all children increases as a result of more efficient
resource use in both public and private schools (with children attend-
ing private school experiencing disproportionately larger increases in
school quality—offset by disproportionately higher tuition costs for
their parents.) It is unlikely, however, that public school resources are
unaffected by increasing activity in the private sector. Several compet-
ing forces emerge: First, to the extent to which policies aimed at mobi-
lizing private school markets induce private-school-attending students’
parents to supplement public subsidies with their own resources, more
resources may be available (on a per-pupil basis) in the public sector.
Second, to the extent to which the political voting equilibrium changes
as private-school-attending students’ parents reduce their political sup-
port of public schools, the overall resources allocated to public schools
may shrink. Depending on which of these forces dominates, public
schools may experience either an increase or a decrease in per-pupil
resources, with simulations suggesting that the latter force dominates
for initial expansions of the private sector while the former domi-
nates for larger expansion (Nechyba 2003a).
In addition, a more complex (and, in some circumstances, more
realistic) model of the politics surrounding public school funding
might grant disproportionately more political power to higher-income
parents. To the extent to which the private sector focuses dispropor-
tionately on higher-income families, public schools that serve the poor
might therefore be much more adversely affected than suggested by a
simply ‘‘one-man, one-vote’’ model. This may arguably be precisely
what has happened in countries like Brazil, where notoriously ineffi-

cient and underfunded public schools serve only the poor while
middle- and high-income families almost universally attend private
schools.
3.3 Pedagogical Targeting and Horizontal Differentiation
Now suppose we return to the simple setting at the beginning of the

previous section but we introduce to this setting ‘‘horizontal’’ hetero-
geneity of children. By ‘‘horizontal’’ heterogeneity, I mean simply that
different children respond differently to alternative pedagogical ap-
proaches or that parental tastes on particular characteristics of schools
(such as the degree of religious education) differ. Placing all children
in the public school might then be inefficient if that school has to com-
mit to a single or a limited set of pedagogical approaches, and private
schools gain a competitive advantage by differentiating themselves
horizontally (Ferreyra 2007; Cohen-Zada and Justman 2005). This
could be viewed as a special case of the resource efficiency advantage
of private schools discussed in the previous section—with the public
school inefficiency now arising from the limited pedagogical ap-
proaches in public schools rather than from rent seeking.
3.3.1 Equilibrium without Peer Effects

Let us begin with the case where there are no cross-type peer effects—
that is, conditional on a particular pedagogical approach being used, a
student of type A neither benefits nor is hurt by having a student of
type B in the same classroom. Separation of types is then unambigu-
ously positive for all students since it allows a more targeted pedagog-
ical approach for each student. Some might benefit more from this
separation than others—implying that some (of the otherwise identi-
cal) parents are willing to pay more for their child to separate than
others. Each parent then makes a choice of whether to attend a private
school targeted to her type of child and pay the corresponding private
school tuition, or to send her child to public school that is less targeted
(and thus of lower quality for her child) but free.
In equilibrium, we would then expect private schools to target those
types of children that benefit the most from separation, with the re-
maining children attending public school. Those children attending
private schools will experience higher school quality given the targeted
pedagogical approaches, while those in the public school will experi-
ence no decline in school quality and potentially an increase if the pub-

lic school can target more effectively given that some student types
have exited to the private sector. The higher ‘‘quality’’ of horizontally
targeted private schools, however, refers to quality as subjectively
judged by parents—and this may be at odds with how policy makers
define quality. To take an extreme example, some parents may view
racial homogeneity of a school as an important dimension of subjec-
tively defined quality, causing schools to erect racial barriers to entry
into the school. Policies might therefore require particular demo-
graphic mixes of private schools if larger social aims (outside maximiz-
ing parental satisfaction) are to be achieved.
3.3.2 Cost Differences of Different Types

It may also be the case, however, that the pedagogical needs of differ-
ent types of children require different levels of per-child resources. For
instance, learning-disabled children, or children with different native
languages, may require more. As a result, the private sector would be
more likely to attract parents of ‘‘low cost’’ children first, leaving ‘‘high
cost’’ children in the public school. Again, however, policies to mobi-
lize the private sector might be designed to take these cost differences
into account, offering, for instance, higher voucher levels for children
with special needs.
3.3.3 Introducing Cross-Type Peer Effects

The efficiency and welfare implications of mobilizing the private sector
become murkier, however, once we introduce the possibility of ‘‘cross-
type’’ peer effects. Such peer effects arise when, conditional on a par-
ticular pedagogical approach, students of type A benefit from the
presence of students of type B (or vice versa), or alternatively if stu-
dents of type A are hurt by the presence of students of type B (or vice
versa). To the extent to which these peer effects are always negative,
the case for separation of types becomes even more compelling—
increasing the potential for efficiency enhancements from private
schools. If the effects are positive in both directions (in ways recog-
nized by parents), some mixing of types may be optimal—but private
schools would have an incentive to structure schools accordingly since
parents of both types would prefer mixing over separation. There are,
however, two types of scenarios under which the private sector might
induce separation of types that is suboptimal. The first arises when
peer effects are positive in one direction and negative in the other, a
case treated in section 3.4. The second arises if the peer effects take the
form of a larger externality that parents themselves have no incentive
to internalize.
Consider, for instance, the possibility that interaction of types results
in a greater awareness of the needs of others, a greater respect for di-
versity, and, in the long run, a better equipped citizenry that can result
in a more harmoniously functioning society. Parents themselves may
place limited value on their own child developing. Put differently,
such externalities result in a classic free rider problem, where each
parent would like to free ride on the benefits from diversity in other
schools while maximizing her own children’s human capital accumula-
tion. Under this scenario, a mobilization of the private sector will result
in too much separation—separation that maximizes production of
human capital at the expense of other social goals. This is not, how-
ever, to say that mobilizing the private sector per se will inevitably re-
sult in too much separation in the presence of these larger externalities.
Rather, to the extent to which policy makers believe that broader social
goals might be achieved, policies aimed at mobilizing the private sec-
tor would have to be designed more specifically with this goal in
mind. For instance, private school voucher amounts could depend on
the demographic mix of students within the private schools that accept
the voucher.
3.4 ‘‘Cream Skimming’’ and Vertical Differentiation
Suppose now that student ability varies in a ‘‘vertical’’ rather than a

‘‘horizontal’’ fashion—in other words, some students have higher
ability than others, with no gains from differentiating pedagogical ap-
proaches but potential gains from offering different curricula to differ-
ent children. Higher-ability students may benefit, for instance, from a
more accelerated curriculum than low-ability students.
3.4.1 Equilibrium without Peer Effects

In equilibrium, we would then expect private schools to differentiate
themselves based on the curriculum they offer, much as was the case
for private schools differentiating themselves based on their pedagogi-
cal approach in the previous section. In the absence of peer effects,
this simply results again in more specialized schools that serve differ-
ent student types more effectively, with parents whose children bene-
fit disproportionately from separation first in line to attend private

schools. As in the previous section, the introduction of household in-
come differences gives rise to equity concerns as high-income parents
become more able to take advantage of targeted private schools, and
this concern is exacerbated when parental income is inversely correlated
with the costliness of curriculum needs. In principle, these concerns
can again be addressed through careful design of voucher policies that
vary voucher levels based on household income and student type.
3.4.2 Introducing Vertical Peer Effects and Cream Skimming

The case for increasing the role of private schools becomes more prob-
lematic, however, in the presence of what I will call vertical peer ef-
fects. Such effects arise when students can be hierarchically ordered in
such a way that students of type A benefit students of type B but stu-
dents of type B hurt the achievement of students of type A. Most often,
this is modeled as students of high ability benefiting students of low
ability and students of low ability hurting the achievement of students
with high ability.
The presence of such peer effects now introduces an incentive for
private schools to erect barriers to ‘‘low peer quality’’ students. Such
barriers are not needed in the absence of peer effects since separation
of types in such a case arises from the decentralized decisions that
individuals make as they choose schools. In the presence of such peer
effects, however, parents with low-ability children may find private
schools aimed at high-ability children attractive because of the positive
peer effects from high peer quality in those schools. Thus, schools have
an incentive to deliberately ‘‘skim the cream’’ off the public schools.
While the concern that voucher policies might lead to a decline in
resources for public schools is often raised, it is not always made ex-
plicit that perhaps the most important of these ‘‘resources’’ in fact have
little to do with financing and much more to do with the types of stu-
dents and parents that are attracted away from public schools.
3.4.3 Different Means of Cream Skimming

Private schools then have two types of approaches at their disposal.
First, they can design tuition policies with the aim of forcing parents of
lower peer quality children to face the costs they are imposing on the
school by placing their children there (Epple and Romano 1998; Cau-
cutt 2001). Under this approach, we would expect private schools to
price-discriminate based on student types, potentially offering scholar-

ships to high-ability children and high tuition to low-ability children.
In equilibrium, this would imply that the only low-ability children that
end up in high peer quality private schools will be those whose parents
have high incomes and can afford the higher tuition—leaving the pub-
lic school with a mix of different student types with parents that are
disproportionately poor. The private sector would then be character-
ized by a hierarchy of schools ordered on the peer quality they offer,
with each of the schools being of higher quality than the public school.
And private school pricing would ‘‘internalize’’ the peer externalities in
the private sector, thus enhancing efficiency (at the expense of raising
genuine equity concerns).
Alternatively, private schools can simply choose to screen students
and admit only those that satisfy some minimum peer quality level
(Nechyba 1999, 2000, 2003a,b,c; Ferreyra 2007). A similar hierarchy of
schools would be expected to emerge, but the composition of students
within any given school would be more homogeneous because pricing
is not used to compensate for the presence of high-cost students.
3.4.4 Ability Tracking as a Response by Public Schools

In the absence of peer effects, the public school may be concerned
about losing students to the private sector and may thus attempt to
use resources more efficiently to minimize such exits. In the presence
of horizontal peer effects (such as those discussed in section 3.3), this
may cause public schools to increase the types of pedagogical ap-
proaches offered. But in the presence of vertical peer effects, a much
stronger incentive emerges for public schools to respond to private
school cream skimming since the exit of high-ability students and
households carries with it a negative externality for the remaining
students.
A possible response by the public sector is then to target resources
more directly on those students that are most likely to exit—that is,
students of high ability and high-income parents. For instance, one
might expect an equilibrium response to result in increased ability
tracking combined with high-income parents exerting disproportional
influence on which track their children are assigned to. This, in es-
sence, could result in ‘‘schools within schools’’ where public schools re-
spond to the private school advantage by selecting peer quality within
the public school tracks (Epple, Newlon, and Romano 2002; McHugh
2005).
In the absence of such tracking, we have said that one would expect
an equilibrium hierarchy of school quality, with private schools differ-
entiated by peer quality, and the public schools offering lower quality
than all private schools. With the emergence of tracking, on the other
hand, this hierarchy of school quality may take on a different form,
with the lowest track in the public schools offering the lowest quality,
but with some private schools offering school quality below the higher
public school tracks. This is worrisome from an equity perspective
since the logic again predicts that those who end up with the lowest
school quality will be poor households with relatively low-ability chil-
dren. And, with the introduction of vertical peer effects, school quality
for such children may well fall with the mobilization of private
schools.
As before, however, such concerns can, in principle be addressed by
the way in which vouchers are designed (Epple and Romano 2002;
Hoxby 2001). For instance, one can design vouchers to be inversely re-
lated to household income and vary depending on student type—thus
offering increased school resources to those who find it disproportion-
ately difficult to afford private school tuition and those whose children
are disproportionately costly.
3.5 Unbundling of Schooling and Housing Choices
In everything we have discussed thus far, we have explicitly assumed

away any geographical considerations—and this in turn has permitted
us to treat the public sector as if it consisted of a single public school
and has allowed us to focus on three types of private school advan-
tages: increased resource efficiency, targeted (horizontal) pedagogical
targeting, and (vertical) cream skimming. As is abundantly clear to
even the most casual observer of school policy debates, however, a
model in which all public schools are treated as if they are the same
does not match empirical reality in contexts such as the United States,
where the public sector is characterized by vastly different schools in
different geographically defined neighborhoods and districts. The pres-
ence of substantial (vertical) differences in public school quality then
suggests the presence of important frictions that result in an equilib-
rium sorting of parents and students into different public schools. And
the main such friction undoubtedly arises from the addition of geo-
graphical constraints that impact the choice set from which parents of
different types can choose.
3.5.1 Sorting into Heterogeneous Public Schools through Housing

Markets
Even if parent A places less value on school quality than parent B, he
would choose a higher-quality school over a low-quality school if both
are equally costly just as consumers will always choose higher-quality
cars over lower-quality cars if all cars are equally priced (even if they
differ in terms of how much value they place on car quality). Thus,
even when parental tastes for school quality differ, some sorting mech-
anism must be in place that results in some children ending up in
better public schools and others in worse public schools. This sorting
mechanism, at least in the United States, is closely related to how
parents gain access to different public schools.3 The right to attend a
particular public school is typically given to those who reside within
some geographically defined region that is served by that public
school. In some states, school districts are small and contain a small
number of schools to choose from. In other states, school districts are
larger, with each district divided into neighborhoods that are served
by particular schools.4 But in each of these cases, the choice of attend-
ing a particular public school is bundled with the choice of residential
location and thus linked to housing markets that price access to
schools. As a result, public schools are priced.
3.5.2 Quasi-Public School Markets and Residential Segregation

Public schools exist in part because of a collective desire to provide ac-
cess to educational opportunities that are unrelated to the economic
circumstances of parents and partly because of a sense that schools
should internalize larger externalities (such as those discussed in sec-
tion 3.3.3). In practice, however, the bundling of public school access
to private housing markets results in a quasi-public system that, while
maintained through taxpayer contributions and shaped by political
processes, contains elements of a private system in that access is
priced—resulting in strong correlations of educational opportunities
for children with economic circumstances of parents as well as consid-
erably less mixing of different types of students than one would expect
if school assignments were random.
Since pricing of public school access is through residential housing
markets, however, the quasi-pricing of public schools has an additional
effect that differs from explicit pricing in private school markets in that
it introduces an explicit economic rationale for residentially segregating
households along income lines. Empirically based simulations of hous-
ing and school markets suggest that this force may be quite powerful,
resulting in levels of income segregation that are substantially greater
than one would expect in the absence of school considerations
(Nechyba 2003b). To the extent that peer externalities operate not only
within schools but also within residential neighborhoods, this raises
additional concerns related to persistent achievement differences.
3.5.3 The Role of Private Schools in Quasi-Public School Markets

Quasi-pricing of school access through housing markets then offers the
final of our four potential competitive advantages for private schools.
Consider, for instance, a middle-income family struggling to pay in-
flated housing prices in a good public school district for the sole reason
of gaining access to that public school. A private school entrepreneur
may then open a school in a neighboring district with low public
school quality in order to permit this household to divorce its housing
choice from its school choice. While the household would now have to
pay school tuition, it is also free to take advantage of bargains in the
housing market of the worse public school district. Thus, the advan-
tage that emerges for private schools arises precisely because of the
quasi-public nature of public schools.
Even in the absence of explicit policies aimed at mobilizing the
private sector, we would therefore expect to find private schools in
lower-income districts in part to allow some households to reside in
larger houses within that district while not being affected by the public
school quality offered. Empirically based simulations in fact suggest
that private school markets may be playing a substantial role in mod-
erating the levels of income segregation we observe—while at the
same time reducing the housing price differences that would persist in
a purely quasi-public system (Nechyba 2003b). Thus, while the quasi-
public nature of public schools in residence-based systems gives rise to
residential segregation, the presence of a simultaneous private sector
tends to ameliorate such residential segregation.
While our previous discussions suggested that an equilibrium can
typically be characterized by a hierarchy of school quality in which the
public school offers the lowest level of school quality, the addition of
the quasi-public nature of public schools now suggests a somewhat
more complicated equilibrium hierarchy. The public school in the low-
est quality school district must still offer the lowest quality in the sys-
tem overall, with all private schools attracting parents through higher
quality. However, it is now quite plausible (and empirically likely)
that an equilibrium will contain some public school districts that offer
school quality higher than what is offered by some private schools
located in lower-quality public school districts, with quasi-prices in
those public schools higher than explicit tuition in lower-quality pri-
vate schools elsewhere. The previous hierarchy of private schools dom-
inating public schools thus continues within a school district but not
necessarily across school districts.
3.5.4 Mobilizing the Private Sector in a Quasi-Public School

Environment
None of this, of course, lessens the importance of issues raised within
a framework that treats the public school sector as a single public
school. Rather, it adds an additional layer of complexity in which those
forces now interact with the additional considerations raised by the
bundling (and unbundling) of school and housing choices. It further-
more focuses attention on the role of school financing on residential
segregation.
Consider a general voucher policy aimed at mobilizing the private
sector in a quasi-public school. In the absence of residential mobility,
we are left with precisely the same forces to consider as we did in the
previous sections—except that these forces would now play out within
each district. In addition, however, one would expect inter jurisdictional
effects to emerge as parents of public school attending students in bet-
ter school districts reevaluate their decision to pay high housing prices
in order to access good public schools. While it might be unrealistic to
assume that a voucher policy will cause households to immediately
move, over time, as households move for reasons related to job loca-
tion or family expansion, we would expect them to make housing
and schooling choices differently than before the introduction of the
voucher policy.
While private schools might therefore initially attract families within
districts in which they form, the theory predicts that increasingly such
private schools would market themselves across districts—causing
inflows of households that, on average, will have higher income than
those currently in the district. Local public school spending may then
rise (because a larger tax base and shrinking public school population
combine to make it easier to raise per-pupil spending, or it may fall if
private-school-attending students’ parents are sufficiently powerful in
the political process (Nechyba 2003a). In addition, local public schools
may suffer from the exit of good peer quality students, but they may
benefit from reduced resource inefficiency due to competitive pres-

sures. In short, a variety of competing effects may cause local public
schools to improve or become worse.
Empirically based simulations suggest that the long-run impact of
the partial unbundling of housing and schooling choices through gen-
eral voucher policies will be more inter- rather than intra-jurisdictional
(Nechyba 1999, 2000, 2003a,b,c,d). This suggests that, while private
schools would emerge disproportionately in lower-income districts,
their clientele will be drawn disproportionately from higher-income
districts. To the extent to which cream skimming by private schools is
an important factor in the emergence of private schools, this further-
more suggests that public schools in higher-income districts may suffer
disproportionately.
3.5.5 Implications for Targeting Vouchers

I have argued in previous sections that equity concerns that emerge as
we think about mobilizing the private sector can be addressed by differ-
entially targeting vouchers to households of different types (and, in some
instances, to schools exhibiting different characteristics). The introduc-
tion of the quasi-public nature of public schools now introduces the
possibility that policy can be aimed at encouraging unbundling of
school and housing decisions by targeting geographically rather than tar-
geting households. If a voucher is targeted to residents who reside in
low-performing (or poor) districts, any household that moves to the
district qualifies for the voucher. Thus, a district-targeted voucher dif-
fers dramatically from a household-targeted voucher in that it can
spread the impact of increased school competition to non-targeted dis-
tricts (Nechyba 2000). In fact, general equilibrium simulations in previ-
ous work suggest that over two thirds of the impact of geographically
targeted vouchers arises from migration—with middle-income house-
holds moving to lower-income (targeted) districts to take advantage of
the vouchers. Such effects cannot arise when targeting is toward low-
income families.
3.5.6 The Political Economy of Voucher Design

The realistic modeling of public schools within the context of hous-
ing markets also has potentially dramatic implications for the political
economy of voucher design. In the absence of modeling the public sec-
tor as quasi-public and linked to housing markets, political economy
considerations are squarely focused on the impact that vouchers would
have on the costs and benefits of accessing quality schools for parents.
But once the link to housing markets is taken into account, these con-
siderations may ultimately be outweighed by considerations related to
capital gains and losses that homeowners are likely to experience in
an environment in which the private sector has been mobilized by un-
bundling housing and schooling choices. The desire of policy makers
to mobilize private schools to enhance educational opportunities there-
fore runs into a political constraint due to the effects of such policies for
the distribution of homeowner wealth (Brunner, Sonstelie, and Thayer
2001).
This suggests that, as different methods of targeting vouchers
funded by higher-level governments are considered, there is an inher-
ent bias toward targeting vouchers to low-income households if poli-
tics is disproportionately influenced by resources of higher-income
households. Such targeting in essence isolates the competitive effect
from a more active private sector to low-income areas—thus reduc-
ing the impact on homeowner wealth through changes in housing
prices. While the most effective way of spreading a competitive effect
throughout the public school system may therefore involve geographic
targeting, it seems unlikely that such targeting by higher-level govern-
ments is politically feasible without other offsetting public policies. At
the same time, such targeting could emerge ‘‘from the bottom up’’—
with districts such as inner cities providing vouchers to residents of
the district in order to draw residents from suburbs back into cities.
3.6 Predicting the Impact of Mobilizing Private Schools
Our preceding discussion suggests a number of competing effects as

the private school sector increases its activity—with the effects depend-
ing on what assumptions we make about the competitive advantages
of private school and the particulars of policies employed to mobilize
the private sector. The question whether competition is good or bad is
therefore much too coarse. The real question is what kinds of competi-
tion are likely to lead to the types of policy outcomes aimed for by pol-
icy makers, and what kind of policy support for private initiative leads
to outcomes in line with policy goals.
Structural economic modeling linked to empirical data can give us
some sense of which predictions regarding vouchers are relatively ro-
bust to changing assumptions and which are sensitive to what we as-
sume about private school formation and public school responses. In
work that takes as its benchmark the quasi-public sector operating

alongside a private sector in New Jersey, I have simulated (in previous
work) different types of private school sectors with competitive advan-
tages analogous to the four types discussed in this chapter. With the
quasi-public sector operating in poor, middle-income, and wealthy dis-
tricts, the model can then simulate the impact of various policies.
While results on different types of vouchers (and other school fi-
nance policies) are available elsewhere, I report here only one set of
simulations in an attempt to illustrate how disagreements regarding
the desirability of vouchers can emerge directly from different assump-
tions about the private and public sector. The table that follows reports
simulated school quality indexes for different voucher amounts (not
targeted in any way) on public school quality in the three different dis-
tricts. The possibility of families unbundling their school and housing
choices is maintained throughout as is a level of peer externality con-
sistent with empirical observations. ‘‘School quality’’ refers to the sub-
jective evaluation of schools by parents (as inferred from house price
differences for identical houses across districts).
In the first third of the table, the only private school advantage
(aside from allowing households to unbundle their housing and
schooling choices) derives from their ability to engage in vertical cream
Table 3.1
Public School Quality as (Nontargeted) Vouchers Are Introduced
Voucher amount
$0 $1,000 $2,500 $4,000 $5,000
Cream skimming only
Poor district 69.97 68.05 65.82 39.83 ***
Middle district 100.00 98.80 89.43 78.93 44.59
Wealthy district 126.31 120.22 112.96 93.19 80.27
Cream skimming and pedagogical targeting
Poor district 70.36 76.46 80.55 81.61 76.85
Middle district 100.00 101.52 104.96 105.99 101.55
Cream skimming and competitive resource efficiency
Poor district 65.72 67.42 69.81 71.08 71.74
Middle district 100.00 101.83 104.90 107.68 109.75
Source: Adapted from tables 5c and 6a in Nechyba (2003a). School quality is normalized
to be equal to 100 in the middle-income district in the absence of vouchers.
skimming. As private schools are fostered through vouchers, the pub-

lic schools therefore lose high peer quality students, and public school
quality for those who remain in public schools drops (at an increasing
rate as voucher amounts increase).5 To the extent to which one believes
that vertical cream skimming is the primary competitive advantage of
private schools, one might therefore be concerned about the impact
that an increasingly active private school market has on public school
quality.
The middle portion of the table then introduces horizontal (peda-
gogical) targeting, the private school advantage discussed in section
3.3. With empirically plausible levels of this advantage, the simulations
suggest that it is plausible for public school quality to increase with a
more active private school market (as public schools can more effec-
tively target their resources on the student types that remain in the
public system). Similarly, the lower portion of the table introduces
resource efficiency on the part of private schools and competition-
induced increases in public school efficiency. Again, plausible levels
of such effects can result in the prediction that public school quality
will increase with greater competition. Altering our assumption about
what private schools do and how public schools respond within em-
pirically plausible ranges therefore can alter dramatically how we
think private school competition will affect the public sector.
The only prediction of the model that appears to be completely ro-
bust to changing assumptions about private and public schools is the
residential desegregating effect of increased private school competi-
tion. To the extent that larger externalites from diversity are present
and operate both within schools and neighborhoods, this may amelio-
rate concerns raised by greater segregation in the school sector.
3.7 Conclusion
This paper suggests that private schools must operate with some
competitive advantages relative to public schools in order to attract
households—and that the nature of these advantages is likely to shape
our view of how the private sector can be most effectively mobilized to
advance academic achievement and other social goals. Rather than
asking whether competition is desirable, the discussion suggests more
nuanced policy questions asking what kinds of policies are most likely
to advance the aims of policy makers depending on the underlying
economic realities. Throughout, it is important to realize that, at least
in the United States, there really is no such thing as a ‘‘public school,’’

given the private school characteristics of public schools that emerge
when access is rationed through housing markets. It is similarly un-
likely that there will ever be such a thing as a fully ‘‘private school,’’
given that private schools are subject to government oversight and
given that explicit government support for private schools will almost
certainly be conditional on certain public aims being met.
In the context of private school vouchers, policy makers can choose
from a variety of methods of targeting vouchers in order to address po-
tential concerns. Broadly speaking, such targeting can be done along
four different dimensions: (1) by parental characteristics (such as in-
come); (2) by student characteristics (such as learning disabilities, apti-
tude, etc.); (3) by school characteristics (such as the demographic
composition of the voucher-accepting private school); or (4) by the
characteristics of the neighborhood in which the household lives (such
as the level of underperformance of the local public school).
While different assumptions about private schools and public school
responses to competition give rise to a number of concerns, it is unclear
at this point to what extent these concerns ought not apply equally
to public schools given the quasi-public nature of schools in which
other prices (such as property values) substitute for explicit tuition.
Furthermore, it appears plausible that such concerns can be addressed
through voucher design (along the four dimensions previously dis-
cussed), perhaps even more effectively than they could be in the ab-
sence of vouchers within the context of a quasi-public school system.
Put differently, policies aimed at mobilizing the private sector can
come in many shapes and flavors, leaving much room for policy mak-
ers to design with an aim of meeting specific policy goals.
Notes
1. Other policies aimed at mobilizing the private sector include elements of the charter
school and home schooling movement, support for privately funded enrichment pro-
grams, Saturday and Sunday programs, Internet supplements to traditional schooling,
industry partnerships, etc.
2. Yet another component of the overall equilibrium that is not emphasized in this chap-
ter involves equilibrium in the teacher labor market. See, for instance, Loeb and Page
(2000).
3. In other contexts, such as some of the European systems, access to different types of
public schools is based on academic achievement—making the ‘‘ability tracking’’ model
of section 3 a more relevant model than the quasi-public school model discussed here.
4. Sometimes some degree of choice within a district exists, but even that is often subject
to capacity constraints that cause those living close to a school within the district to have
the right to attend and those that live closer to another school excluded by those con-
straints or by transportation costs involved in getting to the alternative school.
5. The public school in the poor district actually ceases to exist when the voucher level
reaches $5,000, the minimum per pupil spending level assumed for a school to exist. The
more dramatic drops in quality in each district take place when private school attendance
in the district increases sufficiently to ‘‘tip’’ the political support for public schools.
References
Bayer, P., F. Ferreira, and R. McMillan. Forthcoming. A unified framework for measuring
preferences for schools and neighborhoods. Journal of Political Economy.
Brunner, E., J. Sonstelie, and M. Thayer. 2001. Capitalization and the voucher: An analy-
sis of precinct returns from California’s Proposition 174. Journal of Urban Economics 50, no.
3: 517–536.
Caucutt, E. 2001. Peer group effects in applied general equilibrium. Economic Theory 17:
25–51.
Chubb, J., and T. Moe. 1990. Politics, Markets and America’s Schools. Washington, D.C.:
Brookings Institution.
Cohen-Zada, D., and M. Justman. 2005. The religious factor in private education. Journal
of Urban Economics 57, no. 3: 391–418.
Cooley, J. 2005. Desegregation and the achievement gap: Do diverse peers help? Duke
University working paper, Durham, N.C.
Epple, D., E. Newlon, and R. Romano. 2002. Ability tracking, school competition, and the
distribution of educational benefits. Journal of Public Economics 83: 1–48.
Epple, D., and R. Romano. 1998. Competition between public and private schools, vouch-
ers and peer group effects. American Economic Review 88, no. 1: 33–62.
Epple, D., and R. Romano. 2002. Educational vouchers and cream skimming. Working
Paper No. 9354. National Bureau of Economic Research, Cambridge, Mass.
Ferreyra, M. 2007. Estimating the effects of private school vouchers in multi-district

economies. American Economic Review 97: 789–817.
Hanushek, E. 1999. Some findings from an independent investigation of the Tennessee
STAR experiment and from other investigations of class size effects. In Earning and
Learning: How Schools Matter, ed. S. Mayer, and P. Peterson. Washington, D.C.: Brookings
Institution.
Harris, J. 1998. The Nurture Assumption. New York: The Free Press.
Hoxby, C. 1996. How teachers’ unions affect education production. Quarterly Journal of
Economics 111, no. 3.
Hoxby, C. 1999. The effects of school choice on curriculum and atmosphere. In Earning
and Learning: How Schools Matter, ed. S. Mayer, and P. Peterson. Washington, D.C.: Brook-
ings Institution.
Hoxby, C. 2000a. Does competition among public schools benefit students and tax-
payers? American Economic Review 90, no. 5.
Hoxby, C. 2000b. Do private schools provide competition for public schools? Working
paper. Harvard University, Cambridge, Mass.
Hoxby, C. 2001. Ideal vouchers. Working paper. Harvard University, Cambridge, Mass.
Krueger, A. 1999. Experimental estimates of education production functions. Quarterly

Journal of Economics 114, no. 2: 497–532.
Loeb, Susanna, and Marianne Page. 2000. Examining the link between teacher wages and
student outcomes: The importance of alternative labor market opportunities and non-
pecuniary variation. Review of Economics and Statistics 82, no. 3: 393–408.
Manski, C. 1992. Education choice (vouchers) and social mobility. Economics of Education
Review 11, no. 4: 351–369.
McHugh, C. 2005. School competition and empirical evidence on the strategic behavior of
schools. Ph.D. dissertation, Duke University.
McMillan, R. 2004. Competition, incentives, and public school productivity. Journal of
Public Economics 88: 1871–1892.
Nechyba, T. 1999. School finance induced migration patterns: The impact of private
school vouchers. Journal of Public Economic Theory 1, no. 1: 5–50.
Nechyba, T. 2000. Mobility, targeting and private school vouchers. American Economic Re-
view 90, no. 1: 130–146.
Nechyba, T. 2003a. Introducing school choice into multi-district public school systems. In
The Economics of School Choice, ed. C. Hoxby, 145–194. Chicago: University of Chicago
Press.
Nechyba, T. 2003b. School finance, spatial income segregation and the nature of com-
munities. Journal of Urban Economics 54, no. 1: 61–88.
Nechyba, T. 2003c. Centralization, fiscal federalism and private school attendance. Inter-
national Economic Review 44, no. 1: 179–204.
Nechyba, T. 2003d. Public school finance and urban school policy: General vs. partial
equilibrium analysis. Brookings-Wharton Papers on Urban Affairs, 139–170.
Oates, Wallace. 1969. The effects of property taxes and local public spending on property
values: An empirical study of tax capitalization and the tiebout hypothesis. Journal of Po-
litical Economy 77, no. 6: 957–971.
4 The Practice of Public-Private Partnerships
Norman LaRocque
Education sectors the world over are facing a number of social, educa-
tional, and economic challenges. Many governments have responded
to these challenges by introducing market-based education policies
emphasizing choice, managerial autonomy, and accountability for
results. This chapter provides an overview of international examples
of government agencies contracting for the delivery of education ser-
vices, including contract schools, charter schools, voucher-type pro-
grams, and infrastructure public-private partnerships. It concludes by
drawing some tentative lessons for the design and implementation of
contracting in the education sector.
Contracting has been widely used by government agencies to pro-
cure a range of services such as transportation, refuse collection, and
fire protection. While contracting for social services is less common,
private organizations (and particularly not-for-profit organizations)
have long played a role in the delivery of services such as child care,
employment counseling, and welfare support. In the education sector,
governments have made use of contracting with the private sector for
the delivery of auxiliary services such as school transport, food ser-
vices, and cleaning. Contracting for the delivery of core education
services is at a much earlier stage of development. While such contract-
ing is not widespread, there are a number of examples from around the
world.
Contracting can be defined as a purchasing mechanism used to ac-
quire a specified service, of a defined quantity and quality, at an
agreed-on price, from a specific provider, for a specified period.1 In
practical terms, contracting involves a government agency entering
into an agreement with a private provider to procure an education ser-
vice or bundle of services in exchange for payment.
72 Norman LaRocque
4.1 Potential Benefits of Contracting
Proponents argue that contracting out the delivery of education ser-

vices to the private sector may have a number of benefits over tradi-
tional procurement methods. These include improved efficiency in
service delivery, greater transparency in government spending, and
increased access to services, especially for those who were not well
served under traditional procurement methods. Contracting may also
allow government agencies to secure specialized skills, focus on func-
tions where they have a comparative advantage, overcome operating
restrictions such as inflexible salary scales and work rules, and facili-
tate the adoption of service delivery innovations.
On the other hand, opponents argue that contracting can have a
number of drawbacks, including that it is more expensive than tra-
ditional procurement methods (for example, because of the cost of
awarding and managing contracts and a lack of competition), fosters
corruption, and results in a loss of government accountability and
control.2
4.2 Typology of Contracting Models in Education
Education sector contracting can take several different forms, as out-

lined in table 4.1. Under the management contract model, a government
agency contracts with a private provider to manage an existing gov-
ernment service or set of services using government infrastructure. For
example, a local school district in the United States could contract out
the management of a failing school to a private operator. Such models
typically involve the government paying a fixed amount per student to
the private sector provider. Management contracts can be structured
differently. Under one variant, staff remain employed by the school
district (i.e., the pure management model). Under a second variant, the
private contractor employs its own staff (i.e., the operational contract
model).
Under the service delivery model, a government agency contracts with
a private provider to deliver a specified service or set of services using
privately owned infrastructure. Service delivery contracts can be used
either for ‘‘core’’ education activities (e.g., supplementary tuition,
schooling improvement, school review) or ‘‘non-core’’ activities (e.g.,
school transport, food services).
The Practice of Public-Private Partnerships 73
Table 4.1
Typology of Arrangements for Service Delivery in Education
Who chooses Who manages Who provides Who employs
Delivery arrangement service? service? infrastructure? staff?
Government provision Government Government Government Government
Management model:
n Pure management Government Private sector Government Government
contract
n
Operational contract Government Private sector Government Private sector
Service delivery model:
n
Core services Government Private sector Private sector Private sector
n
Non-core services Government Private sector Private sector Private sector
Infrastructure PPPs Government Private sector Private sector Core staff ¼
model public sector
Non-core
staff ¼ private
sector
Private sector provision Consumer Private sector Private sector Private sector
Under an infrastructure public-private partnership (PPP) model, a gov-

ernment agency contracts with the private sector for the provision of
educational infrastructure such as schools, IT laboratories, or hostels.
The exact nature of infrastructure PPPs can differ. Under the most
common variant—build-operate-transfer (BOT)—the private sector fi-
nances, constructs, and operates a school or other infrastructural asset
for a fixed period (e.g., twenty-five or thirty years). During that period,
the private partner effectively ‘‘leases’’ the school to the government for
a specified rental. At the end of the contract period, the school or other
infrastructural asset is turned over to the public sector. Under this
model, teaching staff are employed by the government, but non-core
staff (e.g., janitors, food service, and maintenance workers) are em-
ployed by the private contractor.
Contracting models differ from both pure government provision and
pure private sector provision in that they involve both the government
and the private sector in some form. For example, under the manage-
ment model, the government contracts out the provision of the service
to private operators, but retains ownership of the facilities (and in
some cases continues to employ staff). Similarly, both the service deliv-
ery model and the management model involve government financing
of the service, although the private sector actually delivers the service.
74 Norman LaRocque
Finally, infrastructure PPP models involve private sector financing and

operation, but ownership reverts to the public sector at the end of the
concession or contract period.
4.3 International Examples of Contracting in Education
There are an increasing number of examples of contracting in educa-

tion, in both developed and developing countries. These include char-
ter schools in the United States; privately managed public schools in
the United States, Colombia, and South America; government contract-
ing with the private sector for the delivery of education in the Philip-
pines, Uganda, and Côte d’Ivoire, and the provision of educational
infrastructure in the United Kingdom, Australia, and Egypt. Several
examples are outlined as follows and summarized in table 4.2.
4.3.1 Private Management of Public Schools

One area of increasing private participation is the private management
of public schools. There are various models of private management of
public schools employed by government agencies in several developed
and developing countries. Four examples are highlighted as follows:
Contract Schools, United States The private management of public

schools in the United States can take either of two forms. The first in-
volves direct contracting, under which a local school board contracts
directly with an Education Management Organization (EMO) to man-
age a public school. The second involves indirect contracting under
which EMOs manage charter schools either as the holder of the school
charter or under contract to the organization that holds the school
charter.
Although contract schools are privately managed, they remain pub-
licly owned and funded. Students usually do not pay fees to attend
these schools. Typically, private sector operators are brought in to op-
erate the worst-performing schools in a given school district. Private
sector school managers may operate under either management con-
tracts or operational contracts. Under the former, the management of a
public school is turned over to a private sector operator, but teaching
staff remain employed by the local school board and are subject to the
teacher union contract. Under the latter, teaching staff are employed by
the private operator and their terms and conditions of employment
may differ from the teacher union contract. The private management
company is paid either a fixed amount per student (usually equal to

the unit cost of schooling in the public sector) or a management fee
and must meet performance benchmarks.
In 2005–06, 521 public schools—with enrollments of 237,000—were
under private management by fifty-one EMOs in twenty-nine states
and the District of Columbia (D.C.). The largest EMO was Edison
Schools, with 101 schools and 60,000 students. Approximately 84 per-
cent of EMO-managed schools are charter schools, while the remaining
16 percent are under direct contracts with the local school board. Both
the number of privately managed public schools and enrollments have
grown since the late 1990s, with a small decline in the most recent year
(figure 4.1).3
In Philadelphia, a state takeover of the city’s schools saw seventy
of the worst-performing schools contracted out to for-profit and not-
for-profit private contractors, including Edison Schools, which was
awarded contracts to manage twenty schools in 2002 (increased to
twenty-two in 2005). The Philadelphia experience remains controver-
sial.4 Chicago Public Schools (CPS) is using contract schools as part of
its Renaissance 2010 initiative. Contract schools are managed by inde-
pendent not-for-profit organizations subject to a performance agree-
ment between the organization and the CPS and are freed from many
CPS regulatory requirements, but not from state school laws. Teachers
and staff of contract schools are employed by the private operator.
Charter Schools, United States Charter schools are secular public

schools of choice that operate with freedom from many of the regula-
tions that apply to traditional public schools, such as geographic en-
rollment restrictions and teacher union contracts. The quid pro quo for
charter schools’ increased autonomy is strengthened accountability.
The charter that establishes a school is a performance contract that
details the school’s mission, program, goals, students served, methods
of assessment, and ways in which success will be measured. Charter
schools may be managed by the community or by a for-profit or not-
for-profit school manager.
School charters may be granted by a district school board, a univer-
sity, or other authorizing agency. The term of a charter can vary, but
most are granted for three to five years. Charter schools are account-
able to their sponsor or authorizing agency to produce positive aca-
demic results and adhere to the charter contract. A school’s charter
can be revoked if guidelines on curriculum and management are not
Table 4.2
76
Summary of International Examples of Contracting for Education Services

Program Jurisdiction Program size Key elements
Private Management of Public Schools
Contract schools United States 521 contract schools with 237,000 students School districts or charter school boards contract
operated by EMOs in 2005–06 with private providers to manage public schools.
Charter schools United States More than 4,000 schools with over 1 million Charter schools operate with fewer regulations
students in 40 states and D.C. than standard state schools, but must meet
increased accountability requirements.
Bogotá concession Colombia 25 schools with 26,000 students Private schools and/or education organizations
schools bid in competitive process for management
contracts of newly built schools in poor
neighborhoods.
Contracting with Private Schools for the Delivery of Education Services
Government sponsorship Côte d’Ivoire 162,000 students in mid-1990s Government purchases secondary school places
of students in private in private schools.
schools Schools must maintain academic standards in
order to retain contracts with government.
Educational Service Philippines 380,000 students in 1,800 schools in ESC in Under both ESC and EVS, the government
Contracting/Education 2005–06 purchases places for students in private schools
Voucher System 100,000 students assisted under EVS where public schools cannot meet demand.
Fe y Alegrı́a Latin America/ Over 1.2 million students in FyA programs Schools operated by NGO in poorest communities
Spain and over 500,000 students in formal in Latin America.
education in 2005 Community provides land, construction, and
maintenance of schools, while Ministry of
Education typically pays teacher salaries.
Norman LaRocque
Public-private partnerships for educational infrastructure
Private finance initiative United Kingdom 166 education projects valued at Educational infrastructure designed, built,
approximately £5.8 billion financed, and managed by a private sector
consortium, under a contract that typically lasts
for 30 years.
New Schools private New South 9 new public schools built between 2002 Private sector financing, design, and construction
finance project Wales, Australia and 2005 and an additional 10 from 2006 of public schools, as well as provision of cleaning,
maintenance, repair, security, safety, utility, and
related services for school buildings, furniture,
fittings, equipment, and grounds
PPP for educational Nova Scotia 39 schools built under P3 program in late Schools are financed, built, and operated by the
infrastructure 1990s private sector and leased to the government for 20
years.
Offenbach Schools/ Germany 92 schools in Offenbach County. Capital Government contracting for the finance,
Cologne Schools Project value of over EUR780 million renovation, and operation of public schools in
7 schools in Cologne with value of EUR125 Offenbach County. Private sector partners will
The Practice of Public-Private Partnerships
million operate schools for 15 years.

Refurbishment and operation of schools in
Cologne. Private sector will operate the schools
for 25 years.
Proyectos para Mexico 28 PPP projects being developed in health, Government contracts with private providers to
Prestación de Servicios education, and transport sectors design, finance, build, operate, and maintain
assets and services.
Government is piloting the program to build a
new campus for the University of San Luis Potosi.
New schools project Egypt Construction of 2,210 new schools Government provides land, while private sector
designs, constructs, finances, and furnishes
schools and provides noneducational services
under 15–20 year agreements.
77
78 Norman LaRocque
Figure 4.1
EMO-managed Schools and Enrollments, United States, 1998–99 to 2005–06
Source: Molnar et al. (2006, 3)
followed or standards are not met. At the end of the charter term, the
authorizing agency may renew the school’s contract.
The first charter school law was passed in 1991 in Minnesota, with
the first charter school opening the following year. The number of
charter schools has increased steadily since the early 1990s. There are
currently about 4,000 charter schools serving over one million students
in forty states and Washington, D.C. State charter school laws differ
significantly in terms of their support for charter schools. Charter laws
vary considerably by state and about one-fifth of states do not have
any charter school law. Not all charter school laws are created equal
and the regulatory framework is critical to the role and effect of charter
schools. Approximately one-half of states with charter school laws
have laws that are considered ‘‘strong’’ or ‘‘medium’’ in that they pro-
vide charter schools with more management freedom, limit red tape,
allow community and for-profit providers to operate, and provide
multiple avenues for charter authorization.5
Colegios en Concesión Program in Bogotá, Colombia In Colombia,

the City of Bogotá has introduced the Colegios en Concesión (Conces-
sion Schools) program, under which the management of some public
schools is turned over to high-quality private schools. The first conces-
sion schools began operating in 2000. In 2004, there were twenty-five
Figure 4.2
Enrollments in Concession Schools, Bogotá, 2000–04
Source: Secretary of Education, Bogotá
schools serving over 26,000 students under private management (see

figure 4.2). The program was expected to grow to approximately
45,000 students in fifty-one schools (about 5 percent of public school
coverage in Bogotá), but this has apparently been put on hold due to a
change in the city’s mayoralty. Other examples of school contracting
also exist elsewhere in Colombia, including Medellin and Cali.
Under the concession schools model, private schools and/or educa-
tion organizations bid in competitive process for management con-
tracts of newly built schools in poor neighborhoods of Bogotá.
Contractors may manage a single school or a group of schools. Schools
must provide educational services to poor children and are paid
Col$1,114,500 per full-time student per year—well below public school
unit costs for half-day schooling. Management contracts are for fifteen
years. Contracts with providers are performance-based and establish
clear standards that must be met. The provider has full autonomy over
school management and is evaluated on results. Failure to meet perfor-
mance targets (e.g., standardized test scores and dropout rates) for two
consecutive years can result in the cancellation of the contract.
Schools are monitored through an inspection carried out by a private
firm to monitor the maintenance of the school facilities and property.
In addition, the Ministry of Education carries out reviews of pedagogi-
cal standards and finances an independent evaluation to determine
whether academic objectives have been met.
80 Norman LaRocque
The concession schools program is designed to overcome many of

the traditional problems faced by public schools. These include weak
leadership, inability of schools to select their own personnel, lack of
labor flexibility, lack of equipment and supplies, bureaucratic red-tape,
and the politicization/unionization of the education sector. Initial re-
sults show it has led to management improvements. There is a high
community demand for more concession schools and educators have
expressed satisfaction with the increased autonomy that schools enjoy.6
Fe y Alegrı́a, Latin America Fe y Alegrı́a (FyA) is a nongovernmen-

tal organization controlled by the Jesuit Order of the Catholic Church
that operates formal preschool, primary, secondary, and technical edu-
cation programs in the poorest communities in Latin America and
Spain. The program began in 1955 and operates in fifteen countries.
FyA’s primary mission is to provide quality education to the poor, en-
sure that students complete at least the basic cycle of schooling, and
establish schools that contribute to community development. Under
the FyA model, the salaries of teachers and the principal are paid by
ministries of education, while the land and school infrastructure is
provided by foundations, international agencies, and voluntary fees
from the local community. FyA trains and supervises teachers, man-
ages the school, and assists it in its operation as a community develop-
ment center.
A national office coordinates the network of FyA schools in each
country, while overall coordination is provided by headquarters in
Venezuela. Most FyA schools are located in rural areas, although some
are found in or near urban slums. FyA schools can be public or private,
although a majority are public. Schools generally enjoy considerable
operational autonomy, including the ability to appoint school directors
and teachers. The country’s central curriculum is supplemented with
locally developed materials. FyA schools do not charge compulsory
fees. In 2005, there were over 1.2 million students in the FyA network,
40 percent of whom were in formal education programs.
4.3.2 Contracting with Private Schools for the Delivery of Education

Services
A second form of private involvement in education is where the gov-
ernment purchases places at nongovernment schools for public school
students, rather than providing the places itself in a government-
owned school. Examples from two countries are highlighted here.
Government Sponsorship of Students in Private Schools, Côte

d’Ivoire In Côte d’Ivoire, the government has addressed a lack of
public school places by sponsoring students to attend private religious
or secular secondary schools and training institutions. Under the spon-
sorship program, private schools receive a payment for each public
student they enroll. The government sponsors students in lower and
upper secondary education and in professional and technical training.
The payment amount varies with the student’s educational level.
Only schools that are chartered can participate in the program. Partici-
pation is contingent on the school achieving good academic results.
The number of students in the sponsorship program grew from
116,000 in 1993 to 223,000 in 2001, an increase of 92 percent. In 1997
the government paid out some U.S.$10 million to sponsor over 160,000
students at the school level (approximately 40 percent of private school
enrollments in that year).
Educational Service Contracting and Education Voucher System, the

Philippines The Educational Service Contracting (ESC) scheme in
the Philippines was introduced in the 1980s as part of a range of pro-
grams known as Government Assistance to Students and Teachers in
Private Education (GASTPE). Under the ESC scheme, the government
contracts with private schools to enroll students in areas where there is
a shortage of places in public high schools. The per-student payment to
private schools can be up to PhP5,000 and cannot exceed the unit cost
of delivery in public high schools. To be eligible for ESC, students
must generally attend schools that charge very low tuition fees.
The program is targeted at students from low-income families. ESC
is administered by the Fund for Assistance to Private Education
(FAPE), a private not-for-profit organization. Participating schools
must be certified by the Department of Education. Institutions that fail
to meet certification requirements can either be put on probation or
disqualified from the ESC program. In 2005–06, over 380,000 students
in some 1,800 private schools were subsidized under ESC (figure 4.3).
In 2007, the government was expected to spend some P2.44 billion
(U.S.$54 million) on GASTPE programs.
More recently the Philippine government has introduced the Educa-
tion Voucher System (EVS), which provides a PhP5,000 subsidy to
grantees. The EVS, which took effect in the 2006–07 school year, pro-
vides a more flexible and less targeted form of student assistance.
Under the EVS, students are allowed to use their voucher to enroll in
82 Norman LaRocque
Figure 4.3
Number of ESC Grantees and Participating Schools, 1986–87 to 2005–06
Source: Fund for Assistance to Private Education and Department of Education
any private school, irrespective of whether it is in an area where public

schools are overcrowded. Schools can charge fees above the voucher
amount and need not be accredited. In 2007, there were some 100,000
vouchers being distributed under the EVS.
4.3.3 Public-Private Partnerships for Educational Infrastructure

Public-private partnerships are an increasingly common form of pro-
curement for large infrastructure projects in the education sector. Infra-
structure PPPs can be structured in a variety of ways. As previously
discussed, under the most common type of PPP arrangement—BOT—
a private operator is granted a franchise (concession) to finance, build,
and operate an educational facility such as a public school, university
building, or hostel. The government in effect leases the facility from
the private sector for a specified period, after which the facility is trans-
ferred to the government.
Infrastructure PPPs can be structured in a variety of ways, although
they do have a number of common characteristics:
n
the government retains responsibility for the delivery of core ser-
vices such as teaching and teaching staff continue to be employed
by the government. Non-teaching staff are employed by the private
operator;
n
the private sector invests in school infrastructure and provides re-
lated non-core services (e.g., building maintenance);
n
arrangements between the government and the private sector are
governed by long-term contracts—usually twenty-five to thirty years.
Contracts specify the services the private sector has to deliver and the
standards that must be met;
n
service contracts are often bundled, with the provider taking on sev-
eral functions such as design, construction, and maintenance; and
n
contract payments are contingent upon the private operator deliver-
ing services to an agreed performance standard.
Infrastructure PPPs differ from traditional procurement methods in

several ways. First, the private sector provides the capital required to
finance the project. Second, the government specifies the contract in
terms of outputs or service level requirements, rather than in terms of
inputs such as the number and size of classrooms. Third, the newly
constructed facility is not turned over to the government upon comple-
tion. Rather, it is operated by the private sector until the end of the
contract period. Several examples are outlined as follows.
Private Finance Initiative, United Kingdom The Private Finance

Initiative (PFI) program was introduced under the Conservative gov-
ernment in 1992 as part of a broader policy of public sector moderniza-
tion, and has been strongly supported by the Labor government since
it came to power in 1997. The government uses PFI only where it is
appropriate and where it expects it to deliver value for money. PFI
uptake in the education sector was slow in the early years, but grew
considerably following the introduction of a number of program im-
provements since the late 1990s. To date, the Department for Educa-
tion and Skills (DfES) has signed some 166 education PFI deals, with a
value of approximately £5.8 billion.
New Schools Private Finance Project, Australia Under the New

Schools Project in the Australian state of New South Wales, the pri-
vate sector is financing, designing, and constructing nineteen public
schools. The project began with nine schools in 2002 and was ex-
panded with a further ten schools in 2006. These new schools are being
built to standards that must meet or exceed Department of Education
and Training (DET) school design standards. The private sector is also
providing cleaning, maintenance, repair, security, safety, utility, and
84 Norman LaRocque
related services for school buildings, furniture, fittings, and equipment

over a twenty-five- to thirty-year period. In return, the private sector
receives performance-related monthly payments from the DET during
the operational phase of the project.
The New Schools Project in New South Wales is part of a broader
move toward PPPs in Australia. PPPs have been used by various gov-
ernments to procure infrastructure across a range of sectors, including
transport, health, and prisons. They have also been used in higher edu-
cation, with the University of Southern Queensland and Swinburne
University of Technology both using PFIs to construct educational
infrastructure.
New Schools Project, Egypt The government of Egypt has recently

embarked on an extensive school infrastructure PPP involving the con-
struction of 2,210 new primary and secondary schools in an attempt to
meet the president’s target of 3,500 new schools by 2011. The project
started in late 2006 with 300 schools in twenty-three governorates. The
positive response of the private sector led to the program’s expansion
in early 2007 to include a further 1,910 schools. Under the PPP, the
government provides land, while the private sector designs, constructs,
finances and furnishes schools and provides non-educational services
under fifteen- to twenty-year agreements. The value of the PPP is esti-
mated at LE11 billion (approximately U.S.$2 billion).
Other examples of education infrastructure PPPs include schools in
Nova Scotia, Canada, the Offenbach Schools, and Cologne School Proj-
ects in Germany, the Montaigne Lyceum in the Netherlands, and the
Proyectos para Prestación de Servicios in Mexico.
4.4 Lessons for Policy Design and Implementation
Contracting for the delivery of education services, although growing,

remains in its infancy. The school sector in all countries is dominated
by a mix of ‘‘traditional’’ private and public schools. To date, there is
little rigorous evidence on the effectiveness of contracting as a tool for
improving educational outcomes. Similarly, there are few rigorous
studies examining how best to design and implement programs that
involve contracting for the delivery of education services. However, a
small number of studies have examined the experience with particular
education-specific contracting initiatives or contracting in the public
sector more generally.7
These studies provide some tentative lessons on the design and im-
plementation of education contracting programs. In particular, they
highlight the importance of several success factors in contracting,
including the existence of an enabling policy and regulatory environ-
ment and a strong legal framework. The government’s role should be
to spell out the desired outputs and performance standards, set penal-
ties for failure to achieve and rewards for success, and then leave pro-
viders to decide the best way of organizing themselves to deliver the
required outputs to the specified standard.
The establishment of appropriate performance measures is a critical
element of contract design.8 Performance indicators may be quantita-
tive (e.g., test scores) or qualitative (e.g., parental satisfaction) in nature
and provide an objective basis for determining whether the service
provider has met the agreed terms and conditions of the contract.
They may also be linked to provider compensation, with providers
who meet the required standard rewarded through higher payments,
while those who fail to do so are penalized—either through reduced
payments or contract termination.
The contracting approach to government procurement of educa-
tion services places much more significant demands on government
agencies than do traditional methods of government procurement. The
need to specify, monitor, and enforce complex contracts means that
contracting agencies must have the information, skills, and capabil-
ity required to manage such programs. Splitting the purchaser and
provider roles within the relevant regulatory authority can help to
minimize conflicts between the government’s role as purchaser of
education services and its role as provider of education through the
public school system. Use of an independent organization to evaluate
the contractor’s performance can ensure further neutrality in the moni-
toring and enforcement of contracts.
A transparent and competitive bidding process is most likely to
build community acceptance for the use of contracting, and is most
likely to deliver better value for money, reduce the scope for corrup-
tion, and encourage growth in the private education services market.
A staged process ensures good specification of desired services and
expected outcomes and a more rigorous assessment of provider capa-
bility.9 Contract length can be tailored to the particular service being
contracted for. A range of factors must be considered in determining
appropriate contract length, including the appropriate period for as-
sessing performance, the maturity of the contracting regime, and the
86 Norman LaRocque
transaction costs of negotiating contracts. Contracts should be long

enough to generate interest and encourage private investment in the
education sector, but not so long as to blunt incentives for performance
and limit potential contracting gains.10
4.5 Conclusion
This chapter has reviewed a number of examples of contracting for the

delivery of education services. Although contracting for the delivery
of education services remains in its infancy in many countries, it is a
growing phenomenon. It is not a panacea, but does offer governments
an additional mechanism for improving educational outcomes and in-
creasing the efficiency of educational provision. Educational contract-
ing has been, and will continue to be, controversial. Its beneficiaries
are often politically weak and unorganized, while its opponents—for
example, teacher unions—are well financed and organized.
The success of educational contracting requires good policy design,
well-managed implementation, effective political management, well-
designed evaluations, and a fundamental redefinition of the role of
public education authorities.
Notes
Revised version of paper presented at the Harvard University/World Bank Conference

‘‘Mobilizing the Private Sector for Public Education,’’ Cambridge, Massachusetts, October
5–6, 2005. Comments and input from Harry Anthony Patrinos, April Harding, Raji Chak-
rabarti, and anonymous referees are gratefully acknowledged.
1. Taylor (2003), 158.

2. Savas (2000), 76–77 and Brook and Petrie (2001), 4.
3. Molnar et al. (2006), 3 and 29.
4. Gill et al. (2007) and Peterson (2007).
5. Center for Education Reform (2006).
6. Rodriguez (2002), 8–9.

7. See Rhim (2005); Hentschke et al. (2003); Bulkley et al. (2004); Center for Comprehen-
sive School Reform and Improvement (2005); and OECD (1999).
8. New South Wales Treasury (2005).

9. For a comprehensive discussion of a competitive process for contracting out the deliv-
ery of public services, see Savas (2000), 174–210.
10. OECD (1999), 32.

References
Brook, P. J., and M. Petrie. 2001. Output-based aid: Precedents, promises, and challenges.
In Contracting for public services: Output-based Aid and Its Applications, ed. P. J. Brook and S.
Smith, 4. Washington, D.C.: World Bank.
Bulkley, K., L. Mundell, and R. Morgan. 2004. Contracting Out Schools: The First Year of the
Philadelphia Diverse Provider Model. Philadelphia: Research for Action.
Center for Comprehensive School Reform and Improvement. 2005. Contracting With Ex-
ternal Education Management Providers. Washington, D.C.
Center for Education Reform. 2006. Charter Schools Today: Changing the Face of American
Education Part 2: Raising the Bar On Charter School Law, 2006 Ranking & Scorecard, 9 th Edi-
tion. Washington, D.C.
Gill, B. et al. 2007. State Takeover, School Restructuring, Private Management, and Student
Achievement in Philadelphia. Santa Monica: RAND Corporation.
Hentschke, G. C., S. Oschman, and L. Snell. 2003. Trends and Best Practices for Education
Management Organizations. San Francisco: WestEd.
Molnar, A. et al. 2006. Profiles of For-Profit Education Management Companies 2005–2006.
Eighth Annual Report. Tempe: Arizona State University.
New South Wales Treasury. 2005. New Schools Privately Financed Project: Post Implementa-
tion Review. Sydney: Office of Financial Management.
OECD. 1999. Performance Contracting: Lessons from Performance Contracting Case Studies, A
Framework for Public Sector Performance Contracting. Paris.
Patrinos, H. A., and N. LaRocque. 2007. Enhancing Accountability in Schools: What Can
Choice and Contracting Contribute. Washington, D.C.: World Bank.
Rhim, L. M. 2005. School Restructuring in Philadelphia: Management Lessons from 2002 to
2005. ECS policy brief. Denver: Education Commission of the States.
Rodriguez, A. 2002. Executive summary. In Case Study: Public School Concession Model of
Bogotá Colombia. Washington, D.C.: World Bank.
Savas, E. S. 2000. Privatization and Public-Private Partnerships. New York: Chatham House
Publishers/Seven Bridges Press.
Swope, J., and M. Latorre. 2000. Fe y Alegrı́a Schools in Latin America: Educational Commu-
nities Where the Pavement Ends. Santiago: Centro de Investigación y Desarrollo de la Edu-
cación, Santiago.
Taylor, R. J. 2003. Contracting for health services. In Private Participation in Health Services
Handbook, ed. A. L. Harding and A. S. Preker, 157–219. Washington, D.C.: World Bank.
World Bank. 2005. Colombia: Contracting Education Services. Report No. 31841-CO. Wash-
ington, D.C.: World Bank.
II Traditional Forms of School Choice
5 Public-Private Schools in Rural India
Karthik Muralidharan and Michael Kremer
While the focus of primary education policy in developing countries

such as India has largely centered on increasing the resource base and
the number of government-run schools, the role of private fee-charging
schools in the primary education sector has not been appreciated as
much. However, as several recent papers point out (Kingdon 1996;
PROBE Team 1999; De, Noronha, and Samson 2001; Tooley and Dixon
2003; and Mehta 2005), there is reason to believe that private fee-
charging schools increasingly cater to a substantial fraction of the
primary-school-going population in India. Most research on this sub-
ject to date comes from small-sample studies at the state or district
levels.1
This chapter presents results from a nationally representative survey
of rural private primary schools in India that the authors conducted
in 2003. Twenty-eight percent of the population of rural India has ac-
cess to fee-charging private schools in the same village. Richer states
have fewer rural private schools. States, districts, and villages with
poor public school performance are each more likely to have private
schools. Nearly 50% of the rural private schools in our sample were
established five or fewer years before the survey, and nearly 40% of
private school enrollment is in these schools. This suggests rapid ex-
pansion of private schooling, although it could also in part reflect turn-
over among schools in the sector.
Private-school teacher salaries are typically one-fifth of the salary of
regular public school teachers (and are often as low as one-tenth of
these salaries). This enables the private schools to hire more teachers,
have lower pupil-teacher ratios, and reduce multigrade teaching. Pri-
vate school teachers are significantly younger and more likely to be
from the same area as their counterparts in the public schools. They
are 2–8 percentage points less absent than teachers in public schools
92 Karthik Muralidharan and Michael Kremer
and 6–9 percentage points more likely to be engaged in teaching

activity at any given point in time. They are more likely to hold a col-
lege degree than public school teachers, but are much less likely to
have a formal teacher-training certificate. Children in private school
have higher attendance rates and superior test-score performance,
the latter true even after controlling for observed family and school
characteristics.
The first section outlines the sampling methodology and how the
data was collected. The second section presents results on the extent of
private school prevalence and correlates of private school existence.
The chapter then discusses the economics of private unaided schools
and their sources of competitive advantage by comparing them with
public schools on various measures including infrastructure, teacher
characteristics, student characteristics, and student performance.
5.1 Sampling Methodology and Data
The data used in this chapter was collected as part of a multicountry

study conducted by us and coauthors on provider absence in schools
and health clinics where India was one of the countries studied (the
detailed results from the cross-country study are presented in Chaud-
hury et al. 2006).2 Within India, 20 states were selected, representing
98 percent of the population, or roughly one billion people. Using geo-
graphically stratified random sampling, 10 districts were selected with-
in each state and 10 primary sampling units (PSUs) were selected in
each district. The PSUs were allocated to rural and urban sectors in
accordance with the population distribution within each sampled dis-
trict.3 Rural PSUs (villages) within a sampled district were selected
randomly without replacement with probability proportional to size
(PPS).4
The survey focused on government-run primary5 schools but also
covered rural private schools in villages where they existed. The defini-
tions of school categories that we use are similar to those detailed in
chapter 6 by Geeta G. Kingdon. The term government school refers to
government-funded schools that are run by the government but does
not include the government-aided schools that are privately managed.
The terms public schools and government schools are used inter-
changeably in this chapter. The private schools referred to in the rest
of this chapter are those that charge user fees and do not receive any fi-
nancial support from the government. This includes both recognized
Public-Private Schools in Rural India 93
and unrecognized private schools, but does not include ‘‘private-aided

schools’’ which are privately managed schools that receive funding
from the government, and are typically forbidden from charging user
fees.
Recognized private schools are required to conform to various gov-
ernment norms; the main benefit of recognition is that only recognized
schools are eligible to issue ‘‘transfer certificates’’ (TCs) to their stu-
dents (see chapter 6 for more details on the requirements for recogni-
tion). These TCs in turn are required for students to move across
schools with credit granted for academic work done in the previous
school. In practice, however, many of the recognized schools do not
meet the stipulated norms (Kingdon, chapter 6 in this volume), and
Tooley and Dixon (2003) argue that it is not uncommon for operators
of private schools to have to pay bribes to obtain recognition status.
One response to the obstacles to obtaining recognition has been an
increasing prevalence of unrecognized private schools that charge fees
but have not obtained recognition and are not authorized to issue TCs.
Unrecognized private schools circumvent this practice in several ways,
the most common of which is double enrollment, whereby children are
enrolled in both the government-run school (which is recognized by
default) and in the unrecognized private school. Note that private
unrecognized schools are more than just supplemental tuition centers
and should be thought of as schools, because they usually run during
the same hours of the regular school, and children typically do not
attend both kinds of schools although they may be enrolled in both.
Double enrollment is a convenient arrangement for all parties because
the government school gets to show high levels of enrollment, parents
and children get textbooks and other free supplies from the govern-
ment school, and new private schools can operate without the burden
of seeking recognition since TCs will be issued by the government
school. However, this does lead to systematic underestimation of the
relative size of the government and fee-charging private school sys-
tems in India, as discussed in Kingdon (chapter 6 in this volume).
In the rural sample, the survey covered all the primary schools in
the village subject to a maximum of three (the maximum number of
schools that could be covered during one day in the field). When the
investigators reached the village, they listed all the schools present
within a radius of two kilometers from the village center. In villages
with fewer than three schools, all the schools were covered. In vil-
lages with more than three schools, three schools were surveyed; one
school was randomly selected in each of the three main categories of

rural schools (government schools, private schools, and nonformal ed-
ucation centers). In cases where there was no nonformal school, but
more than three schools in the village, enumerators selected two gov-
ernment schools and one private school or one government school and
two private schools (the latter was the case only if there was only one
government primary school but more than two private schools in the
village).
Thus in addition to being representative of government-run primary
schools, the dataset is also representative of the universe of private
unaided primary schools in rural India because at least one private
school was surveyed in any village that had at least one private school.
Fifty-three percent of the private schools in our sample are unrecog-
nized, suggesting that official sources of data on private schools sig-
nificantly understate the extent of private school prevalence.6 While
government surveys only include the recognized private schools, the
random selection method is indifferent to the recognition status of the
school and the sample here therefore includes both types of schools.
Furthermore, the random selection of the schools within a village en-
sures that the distribution of school types in the sample is a reflection
of the distribution of school types in the population. The remainder
of this chapter does not distinguish between private recognized and
unrecognized schools because they are both fee-charging schools that
do not receive funds from the government, and this is the school cate-
gory we focus on here.
Enumerators made three unannounced visits to each selected school
over a three- to four-month time period from December 2002 to March
2003. Teacher absence was measured in all surveyed schools by physi-
cally verifying the presence of teachers on the school roster. In addition
to recording teacher attendance, data was also collected on student
attendance, school facilities, and teacher characteristics. Finally, the
enumerators also administered a short test7 to 10 randomly selected
fourth-grade children and collected basic demographic information on
these children in all the schools that we surveyed.
5.2 Private School Prevalence and Its Correlates
Twenty-eight percent of the villages in our sample have a private

school. Since the villages were sampled on a probability proportional
to size (PPS) basis, this implies that 28% of the population of rural
Table 5.1
Private School Prevalence by State
% of % of
villages villages
with a with a
private private
State school State school
Gujarat 0 Andhra Pradesh 30

Maharashtra 1 Uttranchal 30
Orissa 4 Tamil Nadu 31
Kerala 6 Assam 33
Karnataka 12 Rajasthan 52
Chhatisgarh 15 Bihar 54
Himachal Pradesh 15 Uttar Pradesh 57
West Bengal 16 Punjab 65
Jharkhand 17 Haryana 68
Madhya Pradesh 23 All India 28
India has access to a private school in the same village in which they
live. But there is sharp variation in the prevalence of private schools
across states, with Gujarat and Maharashtra having almost no rural
private schools, while over 50% of the sampled villages in Rajasthan,
Bihar, Uttar Pradesh, Punjab, and Haryana have a private school in
the same village (table 5.1). Recent household-survey-based evidence
presented in the Annual Status of Education Report (2005) confirms
the increasing role of private schooling in rural India by showing that
15.5% of children aged 6–10 in rural India attend a private school and
that over 20% of the children in this group attend a private school in
several states.8
Table 5.2 presents results from ordinary least squares (OLS) regres-
sions where the binary variable of private-school existence (at the
village level) is regressed on potential predictors of private school exis-
tence. The first column includes the log of the village population, the
log of the mean pupil-teacher ratio in the public schools in the village,
and the mean level of teacher absence9 in the public schools in the vil-
lage. The second column includes state fixed effects. The third column
replaces the state dummies with the log of state per capita GDP. The
fourth column includes district-level estimates of mean per capita con-
sumption calculated from the fifty-fifth round of the National Sample
Survey,10 and the fifth column includes district-level consumption as
well as state fixed effects.11
Table 5.2
Correlates of Private School Existence at the Village Level
Dependent variable ¼ 1 if village has a private school, 0 if it does not
(1) (2) (3) (4) (5)

Log village population 0.114 0.157 0.125 0.11 0.159
[0.012]*** [0.014]*** [0.037]*** [0.018]*** [0.017]***
Log pupil teacher 0.089 0.042 0.034 0.1 0.037
ratio [0.022]*** [0.026] [0.051] [0.031]*** [0.027]
Mean public school 0.292 0.114 0.214 0.303 0.108
Absence in village [0.065]*** [0.060]* [0.103]* [0.074]*** [0.053]**
Log state GDP/Capita 0.298
[0.157]*
Log district 0.07 0.121
Consumption/Capita [0.076] [0.059]**
Constant 0.962 1.065 0.851 1.39 0.352
[0.101]*** [0.117]*** [0.975] [0.480]*** [0.366]
State fixed effects No Yes No No Yes
Observations 1523 1523 1450 1523 1523
R-squared 0.1 0.33 0.12 0.1 0.33
Notes: Robust standard errors in brackets
*** significant at 1%; ** significant at 5%; * significant at 10%
Villages with larger populations are significantly more likely to have

a private school in all specifications. The most noteworthy result is that
private schools are significantly more likely to exist in villages with a
high rate of teacher absence in public schools. While the relation is
very strong across Indian states, it is still significant at the 10% level af-
ter controlling for state fixed effects, and remains significant in all spec-
ifications. The surprising result is that states with a higher per capita
income are less likely to have private schools in their villages. While a
high pupil-teacher ratio (PTR) in the public schools in the same village
is a predictor of private school existence across India, the correlation is
not significant with either state income controls or state fixed effects,
suggesting that the PTR in public schools is negatively correlated with
the per capita GDP of the states. The final column shows that when we
include state-fixed effects, richer districts are less likely to have a pri-
vate school, though villages with high public-school teacher absence
are more likely to have a private school.
Chaudhury et al. (2006) shows that higher-income countries and
richer Indian states have significantly lower rates of teacher absence in
schools. Thus if private schools arise as a response to public school fail-

ure, we might expect richer states to have fewer private schools. On the
other hand, since private schooling is likely to be a normal good we
might expect the prevalence of private schools to be higher in the
richer states.
The correlation between public school failure (as measured by teach-
er absence and nonteaching activity) and the likelihood of the existence
of private schools can be seen clearly in figures 5.1a and 5.1b. While the
two states with the highest incidence of private schools (Punjab and
Haryana) happen to be among the richer states of India, it is quite
striking that the two states with the lowest level of teacher absence in
public schools (Gujarat and Maharashtra) have almost no rural private
schools, even though these are two of the richest states in India.
Table 5.3 shows more related evidence by comparing teacher ab-
sence rates across different kinds of schools in India. The first column
of table 5.3 shows the weighted average teacher absence by school
type across the full sample of schools. Columns 3–5 show the differ-
ence in teacher absence relative to the government-run schools. While
the weighted average all-India teacher absence in private schools of
22.8% is slightly lower than that of the 25.2% in government schools,
this difference is not significant. However, with the addition of village/
town fixed effects, the teacher absence rate is 3.8% lower in private
schools relative to government schools and this is significant at the
1% level. The addition of school, teacher demographics, and visit-level
controls increases this difference to 7.8%, which is over 30% of the
observed absence rate in government schools (25.2%). This suggests
that private schools are disproportionately located in areas with poorly
performing public schools and that the efficiency of the private school
(at least as measured by teacher absence) is even higher after control-
ling for school facilities (which are negatively correlated with teacher
absence) and teacher demographics.
The higher prevalence of private schools in villages with high ab-
sence among public school teachers could be interpreted as suggesting
that private schools enter where public schools are failing or as evi-
dence that the establishment of private schools reduces political pres-
sure for teacher attendance in public schools. However, to the extent
that one might expect higher-income states to have more private
schools, the finding that richer areas have fewer private schools sug-
gests that poorly performing public schools rather than increasing
incomes are the more important source of demand for private schools.
Figure 5.1
Table 5.3
Absence Rate by School Type
Difference relative to government-run
schools
(1) (2) (3) (4) (5)
Village/ Village/
Number No town town fixed
Teacher of obser- fixed fixed effects þ
absence vations effects effects controls*
Government- 25.2% 34,493 — — —
Run Schools
Non-formal 26.9% 393 1.7% 2.7% 2.4%
Schools
Private Aided 20.1% 3,371 C5.1% 1.3% 0.4%
Schools
Private Schools 22.8% 9,075 2.4% C3.8% C7.8%
Notes: * Controls include a full set of visit-level, teacher-level, and school-level controls
Bold numbers indicate significant differences at the 1% level
Finally, it is noteworthy that there is some evidence that large-scale

prevalence of rural private schools is a recent phenomenon. This is
suggested in previous studies of specific states such as De, Noronha,
and Samson (2001), and Mehta (2005), but we are able to confirm this
on a nationwide basis. Figure 5.2 plots the cumulative distribution
function (CDF) of private school formation and enrollment over time,
and we see that nearly 50% of the private schools in the sample have
been established in the five years before the survey. Nearly 40% of the
total private school enrollment is in schools that were less than 5 years
old and over 60% of total enrollment is in schools that were less than
10 years old in 2003. Of course, these numbers will exceed the net in-
crease in private school enrollment to the extent that other private
schools exited over the period.
5.3 Economics of Rural Private Schools
5.3.1 School Infrastructure

Table 5.4 presents summary statistics on school infrastructure in public
and private schools. While private schools are more likely to have an
electricity connection and toilets for teachers, they are less likely to
have libraries (book banks) and classrooms without mud floors. On
aggregate there doesn’t appear to be a significant difference in the
Figure 5.2
Private School Formation/Enrollment over Time (Cumulative distribution function)
Table 5.4
Private versus Public School Facilities
Difference Difference
with state with village
Public Private Difference fixed effects fixed effects
Fraction of schools with 0.26 0.414 0.154*** 0.198*** 0.191***

electric connection available
Fraction of schools with 0.541 0.273 0.269*** 0.236*** 0.238***
library available
Fraction of schools with 0.943 0.939 0.004 0.030** 0.029
covered classrooms available
Fraction of schools with non- 0.816 0.674 0.142*** 0.184*** 0.197***
mud floors available
Fraction of schools with 0.326 0.447 0.121** 0.052** 0.027
teacher toilet available
Average school infrastructure 2.885 2.745 0.14 0.199*** 0.247***
index (0–5 scale)
Notes: Significance level: *** 1%; ** 5%; * 1%

infrastructure index between private and public schools, but the results
with state and with village fixed effects suggest that conditional on
being in the same village, private schools have poorer facilities and
infrastructure than the public schools.
5.3.2 Sources of Competitive Advantage of Private Schools

Probably the single most distinguishing feature of the private schools
in rural India is the fact that they pay much lower salaries to teachers
than the government schools. While we don’t directly collect data on
teacher salaries, we have data on the various fees charged by each
school in our sample along with the total enrollment, which allows us
to estimate the monthly revenue for the private schools (since they typ-
ically don’t receive any funding beyond what they raise in school fees).
Median monthly revenue of a private school in our sample is around
Rs4,000 per month,12 with the median fee being Rs63 per month and
the median private school having an enrollment of 72 students.
We can calculate an upper bound for teacher salaries in private
schools assuming that all the revenues of the private schools are
used to pay teacher salaries. We calculate the upper bound on median
teacher salary to be less than Rs1,000 per month and the upper bound
on the mean teacher salary to be less than Rs1,750 per month. The
mean salary for a regular government school teacher in a typical state
like Andhra Pradesh (where we have actual salary data13) is around
Rs7,500 per month. We can see that the typical total monthly revenue
of a private school is often less than the monthly salary of one govern-
ment school teacher. Even conservatively, rural private school teacher
salaries are typically around one-fifth of that of regular government
teacher salaries and they are often as low as one-tenth of the salaries
of regular government teachers. The differences are even more pro-
nounced when benefits are included because government teachers are
guaranteed a pension after retirement, while private school teachers
rarely have such provisions. This allows the private schools to hire
more teachers, reduce multigrade teaching, and have significantly
lower pupil-teacher ratios.
Table 5.5 clearly demonstrates these points. The average PTR in the
private schools of 19.2 is less than half the ratio of 43.4 in public
schools. This gap of 24.3 widens to 29.6 with state fixed effects, and to
34.4 with village fixed effects. Thus conditional on being in the same
village, the private school has nearly 35 fewer pupils per teacher than
the government school in the same village. Doing the calculation using
Table 5.5
Sources of Private School Competitive Advantage
Public Private Difference FEs FEs
Mean total enrollment 141.9 98.3 43.6*** 49.6*** 80.7***
Mean number of teachers 3.6 5.2 1.6*** 1.48*** 0.87***
Pupil-teacher ratio 43.43 19.16 24.3*** 29.6*** 34.43***
Log pupil-teacher ratio 3.583 2.783 0.800 0.931*** 1.045***
Multigrade teaching 71% 51% 0.20*** 0.20*** 0.11***
Average grade of starting 2.62 1.67 0.95*** 1.27*** 1.35***
teaching English
Fraction of teachers engaged 44% 50% 5.7%*** 8.6%*** 9.3%***
in teaching activity
Average student attendance 64.4% 75.7% 11.3%*** 12.1%*** 13.4%***
logs, we find that the PTR of a public school is 2.85 times higher than
the PTR of a private school in the same village. The lower PTR in the
private schools also translates into lower levels of multigrade teaching
(the practice of one teacher simultaneously teaching multiple grades in
the same room).
Field interviews with parents of children attending rural private
schools suggest that two of the major attractions of private schools are
the fact that they start teaching English early and that there is more
teaching activity in these schools. The last two rows of table 5.5 con-
firm that these differences do exist. Private schools on average start to
teach English a whole grade earlier, with the effect being even more
pronounced with state and village fixed effects. Private schools also
have significantly more teaching activity going on, and again the mag-
nitude of the difference increases with state and village fixed effects.
One reason for this is likely to be that head teachers in private school
are much more likely (and able) to take disciplinary action against
shirking teachers than their counterparts in the public schools. We
found that only one head teacher in the nearly 3,000 public schools we
surveyed reported ever dismissing a teacher for repeated absence.14
On the other hand, 35 head teachers in a sample of around 600 private
schools reported having at some point dismissed a teacher for repeated
absence, and therefore shirking teachers in the private sector are
around 175 times more likely to have disciplinary action taken against
them!
If we consider the cases with village fixed effects (which is the rele-
vant case when considering the choice faced by a parent with regard
to choosing between a private and public school in the same village),
we see that combining the effects of a lower pupil-teacher ratio and a
higher level of teaching activity leads to a child in the private school
having three to four times more ‘‘teacher contact’’ time than in the pub-
lic school.
The better performance of the private schools is also reflected in the
fact that student attendance rates are also substantially higher in pri-
vate schools (as seen in the last row of table 5.5). Pupil attendance is
11.3% higher in the all-India sample, and 13.4% higher with village
fixed effects. If we think that the true measure of the relative role of the
private and public sectors is attendance as opposed to enrollment, then
the true share of rural children taught in the private sector will be even
higher after adjusting for the differential attendance rates.
5.3.3 Teacher Characteristics

A key issue that follows the discussion on teacher pay in private
schools is to understand who the private school teachers are and the
reasons for their willingness to work at such low salaries. Field visits
suggest that the availability of these inexpensive teachers in the vil-
lages is being driven by local educated young people who are typically
unable to find jobs, unwilling (and usually not needed) to work in ag-
riculture, and not looking at teaching as a long-term career. Teaching
suits these young people well because the short working day of four to
six hours allows them the time for further study via correspondence
(distance-education) courses or in colleges that follow a different shift.
The short working days also allow them to look for other longer-term
jobs on the side. And finally, teaching provides them with both income
and respectability while they look at other long-term options.
Table 5.6 provides summary statistics consistent with this view. The
private school teachers are on average over ten years younger than
their counterparts in the public sector and are twice as likely to be from
the same village where the school is located. They are more likely to
have a college degree but also much less likely to have a professional
teaching certificate, which suggests that even though they are more
educated, they are not looking at teaching as a long-term career option.
This probably helps to explain why teacher absence is not even
lower than it is in the private schools given the high likelihood of
action being taken for repeated absence. Since the private school
Table 5.6
Teacher Characteristics
Average age of teachers 40.28 29.61 10.67*** 11.92*** 12.35***
Fraction of college 39% 49% 0.10*** 0.03* 0.01
graduates among teachers
Fraction of teaching 80% 28% 0.52*** 0.61*** 0.64***
certificate holders among
teachers
Fraction of female teachers 36% 41% 0.05 0 0.02
Fraction of local teachers 23% 46% 0.23*** 0.26*** 0.24***
teachers are being paid a much lower wage and are often looking at
other long-term options, there is little ‘‘efficiency wage’’ cost of being
fired. Thus, if pursuing other opportunities requires a certain level of
absence (and an accompanying probability of action being taken), this
is a tradeoff that the private school teachers probably are willing to
make. However, despite the low wages, we see that private schools
have lower teacher absence and higher teaching activity than the
public schools—especially in the same village.
5.3.4 Parent Characteristics

Given that public schools are free and private schools charge fees,
we would expect that the students attending the private schools come
from more socioeconomically privileged backgrounds. Based on the
random sample of children in the fourth grade whom we test and col-
lect demographic information on, we can compare the family back-
grounds of children in both types of schools. Table 5.7 provides these
comparisons, and as we would expect, the children attending private
schools come from more advantaged family backgrounds. They have
more educated parents and indicate possessing a higher level of assets.
However, it is worth noting that the absolute level of education of the
parents of the children attending private schools is actually quite low.
For instance, 20% of the private school students are first-generation
learners, which while lower than the 30% finding in public schools, is
still quite significant. Thus while private schools cater to the more af-
fluent in the rural areas, many of their students come from disadvan-
taged backgrounds. This is consistent with the results of Tooley and
Table 5.7
Household Characteristics
Average number of rooms 2.423 2.914 0.742*** 0.574*** 0.560***
in house
Average fraction of children 0.169 0.212 0.043*** 0.041*** 0.066***
taking tuition
% of literate fathers 0.71 0.804 9.4%*** 0.118*** 0.146***
% of literate mothers 0.445 0.542 9.7%*** 0.122*** 0.163***
% of fathers with education 0.242 0.432 19%*** 0.208*** 0.236***
10 grades or higher
% of mothers with education 0.087 0.197 11%*** 0.117*** 0.129***
10 grades or higher
Dixon (2003), who mention that the majority of private schools in India
cater to the poor (though their observation is based on an urban study)
and the findings reported by Andrabi, Das, and Khwaja (2002) that pri-
vate schools in rural Pakistan are affordable to middle- and even low-
income groups.
5.3.5 Performance of Private Schools

As discussed earlier, private schools have lower teacher absence and
higher levels of teaching activity. They also exhibit significantly superior
performance on the test that was administered. Table 5.8 shows the
test score performance advantage of private schools (in standard devi-
ations). While controlling for family and other characteristics reduces
the size of the private school effect, it is still strongly significant and of
considerable magnitude (0.4 standard deviations on the test). Of course,
we cannot rule out that some of these results are being driven by unob-
served heterogeneity among the students. Similarly, as discussed earlier,
student attendance is around 11 percentage points higher in the private
schools (75%) relative to the public schools (64%). This could partly be
due to artificially inflated enrollment figures in the government schools.
5.4 Conclusions
We find that private unaided fee-charging schools are widespread

in rural India, particularly in areas where the public system is
Table 5.8
Performance Differentials of Private Schools
Regression of mean student test score (in std. deviations) on school type and controls
(1) (2) (3) (4) (5)

Private School 0.57*** 0.50*** 0.40*** 0.38*** 0.41***
Controls
Family demographics no yes yes yes yes
and private tuition
School facilities no no yes yes yes
State fixed effects no no no yes no
Village fixed effects no no no no yes
Observations 29462 27242 25561 25571 25571
R-squared 0.05 0.09 0.1 0.14 0.43
dysfunctional. The number of such schools appears to be growing

rapidly with both demand-side variables (desire for English-medium
education, less multigrade teaching, smaller classes, more accountable
teachers) and supply-side variables (availability of educated unem-
ployed young people) playing an important role in this rapid growth.
Salaries paid by these schools are only about one-fifth of those paid by
public schools, but these schools have many more teachers relative to
the number of pupils, and the private school teachers are more likely
to be teaching than public school teachers.
Our results have a number of implications. First, efforts to improve
the quality of education in India should consider the private as well as
the public sector—especially since private schools are disproportion-
ately located where the public system is failing. For example, policy
makers might consider the possibility of offering short training courses
to raise skills among private school teachers.
Second, the disparities between private and public schools highlight
some potential areas for reform in the public sector. The huge salary
differential suggests that many public school teachers may be receiving
enormous rents.
Finally, there may be scope for public-private partnerships in edu-
cation, whether in the form of voucher programs or otherwise. One
issue with voucher programs is whether there will be an adequate
supply response, but the evidence suggests that private schools are
already widespread in rural areas and that new schools can be created
rapidly.
There is substantial scope for carefully designed policy experiments

aimed at leveraging the private sector for universal quality education,
and it is important to follow these experiments with rigorous evalua-
tion to provide systematic evidence for future policy decisions in this
regard. The recent draft of the ‘‘Right to Education Bill’’ that is ex-
pected to be introduced in Parliament mandates that 25% of seats in
private educational institutions be reserved for ‘‘weaker sections’’ of
society. It also goes on to say that for each such admitted child, the
‘‘government shall reimburse to the school at a rate equal to the per-
child expenditure in state schools/fully aided schools, or the actual
amount charged per student by such school, whichever is less.’’ The
discussion around this legislation would be an opportune moment to
think about the most efficient institutional forms for delivery of pri-
mary education in India.
Acknowledgments
We thank Nazmul Chaudhury, Jeffrey Hammer, and Halsey Rogers for

their collaboration and insights on the global study that generated the
data that this chapter is based on, and Konstantin Styrin for valuable
research assistance. We offer thanks to the staff of the Social and Rural
Research Institute, New Delhi—and especially to Chhavi Bhargava,
Navendu Shekhar, A. V. Surya, and Aditi Varma—for conducting and
overseeing the fieldwork for the primary surveys. We also thank par-
ticipants at the PEPG conference, two anonymous referees, Rajashri
Chakrabarti, and Paul Peterson for comments and suggestions. All
errors are our own.
Notes
Karthik Muralidharan is Assistant Professor of Economics at the University of Califor-

nia–San Diego. Michael Kremer is Gates Professor of Developing Societies in the Depart-
ment of Economics at Harvard University.
1. Notable among these are Bashir (1994) in Tamilnadu; Kingdon (1996b) in Lucknow
(Uttar Pradesh); Govinda and Varghese (1993) in Madhya Pradesh; Tooley and Dixon
(2003) in Hyderabad (Andhra Pradesh); and Mehta (2005) in Punjab. As Kingdon (1996a)
mentions, ‘‘given inter-state variations in the structure and organization of education in
India, evidence from a single state will be illustrative but not necessarily representative.’’
2. See Chaudhury et al. (2006) for detailed results from the cross-country study.
3. Thus a district with 90% of its population in rural areas would have 9 rural PSUs and
1 urban PSU, whereas a completely urban district (as is the case when the randomly
picked district is the state capital, for example) would have 10 urban PSUs.
4. See appendix A of Kremer et al. (2004) for a detailed description of the sampling
procedure.
5. Covering grades one to five in most states, and grades one to four in some states,
depending on the classification of primary schools in the concerned state. The focus of
the study was completely on primary schools, and so the usage of the term school should
be understood to mean primary school unless stated otherwise.
6. Unrecognized schools are also more recently established, with an average age of 7.6
years as opposed to recognized private schools with an average age of 9.9 years. The frac-
tion of schools in this sample that report being run by a religiously oriented group is
quite small (15 out of 592 or 2.5% of schools). Schools run by religiously oriented groups
form a larger share of the private-aided schools that get government grants and are not
allowed to charge tuition fees (33 out of 152 or over 20%).
7. Since the survey was done across several states with different languages, the test was
weighted towards math as opposed to language. The test was short but the items used
had been pretested for validity. The test consisted of 12 arithmetic questions and 2 verbal
questions (that asked the students names in the local language and English respectively).
See appendix B of Kremer et al. (2004) for a detailed description of the test as well as the
procedure by which it was administered, graded, and coded.
8. These states include Andhra Pradesh, Haryana, Kerala (including private aided
schools), Punjab, Rajasthan, and Uttar Pradesh.
9. A teacher was considered to be absent if, at the time of a random visit during school
hours, he or she could not be found anywhere in the school premises. See Chaudhury
et al. (2006) and Kremer et al. (2005) for details on how absence and teaching activity
were measured and on the various steps we took to measure these accurately.
10. We thank Petia Topalova for making her calculations of district-level consumption
estimates available to us. See Topalova (2005) for details on these calculations.
11. Robust standard errors clustered at the state level are reported for specifications with
state-level right-hand-side variables and likewise for district-level variables, where the
standard errors are clustered at the district level.
12. The approximate exchange rate at the time of publication is Rs45 ¼ U.S.$1.
13. Direct data on teacher salaries in Andhra Pradesh has been collected in a different
ongoing study by one of the authors. The salary figures would be even higher if we
included benefits, the largest portion of which is the present value of a defined benefits
retirement pension. Private school teachers typically receive no benefits.
14. See Kingdon and Muzammil (2001) for more details on the power of public-school
teacher unions and how it has evolved over the years (based on a case study of the state
of Uttar Pradesh).
References
Andrabi, Tahir, Jishnu Das, and Asim Khwaja. 2002. The rise of private schooling in Paki-
stan: Catering to the urban elite or educating the rural poor? Mimeo. Kennedy School of
Government, Harvard University.
Annual Status of Education Report (ASER). 2005. Mumbai: Pratham Resource Center.
Bashir, Sajitha. 1994. Public versus private in primary education: Comparison of school
effectiveness and costs in tamilnadu. PhD dissertation, London School of Economics.
Chaudhury, Nazmul, Jeffrey Hammer, Michael Kremer, Karthik Muralidharan, and F.
Halsey Rogers. 2006. Missing in action: Teacher and health worker absence in developing
countries. Journal of Economic Perspectives (winter): 91–116.
De, Anuradha, Claire Noronha, and Meera Samson. 2001. India: Private schools and uni-
versal elementary education. South Asia Education Sector, Technical Working Paper No.
3. World Bank, Washington, D.C.
De, Anuradha, Manabi Majumdar, Claire Noronha, and Meera Samson. 2002. Private
schools and universal elementary education. In India Education Report—A Profile of Basic
Education, ed. R. Govinda. New Delhi: Oxford University Press.
Govinda, R., and Varghese, N. V. 1993. Quality of primary schooling in India: A case
study of Madhya Pradesh, International Institute for Educational Planning, National In-
stitute of Educational Planning and Administration. New Delhi: NIEPA.
Kingdon, Geeta. 1996a. The quality and efficiency of private and public education: A case
study of urban India. Oxford Bulletin of Economics and Statistics 58, no. 1: 57–81.
Kingdon, Geeta. 1996b. Private schooling in India: Size, nature and equity effects. Eco-
nomic and Political Weekly 31, no. 51 (December).
Kingdon, Geeta Gandhi, and Mohd Muzammil. 2001. A political economy of education
in India- I: The case of UP. Economic and Political Weekly 36, no. 32 (August: 11–18).
Kremer, Michael, Karthik Muralidharan, Nazmul Chaudhury, Jeffrey Hammer, and F.

Halsey Rogers. 2004. Teacher absence in India. World Bank: Washington, D.C.
Kremer, Michael, Karthik Muralidharan, Nazmul Chaudhury, Jeffrey Hammer, and F.

Halsey Rogers. 2005. Teacher absence in India: A snapshot. Journal of the European Eco-
nomic Association 3: 658–667.
Mehta, Arun. 2005. Elementary education in unrecognized schools in India: A study of
Punjab based on DISE 2005 data. NIEPA, New Delhi.
PROBE Team. 1999. Public Report on Basic Education in India. New Delhi: Oxford Univer-
sity Press.
Shah, Parth. 2005. Equity in education. Business Standard, July 13.
Tooley, James, and Pauline Dixon. 2003. Private schools for the poor: A case study from
India. Reading, U.K.: CfBT Research and Development. Available at hhttp://www.cfbt
.com/PDF/91001.pdfi.
Topalova, Petia. 2005. Trade liberalization, poverty, and inequality: Evidence from Indian
districts. NBER Working Paper 11614, Cambridge, Mass.
6 School-Sector Effects on Student Achievement in
India
Geeta G. Kingdon
Analysis of education in India in general and of private and public

schools in particular is hampered by the lack of available data. Despite
recent improvements, there is a serious dearth of reliable educational
data in India. First, the official data collection exercise on schools (both
annually and in the periodic All India Education Survey) collects infor-
mation only on the so-called ‘‘recognized’’ schools. Thus, large num-
bers of private schools are not included in the official data since they
are ‘‘unrecognized’’ (Kingdon 1996a). Second, coverage of even the rec-
ognized schools is incomplete. For instance, coverage of various types
of special schools is patchy across different states, such as central
schools, army schools, education guarantee schools, schools registered
with national examination boards, and so on (Mehta 2005). Third, en-
rollment figures in school-returns data are unreliable because failing
publicly funded schools exaggerate their student numbers to justify
their existence (Drèze and Kingdon 1998). Fourth, no national-, state-,
or district-level data are collected on student learning achievement in
primary and junior education in private and public schools; while
exam boards do have achievement data for the secondary school level,
these are not publicly available to researchers and, in any case, they
are not linked to student, teacher, and school characteristics. The
Annual Status of Education Report (Pratham 2007) collects national
household data on over 300,000 primary-age children’s learning
achievements but does not collect much information on home back-
ground or on schools and teachers.
Partly reflecting this lack of data, there is a paucity of good research
on educational issues in India. Most of the existing research based on
small surveys and using achievement production functions merely
establishes correlations rather than causation between student achieve-
ment and particular school inputs. The inability to deal convincingly
112 Geeta G. Kingdon
with issues of the potential endogeneity of school inputs has been due
to the ubiquitous problems of lack of credible instruments and lack
of panel or experimental data, though some recent studies have used
randomized experiments to study the impact of particular educational
interventions (Banerjee et al. 2005; Duflo and Hanna 2005; Muralid-
haran and Sundararaman 2006; Pandey, Goyal, and Levine 2006) and
others have used statistical techniques such as propensity-score match-
ing methods ( Jalan and Glinskaya 2002), instrumental variable meth-
ods (Kingdon and Teal 2007), pupil fixed-effects approaches (Kingdon
2006), and treatment effect models (Schmid 2006).
The first section of this chapter presents evidence on the relative
sizes of private, aided, and government schooling sectors in India. The
second section examines the relative effectiveness and per pupil costs
of private and public schools in India and the final section discusses
India’s experience with public-private partnerships in education.
6.1 The Relative Sizes of the Private and Public Schooling Sectors
The very first fact about the private and public schools in India is that
even their relative enrollment shares are not known with a degree of
accuracy. This is mainly due to a failure to include all types of schools
in official data collections but also partly due to exaggeration of enroll-
ments in publicly funded schools in these data (Kingdon 1996a; Drèze
and Kingdon 1998).
6.1.1 Typology of School Types in India

There are three main school types in India: government, aided, and pri-
vate. Schools run by the central, state, or local governments are re-
ferred to as government schools. Schools run by private managements
but funded largely by government grant-in-aid are known as private
aided or just aided schools. They charge the same fee levels as govern-
ment schools (which is now mandated to be nil) and pay the same sal-
ary rates to teachers as in government schools. From the early 1970s
onward, their teachers have been paid directly from the state govern-
ment treasury and are recruited by a government-appointed Education
Service Commission rather than by the school. Thus, government and
aided schools are now quite similar in their mode of operation. Schools
run by private managements without state aid are known as ‘‘private
unaided’’ schools. These run entirely on fee revenues and have virtu-
School-Sector Effects on Student Achievement in India 113
ally no government involvement in matters such as teacher recruit-

ment. These are thus the genuinely private schools and we refer to
these simply as ‘‘private’’ schools rather than using their full name ‘‘pri-
vate unaided.’’
Private schools in turn divide into two types: recognized schools and
unrecognized schools. Government recognition is an official stamp of
approval. To be eligible for government recognition, a private school is
by law required to fulfill a number of conditions.1 However, hardly
any private schools that get recognition actually fulfill all the condi-
tions. For instance, many recognized private schools in Uttar Pradesh
run in rented buildings when having an owned building is a mandated
condition of recognition (Kingdon 1994). Indeed, some of the condi-
tions are or have over time become mutually inconsistent.2 The main
benefit of recognition used to be that with recognition a school became
entitled to issue valid transfer certificates (TCs), which are a mandated
requirement for admission into upper primary and secondary schools.
However, the emergence of large numbers of unrecognized primary
schools (as shown later) suggests this requirement is no longer strictly
applied and that, de facto, recognized and unrecognized schools may
not be too different in terms of their physical facilities and modus
operandi.
6.1.2 Private Schooling Share According to Official and Household

Data
Despite the data deficiencies listed above, it is clear that the fee-
charging private schooling sector in India is much larger than thought
in the past. Kingdon (1996a) challenged the prevailing notion in Indian
writings, based on official published data, that the size of the private
sector in primary education was ‘‘infinitesimally small’’ or ‘‘negligibly
small.’’
Table 6.1 shows the enrollment share of private schools in rural and
urban India, according to both official school returns data in 1993 and
2002 and household survey data from 1993 and 2006. The bottom half
of the table shows corresponding figures for Uttar Pradesh, a state
with high levels of private school participation. The latest figures for
the year 2005–06 from the District Information System for Education
(DISE) are not shown because of its incomplete coverage.
Table 6.1 shows that according to official statistics, in 1993, only 2.8%
of all rural primary school students in India were studying in private
Table 6.1
Enrollment Share of Private Schools, 1993–2006
Official Household Official Household
published survey published survey
data data data data
Area School level 1993 1993 2002 2006
All India
Rural Primary 2.8 10.1 5.8 19.5
Junior/middle 6.5 7.9 11.1 20.4
Secondary 6.8 10.1 14.3 22.8
Urban Primary 25.7 26.2* 28.9 NA
Junior/middle 18.8 15.4* 39.1 NA
Secondary 11.5 11.2* 32.4 NA
Uttar Pradesh
Rural Primary 8.8 30.7 15.6 30.5
Junior/middle 28.3 23.3 31.0 35.0
Secondary 10.9 14.4 41.0 37.8
Urban Primary 53.3 49.7* 64.1 NA
Junior/middle 29.6 25.1* 48.2 NA
Secondary 5.3 11.3* 29.7 NA
Source: 1993 official data computed from Sixth All India Education Survey (NCERT 1998).
2002 official data computed from Seventh All India Education Survey, available at hhttp://
gov.ua.nic.in/NScheduleData/main3.aspxi. Rural household survey figures for 1993 are
based on the author’s calculations from 1993–94 NCAER survey. The urban household
survey figures marked * are taken from 1995–96 National Sample Survey published in
NSSO (1998, A69–82). Household survey figures for 2006 for rural India taken from
ASER2006 (Pratham 2007).
Note: In official data I have taken grades 9–12 as secondary school, i.e., corresponding to
students aged about 15–18 years old. ASER household survey collected data only on chil-
dren up to age 16, so children aged 7–10, 11–14, and 15–16 are assumed to be in primary,
middle, and secondary school respectively. In ASER, 18.6% of all children aged 7–10
were in private school and 4.6% were not in school, thus the private school share
of total school enrollment is taken to be ð18:6=ð100 4:6Þ 100 ¼ 19:5%Þ and similar cal-
culations were performed for middle and secondary school ages.
schools but that, according to household survey data for the same year,
10.1% of all rural Indian 6–10-year-old school attendees went to a pri-
vate school, a figure more than three times as high as the official esti-
mate.3 In rural Uttar Pradesh, official estimates for primary education
put the 1993 enrollment share of private schools at 8.8% but according
to the 1993–94 National Council of Applied Economic Research
(NCAER) household survey, the actual share was 30.7%, again more
than three times as high as the official estimate. By the time of the Pub-
lic Report on Education (PROBE) survey in 1996, 36% of all primary-
age students (6–11-year-olds) in rural Uttar Pradesh attended private
schools (PROBE Team 1999). Table 6.1 also shows that the enrollment
share of private schools at the primary level rose from 2.8% in 1993 to
5.8% in 2002. If the extent of underestimation of private enrollment in
2002 is the same as in 1993, then the true private school share of total
primary enrollments in rural India is three times as high as 5.8%, that
is, about 17%. This is close to the only recent national estimate avail-
able: the ASER-2006 national sample survey of over 330,000 house-
holds across 15,800 villages finds that 19.5% of school-going rural 7–
10-year-olds attended private schools in 2006.4 Table 6.1 shows that in
urban India, private schools’ share of total enrolment in 2002 was be-
tween about 30 and 40% at different levels of education, though being
an official figure, this ignores the numerous enrollments in private
unrecognized schools.
Some reasons for the large discrepancy between household survey
estimates and official estimates of the size of the private schooling sec-
tor in India are discussed in Kingdon (1996a) and Drèze and Kingdon
(1998): First, government and aided school teachers have an incentive
to over-report their enrollments when there is low demand for such
schools (since a school with falling rolls would lose teachers), and this
reduces the apparent enrollment share of private schools; second, as
previously stated, all official school ‘‘censuses’’ are carried out only in
the government-recognized schools, and in most Indian states there is
no requirement on private primary schools to be registered, let alone
government-recognized. It seems that rural private schools in particu-
lar do not easily get government recognition, for which many condi-
tions need to be satisfied. As Kingdon (1996a) says, given the exacting
conditions for and scant rewards of recognition, it is not surprising that
many private primary schools remain unrecognized.
The true size of the private schooling sector is greatly underesti-
mated in official data due to enumerating only the recognized schools.
Household survey data give a much more accurate picture since
parents have no incentives to over-report enrollment in publicly
funded schools or to report enrollment in recognized schools only.
Household survey data in table 6.1 suggest the extent to which the
enrollment share of private schools in primary education is underesti-
mated in official data—namely by about 67% in rural areas. Muralid-
haran and Kremer (2006) find that in their national survey of 20 states,
51% of all private primary schools were unrecognized. This accords
with evidence from individual states in other studies.5
Private schooling is utilized even among the poor in India. Findings

from the Micro Impacts of Macroeconomic Policies (MIMAP) survey
show that, of all enrolled children aged 5–10-years-old living below
the poverty line, 14.8% attended private schools (8% in rural and 36%
in urban India). The corresponding figures for ages 11–14 ( junior
school age) and 15–17 (secondary school age) were 13.8% and 7.0% re-
spectively (Pradhan and Subramaniam 2000). That private schools are
used by poor families is also found in five north Indian states (PROBE
Team 1999) and by Tooley and Dixon (2003) in Andhra Pradesh.
6.1.3 Growth in Private Schooling

The most telling statistic, however, is not the share of private schooling
in the stock of total school enrollment, but rather the share of private
schooling in the total recent increase in school enrollment at differ-
ent levels. This shows the relative growth of private schooling in
India (i.e., relative to the growth of government and aided schooling).
Table 6.2 presents the proportion of the total enrollment increase (over
time) that is absorbed by private schools. It is constructed from under-
lying numbers as shown in table 6.A.1 for urban India.6 Although
information in these official statistics excludes the numerous unrecog-
Table 6.2
Share of Recognized Private Schools in Total Enrollment Increase, by Region, Level of
Education, and Time Period
1978–86 1986–93 1993–2002
Rural
Primary 2.8 18.5 24.4
Middle 7.2 12.8 23.2
Secondary 5.8 15.8 30.9
Urban
Primary 56.8 60.5 95.7
Middle 35.7 31.8 71.7
Secondary 17.7 17.7 46.7
Rural & Urban
Primary 13.5 35.3 38.9
Middle 15.0 21.4 37.8
Secondary 10.7 16.8 38.4
Source: Author’s own calculations based on enrollment by school management type in
the All India Education Surveys for various years (NCERT 1982, 1992, 1998, 2006). See
table 6.A.1 for the underlying urban data.
nized schools, even recognized private school growth numbers are

telling.
We learn two things from table 6.2: first, that growth of private
schooling has accelerated over time; second, that in urban areas the
growth of private schooling has consistently been the greatest at the
primary level and progressively smaller at the middle and secondary
school levels, which is perverse from the equity point of view since
children of the poor are most well-represented at the primary school-
ing level.
Table 6.2 shows that in urban India, 56.8% of all the increase in total
primary school enrollment in the period 1978–86 was absorbed by pri-
vate schools; the corresponding figure for 1986–93 was 60.5% and for
the period 1993–2002 was 95.7%. Clearly, the pace of ‘‘privatization’’
increased greatly in the 1993–2002 period. In this nine-year period,
government and aided primary schools together absorbed only 4.3%
of the total net increase in primary school enrollments, that is, their
numbers or enrollments grew very slowly. Nearly 96% of the total in-
crease in urban primary enrollment was due to the growth of private
schooling! It bears emphasizing that even this dramatic statistic is an
underestimate since it takes no account of enrollment growth in the
numerous unrecognized private schools that are excluded from the of-
ficial statistics. The recent growth of private primary schooling in
urban India has been nothing short of massive. In rural India the rate
of expansion of private primary schooling has been much slower but
even here the pace of privatization picked up over time: only 2.8% of
total rural growth in primary enrollment in the 1978–86 period was
absorbed by private schools, but the corresponding figure for the
1986–93 period was 18.5% and for 1993–2002, 24.4%. Again, these
figures are all underestimates since they do not include growth in
enrollments in the unrecognized private primary schools. It is also
worth stating that any increase in aided school enrollments—shown in
table 6.A.1—(if it comes from the establishment of new aided schools
rather than merely from expansion in enrollment size in existing aided
schools) represents in fact an increase in private schools since
aided schools are private schools that start receiving government
grant-in-aid.
In some states, acceleration in the growth of private schooling
was spectacular even in the 1986–93 period. For instance, in urban
Uttar Pradesh (not shown in table 6.2), 94% of all new primary school
enrollment over the period 1986–93 occurred in private schools. The

growth of private schooling, particularly at primary and middle levels
of education, signals growing inequality of educational opportunity.
The growth of private schooling offers a possible explanation for the
fact that despite falling or virtually static per-capita public education
expenditure in several Indian states and falling share of elementary
education expenditure in state domestic product (Drèze and Sen 2002),
these states have improved their educational outcome indicators in the
1990s (Kingdon et al. 2004).
In the next section I examine evidence on the relative effectiveness
of private, aided, and government schools in India. This may help to
explain—at least in part—the relative popularity and growth of differ-
ent school types.
6.2 Internal Efficiency of Private and Public Schools
6.2.1 Relative Effectiveness of Private and Public Schools

Until recently, due to the lack of achievement data linked to school and
teacher characteristics, studies of the relative effectiveness of public
and private schools in India have had to rely on achievement tests car-
ried out by the researchers themselves in small samples of schools
(Bashir 1994, 1997; Govinda and Varghese 1993; Kingdon 1994, 1996b;
Tooley and Dixon 2003). These studies have been carried out in differ-
ent parts of India (Tamil Nadu, Madhya Pradesh, Uttar Pradesh, and
Andhra Pradesh respectively) and differ in several respects7 but they
share the common conclusion that private school students generally
outperform their public school counterparts in learning achievement
even after controlling for schools’ student intakes. Recently, Muralid-
haran and Kremer (2006) corroborate the findings in earlier studies
with nationally representative data on rural primary schools.
Bashir (1994, 1997) found that in the southern Indian state of Tamil
Nadu, private primary school students performed significantly better
in mathematics than government school students, though this was not
true in Tamil language achievement (although many of the private
schools were English-medium schools, unlike government schools,
which were Tamil-medium). She also found aided schools to be
more effective than government schools. Govinda and Varghese (1993)
found that in the central Indian state of Madhya Pradesh, achievement
levels of primary school students in private unaided schools were con-
siderably higher—in both mathematics and language—than those of
pupils in either aided or government schools. A private school effect

remained even after controlling for differences in home background
and school inputs. Kingdon (1994) found that in the northern Indian
state of Uttar Pradesh, private school students outperformed their
aided and government school counterparts and that aided and govern-
ment schools were similar in terms of their effectiveness in imparting
learning. Muralidharan and Kremer (2006) bring national data to bear
on this issue. They find that in rural India, private school students out-
perform public school students.
As is well known, even in studies that have information on measur-
able student characteristics, a major problem in studying the impact of
school type on student achievement is that students may choose school
type on the basis of unobserved traits such as ability and motivation.
If so, then any private school achievement advantage over public
schools—after controlling for observed student characteristics—cannot
simply be attributed to school type. To have a clean impact evaluation,
one needs either an experiment with students randomly assigned to
private and public schools, or a convincing way of dealing with endog-
enous sample selection into private and public schools. There are no
randomized experiments available in India to study the relative effec-
tiveness of private and public schools. Kingdon 1996 is the only peer-
reviewed journal study for India that attempts to control for potential
endogenous selection into different school types on the basis of un-
observed characteristics using the Heckman procedure.
As an illustration, table 6.3 summarizes this study’s findings from
Uttar Pradesh. The method of comparing the relative effectiveness of
the different school types is as follows: Choose a pupil at random from
the entire student population in the district and give her the average
characteristics of the full sample of pupils, say X. Then, using the fitted
selectivity-corrected achievement (ACH) equations for government (G),
private aided (PA) and private unaided (PUA) schools, predict a score
for this representative student if she were to attend a G school, another
score if it were a PA school, and a third score if it were a PUA school.
That is, predict an achievement score in each school type as:
ACHG ¼ ^bG X ð6:1Þ
ACHPA ¼ ^bPA X ð6:2Þ
ACHPUA ¼ ^bPUA X ð6:3Þ

Table 6.3
Raw and Standardized Achievement Scores and Relative Advantage Points by Sector
and Subject: G, PA, and PUA Schools
(A) (B)
Achievement points Achievement advantage points
G PA PUA PUA-G PUA-PA PA-G
(a) (b) (c) (c a) (c b) (b a)
Mathematics
Raw 8.97 8.36 17.09 8.12 8.73 0.61
Standardized (d) 11.38 10.09 12.80 1.42 2.71 1.29
[18] [31] [211]
Reading
Raw 9.77 10.86 16.85 7.08 5.99 1.09
Standardized (e) 13.78 13.73 13.82 0.04 0.09 0.05
[1] [2] [5]
Achievement
Raw 18.74 19.22 33.94 15.20 14.72 0.48
Standardized (d þ e) 25.16 23.82 26.62 1.46 2.80 1.34
[10] [19] [279]
OLS standardized 20.57 22.60 27.56 6.99 4.96 2.03
achievement points
Note: The maximum marks possible in the math and reading tests were 36 and 29 respec-
tively. Thus, the maximum achievement mark was the total of the two, i.e., 65. The fig-
ures in brackets are the standardized achievement advantages as a percentage of the raw
achievement advantages. The negative signs imply achievement disadvantages.
where the ^bs are the estimated coefficient vectors in the three sectors
and X is a vector of mean values of the explanatory variables, averaged
over the entire sample. Now PUA schools’ achievement advantage
over G schools, for example, can be calculated as (6.3) (6.1), PA
schools’ relative advantage over G schools as (6.2) (6.1), and so on.
The achievement scores thus calculated and the relative achievement
advantages of different school types are presented in table 6.3.
Table 6.3, column B, shows that the unadjusted (raw) mean achieve-
ment advantage of private unaided schools over government and
aided schools in all subjects falls greatly when personal endowments
and sample selectivity of pupils are controlled for. For example, PUA
schools’ raw mathematics-score premium over G schools of 8.12 points
falls to 1.42 points (still a large achievement advantage of 0.21 standard
deviations). This implies that, of the PUA schools’ mathematics advan-
tage of 8.12 points vis à vis G schools, 82% is explained by student in-
take and only 18% can be attributed to school influences. The PUA
schools’ raw mathematics advantage over PA schools falls from 8.73
points to 2.71 points, so that 31% of the observed PUA math advantage
is due to school-related factors and 69% due to student intake. The pre-
dicted mathematics score of a child in a PUA school (12.80 points) is
27% higher than her predicted math score in a PA school, where it
would be 10.09 points. In other words, PUA schools are 27% more
effective than PA schools in their math teaching.8
G schools’ tiny mathematics advantage over PA schools increases
after controls, suggesting that G schools are more effective in impart-
ing numeracy skills than PA schools. It is notable that all three school
types are roughly equally effective in imparting reading skills. The raw
reading-score premiums virtually disappear when student background
and selectivity are controlled.
The finding in econometric studies—that private schools are gener-
ally more effective than public schools in India—is broadly corrobo-
rated by the qualitative findings of the PROBE report, based on a
survey of 242 villages in 5 north Indian states (PROBE Team 1999).
The authors emphasize low teaching activity in public schools. The re-
port states that the extreme cases of teacher negligence were ‘‘less dev-
astating than the quiet inertia of the majority of teachers. . . . In half of
the sample schools, there was no teaching activity at the time of the
investigators’ visit. . . . Inactive teachers were found engaged in a vari-
ety of pastimes such as sipping tea, reading comics, or eating peanuts,
when they were not just sitting idle. Generally speaking, teaching
activity has been reduced to a minimum in terms of both time and
effort. And this pattern is not confined to a minority of irresponsible
teachers—it has become a way of life in the profession’’ (PROBE Team
1999, 63). While it does not aim to make a case for private schools, the
PROBE report contrasts such teacher behavior in government schools
with that in private schools. It notes (p. 64) ‘‘the high level of teaching
activity in private schools, even makeshift ones where the work envi-
ronment is no better than in government schools.’’ Again on page
102 the report notes, ‘‘In most of the private schools we visited,
there was feverish classroom activity.’’ Also: ‘‘This feature of private
schools brings out the key role of accountability in the schooling
system. . . . In a government school the chain of accountability is
much weaker as teachers have a permanent job with salaries and
promotions unrelated to performance. This contrast is perceived with
crystal clarity by the vast majority of parents’’ (p. 64). Other authors
Table 6.4
Salary Expenditure as a Proportion of Total Education Expenditure
Recurrent as a
percentage of
total educational Salary as a percentage of total recurrent
Year expenditure educational expenditure (%)
Primary Junior Secondary
1960–61 74.7 87.9 85.1 72.3
1965–66 79.4 90.7 89.2 75.3
1969–70 85.0 92.3 90.4 85.6
1974–75 87.1 96.6 94.3 87.1
1981–82 94.8 96.7 93.8 89.9
1987–88 97.3 NA NA 90.7
Source: Table 13.13 from Kingdon and Muzammil (2003).
Note: The system of reporting changed after 1988.
too have noted lax attitudes and low teacher accountability (Weiner
1990). This, in turn, seems to have its roots, at least partly, in teachers’
own demands for a centralized education system (Kingdon and
Muzammil 2003).
It is thought that in explaining the increased popularity of pri-
vate education, the breakdown of government schools is often more
decisive than parental ability to pay. ‘‘In rural Himachal Pradesh, for
instance, there is a good deal of purchasing power but the government
schools function well, so that there are few private schools. In central
Bihar, by contrast, poverty is endemic, yet private schools can be found
in many villages due to the dysfunctional state of government schools’’
(PROBE Team 1999, 102).
6.2.2 Relative Costs of Private and Public Schools

Next I turn to the relative unit costs of private and public schools, that
is, the monthly cost of teaching each student. School expenditures in
India are dominated by salaries. For example, in government-funded
primary schools, salary expenditure as a proportion of total recurrent
expenditure was 96.7% in 1981–82 (table 6.4). Comparable expenditure
breakdowns are not available for private schools since official statistics
do not collect financial data on private schools.
However, table 6.5 shows a comparison of per-pupil expenditures in
public and private schools in the Kingdon (1996) microstudy for Uttar
Pradesh, showing that in private schools, salaries account for a much
Table 6.5
Annual Per-Pupil Expenditures by School Type (Rupees)
Recurrent expenditure per pupil
Salary as a percentage
School type Salary Non-salary Total of total expenditure
Government (G) 1958.40 50.00 2008.40 97.5
Aided (PA) 1780.93 46.87 1827.80 97.4
Private (PUA) 735.94 262.96 998.90 73.7
Source: Kingdon (1994), chapter 6.
Table 6.6
Average Monthly Salary of Teachers by School Type
School type Average gross salary of sample
( junior schools) teachers (rupees per month)
Government (G) 2449.04

Aided (PA) 2429.48
Private (PUA) 1036.73
Source: Kingdon (1994), chapter 6.
lower proportion of total spending (74%) than in government and

aided schools (97%). Table 6.5 also shows that recurrent per-pupil
expenditure in private schools was only 50% of that in government
schools and 55% of that in aided schools. The relatively low per-pupil
expenditure in private schools is due largely to the fact that teacher sal-
ary rates are far lower in private than government schools. Table
6.6 shows that the average teacher salary in private junior schools
was only 42% of that in government schools and 43% of that
in aided schools. This is consistent with findings from different parts
of India in the early- to mid-1990s (table 6.7). More recent figures in
the last two columns of table 6.7 show that the private-public salary
gap has increased greatly since the early- to mid-1990s. Private schools
pay teachers market-clearing wages that have grown only slowly,
whereas government and aided schools pay teachers prescribed mini-
mum wages that have risen inexorably and contain large economic
rents.
Table 6.8 presents cost per unit of output by school type. The first
row shows that, on average, PUA schools are about twice as cost-
advantageous as G and PA schools. It also shows that there is in math-
ematics (but not in reading) an achievement advantage associated with
Table 6.7
124
Evidence from Indian Studies on Private Unaided and Government School Teachers’ Average Monthly Salaries
Kansal’s Govinda/ Jain’s Bashir’s Singh/ Murali-
PUA pay as a Kingdon’s study Varghese study study Sridhar dharan
School level percentage of study 1994 1990 1993 1988 1994 2002 & Kremer
Lucknow Many
district City of 5 districts Baroda districts 2 districts
of Uttar New of Madhya district of of Tamil of Uttar 20 states
Pradesh Delhi Pradesh Gujarat Nadu Pradesh of India
Primary/junior level G pay 42 39 49 47 47 20 20
PA pay 43 39 66 — 50 — —
Secondary level G pay 74 76 — — — — —
PA pay 79 76 — — — — —
Note: The Kingdon study sampled 182 teachers, Kansal 233 teachers, Govinda and Varghese 111 teachers, Bashir 419 teachers, and Singh and Srid-
har 467 teachers. We do not know the number of teachers sampled in Jain.
Sources: Jain (1988); Kansal (1990); Govinda and Varghese (1993); Bashir (1994); Kingdon (1994); Singh and Sridhar (2002); Muralidharan and
Kremer (2006).
Geeta G. Kingdon
Table 6.8
Unit Costs, Achievement, and Cost per Achievement Point (G, PA, and PUA Schools)
G PA PUA PUA:G PUA:PA PA:G
(a) (b) (c) (c/a) (c/b) (b/a)
Cost per student (C) 2008.00 1827.00 998.00 0.50 0.55 0.91
Predicted mathematics 11.38 10.09 12.80 1.13 1.27 0.89
score (M)
Cost per mathematics 176.00 181.00 78.00 0.44 0.43 1.03
point (C/M)
Predicted reading 13.78 13.73 13.82 1.00 1.00 1.00
score (R)
Cost per reading point 146.00 133.00 72.00 0.50 0.55 0.91
(C/R)
Predicted total score 25.16 23.82 26.62 1.06 1.12 0.95
(T ¼ M þ R)
Cost per score point 80.00 77.00 38.00 0.47 0.49 0.96
(C/T)
Source: Kingdon (1996).
attending a PUA school. Combining PUA schools’ 100% unit cost

advantage over G schools with their 13% mathematics advantage leads
to the conclusion that PUA schools are much more cost-effective than G
schools in their mathematics teaching. Another way of saying this is
that they produce the same level of numeracy skills as G schools at a
mere 44% of the cost of G schools. They produce the same level of
reading achievement as in G schools at half the cost. The comparison
of PUA schools with PA schools is of similar magnitudes.
To summarize, the results show that PUA schools’ ability to pay
market-clearing wages and, thus, their far more thrifty use of teachers
implies a large unit cost advantage over government-funded (G
and PA) schools. This reinforces their achievement advantage over the
other school types so that they are unambiguously and substantially
more cost-effective or internally efficient than both G and PA schools,
which are roughly equally efficient.
However, teachers’ objections to private school salary levels is that
market wages are not commensurate with the cost of (decent) living.
Whether one favors low market wages to achieve cost efficiency in ed-
ucation, or high minimum wages that protect teachers at the expense
of cost-efficiency, is not an ideologically neutral question. However,
it seems that in India, teacher salaries relative to per-capita income
are higher than in many other countries9 and that government-paid
teachers’ salaries have increased impressively in real terms: Drèze and
Saran (1993, 32a) report that in 1993 a teacher’s monthly salary in

Palanpur (Uttar Pradesh) could buy very nearly twice the amount of
wheat that his monthly salary could buy in 1983. Kingdon and
Muzammil (2003, chapter 13) calculate that in the 22-year period from
1974 to 1996, teacher salaries in Uttar Pradesh grew by about 5% per
annum in real—that is, inflation-adjusted—terms. This is significantly
higher than growth of per-capita real GDP in India over this period
which, according to Penn World Tables, was on average 3% per
annum.
6.3 Public-Private Partnership in Education in India
6.3.1 Historical Experience of PPPs

If private schools attract households, it suggests that parents perceive
them to be more advantageous relative to public schools. As Nechyba
(2005) states, ‘‘the nature of these advantages is likely to shape our
view of how the private sector can be most effectively mobilized to
advance academic achievement and other social goals.’’ The main
avowed advantage of public-private partnerships (PPPs)—publicly
funded but privately produced/delivered education—is that they har-
ness the energy, expertise, financial acumen, management skills, and
(sometimes) resources of the private sector to create better value for
money for taxpayers (LaRocque 2004). It is thought that PPPs provide
a more flexible way of producing education, since the private entity
running the school has more discretion about the running of the school
than is possible in public schools. Decentralized decision making at the
level of the school is thought to be more responsive to parents and to
foster local accountability.
In recent years there has been increased advocacy in favor of PPPs in
education. Any collaboration between public bodies such as local or
state government and private operators is referred to as PPP and there
are a wide variety of different types of PPPs in education in different
countries.
A substantial PPP system does operate in India, at least at the sec-
ondary and higher levels of education. This is the system of govern-
ment grant-in-aid to privately managed schools. According to the
Ministry of Human Resource Development (MHRD) cited in Bashir
(2005), in 1995–96 the percentage share of aided schools in total
schools was 34.0% and 44.3% respectively at the secondary and higher
secondary levels, though at the primary and middle levels, it was only
3.4% and 10.1% respectively. According to University Grants Council

data in 2000–01, 42% of all higher education institutions in India were
aided, which closely match the MHRD figures.
There is great interstate variation within India, in the extent to which
aided schools are utilized at different levels of education. For instance,
in 2000–01, in Kerala 59.7% but in Uttar Pradesh only 1.6% of all pri-
mary schools were aided, although at the higher secondary level the
picture was very different: 42.6% of all Keralan but 74.7% of all Uttar
Pradesh higher secondary schools were aided (MHRD, quoted in
Bashir 2005). Grants to aided schools account for a substantial propor-
tion of the education budget, though again there is large interstate
variation, for example, in elementary education from 0% in Gujarat to
84.4% in West Bengal; in secondary education from 1.1% in Himachal
to 94.2% in West Bengal; and in higher education from 0% in Bihar to
87.2% in Maharashtra (Bashir 2005).
When India inherited this PPP system from the British in 1948, aided
schools avoided many government regulations and interference. For
instance, they had far more autonomy than public schools in deter-
mining staff disciplinary policies. Any recognized private school could
apply for government grant-in-aid and, once granted aided status, it
received a subsidy from the state government. Its teachers were paid
out of school revenues and were thus accountable to fee-paying
parents and to the school manager. They could be disciplined and
hired/fired at the level of the school.
However, teachers of aided schools became increasingly unionized
and lobbied hard in the mid- to late 1960s to be paid directly by the
state government rather than via their private management which,
they claimed, engaged in unfair practices such as not paying fair
wages. Their intense lobbying and strikes helped the passage of the
momentous Salary Distribution Act (1971) in Uttar Pradesh and similar
acts in other states, such as the Direct Payment Agreement (1972) in
Kerala. These acts stipulated that aided school teachers’ salaries would
be paid directly to them rather than first going to school management.
The acts represented a massive centralization of school management
and they reduced aided school teachers’ accountability to their local
managers (Kingdon and Muzammil 2003). Thus, over time, aided
schools have become increasingly similar to public schools because
their modus operandi has become more and more like that of public
schools. In addition to their teachers’ salaries now being paid directly
by the state government, their teacher appointments are made by an
Education Service Commission of the state government, as for public

school teacher appointments. Given the similarities in the institutional
arrangements and teacher incentives in aided and government schools
in Uttar Pradesh, perhaps it is not surprising that, as seen earlier, there
is little difference between government and aided schools in terms of
either their effectiveness in imparting learning or in terms of their per-
pupil salary expenditures and per-pupil education expenditures.
Loss of local-level accountability (via centralizing legislation) is not
the only factor behind what is often seen as lax attitudes of public
school teachers toward their schools and students. One manifestation
of poor attitudes is their significantly higher within-village teacher ab-
senteeism rates than in private schools (Kremer et al. 2005), despite
getting salaries on average five times the private teachers’ salary rates
in the early 2000s (see table 6.7). The National Commission on
Teachers (NCT), the only government commission on teachers in post-
independence India, in a report written with much sympathy for the
teaching profession, blames teacher unions, suggesting that union-
backed teachers do not fear adverse repercussions if they are slack in
their work. The report of the NCT notes that ‘‘some of the Principals
deposing before it [i.e. before the Commission] lamented that they had
no powers over teachers and were not in a position to enforce order
and discipline. Nor did the district inspectors of schools and other offi-
cials exercise any authority over them as the erring teachers were often
supported by powerful teachers’ associations. We were told that there
was no assessment of a teacher’s academic work and other duties and
that teachers were virtually unaccountable to anybody’’ (National
Commission on Teachers 1986, 68).
This type of behavior is possibly further strengthened by the fact
that teachers (or rather, mostly their union leaders) are also legislators
in the state parliament, both as Members of the Legislative Assembly
(MLAs) and as Members of the Legislative Council (MLCs).10 In other
words, teachers have their sympathizers in the corridors of power,
who tend to shelter them in case any disciplinary issues arise. Aided
school teachers are in a politically particularly advantageous position
vis à vis government school teachers: although they are publicly paid
workers, they are not debarred from contesting political elections be-
cause they are not deemed to hold an ‘‘office of profit’’ under the gov-
ernment (unlike government school teachers). As a result, aided school
teachers freely contest elections. The National Commission on Teachers
(1986, 68) stated that ‘‘the most important factor responsible for viti-
ating the atmosphere in schools, we were told, has been the role of
teacher politicians and teachers’ organisations.’’
A further possible reason why aided schools—the Indian form of
educational PPP—perform no better than government schools in Uttar
Pradesh is that the government grant to aided schools is devoid of any
performance incentives. Despite the existence of certain rules and con-
ditions, the system of grants-in-aid in Uttar Pradesh is not linked to
the qualitative performance of schools. Even when the school’s grant-
in-aid was made conditional on satisfactory examination performance
of the school’s students, the pass rate was fixed at a paltry 45%—that
is, it was required that only 45% (or more) of the students pass the
high school exam (and a student needs to get a mere 33% average
grade across all subjects to pass high school)! Similarly, low standards
are required for the minimum number of days the school must be open
in order to remain eligible for grants. However, there is little monitor-
ing or verification of compliance with even these undemanding condi-
tions. On the whole, the system still leaves much to be desired and it is
not surprising because in practice, political maneuvers often overrule
the provisions laid down by the state government to sanction and reg-
ulate recurring and nonrecurring grants. The following observation of
Rudolph and Rudolph (1972, 105) with regard to the flouting of condi-
tions of grants-in-aid still holds: ‘‘these grants in aid are technically
conditioned upon the maintenance of certain academic and adminis-
trative standards, but in reality an educational entrepreneur who
enjoys political favour has little difficulty in establishing his institu-
tion’s qualification.’’
While the number of aided schools expanded tremendously in India
in the post-independence period, the system of grants-in-aid has re-
mained essentially the same as that introduced by the British 150 years
ago. By contrast, the British system of grants itself underwent revolu-
tionary changes and became more objective, particularly from the
1920s onward. What incentives can be built into public grants to pri-
vate schools is an area that deserves detailed study. A per-student (as
opposed to block) grant system may be desirable that relates a PPP
grant to various school performance indicators such as percentage of
total expenses spent on nonsalary costs (to encourage quality improve-
ments), percentage of total funds raised from nonfee sources such as
parental donations (to encourage equitable resource-generation), per-
centage of parents who are satisfied with the school (to encourage ac-
countability), and average number of students per class (to encourage
cost-consciousness), and so on. A more rational grant structure could

be a policy correction that has potentially the biggest payoffs in terms
of improved cost-efficiency in Indian education.
In summary, while PPPs are in theory supposed to lead to better
quality schooling than publicly produced education, educational PPPs
in India—the private aided schools—mostly function no better than
public schools, at least at the junior and secondary levels in Uttar Pra-
desh where the author has done most of her research. An important
reason for this appears to be that, over time, in response to their teach-
ers’ demands, aided schools have become like public schools, with few
performance incentives and a lack of local accountability. Govern-
ments have lacked the courage to increase local accountability of teach-
ers, who constitute a well-organized group with powerful political
representation and strong unions: Kingdon and Muzammil (2003, ch.
10 and 11) show that teacher unions in Uttar Pradesh have opposed
government proposals to introduce local-level accountability.
This experience of PPPs in education in India has important lessons
for future education policy in India as well as for other countries. One
thing it suggests is that when PPPs in education operate side by side
with government schools, political pressure can mount over time for
comparable treatment of teachers across the two school types, and any
advantages of PPPs over government schools—if real—may not be
enduring. However, the experience of PPPs in education in other coun-
tries, for example, in the Netherlands and elsewhere, shows that the
build-up of such political pressure is not inevitable and that PPPs can
work well in education. Why PPPs function well in some countries
and apparently not in others is a research question that deserves atten-
tion. The devil seems to be in the detail of the PPP scheme—the design
features that distinguish one PPP scheme from another.
6.3.2 Proposed New Form of PPP in Education

One of the main provisions of the current draft Right to Education Bill
is that the national government will pay private schools for some pub-
licly paid places. This effectively proposes to introduce a new form of
PPP involving a per-student public subsidy to private schools, quite
different to the way Indian states have financed private (so-called
aided) schools thus far, which is by paying block grants in the form of
salaries of all teachers of the aided school. The draft bill proposes to
oblige all private schools to give 25% of school places to students from
‘‘the weaker sections of society’’ and the government promises to reim-

burse the private schools for these places ‘‘at a rate equal to the per
child expenditure in state schools/fully aided schools and state funded
pre-schools, or the actual amount charged per student by such school,
whichever is less, in such manner as may be prescribed’’ (clause 14.2,
chapter 4, Right to Education Bill, August 2005).
This is the first time a post-independence Indian government has
sought to utilize the private sector to provide publicly funded educa-
tion (the aided school scheme was inherited from the British). Interest-
ingly, the scheme is championed not by the right wing, the usual
advocates of private education, but rather by those concerned with eq-
uity in education. Far from being the result of lobbying by the private
school sector for government funds, the scheme is rather generally
opposed by private schools on the grounds that mixing disadvantaged
children with those from well-off homes will be psychologically dam-
aging for disadvantaged children.
The bill and its provisions raise a number of important issues in
elementary education that have not been widely aired or seriously
debated. First, it has not been clarified how ‘‘weaker sections’’ will be
defined and chosen and how all disadvantaged children will get an
equal chance of access to private schools. Second, the choice of this par-
ticular way of providing ‘‘education of equitable quality’’ has not been
justified in comparison to other potential designs. Different designs,
depending on the alternative incentive structures inherent in them, can
address different educational efficiency and equity goals. For instance,
the bill proposes to give money directly to private schools to accept dis-
advantaged students rather than giving the same money as vouchers
(entitlements of a particular monetary value) to disadvantaged children.
The efficiency implications of these two ways of setting up the PPP
could be very different due to the differing potential for school compe-
tition under these two ways of providing the same amount of funding.
Whether money is given directly to the school (supply-side funding) or
via the students (demand-side funding) also has potentially different
equity implications because the matching of students to particular
schools is likely to be different under these two models. Third, the draft
bill could have major implications for the overall number of private
schools and their fee levels. It is unclear, for instance, whether pri-
vate schools’ response to the bill will be to create new places to accom-
modate publicly paid students or to replace 25% of existing students, or
a bit of both. Moreover, since per-pupil expenditure in public schools is

much larger than fee levels in most private schools (which now pay
teachers on average one-fifth the salary level of public schools, as seen
in table 6.8), the bill’s stipulations could well generally increase private
school fee levels.
This raises the question of why recommendations for decentralizing
reform in India, including the current draft Right to Education Bill,
have never included serious consideration of the possibility of provid-
ing school choice to students via vouchers, as a way of addressing ac-
countability of schools and teachers towards parents, unlike in other
countries such as Chile, Colombia, New Zealand, the United States,
and the United Kingdom, where there has been vigorous debate about
and experimentation with vouchers as well as charter and concession
schools. There are several potential explanations for this omission, as
well as several concerns about school voucher schemes.
First, in India (and other poor countries), the most obvious failure
of public schools is their very visible lack of resources, infrastructure,
facilities, books, and teaching materials, and the obvious remedy is
seen to be for government to fix these physical deficiencies. In many
other countries, the focus of school reform has moved to improving
incentives rather than inputs. Hanushek (2003) shows that while inputs
matter somewhat more in developing than in developed countries, the
provision of more resources does not raise student achievement levels
in the majority of studies.
A second plausible reason for India’s lack of consideration of a radi-
cal voucher-type reform is the fear of upsetting powerful vested inter-
ests such as teacher unions, which are likely to vehemently oppose
such proposals to increase local accountability. Unions have fought
hard over decades for legislation that shelters teachers from having
to be locally accountable, and successive Indian governments have
judged it politically infeasible to upset this powerful group that staffs
polling booths at election time.
Third, while the issues are complex and much debated, some
authors have raised concerns about adverse equity effects of vouchers
(Hsieh and Urquiola 2003; Ladd 2002; LaRocque 2004). They find that
voucher schemes can encourage the relatively better-off students to
abandon public schools, supplement the voucher with private funds,
and take private school places, thus leaving public schools with the
less-well-off and often less-motivated students. These are serious con-
cerns worth investigation. However, Nechyba (2005) argues that the

equity effects of school choice and PPP schemes can be addressed by
the way in which vouchers are designed. ‘‘In particular, one can design
vouchers to be inversely related to household income and to vary
depending on student type thus offering increased school resources to
those who find it disproportionately difficult to afford private school
tuition and those whose children are disproportionately costly.’’
Fourth, there would be concerns about implementation of school
choice schemes in the Indian context, such as: (1) the need to pro-
vide transport to nearby villages in order to offer real school choice/
competition in rural (low-population density) areas, which has its
attendant administrative and cost implications; and (2) the issue of
whether uneducated/illiterate parents are able to make informed
school choices. It may be argued that a voucher scheme will also be
problematic because of the lack of a strong regulatory system to ensure
schools’ compliance with standards and the scope for corruption in the
presence of weak monitoring and high costs of verification. However,
it is well known that the current system also suffers from weak regula-
tion and widespread corruption (e.g., see Dixon 2005) so the question
is whether these difficulties would increase in the presence of voucher
funding of education, and the answer is unclear. The point here is not
to make a case for or against vouchers or any particular way of giving
funds to private schools but rather to say that all the above concerns
and issues are worthy of detailed consideration before the legislation
is finalized.
6.4 Summary and Conclusions
Analysis of education issues in India is hampered by the absence of

data on student achievement and partial coverage of schools in official
data. Nevertheless, it is clear that private schooling has mushroomed
in India, particularly at the primary level, where the government
does not exert control as much as it does at the higher levels. Private
schooling is also popularly utilized by families below the poverty line.
According to qualitative accounts, the growth of private schools is
greatest in areas where public schools do not function well.
Evidence suggests that private schools are more than twice as cost-
effective as government schools in the large northern state of Uttar Pra-
desh. In other states where this issue has been explored (Tamil Nadu,
Madhya Pradesh, and Andhra Pradesh), private schools have also been
found to be generally more effective than government schools in
imparting learning, after controlling for student intake.
While aided schools—a form of public-private partnership in educa-
tion—are no more cost-effective than government schools in Uttar
Pradesh, this appears to be because over time they have become more
and more like government schools owing to aided school teachers’
successful lobbying for comparability of treatment vis à vis govern-
ment school teachers.
The draft Right to Education Bill proposes to introduce a new form
of public-private partnership in the form of a per-student subsidy to
private schools, but the implications of this measure have not been
vigorously debated yet. Nor have issues of school choice and competi-
tion via vouchers to families been considered in terms of their quality
and equity effects, as compared with the current PPP proposal in the
bill that intends to give funds directly to schools. The drawbacks of
voucher schemes including problems of implementation were dis-
cussed, noting that some of the same concerns would also apply to the
currently proposed form of PPPs in the draft bill and noting that equity
concerns may be addressed by making the voucher amount inverse to
family income.
It is critical to have a full national debate about the merits and
drawbacks of the draft bill’s proposed way of giving funds to private
schools, in comparison with alternative PPP designs. In such a discus-
sion, it would be useful to learn from the mistakes and successes of
other countries that have tried alternative schemes for allocating public
funds to private schools. Moreover, there may be a case to make for
introducing the proposed measures on a pilot basis in one part of the
country—to observe their effects for a specific period, and then to
hone and improve what will potentially be a far-reaching and long-
standing measure.
6.5 Appendix
See facing page.

Table 6.A.1
Growth in Enrollments by Level and School Type in India, 1978–2002 (Recognized Schools Only) India (Urban Only)
Level 1978–86 1986–93 1993–2002
Number of students enrolled Absolute % share of Absolute % share of Absolute % share of

increase in the total increase in the total increase in the total
1978 1986 1993 2002 enrollment increase enrollment increase enrollment increase
School type (a) (b) (c) (d) (e) ¼ b a (f) ¼ e/x1 (g) ¼ c b (h) ¼ g/x2 (i) ¼ d c ( j) ¼ i/x3
Primary
Government 10,270,760 11,189,956 12,836,933 12,766,950 9,19,196 26.7 1,646,977 37.1 69,983 1.3
Aided 4,735,795 5,304,932 5,414,067 5,710,967 5,69,137 16.5 109,135 2.5 296,400 5.6
Private 1,663,969 3,617,791 6,305,253 11,339,424 19,53,822 56.8 2,687,462 60.5 5,034,171 95.7
Total increase 34,42,155 4,443,574 5,261,088
in all types (x1) (x2) (x3)
Junior
Government 3,173,594 4,272,930 5,229,084 5,581,666 10,99,336 43.2 956,154 31.3 352,582 10.3
Aided 3,336,413 3,874,078 4,999,795 5,612,649 5,37,665 21.1 1,125,717 36.9 612,854 18.0
Private 488,266 1,395,610 2,367,067 5,084,580 9,07,344 35.7 971,457 31.8 2,447,513 71.7
Total increase 2,544,345 3,053,328 3,412,949

School-Sector Effects on Student Achievement in India
Secondary
Government 1,808,870 2,679,760 3,996,181 5,282,214 8,70,890 34.3 1,316,421 44.6 1,286,033 21.6
Aided 2,687,164 3,906,889 5,016,267 6,905,070 12,19,725 48.0 1,109,378 37.6 1,888,803 31.7
Private 195,969 645,442 1,168,160 3,944,952 4,49,473 17.7 522,718 17.7 2,776,792 46.7
Total increase 25,40,088 2,948,517 5,951,628
Source: NCERT (1982) Fourth All India Education Survey of 1978–79; NCERT (1992) Fifth All India Education Survey of 1986–87, p. 1116–1138;
135
NCERT (1998) Sixth All India Education Survey of 1993–94; figures from the Seventh All India Education Survey of 2002–03 obtained from
hhttp://gov.ua.nic.in/NScheduleData/main3.aspxi
Notes
1. In the state of Uttar Pradesh, to gain government recognition a private school must be
a registered society, have an owned rather than a rented building, employ only trained
teachers, pay salaries according to government prescribed norms, have classrooms of a
specified minimum size, and charge only government-set fee rates. It must also instruct
in the official language of the state and not be situated within five kilometers of a govern-
ment school (Kingdon 1994, chapter 2).
2. For instance, the condition to charge only government school-tuition-fee rates is
now incompatible with the condition to pay the government-prescribed salary rates to
teachers, since government school-fee rates have been cut consistently since the 1960s
and were abolished altogether in the early 1990s in all elementary schools, and since
government-prescribed minimum salaries to teachers have risen inexorably over time:
Kingdon and Muzammil (2003, chapter 13) estimate that average teacher salary rates
rose by 5% per annum in real terms in the 22-year period between 1974 and 1996.
3. The two sources are not exactly comparable since some school-going 6–10-year-olds
may attend preprimary or upper primary grades, i.e., be over- or underaged for their
grade.
4. Although ASER merged aided and unaided private schools into a single category, pri-
vate, at the primary level, there are few aided schools in most states so that the private
enrollment rates in ASER can be taken to mean mostly private unaided school enroll-
ments. ASER2006 found that 20.4% of boys and 16.8% of girls enrolled in grades one to
eight attended private schools. This 21% gender gap suggests one way in which girls are
discriminated against, namely via being substantially less likely to be sent to private
schools than boys (see Kingdon 2005).
5. Aggarwal (2000) found that in his four surveyed districts of Haryana in 1999, there
were 2,120 private primary schools of which 41% were unrecognized. The PROBE survey
of 1996 in five north Indian states did a complete census of all schools in 188 sample vil-
lages. It found 41 private schools of which 63% were unrecognized. Mehta (2005) found
that in seven districts of Punjab, there were 3,058 private elementary (primary and ju-
nior) schools, of which 86% were unrecognized.
6. Take the example of the junior (or upper primary) education level in urban India. Be-
tween 1993 and 2002, according to table 6.A.1, junior enrollment increased by 3,412,949.
Out of this, the enrollment increase in private schools was 2,447,513, which is 71.7% of
the total increase in junior enrollments.
7. While Kingdon’s study is based on students in the final year of upper primary educa-
tion (grade eight), the other studies are based on students in the final year of lower
primary schooling (grades four or five). The methods used differed too. Bashir used hier-
archical linear modeling, Govinda and Varghese used OLS regression, and Kingdon used
sample selectivity correction models. The extent of controls for home background dif-
fered across the studies too, as well as whether school and teacher characteristics were
included in the achievement equations. Finally the costs of private and public schooling
were calculated differently in the different studies. In all three studies, the stratified ran-
dom samples of private schools consisted of schools of all types—nonprofit, proprietary,
faith-based, high-fee and low-fee schools, etc.
8. The correction for sample selectivity reduces the private school achievement advan-
tage over government schools by a very large amount (compared to the OLS results in
the last row of table 6.4). This large reduction is somewhat surprising since one would
not expect the unobserved factors (that remain after controlling for the child’s score in
the Raven’s test of ability and for a rich set of home background characteristics) to make
such a large difference to a child’s predicted achievement score.
9. For example, the ratio of (public primary school) teacher salary to per capita GDP in
the late 1990s was 1.15 in OECD countries, 4.4 in Africa, 2.3 in Latin America, and 2.9 in
Asia (UNESCO statistics, available at hportal.unesco.org/education/en/file_download
.php/i) but 8.5 in Uttar Pradesh, India (author’s own calculation).
10. The constitution of India guarantees representation to teachers in the Upper Houses
of state legislatures. Thus, uniquely among all worker groups, the teaching profession
has been singled out for this political privilege (see Kingdon and Muzammil 2003),
though Upper Houses now exist in only four large states in India.
References
Aggarwal, Yash. 2000. Public and private partnership in primary education in India: A
study of unrecognised schools in Haryana. New Delhi: National Institute of Educational
Planning and Administration.
Banerjee, A., S. Cole, E. Duflo, and L. Linden. 2005. Remedying education: Evidence from
two randomized experiments in India. NBER Working Paper No. 11904, December.
Bashir, Sajitha. 1994. Public versus private in primary education: Comparisons of school
effectiveness and costs in Tamil Nadu. Unpublished Ph.D. thesis, London School of
Economics.
Bashir, Sajitha. 1997. The cost-effectiveness of public and private schools: Knowledge
gaps, new research methodologies and an application in India. In Marketizing Education
and Health in Developing Countries: Miracle or Mirage?, ed. C. Colclough, chapter 5. Oxford:
Clarendon.
Bashir, Sajitha. 2005. Grant-in-aid mechanism in India: Policy note. World Bank.
Dixon, P. 2005. Private aided primary and secondary schools in Karnataka, India: A cau-
tionary tale of a public-private partnership. Mimeo, School of Education, Newcastle
University.
Drèze, J., and M. Saran. 1993. Primary education and economic development in China
and India: Overview and two case studies. Discussion Paper No. 47, Development Eco-
nomics Research Programme, London School of Economics.
Drèze, Jean, and Geeta Gandhi Kingdon. 1998. Biases in education statistics. The Hindu,
March 6.
Drèze, Jean and Geeta Gandhi Kingdon. 2001. Schooling participation in rural India. Re-
view of Development Economics 5, no. 1 (February): 1–24.
Drèze, Jean, and Amartya Sen. 2002. India: Development and Participation. Oxford and
Delhi: Clarendon Press.
Duflo, Esther, and Rema Hanna. 2005. Improving teacher attendance in rural India.
Mimeo, Poverty Action Lab, MIT, Cambridge, MA.
Govinda, R., and N. V. Varghese. 1993. Quality of Primary Schooling in India: A Case Study
of Madhya Pradesh. Paris: International Institute for Educational Planning, and New Delhi:
National Institute of Educational Planning and Administration (NIEPA).
Hanushek, Eric. 2003. The failure of input-based schooling policies. Economic Journal 113,
no. 485: F1–F120.
Hsieh, C. T., and M. Urquiola. 2003. When schools compete: How do they compete? An
assessment of Chile’s nationwide school voucher program. Working Paper No. 10008.
National Bureau of Economic Research.
Jain, S. C. 1988. Non-official initiatives for catering to the educational needs of the urban
poor: A case study of Baroda slums. Unpublished typescript. Baroda: Baroda Citizens
Council.
Jalan, Jyotsna, and Elena Glinskaya. 2002. Small bang for big bucks? An evaluation of a
primary school intervention in India. Mimeo, World Bank, August.
Kansal, S. M. 1990. Disparity in income and levels of living among teachers in Delhi. Eco-
nomic and Political Weekly 25, no. 46 (Nov. 17): 2547–2554.
Kingdon, Geeta G. 1994. An economic evaluation of school management-types in India:
A case study of Uttar Pradesh. Unpublished D.Phil. thesis, Economics Department, Ox-
ford University.
Kingdon, Geeta G. 1996a. Private schooling in India: Size, nature and equity effects. Eco-
nomic and Political Weekly 31, no. 51: 3306–3314.
Kingdon, Geeta G. 1996b. The quality and efficiency of public and private schools: A case
study of urban India. Oxford Bulletin of Economics and Statistics 58, no. 1 (Feb.): 55–80.
Kingdon, Geeta G. 2005. Where has all the bias gone? Detecting gender bias in the intra-
household allocation of educational expenditure in rural India. Economic Development and
Cultural Change 53, no. 2: 409–452.
Kingdon, Geeta Gandhi. 2006. Teacher characteristics and student performance in India:
A pupil fixed effects approach. GPRG-WPS-059, Global Poverty Research Group, Depart-
ment of Economics, University of Oxford.
Kingdon, G., R. Cassen, K. McNay, and L. Visaria. 2004. Education and literacy. In
Twenty-First Century India: Population, Economy, Human Development, and the Environment,
ed. T. Dyson, R. Cassen, and L. Visaria. New Delhi: Oxford University Press.
Kingdon, Geeta G., and Mohammad Muzammil. 2003. The Political Economy of Education
in India: Teacher Politics in Uttar Pradesh. Delhi: Oxford University Press.
Kingdon, Geeta Gandhi, and Francis Teal. 2007. Does performance related pay for teach-
ers improve student achievement? Some evidence from India. Economics of Education Re-
view 24, no. 4 (August): 473–486.
Kremer, Michael, Nazmul Chaudhury, Hasley Rogers, Karthik Muralidharan, and Jeffrey
Hammer. 2005. Teacher absence in India: A snapshot. Journal of the European Economic As-
sociation 3: 658–667.
Ladd, Helen F. 2002. School vouchers: A critical view. Journal of Economic Perspectives 16,
no. 4: 3–24.
LaRocque, Norman. 2004. School choice: Lessons from New Zealand. Paper presented to
the conference on ‘‘What Americans Can Learn from School Choice in Other Countries.’’
Cato Institute, Washington D.C., May.
Mehta, Arun. 2005. Elementary education in unrecognized schools in India: A study of
Punjab based on DISE 2005 data. New Delhi: NIEPA.
Muralidharan, Karthik, and M. Kremer. 2006. Private and public schools in rural India.
Mimeo, March, Harvard University, Cambridge, MA.
Muralidharan, Karthik, and V. Sundararaman. 2006. Teacher incentives in developing
countries: Experimental evidence from India. Mimeo, Harvard University.
National Commission on Teachers. 1986. The teacher and society: Volume I. Report of the
National Commission on Teachers, 1983–85. New Delhi: Government of India Press.
National Council for Educational Research and Training (NCERT). Various years. All
India education survey. New Delhi: NCERT.
National Sample Survey Organisation. 1998. Attending an educational institution in In-

dia: Its level, nature, and cost. Report no. 439. New Delhi: NSSO.
Nechyba, Thomas. 2005. Mobilizing the private sector: A theoretical overview. Working
Paper PEPG 05-06, Program on Education Policy and Governance, Kennedy School of
Government, Harvard University.
Pandey, Priyanka, M. Goyal, and D. Levine. 2006. Does information matter for service
delivery and governance? Mimeo, Johns Hopkins School of Medicine, Baltimore, MD.
Pradhan, B., and Subramanian, A. 2000. Education, openness and the poor: Analysis
of an all-India survey of households. National Council of Applied Economic Research
(NCAER) Discussion Paper, no. DP020015. New Delhi: NCAER.
Pratham. 2007. Annual status of education report 2006. New Delhi: PRATHAM Resource
Centre.
PROBE Team. 1999. Public report on basic education in India. New Delhi: Oxford University
Press.
Right to Education Bill. 2005. Right to Education Bill, draft, August 25, 2005, at hhttp://
www.education.nic. in/htmlweb/RighttoEducationBill2005.pdfi.
Rudolph, L. I., and S. H. Rudolph. 1972. Education and politics in India. Cambridge, MA:
Harvard University Press.
Schmid, Juan Pedro. 2006. Was the District Primary Education Programme in India effec-
tive? Draft Ph.D. thesis chapter, Swiss Federal Institute of Technology, Zurich, August.
Singh, Shailendra, and Kala Seetharam Sridhar. 2002. Government and private schools:
Trends in enrollment and retention. Economic and Political Weekly 37, no. 41 (October 12):
4229–4238.
Tooley, James, and Pauline Dixon. 2003. Private schools for the poor: A case study from
India. Reading, U.K.: CfBT.
Weiner, Myron. 1990. The Child and the State in India: Child Labor and Education Policy in
Comparative Perspective. Princeton: Princeton University Press.
III Public Funding of Privately Managed Schools
7 School Vouchers in Colombia
Eric Bettinger
In the early 1990s, secondary school enrollments among the poorest

20% of Colombia’s population were only 55%. By contrast, 89% of the
richest 20% of Colombia’s population were attending secondary school
and 75% of the overall population were enrolled (Sanchez and Mendez
1995). As policy makers grappled with how to increase poor student
enrollments and lessen the enrollment gap, they also faced an over-
burdened and overcrowded public school system. In Colombia, the
average school day is four hours, and most public schools hosted mul-
tiple sessions per day. Only 2% of public secondary schools were host-
ing only one session per day, and almost 20% of public secondary
schools were hosting three sessions per day. In lieu of these multiple
sessions at each school, the World Bank (1993) noted that many schools
could not facilitate additional enrollments despite projected enrollment
growth.
In 1991, Colombia attempted to improve enrollment rates through a
unique partnership between the public and private sectors (King, Ora-
zem, and Wohlgemuth 1998). The program, called the Plan de Amplia-
ción de Cobertura de la Educación Secundaria (PACES), sought to
take advantage of excess capacity in the private sector. The Colombian
government issued private school vouchers for students entering sixth
grade, the start of Colombian secondary school. The vouchers tar-
geted the poorest third of the population and were renewable so long
as the recipient made adequate progress towards secondary school
graduation.
By 1997, PACES had grown into one of the world’s largest private
school voucher programs. Over 125,000 PACES vouchers had been
awarded. While the program was large relative to other voucher pro-
grams, it was small relative to the overall secondary school system. In
144 Eric Bettinger
1995, approximately 3.1 million students attended secondary schools

in Colombia with about 37% of them in private schools.
One of the distinctive elements of PACES is its use of lotteries. From
the beginning, the demand for PACES vouchers far exceeded the sup-
ply. PACES required the use of lotteries to allocate vouchers when
there was excess demand. These lotteries created natural ‘‘control’’ and
‘‘treatment’’ groups similar to a randomized trial. Students who ap-
plied unsuccessfully to the voucher lottery form an unbiased com-
parison group for students who won the voucher lottery. Comparing
the academic and nonacademic outcomes of students involved in the
voucher lottery can show the effects of the voucher program.
There have been two major studies utilizing these voucher lotteries
to measure PACES’ effects. The first study was conducted by Joshua
Angrist, Erik Bloom, Elizabeth King, Michael Kremer, and me (Angrist
et al. 2002). During 1998 and 1999, we attempted almost 3,000 surveys
of students who had applied for PACES vouchers in selected cities
throughout Colombia. The data from these surveys showed that after
three years, voucher lottery winners scored about 0.2 standard devia-
tions higher on standardized exams, were 15 percentage points more
likely to have attended private school, and were about 5 percentage
points less likely to have repeated a grade in secondary school. Because
of the reduced grade repetition, voucher winners had completed 0.1
more years of schooling. The vouchers, however, did not significantly
affect dropout rates.
While the results of the first study were compelling, as I will discuss,
there were reasons to doubt whether the voucher program had led to
long-term differences in outcomes for voucher students. As a follow-
up to the first study, Joshua Angrist, Michael Kremer, and I pursued
a longer-run follow-up focusing on high school graduation and the
college entrance exam (Angrist, Bettinger, and Kremer 2006). In this
study, we matched PACES application data to administrative records
from Colombia’s college entrance exam, the ICFES (Instituto Colom-
biano Para El Fomento De La Educación Superior). Using these data,
we again compared outcomes of voucher applicants. The results were
striking. Voucher lottery winners were about 20% more likely to take
college entrance exams than unsuccessful voucher applicants; and not
only were they more likely to take the exam, but they also scored
higher on it.
This chapter seeks to review this evidence from Colombia. Section
7.1 provides additional background information on the Colombia
School Vouchers in Colombia 145
voucher program. Section 7.2 describes the data sources and methodol-
ogies used in these evaluations. Section 7.3 provides an overview of
evidence on the vouchers’ effectiveness after three years. Section 7.4
discusses the impact of the voucher on college entrance exams. Section
7.5 discusses possible mechanisms by which the voucher may have af-
fected student outcomes, including possible puzzles raised in the eval-
uation of the Colombian voucher project. Section 7.6 concludes.
7.1 PACES Background
The Colombian government established the PACES program in No-

vember 1991. The program was part of a larger effort to decentralize
public services and to expand private sector provision (King et al.
1997). The Colombian government advertised the program in local
newspapers and on the radio, and the program immediately proved
popular. In its first year, in Bogotá alone, 14,607 students applied for
the program.
In order to improve enrollment rates among the poorest families in
Colombia, PACES targeted low-income families (King, Orazem, and
Wohlgemuth 1998). To qualify for the voucher, parents had to present
a utility bill proving that they lived in one of the two lowest socioeco-
nomic strata (out of six possible strata). Research by Morales-Cobo
(1993) suggests that this targeting was effective in Bogotá.
To be eligible for the voucher, children had to be entering sixth
grade, the start of Colombian secondary schools, and be under the age
of 16. Children were also only eligible if they had been attending pub-
lic school in the previous year and had already arranged admission at
a participating private secondary school. Not all private schools par-
ticipated in the program. Only about 40% of private schools actually
accepted the voucher, and schools that typically participated were not
elite schools but rather low-tuition schools serving low-income popula-
tions. King, Rawlings, Gutierrez, Pardo, and Torres (1997) investigated
differences between public secondary schools and participating private
schools. They found that pupil-teacher ratios, test scores, and access to
technology were similar across schools. The schools also had similar
median scores on the ICFES exam.
Students could use the vouchers at both academic and vocational
schools. Vocational schools, including some for-profit schools, were
over-represented in the group of participating private schools, al-
though after 1996 for-profit schools were excluded from the voucher
146 Eric Bettinger
program. Because students were accepted at a private school prior to

the lottery, we can actually dichotomize the lottery into two parts: stu-
dents who had already been accepted at a vocational school and
students who had already been accepted at an academic school. As
I will discuss, Michael Kremer, Juan Saavedra, and I use this infor-
mation to shed light on mechanisms by which the voucher may have
affected student outcomes.
So long as students were promoted at the end of a grade, they could
automatically renew their voucher through eleventh grade, the end of
Colombian high school. Students failing a grade were supposed to be
dropped from the PACES program. Calderon (1996) shows that about
77% of recipients renewed their vouchers. Additionally, the rules of the
voucher program allowed students to transfer to other schools and re-
tain their voucher; however, our data suggests that few students who
transferred kept their vouchers.
The voucher initially covered full tuition in a participating pri-
vate school, but the value did not keep pace with tuition. By 1998,
the voucher covered a little over half of tuition fees. The funds for the
voucher came from both the municipal (20%) and federal (80%) gov-
ernments. Municipalities determined the appropriate number of
vouchers, and each municipality conducted its own lottery if demand
for the vouchers exceeded supply. We obtained computerized or paper
copies of lists of lottery winners and losers from local municipalities.
In the applicant lists, we observed the students’ names, contact infor-
mation, national identification number, and school of application. The
most important piece of information was students’ national identifica-
tion numbers. An identification number consists of 11 digits, the first
6 indicating date of birth. The eleventh digit in the ID number has
a mathematical relationship with the other four digits, which we can
check to verify that the ID number is valid. About 9.5% of applicants
had invalid birth dates. This was the prevailing reason why IDs were
invalid. If students reported valid birth dates, 97% of the time their ID
number was valid.
Using the application data, we can verify if the lottery was indeed
random. If it was truly random, we should find few differences be-
tween the characteristics of voucher lottery winners and losers. Table
7.1 shows data from the applicant lists on age, gender, having a phone,
and having a valid ID. We report data for multiple cohorts including
applicants from Bogotá who applied in 1992 and 1995. The first col-
umn of table 7.1 shows the average characteristic among voucher lot-
Table 7.1
Student Characteristics by Voucher Status
Losers’ mean Difference by voucher status
(1) (2)
A. Bogotá 1995
Age 12.78 .137
[2.22] [.064]
Male .484 .004
[.016]
Has phone .874 .013
[.010]
Has valid ID .882 .010
[.010]
B. Bogotá 1992
Age 12.83 .093
[1.23] [.029]
Male .533 .042
[.010]
Has phone .397 .184
[.009]
Has valid ID .681 .053
[.009]
Notes: The table reports voucher losers’ mean and difference for voucher winners.
Standard deviations are in the first column for nonbinary variables. Standard errors are
included in the second column.
tery losers and its standard deviation. The second column shows the
difference between voucher winners and losers and the corresponding
standard error. In terms of gender and having a phone, we find no sig-
nificant differences between voucher winners and losers in the 1995 co-
hort. When we look at age, we find that younger students are more
likely to win the vouchers. This is significant and may suggest some
type of nonrandomness.
Even if the lottery was random, however, there may be some reasons
for this finding. First, in the Bogotá 1995 cohort, there are a couple of
significant outliers (e.g., a reportedly 92-year old fifth grader) among
voucher lottery losers. If we exclude these individuals or compare
medians rather than means, the difference in ages is much smaller. An-
other possible reason for this difference involves the accuracy of the
records. In most cases, we received two separate lists—one with all lot-
tery losers and another with lottery winners. One of our worries was
that information for lottery winners was updated while lottery losers’
148 Eric Bettinger
information was not. This could lead to more accurate ages, addresses,
and ID numbers. In the Bogotá 1992 lottery, there appears to be some
evidence of this. Voucher winners were about 5 percentage points
more likely to have a valid ID than voucher losers and about 18 per-
centage points more likely to have reported a phone number. How-
ever, in the Bogotá 1995 cohort, we did not find differences in the
likelihood that students had valid ID numbers.
As a final check on the data, we looked at the ‘‘win’’ rate of each of
the schools. In theory, schools should have had ‘‘win’’ rates among lot-
tery applicants from their school that were similar to the overall lottery
average. For the cohorts represented in table 7.1, there were few out-
liers, and the existing outliers were typically the result of a low number
of overall applicants to that school. However, in one cohort not in-
cluded in table 7.1, we found significant outliers. In the excluded co-
hort, we found one school in particular where 100% of applicants had
won the voucher. Given the school’s reputation as a politically con-
nected school and the large number of students who had applied, we
could not rule out nonrandomness.
7.2 Data and Methodologies
7.2.1 Data Sources

There were three sources of data used in the analyses. First, as I have
explained, our studies relied on information from the applicant list.
Using contact information from the list, we attempted to interview a
random sample of voucher applicants from Bogotá in 1995.1 Generally,
we stratified this sample so that we were contacting equal numbers of
lottery winners and losers. The resulting household surveys are the
second source of data. The surveys included comprehensive details of
students’ schooling including a grade-by-grade summary of schools
attended and grade promotion. The surveys also gathered information
about students’ parents and siblings. Our response rate in the surveys
was 55% among voucher winners and 53% among voucher losers. The
difference was not significant.
In conducting our interviews, few applicants actually refused to
respond. The most frequent reasons for lack of response were bad
addresses or moves. Our response rate is slightly lower than that in
other voucher studies (e.g., Mayer et al. 2002). Although we would
have liked the response rate to be higher, the symmetric response rates
Table 7.2
Descriptive Statistics by Voucher Status
Difference by
Losers’ mean voucher status N
(1) (2) (3)
A. Bogotá 1995
Age at time of survey 15.0 0.013 1,172
[1.4] [0.078]
Male 0.501 0.004 1,139
[0.029]
Mother’s highest grade completed 5.9 0.079 1,075
[2.7] [0.166]
Father’s highest grade completed 5.9 0.431 911
[2.9] [0.199]
Mother’s age 40.7 0.027 1,123
[7.3] [0.426]
Father’s age 44.4 0.567 940
[8.1] [0.533]
Father’s wage (>2 min wage) 0.100 0.005 861
[0.021]
Notes: The table reports voucher losers’ mean and difference for voucher winners. Stan-
dard deviations are in the first column for nonbinary variables. Standard errors are
included in the second column.
across winners and losers suggests that any bias resulting from non-
response is likely to be minimal (Angrist 1997). Because response prob-
abilities are uncorrelated with voucher win/loss status, there should
be little bias from our failure to interview all applicants.
Table 7.2 shows some basic student-level characteristics used in the
analysis. We only report descriptive characteristics in this table and re-
port student outcomes in other tables. As in table 7.1, the first column
shows the average characteristic of voucher lottery losers and the cor-
responding standard deviation. The second column of table 7.2 shows
the differences for voucher winners with the corresponding standard
error. The third column shows the sample size for the specific variable.
As before, we find few significant differences
The typical applicant was about 15 years old. About half of the
applicants were male. The average education of both the mothers and
fathers in the sample was slightly less than six years. We detected some
differences in the education levels of fathers in our data. We find that
fathers of voucher winners had about 0.4 years less of schooling com-
pleted, although we have a smaller sample size in these regressions.
150 Eric Bettinger
We also find that about 10% of the fathers in our data were earning
more than two minimum wages. This does not vary across voucher
status.
Our final source of data came from matching student applications
to the ICFES exam, the national college entrance exam. Ninety percent
of students graduating from Colombia’s secondary school system take
the ICFES exam. It is the most common exam used in college admis-
sions and 75% of test takers go on to college (World Bank 1993). Be-
cause of the high proportion of high school students who take the
exam, it is likely a better proxy for high school graduation than for col-
lege entrance.
The primary variables used in the matching were the student’s name
and identification number. If applicant lists had been updated so that
voucher winners had more correct names and ID numbers than
voucher losers, then we could detect spurious effects of the voucher
solely because winners would have more accurate information. Be-
cause our analysis found that voucher winners and losers in the Bogotá
1995 lottery had similar likelihood of having a valid national ID, we
focus our matching solely on them. We also matched ICFES records
for the 1992 cohort. Similar to the results I show below for the 1995
cohort, we found that voucher winners in the 1992 cohort were more
likely to take the college entrance than voucher losers. However,
because of the possibility that voucher winners’ records for the 1992
cohort had been updated, our finding could be spurious.
ICFES exams are offered twice a year, and for the 1995 cohort, we
searched for matches among all test-takers in 1999, 2000, and 2001.
Assuming that students had not repeated, they should have taken the
ICFES exam in 2000. The ICFES scores used here are from the rede-
signed scoring system introduced in March 2000. Our scores are for
the mathematics and language components of the Common Core of
Basic Competence (Nucleo Comun de Competencias Basicas), which in-
cludes modules in biology, chemistry, physics, mathematics, language,
history, geography, and a foreign language test chosen by the student.
The ICFES is given over a two-day period with two morning sessions
and an afternoon session on the first day. The mathematics and lan-
guage components of the Common Core each take one hour and have
35 questions. Test scores are reported on a scale of 0–100, with the
score distribution highly concentrated in the 30–70 range. The distribu-
tions of mathematics and language scores for all those tested in Bogotá
in March 2000 appear in figure 7.1 (for 6,868 examinees). We discuss
Figure 7.1
Distribution of Test Scores in Bogotá versus PACES Sample
(a) Language
(b) Math
152 Eric Bettinger
the specific variables from the ICFES and the matching strategy in sec-
tion 7.4 of this chapter.
7.2.2 Empirical Methodologies

Because of randomization, simple t-tests comparing the outcomes of
winners and losers can identify the effects of the voucher (Angrist and
Krueger 1999). We also use the following regression model to identify
the effects of the voucher:
yi ¼ Xi0 b 0 þ a0 Di þ ei ; ð7:1Þ
where yi is a measure of some type of outcome for student i, Di is an

indicator for whether a student won the voucher lottery, and Xi in-
cludes covariates such as age, gender, and controls for neighborhood.
The parameter a shows the effect of winning the voucher lottery on
student outcomes. This is often called the ‘‘intention to treat’’ parame-
ter. It shows the effect of offering the voucher. The intention to treat
reflects both the ‘‘effect of the treatment on the treated’’ (i.e., the effect
of using a voucher) and the probability of being treated. If everyone
who is offered the voucher uses it and no one in the control group
does, then the intention to treat is equivalent to the effect of the treat-
ment on the treated. The randomization of the vouchers enables us to
identify the intention to treat. While we would like to identify the effect
of using a voucher, we do not have a way of controlling for selection
into using the voucher. While we can identify the people who were
offered but declined the voucher, we cannot identify the individuals
who were not offered the voucher and would have declined had they
been offered it.
Some have suggested that voucher experiments might facilitate iden-
tification of the effect of private schools (e.g., Rouse 1998). To be a good
instrument for private schooling, the voucher lottery should be corre-
lated with the likelihood that a student attends private school but
uncorrelated with student outcomes except through its influence on
private schooling. It is the latter restriction that is likely not satisfied in
the Colombian voucher experiment. In Colombia, there were a number
of reasons why the voucher could have directly influenced test scores
besides through private schooling. For example, the voucher could
have been an income shock. As I will show, most of the voucher appli-
cants attended private school in the year immediately after the lottery.
The voucher could have just been a subsidy to families already com-
mitted to attending private school. Additionally, the voucher program

may have changed student incentives. If a student failed a grade, they
lost the voucher. This may have influenced voucher winners to try
harder in school than they otherwise would have. I discuss possible
mechanisms in section 7.5.
For most of the outcomes of interest, we can measure the effect using
equation 7.1. However, when we look at test scores of voucher appli-
cants on the ICFES exam, estimates based on equation 7.1 are likely
biased. One of the outcomes we evaluate is whether or not students
take the ICFES exam. Because the voucher may influence who takes
the exams, it likely biases any comparisons of the average test scores
of test-takers. This is because we observe students who take the exam
because of the voucher while we do not observe students who do not
take the exam but would have had they received a voucher. We dis-
cuss some ways of dealing with this selection in section 7.4.
7.3 Effects after Three Years
Table 7.3 shows estimates of equation 7.1 for a variety of educational

outcomes. The first column shows the average outcome for students
who lost the voucher lottery. As we mentioned earlier, almost 90% of
students who applied unsuccessfully for the voucher still attended pri-
vate school in sixth grade. Voucher winners were about 6 percentage
points more likely to attend sixth grade in private school. By seventh
grade the proportion of voucher lottery losers in private school drops
to about 67% and voucher winners are about 17 percentage points
more likely to be in private school. This difference in private school
attendance rates persists up to the time of our survey. Since most of
the students who applied to the program attended private school im-
mediately after the voucher application, the results show the effects
of vouchers on students who already had strong interest in private
schooling.
When we examine the highest grade completed, we find that
voucher winners have completed about 0.1 years of schooling more
than voucher losers within three years of the voucher lottery. This dif-
ference does not come from differences in dropout rates. Voucher lot-
tery winners and losers are equally likely to be enrolled in school at
the time of our survey. The difference arises from grade repetitions.
About 22% of voucher lottery losers had ever repeated a grade and
154 Eric Bettinger
Table 7.3
Descriptive Statistics and Estimates of the Voucher Effect
Bogotá 1995
Losers’ Coefficient on voucher status

means Basic controls
Dependent variable (1) (2)
Started 6th grade in private 0.877 0.057
[0.328] [0.017]
Started 7th grade in private 0.673 0.168
[0.470] [0.025]
Currently in private 0.539 0.153
[0.499] [0.027]
Highest grade completed 7.5 0.130
[0.960] [0.051]
Currently in school 0.831 0.007
[0.375] [0.020]
Finished 6th grade 0.943 0.023
[0.232] [0.012]
[0.360] [0.019]
[0.483] [0.027]
Ever repeated a grade 0.224 0.055
[0.417] [0.023]
Number of repetitions of 6th grade 0.194 0.059
[0.454] [0.024]
Math scores [n ¼ 282] 0.178 0.153
[0.120] [0.114]
Reading scores [n ¼ 283] 0.204 0.203
[0.115] [0.114]
Writing scores [n ¼ 283] 0.126 0.128
[0.116] [0.105]
Total test scores [n ¼ 282] 0.217 0.205
[0.116] [0.108]
Applicant is working 0.1690 0.0297
[0.3751] [0.0205]
Married or living with companion 0.0160 0.0087
[0.1256] [0.0059]
N 562 1,147
Notes: The table reports voucher losers’ means and the estimated effect of winning a
voucher. Numbers in brackets are standard deviations in the column of means and
standard errors in columns of estimated voucher effects. The regression estimates are
from models that include controls for phone access, age, type of survey and instrument,
strata of residence, and month of interview.
voucher lottery winners were about 5.5 percentage points less likely to
have repeated a grade.
This difference in repetitions is also manifested in looking at the like-
lihood of completion of the sixth, seventh, and eighth grades respec-
tively. About 94% of lottery losers had finished sixth grade but only
85% and 63% had finished seventh and eighth grades respectively. The
difference between voucher winners also increases over time so that by
eighth grade, voucher winners are about 10 percentage points more
likely to have finished the grade.
While grade repetition is often used a measure of the quality of edu-
cation in developing countries (e.g., Psacharopolous, Tan, and Jimenez
1986), grade repetition may not fully signal academic achievement in
the PACES setting. As part of the PACES program, students’ vouchers
were only renewed if students passed their grade. One explanation for
lower repetition rates is that schools may have had an incentive to pro-
mote voucher students in an effort to keep tuition monies flowing to
their schools.
To test whether the grade repetition result reflected higher academic
achievement, we administered a standardized exam to a sample of
ICFES applicants. On average, lottery losers scored about 0.2 standard
deviations above the population mean in both math and writing while
voucher winners scored about 0.2 standard deviations higher in writ-
ing. While voucher winners score higher in math and reading, the re-
sults are not statistically significant unless we combine the various test
scores. Although the sample we tested was fairly small, the fact that
voucher lottery winners scored higher than losers suggests that the
voucher had impacted student achievement within three years.
Finally, our results in table 7.3 show that voucher students were less
likely to be working at the time of survey. They were also less likely to
be married or cohabiting.
7.4 College Entrance Exams
After three years, we found that students had more years of school
completed, less repetitions, and higher standardized exam scores, yet
it was unclear if these effects could turn into long-term effects. For
example, by the third year after the lottery, more than half of the stu-
dents were no longer using the voucher. Additionally, the group of
students who took the exam we administered was small and may not
have been fully representative of the population of lottery winners and
156 Eric Bettinger
Table 7.4
Voucher Status and the Probability of ICFES Match
Dependent Coefficient on voucher status
variable mean Basic controls
Matching strategy (1) (2)
Exact ID match .354 .059
[.015]
ID and city match .339 .056
[.014]
ID and seven-letter name match .331 .059
[.014]
ID, city, and seven-letter match .318 .056
[.014]
Notes: Robust standard errors are shown in brackets. The sample includes all applicants
to the Bogotá 95 voucher lottery with valid ID numbers and valid age data (i.e., ages 9 to
25 at application).
losers since only 60% of the students we invited to the exam actually
attended.
To test whether the voucher led to long-term educational differences
between voucher winners and losers, we gathered additional data on
PACES applicants’ ICFES exams. From the ICFES records, we know
whether a student took the ICFES exam and their test scores in math
and language if they did take the test.
Table 7.4 shows estimates of the effects of vouchers on the likeli-
hood that students take the ICFES exam. We report the coefficient on
voucher from equation 7.1 when we include covariates for gender and
age. We report estimates based on four different matching strategies. In
the first strategy, we matched students’ national identification num-
bers alone. On average we were able to match 35.4% of voucher appli-
cants to their college entrance exam. Voucher winners, however, were
5.9 percentage points more likely to be matched than voucher losers.
We also report estimates based on matching both the identification
number and the city of residence. Our match rate drops about 1.5 per-
centage points, but we still estimate a 5.6 percentage point effect of the
voucher. In relative terms, this is about a 20% effect of the voucher on
the likelihood of taking the college entrance exam. In the final row of
table 7.4, we report estimates based on the identification number, the
city of residence, and the first seven letters of a student’s last name.
Our match rate drops to 31.8% with the more stringent match criteria.
We still find a 5.6 percentage point difference between voucher win-
ners and losers. Clearly, the voucher improved recipients’ likelihood of

taking the exam. Given that taking the exam is a good proxy for high
school graduation, receiving the voucher dramatically improves the
likelihood that students finish secondary school.
Having established the voucher’s effect on test-taking, we now turn
to the effect of the voucher on test performance. As mentioned earlier,
measuring the effect on test performance is difficult since the vouchers
affect who takes the exam. Consider the following example: Suppose
that Juan is a student on the margin of dropping out of secondary
school. If Juan received a voucher, it might have been enough to help
him persist in secondary school and take the college entrance exam.
If Juan is on the margin of dropping out, he is likely not the top-
achieving student in his grade. Now suppose that the voucher has no
effect on a student’s ICFES exam scores. The average ICFES score of
voucher winners is a weighted average of exam scores of students
who would have taken the test in the absence of the voucher, and also
students like Juan who would not have taken the ICFES exam without
the voucher. Because Juan was a low-achieving student, his score is
likely less than the average exam scores of students who would have
taken the ICFES even without a voucher, and hence, the average ICFES
score of lottery winners will be less than the average ICFES score of lot-
tery losers (which is just the average of exam scores of students who
would have taken the ICFES exam without the voucher). If Juan’s story
is typical, then comparisons of the average test scores of winners and
losers will be biased downward.
Rows 1 and 5 of table 7.5 make this comparison. When we compare
the average test scores of voucher winners and losers who took the
exam, we find that voucher winners score 0.70 percentage points
higher in language and 0.40 percentage points higher in math. The esti-
mated effect on language scores is statistically significant. If indeed the
comparison of ICFES test scores is biased downward, then the esti-
mated effects are smaller than the true effects of the voucher. In Angrist
et al. (2006), we show that assuming that the voucher does not harm
students (which seems reasonable given that students could quit using
the voucher without penalty), then the estimated effects in rows 1 and
5 of table 7.5 are lower bounds for the true treatment effect.
In table 7.5, we also employ two other strategies to estimate the ef-
fect of the voucher. In the remaining rows, we censor the sample by
assigning values to students who did not take the exam. The motiva-
tion for this specification is simple. Suppose that students’ latent (or
158 Eric Bettinger
Table 7.5
OLS and Tobit Estimates of Voucher Effect on ICFES Exams
Dependent Coefficient on voucher status
variable mean Basic controls
Specification (1) (2)
Language scores
OLS with score > 0 47.4 .70
[5.6] [.33]
OLS censored at 1% 37.3 1.14
[8.0] [.24]
Tobit censored at 1% 37.3 3.29
[8.0] [.70]
[4.7] [.46]
Math scores
OLS with score > 0 42.5 .40
[4.9] [.29]
OLS censored at 1% 35.7 .79
[5.8] [.18]
[5.8] [.51]
[4.6] [.45]
from models that include controls for age and gender. Censoring point is the indicated
percentile of the test score distribution, conditional on taking the exam.
expected) ICFES exam is related to the probability of taking the exam.

Students who expect to score low will likely not take the exam. If non-
test-takers have low exam scores, we may be able to estimate the effect
of the program by assigning them a test score. In rows 2, 3, 6, and 7, we
assign non-test-takers the test score of the first percentile of test-takers.
In rows 4 and 8, we assign non-test-takers the test score of the tenth
percentile. With the censoring, we use both OLS and Tobit models. In
the Tobit models, we make the additional assumption that the underly-
ing distribution of test scores is normally distributed.
When we censor, the estimated effects of the voucher on test scores is
much larger than the raw comparisons of means. In the censored OLS
models, we find estimated effects of 1.14 in language and 0.79 in math.
Both estimates are statistically significant. In the Tobit models, we find
estimated effects of about 2 percentage points in both language and

math. The estimates are also statistically significant.
It is not surprising that the censoring leads to larger and significant
effects of the voucher. We are essentially giving low test scores to
non-test-takers who are disproportionately voucher lottery losers. One
might wonder how sensitive the results are to the censoring point. In
figure 7.2, we show the estimated effects using Tobit when we move
the censoring to different percentiles of the test score distribution.
Consistently, regardless of the censoring point, we find effects of the
voucher near 2 percentage points. The standard error bands show that
these estimates remain significant.
One of the surprising results in figure 7.2 is the fact that even at high
censoring points (e.g., the 80th percentile) we still find that the voucher
led to improvements in students’ test scores. This may imply that
even among high-achieving voucher applicants, voucher winners’ test
scores have improved. In Angrist, Bettinger, and Kremer (2006), we
explored this in greater detail to test this hypothesis. Using nonpara-
metric strategies, we demonstrated that even at the top of the distribu-
tion of test scores voucher lottery winners scored higher than losers.
7.5 Mechanisms for the Voucher Effect
The results thus far suggest that voucher winners had higher academic
achievement after three years and through the end of secondary
school. The result reflects the effect of winning the voucher and not the
effect of using a voucher. As already discussed, the voucher could have
improved student outcomes for a variety of reasons. It could have been
an income shock. The voucher could have strengthened the incentives
for students to work hard. The voucher could have also changed the
type of schools and peers that students had.
To shed light on some of the possible mechanisms, Bettinger,
Kremer, and Saavedra (2006) considered the effects of the voucher on
students who applied to vocational schools. As part of the PACES lot-
tery, students had to be accepted at a participating private school before
applying for the voucher. Many students applied to vocational schools.
These vocational schools were of inferior quality to the other private
schools. They had a smaller proportion of students taking the ICFES
exam and their students typically scored worse on the ICFES exam
than students in non-vocational schools.
160 Eric Bettinger
Figure 7.2
Tobit Coefficients by Censoring Percentile in Score Distribution. The figure plots Tobit
estimates of the effect of vouchers on test scores, using data censored at the point indi-
cated on the x-axis (i.e., values below the indicated percentile are assigned a value of
zero). For the purposes of this exercise, non-test-takers are also coded as having a score
of zero.
(a) Language
(b) Math
Table 7.6
OLS Estimates of Voucher Effect on ICFES Exams for Vocational Students
Coefficient on voucher status
Vocational Nonvocational
(1) (2) (3) (4)
Losers’ Basic Losers’ Basic
Dependent variable means controls means controls
A. Probability of taking ICFES
ID match .274 .056 .318 .057
[.030] [.466] [.017]
ID and city match .265 .052 .301 .061
[.030] [.459] [.017]
ID and name match .202 .048 .235 .033
[.028] [.424] [.016]
N 336 802 1077 2578
B. Performance outcomes on the ICFES
Math score conditional on taking 41.47 .844 42.37 .420
[4.811] [.621] [4.655] [.332]
Reading score conditional on taking 45.73 2.19 47.04 .487
[5.890] [.768] [5.373] [.373]
N 89 254 334 891
from models that include controls for phone access, age, type of survey and instrument,
strata of residence, and month of interview. Columns 1 and 2 report estimates for
students who had applied to and been accepted at a vocational school prior to the
voucher lottery.
However, among the students who applied to vocational schools,

the voucher seemed to have odd effects on the types of schools that
students attended. Voucher winners used their voucher to attend voca-
tional schools. Voucher losers, by contrast, changed schools and went
to academic schools instead. After three years, voucher winners were
18 percentage points more likely to be attending a vocational school,
and as a result, they were more likely to attend schools of inferior qual-
ity as measured by academic performance on the ICFES.
Despite this fact, voucher winners who had initially applied to vo-
cational schools had better academic outcomes than voucher losers
who had applied to the same schools. Table 7.6 shows these results.
Voucher winners were 4–5 percentage points more likely to take the
ICFES exam and had higher reading scores on the exam (which is
162 Eric Bettinger
likely a lower bound for the true effect on reading scores as discussed
in section 7.4).
The fact that voucher winners attended inferior schools and yet had
more positive outcomes suggests that the schools and peers may not
have been the operative channel by which the voucher affected student
outcomes. Bettinger, Kremer, and Saavedra (2006) show that peer ef-
fects alone cannot explain the voucher effects.
7.6 Conclusions
The Colombian voucher program was one of the largest voucher pro-
grams in the world, and the program seems to have had a positive ef-
fect on student outcomes after three years and through the end of
secondary school. After three years, students winning the voucher had
higher test scores, less grade repetition, and more years of schooling
completed than students who had lost the voucher lottery. Addition-
ally, voucher winners were more likely to attend private school, less
likely to be working, and less likely to be married or cohabiting. By
the end of high school, voucher winners were more likely than voucher
lottery losers to have taken the college entrance exam and had higher
college entrance exam scores.
The voucher program was a unique partnership between the public
and private sectors in Colombia. Thousands of students in Bogotá and
about 125,000 students nationwide took advantage of the program.
Angrist et al. (2002) shows that public expenditure increased only
slightly in funding the program, yet the benefits accrued to voucher
winners more than justified the costs.
One remaining puzzle in the Colombian voucher experience is how
the voucher program affected the students. If the Colombian program
affected students solely through private schools, then it may have dif-
ferent policy implications than a voucher program that affected stu-
dents by changing their incentives. The preliminary evidence, at least
for the subset of voucher winners who applied to vocational schools,
suggests that the academic quality of the schools may not have been
the mechanism by which the voucher affected students’ outcomes. The
voucher winners who had applied to vocational schools attended
schools with inferior peer quality yet had better academic outcomes
than voucher lottery losers. Future research hopefully will identify the
specific channel(s) by which vouchers affect students and hence pro-
vide a clearer picture of why the private-public partnership in the case
of Colombian vouchers generated such dramatic improvements in stu-

dents’ academic performance.
Notes
This paper is largely based on ‘‘Vouchers for Private Schooling in Colombia: Evidence
From a Randomized Natural Experiment’’ by Joshua D. Angrist, Eric Bettinger, Erik
Bloom, Elizabeth King, and Michael Kremer (American Economic Review, 2002) and
‘‘Long-Term Educational Consequences of Secondary School Vouchers: Evidence from
Administrative Records in Colombia’’ by Joshua D. Angrist, Eric Bettinger, and Michael
Kremer (American Economic Review, 2006). Figures 7.1 and 7.2 are reprinted with permis-
sion from the American Economic Review.
1. In Angrist et al. (2002), we also interviewed cohorts of students who applied for a
voucher in Bogotá in 1997 and in Jamundı́ in 1993. I only focus on the Bogotá 1995 cohort
in this paper.
References
Angrist, Joshua D. 1997. Conditional independence in sample selection models. Economics

Letters 54, no. 2 (February): 103–112.
Angrist, Joshua D., Eric Bettinger, Erik Bloom, Elizabeth King, and Michael Kremer. 2002.
Vouchers for private schooling in Colombia: Evidence from a randomized natural exper-
iment. American Economic Review 92, no. 5 (December): 1535–1558.
Angrist, Joshua D., Eric Bettinger, and Michael Kremer. 2006. Long-term educational con-
sequences of secondary school vouchers: Evidence from administrative records in Co-
lombia. American Economic Review (May).
Angrist, Joshua D., and Alan Krueger. 1999. Empirical strategies in labor economics. In
Handbook of Labor Economics, vol. III, ed. Orley Ashenfelter and David Card, 1277–1366.
Amsterdam: Elsevier.
Bettinger, Eric, Michael Kremer, and Juan Saavedra. 2006. Are educational vouchers
Redistributive? Evidence from Colombia. Mimeo, Case Western Reserve University,
Cleveland, Ohio.
Calderon, Alberto. 1996. Voucher programs for secondary schools: The Colombian expe-
rience. Human Capital Development and Operations Policy Working Paper. Washington,
D.C.: World Bank. Available at hhttp://www.worldbank.org/education/economicsed/
finance/demand/related/wp_00066.htmli.
King, Elizabeth, Peter Orazem, and Darin Wohlgemuth. 1998. Central mandates and
local incentives: The Colombia education voucher program. Working Paper No. 6, Series
on Impact Evaluation of Education Reforms, Development Economics Research Group,
the World Bank, February.
King, Elizabeth, Laura Rawlings, Marybell Gutierrez, Carlos Pardo, and Carlos Torres.
1997. Colombia’s targeted education voucher program: Features, coverage and participa-
tion. Working Paper No. 3, Series on Impact Evaluation of Education Reforms, Develop-
ment Economics Research Group, the World Bank, September.
164 Eric Bettinger
Mayer, Daniel, Paul Peterson, David Myers, Christina C. Tuttle, and William G. Howell.
2002. School choice in New York after three years: An evaluation of the School Choice
Scholarships Program. Program on Education Policy and Governance Occasional Paper
Series, Harvard University.
Morales-Cobo, Patricia. 1993. Demand subsidies: A case study of the PACES voucher
program. Working paper, Universidad de los Andes, Economics Department, Bogotá.
Psacharopolous, George, J. Tan, and E. Jimenez. 1986. Financing Education in Developing
Countries: An Exploration of Policy Options. Washington, D.C.: World Bank.
Rouse, Cecilia Elena. 1998. Private school vouchers and student achievement: An evalua-
tion of the Milwaukee Parental Choice Program. Quarterly Journal of Economics 13, no. 2:
553–602.
Sanchez, Fabio, and Jairo Mendez. 1995. Por que los niños pobres no van a la escuela?
(Determinantes de la asistencia escolar en Colombia). Mimeo, Departmento Nacional de
Planeacion Republica de Colombia.
World Bank. 1993. Staff appraisal report: Colombia, secondary education project. Latin
America and the Caribbean Region, Report 11834-CO.
8 The Public-Private School Controversy in Chile
Cristian Bellei
Marked-oriented strategies have increasingly been proposed as an ef-

fective and efficient way to increase both quality and equity in educa-
tion. Academic and political debates have attempted to predict the
most probable consequences that market incentives might have on
educational systems. A key issue in those analyses has been the com-
parative study of the public and private schools’ effectiveness, in terms
of students’ academic achievement. In this chapter, I critically review
the research about whether Chilean students attending private schools
obtain larger learning outcomes than their peers studying at public
schools.
Chile constitutes a paradigmatic case in the debate about the relative
efficacy of private and public schools, and research on its experience
might shed light on such a controversy. Its nationwide school choice
system finances both public and private subsidized schools under the
same funding system, a type of voucher program. Considering the
small scale of most U.S. voucher and school choice programs, Chile is
a particularly interesting case to study. Strikingly, previous research
on the Chilean case has obtained very contrasting findings.
The chapter begins with (section 8.1) a brief description of the Chil-
ean education; then, section 8.2 reviews the research on systemic effects
of school choice and section 8.3 on private versus public schools’ effec-
tiveness. Section 8.4 analyzes some key methodological issues that ac-
count for the contrasting findings of previous research; section 8.5
describes the data I used, and sections 8.6, 8.7, and 8.8 provide empiri-
cal evidence about the consequences of the identified methodological
limitations. A final section summarizes the main conclusions of the
analysis, elaborates some interpretative hypothesis, and states some
educational policy implications.
166 Cristian Bellei
8.1 School Choice and Market-oriented Institutions in Chilean

Public Education
For more than two decades, Chilean education has operated under an
institutional design in which fundamental decisions do not rely on na-
tional authorities, but on the combination of family preferences (school
choice) and (public and private) school competition for attracting such
preferences.1 This system—which is an attempt to make education
a self-regulated market—was created during the 1980s as a part of a
large economic and institutional neoliberal reform.
The core of the reform was the inception of a common public fund-
ing system for all public and private subsidized schools: the educa-
tional voucher, which is a monthly payment to the schools, of a fixed
fee per each student who is enrolled in and regularly attends the
school. Additionally, the reform included several changes oriented to
promote competition among public and private schools. Families do
not have restrictions on choice of a school (whether public or private,
near to or distant from home), and schools must compete to attract
families’ preferences. The administration of public schools was decen-
tralized at the municipal level, and the management of schools (includ-
ing the curriculum and human and financial resources) was strongly
deregulated. Finally, reformers created a national evaluation system
of students’ learning (SIMCE, Spanish acronym for System of Mea-
surement of Educational Quality), aimed at informing families about
school quality.
Since 1990, the Chilean government has attempted to combine the
market institutions with stronger state regulations and interven-
tions. Thus, while some educational policies have been designed to
strengthen the market-oriented model (e.g., creation of a ‘‘price dis-
crimination’’ system among subsidized private schools, and publica-
tion of SIMCE results in the newspapers), others have tried to restrict
it, through the active promotion of social equity and educational qual-
ity (e.g., implementation of compensatory programs, creation of a
teacher labor statute, and the universal provision of computers, text-
books, and teacher training).
There are three kinds of schools in Chile: public schools, private
voucher schools, and private nonsubsidized schools. All primary and
most secondary public schools are free; about 90% of private voucher
schools are not free (they have a copay system), and private non-
subsidized schools are elite schools, totally paid for by families.2 This
The Public-Private School Controversy in Chile 167
chapter focuses on the comparison between public and private voucher

schools (although occasional references will be made to the private
nonsubsidized schools). The evolution of enrollment has shown a
highly responsive system. The proportion of Chilean students attend-
ing a public school has decreased systematically, from 78% in 1981, to
58% in 1991, and 47% in 2006; simultaneously, students attending pri-
vate voucher schools have increased from 12% in 1981, to 32% in 1991,
and 45% in 2006. The enrollment in nonsubsidized private schools has
remained a minor part of the Chilean school population.
Finally, the testing system has pointed out a systematic pattern: pri-
vate voucher students’ score—on average—higher than public school
students by about 0.3 to 0.4 S.D. Whether or not this observed gap rep-
resents a genuinely greater private school effectiveness has been a con-
troversial academic and political question in the Chilean educational
debate.
8.2 Systemic Effects of School Choice
There are two competing theories about systemic effects (i.e., the impact
of the voucher mechanism on the effectiveness of the educational sys-
tem as a whole) of the market mechanisms of Chilean education. One
states that subsidized private schools help to improve public schools
through a competition effect, predicting an overall improvement of
Chilean education. The other theory proposes that the potential pro-
ductivity effect may be canceled out by unexpected negative effects of
sorting on public schools (private schools ‘‘skim’’ the best public stu-
dents), with no systemic improvement. Unfortunately, there is very
little research on this issue.
Gauri (1998) studied the school choice process using a sample of
households from the Santiago metropolitan area. He found that higher
socioeconomic status was positively associated with the probability
of attending a school in the top third of the students’ achievement
distribution. Gauri also found that the probability of studying in a
high-performing school significantly increased when the student was
required to take a cognitive and/or academic test as an admission re-
quirement. Hseih and Urquiola (2003) attempted to evaluate whether
the introduction of school choice in Chile improved the productivity of
the school system at aggregated levels, and whether it increased the
educational and socioeconomic differences between private and public
schools. They analyzed fourth-grade mathematics and language test
168 Cristian Bellei
scores, repetition rates, and years of schooling among 10–15-year-olds,

between 1982 and 1988, at school and commune level. After controlling
for several socioeconomic school and commune factors, they found
that communes with a higher proportion of private enrollment tended
to have lower public school test scores, higher private/public test-score
and repetition-rate gaps, as well as higher student SES private/public
differences at the commune level. Finally, at the commune level, nei-
ther the level of 1990 private enrollment nor the 1982–90 increase in
private enrollment were associated with improvements in students’
outcomes. Gauri and Hseih and Urquiola interpreted those findings as
evidence of both negative effects of private school expansion on public
schools, and no positive effect on the quality of the system as a whole.
Nevertheless, since these studies are mainly based on cross-sectional
data and the information prior to the voucher reform is very limited, it
is difficult to make causal claims about the hypothesized relationships.
Gallego (2002) defined each commune as a different school market.
He used fourth- (1994, 1996) and eighth-grade (1995, 1997) language
and mathematics school means as the outcome variables, and con-
trolled for schools’ SES composition and other commune variables
(level of urbanization, size). Gallego found that the proportion of pri-
vate enrollment at the commune level was positively associated with
school performance, and that this competition effect was stronger for
private schools. The key limitation of this approach is that the level of
private enrollment is not an exogenous variable to students’ perfor-
mance, because private schools tend to serve geographical areas with
characteristics positively associated with students’ achievement. Gal-
lego (2004) attempted to overcome this limitation: he used ‘‘priest per
capita’’ as an instrumental variable to identify exogenous variation in
private subsidized enrollment at the commune level. He estimated
that positive differences in private enrollment were associated with
higher student test scores, and higher educational inequity. Neverthe-
less, priest per capita might not be a valid instrument for private
schools in Chile, because priests are actively involved in the expansion
of private schools and they might do that selectively, based on the
expected educational outcomes of a given educational market. Addi-
tionally, Catholic schools account for only a third of the total private
subsidized enrollment, and Catholic schools were precisely those pri-
vate schools that existed in Chile prior to the introduction of market-
oriented reforms.
Finally, Auguste and Valenzuela (2004) also estimated the effect of

competition at the commune level; they used SIMCE-2000 eighth-grade
test scores and implemented an instrumental variable approach (they
used market size and cost of entry as instrumental variables). They
found a positive but small effect of competition on test scores; they also
concluded that higher levels of school competition were associated
with higher school segregation, which harmed public schools.
Overall, the available evidence is not sufficient to evaluate the
theories about the systemic effects of school choice in Chile; neverthe-
less, the evidence strongly suggests that the size of that potential effect
has been extremely small at best (in fact, not noticeable at the national
level). Additionally, the evidence suggests that school choice and com-
petition are linked to an increase in educational inequity. Future re-
search in this area should combine longitudinal studies, analyses of
institutional and educational policy contexts, and a deeper understand-
ing of parental choice and school selection processes.
8.3 Research on Private versus Public Schools’ Effectiveness
A common argument to support market-oriented reforms (including

public funding of private schools) is that private schools are more ef-
fective than public schools because they are more innovative, more
sensitive to the demand, and less bureaucratized. In fact, most studies
focusing on the case of Chile in this area are comparisons of public
and private school effectiveness.
In general terms, this research has evolved following three stages.
Rodriguez (1988), Aedo and Larrañaga (1994), and Aedo (1997) are
part of the first phase: they studied unrepresentative samples, ana-
lyzed exclusively school-level data (from the 1980s or early 1990s), and
focused on urban primary schools. Their studies concluded that—after
controlling for school characteristics—private schools scored signifi-
cantly higher than public schools, on average. Because of their lack of
representativeness, it is not possible to generalize those findings for
the overall Chilean school population.
Bravo, Contreras, and Sanhueza (1999), Mizala and Romaguera
(1999), Carnoy and McEwan (2000), Vegas (2002), and Sapelli (2003)
also analyzed exclusively school-level data, but they studied large,
nationally representative samples.3 This research is also focused on
primary education (mainly fourth grade), and all of them applied
170 Cristian Bellei
ordinary least squares estimates. These five studies constitute the sec-
ond phase of this kind of research.
Finally, McEwan (2001), Mizala and Romaguera (2003), Sapelli and
Vial (2002), Gallego (2002), and Mizala, Romaguera, and Ostoic (2004)
constitute a third, more sophisticated stage. These five studies ana-
lyzed student-level test scores as the outcome variable, and also in-
cluded student-level predictors. They used the nation-level database,
and included both primary and secondary education. They incorpo-
rated more sophisticated research methods, including hierarchical lin-
ear models, probabilistic models of choice, and instrumental variables.
Although some studies analyzed more than one year of students’
test scores, none of them is a longitudinal analysis but rather a series
of cross-sectional estimates. All studies have analyzed mathematics
and/or language test scores as the outcome variable. Most studies
compared public schools with two categories of private schools—
voucher and nonsubsidized schools—although some of them distin-
guished between Catholic and nonreligious voucher schools.
Surprisingly, there is a noticeable variation in the estimates of the
private voucher versus public schools test-score gap: while some
authors have found private voucher school advantage (0.05 S.D. to
0.27 S.D.), others have found public school advantage (0.06 S.D. to 0.26
S.D.), and some have found no statistically significant difference be-
tween them.4 An additional puzzle is that several studies differ in their
findings even when they analyze the same database. In order to under-
stand those discrepancies, the following sections deepen into method-
ological issues of the mentioned second- and third-phase studies.
8.4 Methodological Issues in the Research Comparing Private and

Public Schools
Selection bias is the most serious limitation that affects the estimates of
the relative efficacy of private and public schools in Chile. As men-
tioned, the supply of private schools is neither randomly distributed
among geographical areas nor among social classes, and finding a
good instrument for the supply of private schools has proved to be ex-
tremely difficult. Additionally, the unregulated school choice and ad-
mission processes are highly complex and there is little information
about them.
Based on their preferences and their capacity to pay tuition, parents
can select any school. However, schools may also select their students.
According to the nationwide SIMCE-2003 survey of tenth graders’

parents, 85% of the private nonsubsidized schools, 73% of the private
voucher schools, and 59% of the public schools respondents stated that
their child was selected by the school through an admission process
that included some kind of examination or minimum academic re-
quirement. Gauri (1998) found that, in Santiago, 82% of private non-
subsidized, 37% of private voucher, and 18% of public students had
been compelled to take a test to be admitted to their schools. Those
tests—focused on basic language, reasoning, psychomotor, and social
skills—are applied even to preschool applicants. Finally, student selec-
tion is a continuing process, which operates at any time during the stu-
dents’ career. In fact, many schools expel students who have a low
academic achievement or behavioral problems; almost all expelled stu-
dents are subsequently enrolled in a public school. In these cases,
student selection is based on demonstrated performance.
Selection bias is a crucial problem because student characteristics re-
lated to student performance are also relevant predictors of the type of
schools students attend. Thus, cognitive skills, motivation, and disci-
pline are some relevant unobserved student characteristics affecting
the estimates of private and public school effectiveness. Unfortunately,
there is no information about students’ initial characteristics or previ-
ous test scores. Nevertheless, the contrasting findings about the rela-
tive efficacy of public and private schools in Chile are explained not
only by data limitations, but also by methodological divergences
among researchers about how to tackle the confounding effect of selec-
tion bias. Moreover, the literature has increasingly recognized the
potential role of peer effects on student achievement. In highly segre-
gated environments (like the Chilean educational system), peer effect
might play an even more influential role in students’ performance.
Studies also differ noticeably in the quantity and quality of the co-
variates. Additionally, researchers measure the same phenomenon in
very different ways. Finally, structural or economic variables (e.g., fam-
ily income) have frequently been included in the analyses, but cultural
or social variables (books at home, peer effects, etc.) have been almost
absent.
The appropriate level of data aggregation has been a source of diver-
gence among authors too: while some apply commune-level analysis,
others prefer school-, classroom-, and student-level analyses. This
issue is also linked to the covariates: some authors assume that con-
trolling for student-level variables suffices, but others argue that school
172 Cristian Bellei
compositional effects are relevant as well, so that school-level controls

should be simultaneously included. Notably, little attention has been
given to the multilevel nature of the educational data. Finally, potential
heterogeneous effectiveness between private and public schools,
according to different contextual (e.g., geographical location), educa-
tional (e.g., grade level), or student characteristics (e.g., initial ability)
are almost absent.
8.5 Data
In the remaining sections, by conducting various regression analyses I

will empirically demonstrate the sensitivity of the findings to the iden-
tified methodological issues.
With that purpose, I analyzed two datasets: SIMCE-2002 and
SIMCE-2003. These databases contain individual mathematics and lan-
guage test scores of 253,463 fourth-grade and 239,649 tenth-grade stu-
dents respectively, who are the 95% of the corresponding Chilean
student population. The findings in both subject matters were very
similar; therefore, I will focus on the mathematics results. The datasets
include 6,145 primary schools and 2,117 high schools, respectively.
Several student-level (based on a parents’ survey) and school-level
control variables were also included. Table 8.1 provides a description
of every variable used in the analyses.
8.6 The Confounding Role of Student Selection
8.6.1 Student Selection in the Admission Process

Comparisons between public and private schools in Chile are difficult
initially because public and private school students differ significantly
in almost all variables associated with social-class origin and, accord-
ingly, academic achievement. Table 8.2 shows that students in private
voucher schools have—on average—more educated parents, higher
family income, and more books at home. The key issue is that those
observed differences are almost certainly linked with unobserved dif-
ferences that also affect student achievement. For example, private
schools have a higher proportion of students selected through an ad-
mission process than public schools. Through those processes, schools
typically evaluate the academic potential of the applicant. In order to
account for those differences, researchers normally control for stu-
dents’ SES characteristics; nevertheless, if during the admission process
Table 8.1
Variable Definitions
Variable Definition
Student-level variables
Mathematics Standardized IRT test score
Language Standardized IRT test score
Mother’s education Years of education of the student’s mother
Father’s education Years of education of student’s father
Family income Natural LOG of student’s family income
Books Number of books at student’s home, scale ranging from 0
(0 books) to 5 (>200 books)
Gender Dummy variable for student’s gender (omitted category:
woman)
Repetition Dummy variable indicating whether the student has
repeated a grade
Selection Dummy variable indicating whether the student was
selected by the school through an admission process (e.g.
tests, grades requirements)
Parents’ expectation* Parents’ expectation about the future student’s educa-
tional attainment, scale ranging from 1 (4th grade) to 8
(graduate studies)
School-level variables
Type of school Dummy variables indicating whether the school is public
(omitted category), private voucher, or private non-
subsidized
Mean mothers’ education School average of years of education of students’ mothers
S.D. mothers’ education School standard deviation of years of education of
students’ mothers
Mean parents’ education School average of years of education of both students’
parents
Mean books at home School average of the individual variable ‘‘books at
home’’
Selected students Proportion of students who sere selected by the school
through an admission process
School SES level 5 dummy variables that classify schools in Low, Middle-
Low, Middle, Middle-High, High students’ socioeconomic
status (categories are based on parent’s education, family
income, and proportion of at-risk students in the school)
Quintile income Quintile classification of schools based on the school
average of family income
LOG school families’ Natural LOG of the school average of student’s family
income income
S.D. families’ income School standard deviation of the students’ family income
% repitent students in Percentage of students in the school who have repeated at
school* least a grade
174 Cristian Bellei
Table 8.1
(continued)
Variable Definition
School expels repitent School that (according to parents) expels students who
students* repeat a grade
Students always in this Proportion of students who have been in the same school
school* since 1st grade
School mean years in this School average of years that students have studied in this
school* school
Note: * Only available in SIMCE-2002, 4th grade.
Table 8.2
Comparing Public and Private Voucher School Students
Public Private voucher

N ¼ 109,624 N ¼ 96,585
Mathematics 230.1 (55.2) 250.3 (57.4)***
Language 241.5 (48.0) 257.4 (48.5)***
Mother’s education 9.3 (3.6) 10.8 (3.7)***
Father’s education 9.7 (3.8) 11.1 (3.9)***
Books at home 1.7 (1.2) 2.1 (1.2)***
Selected student 59.1% (47.4%) 71.9% (43.4%)***
Log family income (original scale: 1 to 13) 0.5 (0.6) 0.7 (0.7)***
Gender (male) 49.3% (49.9%) 49.8% (49.9%)
Notes: Student-level characteristics: Mean (S.D.).
@
p < :10; *p < :05; **p < :01; ***p < :001
Source: Author elaboration, based on SIMCE-2003.
schools have additional information about the academic potential of

applicant students, controlling for those covariates may be insufficient.
To explore this issue, I used the information presented in table 8.2 to
predict the probability that a student will be enrolled in a private
voucher school (versus in a public school). I applied logistic regression
to conduct this analysis. The results are presented in table 8.3. Accord-
ing to model 1, the fitted-odds ratio that a student who was admitted
through a selection process will attend a voucher school (versus a pub-
lic school) is 2.07 times the odds for a student who was not selected by
an admission process; model 2 shows that, after controlling for SES
characteristics, the mentioned fitted odds decreased to 1.78 times, but
it remained statistically significant. This finding indicates that, al-
though SES characteristics are associated with the probability of being
Table 8.3
Predicting Private Voucher School Attendance: The Effect of Student Selection through
an Admission Process
Dependent variable: Private voucher school
attendance (vs. public school attendance)
Model 1 Model 2
Selected student 0.73*** 0.58***
(0.01) (0.01)
Mother’s education 0.04***
(0.00)
Father’s education 0.02***
(0.00)
Family income 0.04***
(0.01)
Log family income 0.25***
(0.03)
Books at home 0.11***
(0.01)
Constant 0.47*** 1.28***
(0.01) (0.02)
Max-Rescaled R 2 0.04 0.09
N (students) 237,492 237,492
Notes: Logistic regression models for the relationship between whether a student attends
a private voucher school (vs. public school) as a function of whether he was selected by
the school, and some family characteristics.
@
p < :10; *p < :05; **p < :01; ***p < :001
Source: Author elaboration.
a selected student in a private school, they are not enough to account

for the unobserved differences between selected and nonselected stu-
dents. This strongly suggests that private schools use additional in-
formation (plausibly associated with students’ ability) to make their
admission decisions. In fact, I will later show that, even after control-
ling for several school, family, and student characteristics, selected stu-
dents score significantly higher than nonselected students.
8.6.2 Student Selection during the Schooling Process

None of the analyzed studies have considered the student selection
that affects some students during their schooling process. In this sec-
tion, by using the SIMCE-2002 database, I will analyze some recently
available information related to this issue.
176 Cristian Bellei
About 27% of Chilean fourth-graders in public schools and 33% in

private voucher schools did not start their primary education in their
current school. Additionally, while the proportion of students who
have repeated a grade is about 15% in public schools, in private
schools it is only 8%.5 Since some Chilean schools do not admit stu-
dents who have previously repeated a grade and others expel their
students who are repeating the grade (thus, this practice accounts for
some proportion of the students who have moved to a different school
since first grade), it would be misleading to infer about school quality
based on the current proportion of retained students. In fact, while
31% of private voucher school parents affirm that their school expels
retained students, only 14% of the public school parents assert so.
In order to explore whether the higher exclusion of retained students
among private schools explains part of their observed advantage, table
8.4 contains the parameter estimates of multiple regression models that
relate this information with student test scores. Baseline model 1 shows
that the raw difference in mathematics achievement between public
school and private voucher school fourth-grade students is about 0.34
S.D.6 Model 2 controls for some basic student and school characteris-
tics. As expected, students who have repeated a grade score signifi-
cantly lower than their nonretained peers; additionally, students who
attend a school with a higher proportion of retained students tend to
score significantly lower. Unfortunately, we cannot be confident about
the causal relationship between both measures of student grade reten-
tion and the private-public test-score gap, because they can be either
measures of students’ previous ability or of school quality. Finally,
schools that expel retained students score about 0.19 S.D. higher than
schools that do not expel them, which suggests that the selection of the
abler students and the rejection of the less-skilled students during
the schooling process account for some of the private voucher school
advantage, after controlling several student and school characteristics.
Models 3 and 4, and also 5 and 6 in table 8.4 replicate this analysis
for the subsample of students who have always studied in the same
school and for those students who have moved to a different school, re-
spectively. Although the general pattern is similar in both subsamples,
there are some interesting differences. The positive association between
attending a selective school that expels retained students and test
scores is stronger for students who have changed school than for
students who have remained in the same school (0.26 S.D. vs. 0.14 S.D.
respectively). Conversely, the negative association between the per-
Table 8.4
Identifying the Effect of Student Selection during the Schooling Process
Outcome variable: fourth-grade students’ mathematics test
scores, SIMCE-2002
Students who have Students who have
All students not changed schools changed schools
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6
Private voucher 18.83*** 0.31 20.55*** 0.28 16.34*** 1.75***
(0.24) (0.27) (0.29) (0.33) (0.45) (0.49)
Private nonsubsidized 61.11*** 7.33*** 62.47*** 8.22*** 55.99*** 4.51***
(0.45) (0.59) (0.52) (0.72) (0.92) (1.08)
Student repeated at 21.66*** 23.77*** 19.30***
least a grade (0.38) (0.52) (0.63)
% repeating students 0.33*** 0.26*** 0.40***
in school (0.02) (0.02) (0.03)
School expels 10.54*** 7.82*** 14.74***
repeating students (0.72) (0.86) (1.40)
% students always in 15.99*** 32.11*** 0.06
this school (1.40) (1.93) (2.43)
School mean years in 1.30*** 0.05 2.90***
this school (0.39) (0.48) (0.71)
Constant 238.2*** 188.2*** 239.9*** 180.7*** 234.8*** 192.6***
(0.16) (1.02) (0.19) (1.37) (0.32) (1.82)
Additional control no yes no yes no yes
variables
R2 0.10 0.25 0.11 0.26 0.07 0.25
N (students) 199,112 199,112 137,181 137,181 54,895 54,895
Notes: Relationship between school type and mathematics achievement. Additional con-
trol variables include gender, father’s and mother’s education, LOG family income,
books at home, parent’s expectation, and school SES level. Omitted category: public
schools.
@
p < :10; *p < :05; **p < :01; ***p < :001
centage of retained students in the school and test scores is stronger for
students who have moved to a different school than for students who
have remained in the same school. Both results point in the same direc-
tion: students who change school seem to be more sensitive to the
selective nature of their new school. Note that this can be the result of
schools selecting the best students and families choosing more selective
schools.
As shown in table 8.4, while there is not a statistically significant
difference between public and private school students within the
178 Cristian Bellei
population who have not changed school, among students who have
changed school, students in public schools score slightly higher than
their peers in voucher schools (0.03 S.D.). Consequently, part of the pri-
vate schools’ advantage can be based on their capacity to select and
attract more skilled students. Finally, although the gap between private-
nonsubsidized and public schools is reduced for both groups of stu-
dents, it remains statistically significant, suggesting that there is little
transfer of students between public and private nonsubsidized schools.
8.7 Alternative Ways of Controlling for Differences in Student and

School Characteristics
8.7.1 Controlling for Parents’ Education Level

Because public and private schools serve students with markedly dif-
ferent levels of parent education, all studies control for this aspect; nev-
ertheless, they diverge noticeably in the way this variable is introduced
into the analysis. In order to show how these discrepancies may affect
the estimates of the private/public test-score gap, table 8.5 shows six
regression models, all of them present in the reviewed literature.
Model 1 is a baseline model: private voucher schools score about
0.36 S.D. higher than public schools. Models 2 and 3 incorporate
student-level parents’ education covariates. As expected, controlling
for mother’s education—model 2—reduces the private/public test
score gap (to 0.27 S.D. and 1.16 S.D. respectively); but this gap remains
statistically significant. Also, when father’s education is added—model
3—the private school advantage is reduced slightly and remains statis-
tically significant.
Models 4, 5, and 6 also control for parents’ education, but measured
at school level. Model 4 estimates the private/public gap by control-
ling for the school average of mothers’ years of education. The results
indicate that students in private voucher schools earn lower test scores
than students in public schools; although statistically significant, the
difference is very small (0.02 S.D.). More recently, some researchers
have introduced the heterogeneity of the student population as a dif-
ferent control variable for parents’ education. Thus, model 5 controls
only for the school standard deviation of mother’s years of education:
interestingly, compared to model 1, this variable per se has almost no
effect on the estimate of the private/public school gap. Nevertheless,
model 6 shows that when the school mean of mother’s education is
also present in the model, the effect of the school standard deviation of
Table 8.5
Regression Models that Relate School Type and Students’ Mathematics Achievement,
Controlling for Parents’ Education Variables
Dependent variable: 10 th -grade students’ mathematics test
scores, SIMCE-2003
Private voucher 20.35*** 14.82*** 13.66*** 1.21*** 20.41*** 1.69***
(0.25) (0.24) (0.24) (0.24) (0.25) (0.24)
Private nonsubsidized 86.77*** 64.99*** 59.11*** 0.30 87.01*** 2.27***
(0.45) (0.47) (0.47) (0.55) (0.46) (0.57)
Mother’s education 3.82*** 2.58***
(0.03) (0.04)
Father’s education 2.05***
(0.04)
School mean mother’s 13.49*** 13.55***
education (0.06) (0.06)
School S.D. mother’s 0.67* 4.32***
education (0.33) (0.30)
Constant 230.07*** 194.56*** 186.32*** 107.25*** 227.92*** 120.50***
(0.17) (0.33) (0.36) (0.55) (1.06) (1.06)
R2 0.14 0.20 0.21 0.30 0.14 0.30
N (students) 237,492 237,492 237,492 237,492 237,492 237,492
Notes: Omitted category: public schools.

@
p < :10; *p < :05; **p < :01; ***p < :001
mother’s education changes its sign and increases its effect (i.e., at sim-
ilar levels of mother’s education, more homogeneous schools tend to
score higher than more heterogeneous schools). According to model 6,
public schools score significantly higher than private voucher schools
(0.03 S.D.) when both school-level variables are simultaneously in-
cluded, and this difference is larger than that estimated in model 4,
which only controls for the level, not the variation in mother’s educa-
tion (observe that a similar pattern is identified when comparing pub-
lic and nonsubsidized private schools).
8.7.2 Controlling for Schools’ Socioeconomic Status

There are several hypotheses about how the socioeconomic characteris-
tics of the student population at the aggregated level can affect student
academic achievement: the socioeconomic status of the student popu-
lation might affect teachers’ expectations and teaching practices; it
might also represent a measure of the available material and symbolic
180 Cristian Bellei
resources at school level; and, finally, it might be a measure of peer

effects (students might benefit from their peers’ family resources and
personal abilities through their interaction). Since Chilean public and
private schools differ in the socioeconomic status of their student pop-
ulations, researchers have controlled for this aspect in order to reduce
the bias of the estimates of the private/public schools gap. Neverthe-
less, there is huge disparity in the appropriate level of data aggrega-
tion, the specific covariates, and the type of measurement of school
socioeconomic status. I will illustrate how those divergences affect the
results by analyzing alternative regression models, all of them present
in the literature.
Table 8.6 shows the estimates of ten regression models containing
exclusively school-level measurements of students’ SES (models 2 to
11). These models combine six different school-level controls.7 Model 1
is the baseline.
Models 2 to 7 were estimated by including a single control variable
each time. The introduction of these control variables significantly
increases the capacity of the models to predict students’ test scores: the
proportion of the variation explained for the models ranges from 0.14
in the baseline model to 0.25–0.29 in models with a school SES control
variable added. Based on the R 2 statistics, the six controls have similar
effects on the regression models; nevertheless, their effects on the esti-
mated test-score gaps are markedly different.
Model 2 controls for schools’ SES by using an official classification,
which sorts schools in five SES groups (four dummies were incor-
porated; omitted category: Low SES). This classification has been
regularly used by researchers. The introduction of this covariate dra-
matically reduces the positive difference between private voucher
over public schools (to 0.04 S.D.), although it remains statistically
significant.
Models 3, 4, and 5 control for family income, but measured in three
different school-level indicators: income quintiles, the income natural
logarithm, and the income standard deviation. As shown, the size of
the estimated gap is noticeably different depending on the covariate:
private voucher schools’ advantage over public schools ranges from
0.01 S.D. in model 3 to 0.12 S.D. in model 5. Finally, model 8 shows
that when the level of family income is taken into account (i.e., log of
school mean of family income), the introduction of a variability mea-
sure (i.e., S.D. of income) has almost no impact on the private voucher
effect estimate (compare models 8 and 4).
Models 6 and 7 introduce control variables referred to as cultural

capital (as opposed to economic capital, included in previous models):
school mean of parents’ years of education, and school mean of books
at home. When parents’ education is incorporated as a covariate, pri-
vate voucher schools score lower than public schools, by 0.01 S.D.; in
turn, when books at home is the covariate, private voucher students
obtain lower test scores than public students (0.03 S.D.). Finally, when
both control variables are added simultaneously (model 9), private
voucher schools obtain statistically significant lower student achieve-
ment than public schools, and—compared to model 6—this difference
increases (0.04 S.D.).
Models 10 and 11 evaluate the impact of using simultaneously eco-
nomic and cultural capital covariates. The estimated negative differ-
ences in academic achievement between private voucher schools and
public schools in models 10 and 11 are almost the same as estimated
by model 9. This suggests that cultural differences between the student
populations are the key factors that explain the observed advantage
of private voucher over public schools (note that models 9, 10, and 11
also estimate public schools’ advantage over private nonsubsidized
schools).
In summary, the divergences in the way researchers have attempted
to control for parent education and school SES have had decisive ef-
fects not only on the size but also on the sign of the estimated test-score
gap between private and public schools in Chile.
8.8 Level of Data Aggregation and Data Analysis
8.8.1 Student-level versus School-level Covariates

Although since 1997 student-level data (both test scores and back-
ground information) is available in Chile, researchers still disagree on
what is the best level of aggregation of control variables: while some
use exclusively school-level controls, others use exclusively student-
level controls, and some use both. In table 8.7, I present six regression
models to demonstrate the consequences that this disagreement has on
the estimates of the test-score gap between private and public schools.
Model 1 is the baseline model. A set of student-level covariates was
added to model 2; all those control variables significantly predict stu-
dents’ test scores. Student-level control variables noticeably reduce the
gap between private and public schools, but this gap remains positive
and statistically significant (0.21 S.D. for private voucher schools).
Table 8.6
182
Regression Models that Relate School Type and Students’ Mathematics Achievement
Controlling for School SES Variables
Dependent variable: 10 th -grade students’ mathematics test scores, SIMCE-2003
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Model 9 Model 10 Model 11
Private voucher 20.42*** 2.30*** 0.53* 1.34*** 6.79*** 0.52* 1.59*** 1.54*** 2.09*** 2.58*** 1.93***
(0.25) (0.24) (0.25) (0.25) (0.24) (0.24) (0.25) (0.25) (0.24) (0.24) (0.24)
Private nonsubsidized 86.75*** 3.95*** 41.44*** 8.56*** 10.70*** 1.95*** 3.54*** 4.04*** 4.05*** 11.16*** 7.03***
(0.45) (0.79) (0.47) (0.56) (0.59) (0.56) (0.55) (0.58) (0.56) (0.79) (0.58)
Middle-low SES 10.46*** 14.61***
(0.30) (0.38)
Middle SES 49.95*** 9.29***
(0.34) (0.67)
Middle-high SES 82.04*** 4.79***
(0.53) (1.02)
High SES 105.91*** 6.33***
(0.92) (1.43)
Quintile school 22.33***
income (0.11)
Log school family 37.57*** 30.13*** 8.01***
Incom. (0.18) (0.33) (0.52)
S.D. school family 35.24*** 9.27*** 7.13***
Income (0.19) (0.34) (0.34)
Mean school parents’ 12.99*** 7.80*** 7.65*** 7.68***
education (0.06) (0.13) (0.17) (0.16)
Mean books at home 49.12*** 22.15*** 24.90*** 25.18***
(0.22) (0.51) (0.51) (0.55)
Cristian Bellei
Constant 230.1*** 215.1*** 186.1*** 227.9*** 184.1*** 109.8*** 148.5*** 216.2*** 121.0*** 127.1*** 108.3***
(0.17) (0.25) (0.16) (0.30) (0.56) (0.40) (0.46) (0.61) (1.07) (1.30)
R2 0.14 0.27 0.27 0.27 0.25 0.29 0.29 0.27 0.30 0.31 0.30
N (students) 237,427 237,427 237,427 237,427 237,427 237,427 237,427 237,427 237,427 237,427 237,427

@
p < :10; *p < :05; **p < :01; ***p < :001
The Public-Private School Controversy in Chile
183
184 Cristian Bellei
Table 8.7
Regression Models that Relate School Type and Students’ Mathematics Achievement,
Controlling for Students’ and School Variables
Dependent variable: 10 th -grade students’ mathematics test
scores, SIMCE-2003
Private voucher 20.35*** 11.89*** 10.06*** 2.40*** 3.15*** 3.51***
(0.25) (0.24) (0.24) (0.25) (0.24) (0.24)
Private nonsubsidized 86.77*** 49.85*** 47.64*** 5.45*** 1.29* 0.52
(0.45) (0.51) (0.50) (0.58) (0.58) (0.59)
Mother’s education 1.90*** 1.71*** 0.74***
(0.04) (0.04) (0.04)
Father’s education 1.34*** 1.24*** 0.35***
(0.04) (0.04) (0.04)
Log family income 4.81*** 4.86*** 1.20***
(0.20) (0.20) (0.19)
Books at home 6.71*** 6.29*** 3.65***
(0.10) (0.10) (0.10)
Gender (male) 7.66*** 7.77*** 9.11***
(0.22) (0.22) (0.20)
Selected student 18.84*** 3.96***
(0.24) (0.27)
School mean parents’ 7.91*** 7.17*** 5.97***
education (0.13) (0.13) (0.13)
School S.D. mother’s 3.47*** 0.88** 0.48
education (0.30) (0.30) (0.29)
School mean books at 21.79*** 16.17*** 13.93***
home (0.51) (0.51) (0.51)
% selected students in 33.44*** 29.52***
school (0.46) (0.51)
Constant 230.1*** 181.8*** 174.0*** 131.7*** 119.99 112.0***
(0.17) (0.37) (0.38) (1.08) (1.08) (1.08)
R2 0.14 0.23 0.25 0.30 0.32 0.33
N (students) 237,492 237,492 237,492 237,492 237,492 237,492
@
p < :10; *p < :05; **p < :01; ***p < :001
Model 3 incorporates an additional student-level covariate, which is an

indirect measure of student’s ability: whether the student was selected
through an admission process. I incorporated this variable separately
for two reasons: first, to estimate whether it adds information to the
traditionally used student-level controls; second, given that better
schools can attract abler students, this variable is potentially endoge-
nous and its effect can be confounded with school quality. This indirect
measure of student ability has a strong relationship with student
achievement, even after controlling for the mentioned student char-
acteristics: on average, selected students score 0.34 S.D. higher than
students who were not selected through an admission process.
Additionally, the introduction of this covariate slightly reduces the
positive test-score difference between private voucher schools and
public schools (to 0.18 S.D.).
Model 4 uses exclusively school-level control variables. As noted,
those covariates have a huge effect on the estimate of the private/
public test-score gap; in fact, after controlling for them, public school
students score higher than private voucher students (0.04 S.D.). Corre-
spondingly, the school-level selectivity measure was added to model 5,
which slightly increased this estimated public school advantage over
private voucher schools to 0.06 S.D.
Finally, model 6 includes covariates at the student and school levels.
When both types of controls are simultaneously present, school-level
predictors’ parameter estimates tend to be more stable than student-
level predictors (especially pronounced is the decrease in the selected
student coefficient). As shown, this full model estimates that—on
average—public school students score higher than private voucher
students by 0.06 S.D. (similar to the estimate using exclusively school-
level covariates). Note that according to this full model, there is no
statistically significant difference between private nonsubsidized and
public school students. Finally, the R 2 of the full model is slightly
larger than the R 2 obtained by using only one-level control variables,
suggesting that both kinds of predictors are needed to better explain
students’ test scores.
8.8.2 Between versus within Schools’ Test-Score Variation

The sensitivity of the findings to the choice of covariates at the student
versus school level can be further explored by using a multilevel analy-
sis. Multilevel models are recommended to study student achievement
because the regression assumption that residuals are independent is
186 Cristian Bellei
not satisfied in educational settings. There are two hypotheses to ex-

plain that: school effectiveness (within their schools, students share
common educational experiences that significantly affect their out-
comes), and school segregation (students enrolled in the same school
share unobserved previous characteristics, which are related to their
academic performance). My aim is to propose that the highly segre-
gated nature of Chilean education entails an additional challenge to
study the test-score gap between private and public schools.
Table 8.8 contains multilevel regression models for mathematics and
language test scores. The multilevel analysis allows me to separate the
total variation of students’ test scores in between-schools and within-
schools variations. As reported, there is a very large between-schools
variation in Chilean education: about half of the mathematics test-
scores variation and more than a third of the language test-scores
variation occur between schools.8 This very large between-schools
variation explains why school-level predictors are so successful in esti-
mating students’ test scores: Chilean students’ academic achievement
is highly predictable depending on the schools they attend.
Models 1 and 4 incorporate exclusively school-level variables; as
shown, they explain 72% of the mathematics and 80% of the language
total between-schools variation.9 When school-level predictors are
taken into account, public schools score higher than private voucher
schools in both language and mathematics.
In order to explain some of the within-schools variation, models 2
and 5 use exclusively student-level covariates (note that type of school is
a school-level variable). According to those models, private voucher
schools score higher than public schools in both language and mathe-
matics. Nevertheless, these variables account for an extremely small
proportion of the within-schools variation: 3% in mathematics and 2%
in language. Thus, once the within-schools variation is distinguished,
it becomes apparent that the available standardized information is in-
sufficient to understand individual achievement in a context of highly
segregated schools.
Finally, full-models 3 and 6 include both student-level and school-
level predictors. The findings indicate that public schools score higher
than private voucher schools in language and mathematics. Note that,
compared to models 1 and 3, student-level variables do not increase
the proportion of between-schools explained variation.
To sum up: the size and the sign of the estimated test-score gap be-
tween private and public Chilean schools are highly sensitive to the
Table 8.8
Multilevel Regression Models. Relationship between School Type and Students’ Mathe-
matics and Language Achievement, Controlling for Student and School Variables
Dependent variable: 10 th -grade students’ test scores,
SIMCE-2003:
Mathematics Language
Initial between- 49% 37%
schools variation
Initial within- 51% 63%
schools variation
Private voucher 2.89* 24.26*** 3.30* 1.78* 18.56*** 1.87*
(1.42) (1.65) (1.43) (0.89) (1.12) (0.88)
Private nonsubsidized 4.32@ 70.25*** 3.56 4.35** 47.69*** 3.42*
(2.49) (2.09) (2.49) (1.58) (1.44) (1.56)
School mean parents’ 4.97*** 4.39*** 4.23*** 3.63***
education (0.56) (0.56) (0.36) (0.35)
School mean books at 17.35*** 14.23*** 15.00*** 10.89***
home (2.13) (2.14) (1.37) (1.36)
% selected students in 39.00*** 35.79*** 25.25*** 22.79***
school (2.31) (2.32) (1.46) (1.45)
Mother’s education 0.87*** 0.84*** 1.04*** 1.00***
(0.03) (0.03) (0.03) (0.03)
Log family income 0.69*** 0.78*** 1.06*** 1.20***
(0.17) (0.17) (0.16) (0.16)
Books at home 4.09*** 4.02*** 4.10*** 4.00***
(0.09) (0.09) (0.08) (0.08)
Gender (male) 10.36*** 10.39*** 4.29*** 4.28***
(0.21) (0.21) (0.19) (0.19)
Selected student 4.01*** 3.63*** 2.99*** 2.49***
(0.24) (0.24) (0.22) (0.22)
Percentage of
explained variation:
Between schools 71.6% 48.1% 71.6% 79.7% 55.0% 80.2%
Within schools 0% 2.8% 2.8% 0% 2.4% 2.4%
Number of students 237,629 237,629 237,629 237,629 237,629 237,629
(schools) (2,105) (2,105) (2,105) (2,105) (2,105) (2,105)
@
p < :10; *p < :05; **p < :01; ***p < :001
188 Cristian Bellei
level of aggregation of covariates included in the analyses. Addition-

ally, the segregated feature of the Chilean educational system causes
a very large between-schools variation in students’ performance; as
a result, Chilean students’ academic achievement is highly predictable
depending on the schools they attend. This explains the very strong
statistical relationship between school-level predictors and students’
achievement.
8.9 Conclusions and Discussion
Chile is a paradigmatic case of school choice and a market-oriented

educational system: private and public schools openly compete to cap-
ture family preferences and public subsidies. Unfortunately, several
data limitations and methodological divergences have affected the re-
search on the comparison between private and public school effective-
ness and on the systemic effects of school choice; as a result, previous
research has obtained noticeably contrasting conclusions. This chapter
has identified some threats to the validity of this research. The most
important one is selection bias: the parents’ school-choice process and
the schools’ students-selection process introduce severe biases that
researchers have not been able to overcome satisfactorily. Researchers
also diverge on how to control for the enormous differences in school
and student characteristics between private and public schools, what
the appropriate level of data aggregation and data analysis should be,
and how to tackle the multilevel nature of educational data.
By conducting exemplary data analyses, this chapter demonstrated
that the discussed methodological issues can affect not only the size
but also the sign of the estimated comparative efficacy of private and
public schools in students’ academic achievement. For that reason, the
answer to the question of whether private or public schools are the
most effective in Chile is extremely sensitive to those methodological
decisions. Although my analyses were based on OLS estimates of
cross-section data, they introduced new measures of the sorting pro-
cesses, the key source of bias in these studies. In fact, as hypothesized,
both sorting mechanisms—selective admission processes and rejection
of retained students—were significant predictors of student test scores
and were more disseminated among private schools, accounting for
some proportion of their observed advantage. Based on those analyses,
my most precise estimates (see models 3 and 6 in table 8.8) indicate
that private schools are not more effective than public schools, and
that they may be less effective.
The reasons why private schools are not more effective than public
schools in Chile are beyond the scope of this study. One hypothesis is
that although in the past voucher schools were more effective, public
schools have reacted to the competition by improving their quality,
and consequently have closed the previous gap. As explained, there is
little—if any—evidence to support this hypothesis. An alternative
hypothesis is that the institutional design of the Chilean educational
system has structural deficiencies, because schools can improve their
market position without improving the quality of their educational
service.
The theory underlying this last hypothesis is as follows: competition
among schools has not improved educational quality, because schools
(mainly the private ones) have competed to attract the best students
rather than to improve their educational service, creating a ‘‘zero-sum
game’’: improvements of some schools are annulled by deterioration
of others. Additionally, because parents’ choices have not necessarily
been oriented to educational quality (owing to information deficiencies
and parents’ use of nonacademic criteria), schools have not received
from their customers signals guiding them toward educational im-
provement, but rather toward the use of status symbols and social seg-
regation. Finally, deregulation and free competition have tended to
increase school segregation through a process of mutual reinforcement
between schools and families. From the supply side, schools have
responded to the incentives of competence, by distorting the indicators
of quality through the rejection of students who are less likely to suc-
ceed in school (applying admissions tests), and those who have dem-
onstrated low capacities (expelling them). These sorting and re-sorting
mechanisms, massively applied for two decades, have shaped the Chil-
ean school system in its current segregated features. From the demand
point of view, middle and high social-class families have found that
schools’ social and academic selectivity provide them a large profit of
peer effects within schools: given the high correlation between learning
outcomes and students’ social background, when Chilean families aim
at social selectivity, they obtain academic selectivity by extension.
The current evidence and the main findings of this chapter provide
partial support to this theory. Nevertheless, this study also has some
limitations. First, although it exploited new data on the student selec-
tion processes to eliminate the selection bias, we cannot be completely
confident that the regression models overcame this threat to validity.
Second, because SIMCE does not evaluate the same students over
time, it was not possible to develop longitudinal analyses or create
190 Cristian Bellei
value-added models that control for previous student achievement. Fi-

nally, the lack of data on students’ academic achievement prior to the
voucher reform severely constrains the analysis about potential effects
on the system as a whole.
If my conclusions are correct, they do not imply that private schools
cannot be positive partners of Chilean public education, but it does
suggest that, in order to contribute to improve educational quality and
equity, voucher programs must be carefully designed. In this sense,
the Chilean experience provides some relevant lessons from an educa-
tional policy perspective.
First, every school that receives public resources should guarantee
nondiscrimination to applicant students; thus, admission tests, aca-
demic and economic requirements, and other forms of sorting should
be prohibited. Second, bad information can be as harmful as no infor-
mation: if the evaluation system does not estimate the actual effective-
ness of the school, it can orient families and policy makers to a wrong
direction. Third, funding systems, public policies, and other institu-
tional regulations should recognize that some students (e.g., low-
income students, ethnic minorities) are more challenging to educate
than others. This implies that schools that serve more disadvantaged
students should receive additional resources. Fourth, it is overoptimis-
tic to expect that families’ demand will improve educational quality by
itself; conversely, some public incentives, pressures, and regulations
should also be in place in order to push schools toward genuine pro-
cesses of school improvement. Finally, Chilean private schools include
for-profit and nonprofit institutions, but the current system does not
make this distinction at all, and parents do not have this information.
Legislation and educational policies should differentiate these two
kinds of schools, in order to give priority in access to publicly funded
school improvement programs and other public resources (texts, com-
puters, teaching materials, teacher training, etc.) to schools serving the
public good.
Notes
I would like to thank Paul Peterson, Rajashri Chakrabarti, Marcela Pardo, Brian Jacob,
Roland Fryer, and Richard Murnane for their valuable commentaries and suggestions.
1. See Bellei and Mena (2000) for details.

2. The average tuition of the copay system is about half the cost of the public voucher,
while the average tuition of the elite schools is about four to five times the cost of the
public voucher.
3. Vegas (2002) is an exception: she analyzed 1999 data, and used a unique database on
teachers’ characteristics.
4. Catholic schools have been estimated to be—on average—more effective than public
schools. Unfortunately, in my analysis I cannot distinguish between Catholic and non-
Catholic private schools.
5. Note that this repetition rate is at the middle of fourth grade. According to the Chilean
rules, students should not repeat first grade; thus at that point, students could have re-
peated only second and third grades.
6. I will report the test-score gap in S.D. units. I divided the regression coefficients by the
population S.D. (mathematics S.D. ¼ 56; language S.D. ¼ 48).
7. I did not include the most used school SES index, which is the percentage of students
‘‘at risk.’’ This index is an administrative tool used by the Ministry of Education to dis-
tribute free lunches among schools. The index is based mainly on physical health indica-
tors; it uses only information from the first-grade students, and is self-reported by the
schools. All private nonsubsidized and many private voucher schools have no informa-
tion on this index (researchers assign 0% to these schools). Thus, I do not consider this
index a good measure for research purposes.
8. As a point of reference, consider that the PISA (OECD/UNESCO 2003)—an inter-
national survey of students’ performance in language, mathematics, and science—found
a negative relationship between the level of student achievement and the level of be-
tween-schools variation. For example, the three countries with the highest student
performance—Finland, Canada, and New Zealand—had between-schools variations of
12%, 18%, and 16% respectively; the United States had 30% between-schools variation,
and Chile had one of the highest levels of between-schools variation: 57%.
9. It is important to note that the proportion of explained variation is relative to the re-
spective proportion of explainable between- and within-schools variations.
References
Aedo, Cristian. 1997. Organización industrial de la prestación de servicios sociales.

Working Paper Series R-302, Washington D.C.: Inter-American Development Bank.
Aedo, Cristian, and Osvaldo Larrañaga. 1994. Sistemas de entrega de los servicios
sociales: La experiencia chilena. Washington D.C.: Inter-American Development Bank.
Auguste, Sebastián, and Juan Pablo Valenzuela. 2004. Do students benefit from school
competition? The Chilean experience. Ph.D. dissertation, University of Michigan, Ann
Arbor, MI.
Bellei, Cristian, and Isidora Mena. 2000. The new challenge: Quality and equity in educa-
tion. In Chile in the nineties, ed. Cristian Toloza and Eugenio Lahera, 349–391. Palo Alto,
CA: Stanford University Libraries.
Bravo, David, Dante Contreras, and Claudia Sanhueza. 1999. Rendimiento educacional,
desigualdad, y brecha de desempeño privado público: Chile 1982–1997. Working paper,
Department of Economics, University of Chile.
Carnoy, Martin, and Patrick McEwan. 2000. The effectiveness and efficiency of private
schools in Chile’s voucher system. Educational Evaluation and Policy Analysis 22, no. 3:
213–239.
192 Cristian Bellei
Gallego, Francisco. 2002. Competencia y resultados educativos. Teorı́a y evidencia para

Chile. Cuadernos de Economı́a 39, no. 118: 309–352.
Gallego, Francisco. 2004. School choice, incentives, and academic outcomes: Evidence
from Chile. Mimeo, Massachusetts Institute of Technology, Cambridge, MA.
Gauri, Varum. 1998. School Choice in Chile. Two Decades of Educational Reform. Pittsburgh,
PA: Pittsburgh University Press.
Hseih, Chang-Tai, and Mighel Urquiola. 2003. When schools compete: How do they com-
pete? An assessment of Chile’s nationwide school voucher program. Working Paper
1008, National Bureau of Economic Research, Cambridge, Mass.
McEwan, Patrick. 2001. The effectiveness of public, Catholic, and non-religious private
schools in Chile’s voucher system. Education Economics 9, no. 2: 103–128.
Mizala, Alejandra, and Pilar Romaguera. 1999. Schools performance and choice. The
Chilean experience. The Journal of Human Resources 35, no. 2: 392–417.
Mizala, Alejandra, and Pilar Romaguera. 2003. Equity and educational performance.
Center for Applied Economics, Department of Industrial Engineering, University of
Chile.
Mizala, Alejandra, Pilar Romaguera, and Carolina Ostoic. 2004. A hierarchical model for
studying equity and achievement in the Chilean school choice system. Center for Applied
Economics, Department of Industrial Engeneering, University of Chile.
Rodriguez, Jorge. 1988. School achievement and decentralization policy: The Chilean
case. Revista de Analises Económico 3, no. 1: 1.
Sapelli, Claudio. 2003. The Chilean voucher system: Some new results and research chal-
lenges. Cuadernos de Economı́a 40, no. 121: 530–538.
Sapelli, Claudio, and Bernardita Vial. 2002. The performance of private and public
schools in the Chilean voucher system. Cuadernos de Economı́a 39, no. 118: 423–454.
Vegas, Emiliana. 2002. School choice, student performance, and teacher and school char-
acteristics: The Chilean case. World Bank Policy Research Working Paper 2833, April.
9 The Concession Schools of Bogotá, Colombia
In 1999, the city of Bogotá, Colombia, launched an educational pro-

gram designed to broaden the coverage and quality of basic education.
This program to establish concession schools is a partnership between
the public and private education sectors, with private schools provid-
ing public education in twenty-five schools for a period of fifteen years.
The state provides the infrastructure, selects the students, and pays
a predetermined sum per full-time student per year (approximately
Col$1.2 million (U.S.$520), according to Villa and Duarte 2005), which
is higher than what most regular public schools receive (approximately
Col$1 million or U.S.$430). The concession schools must provide edu-
cation to the population assigned to them by the state, but they are
allowed relative flexibility to contract administrative and teaching staff
and can freely implement their pedagogic models. The concession
schools must also meet performance standards (on quality and quan-
tity) set by the secretary of education. For instance, each school com-
mits to surpass the mean score of standardized tests in similar schools.
Over 25,000 students participate in the concessions program, repre-
senting close to 3% of the total public enrollment in the city (World
Bank 2005). The schools were built following two main criteria. First,
they were located in extremely poor areas of the city. Second, they
were built in areas where the demand for primary and secondary edu-
cation was higher than the number of student placements supplied by
city public schools. Any student from the affected neighborhood can
apply for enrollment in the concession schools. The secretary of educa-
tion of the district uses means testing to allocate places in the conces-
sion schools among the applicants. The students who are not admitted
into the concession schools are placed into nearby public schools.
During visits to schools in the public system done in 2004 and 2005,
it became clear that the differences across regular public school and
194 Felipe Barrera-Osorio
concession schools were very sharp. In contrast to regular public

schools, concession schools had very good infrastructure, similar to
good private schools. Also, the relationship between teachers and stu-
dents in the concession schools was open and very constructive. Fi-
nally, the concession school students were very proud of their schools.
Concession schools are located in areas ranking at the lowest end of
the income distribution. Children in these zones in general face serious
challenges, including lack of affection and other family problems, as
well as poor nutrition. According to interviews with different head-
masters, concession schools, in their objective to provide high-quality
education, offer psychological counseling (daily for several students)
and an environment of affection to students. Children who are sus-
pected of having family problems are subject to home visits by social
workers and psychologists from the school. Children who miss classes
regularly are subject to home visits as well. To counter poor nutrition,
several concession schools have their own food programs, which com-
plement the public program. The regular public school food program
consists mainly of fruit, a piece of cake, and a yogurt or a juice. Some
concessions add a complete lunch, which generally includes a protein.
Another clear objective of the concession schools is to work with the
communities in which they operate. Several have an open-door policy
during weekends, and they encourage teacher-parent meetings on a
regular basis. They work with the community through programs to
resolve family problems and provide adult education. Finally, many
of the headmasters consider the academic impact of the concession
schools a byproduct of the other measures, believing they need to pro-
vide psychological and nutritional balance to students before they can
address academic achievement.
Clearly, family problems and poor nutrition also plague students
in regular public schools. However, those schools do not have the
resources to implement the strategies that concession schools use. In
fact, non-concession public schools often do not have the resources to
maintain what minimal infrastructure they do have. For example, in
both of the public schools that the World Bank team visited, the bath-
room facilities were limited and out of service on some days.
Despite the apparently good reputation of concession schools, in
2005 the program was under debate. Some members of Bogotá’s City
Council claimed that the program did not yield the expected results
because the average standardized test scores in the concessions were
The Concession Schools of Bogotá 195
lower than the average scores for regular public schools. However, as
will be discussed, though the test scores for concession schools were
lower according to a general comparison of means, in fact once con-
founding variables are controlled for, it appears that the concession
schools do have a positive impact.
The qualitative evidence compiled during the visits to the concession
schools and the public debate in the council induced three main
hypotheses that are the subject of this chapter:
1. Dropout rates are lower in concession schools than in similar public

schools. The dropout rate is another measure of schooling and, pre-
sumably, it responds faster to interventions than do test scores.
Through the various interventions previously described, concession
schools attempt to keep children in school.
2. Regular public schools near the concession schools have lower dropout
rates than public schools outside the influence of concessions. Concession
schools have had positive externalities, through their community
work, on nearby public schools. In addition, increased competition in
the educational market due to the existence of concession schools has
caused improvements in nearby schools. Previous literature on school
choice (Hoxby 2002) supports this hypothesis. I test this hypothesis on
dropout rates rather than on test scores, under the presumption that
dropout rates respond faster than test scores to the indirect effect of
the program, and presumably it is very unlikely that one can observe
effects on test scores yet.
3. Concession school test scores are at least equal to or higher than test scores
for regular public schools. Measuring the impact of any program on
test scores can be complicated, and theoretically we should observe
changes in the schools after long-term exposure to the program. Given
its recent implementation, the longest period that any students have
attended a concession school is six years, which means their early
years were spent in regular public schools.
Section 9.2 of this chapter discusses methodological issues, under-

lining the reasons why concession schools may produce better educa-
tional outcomes than regular public schools. Section 9.3 outlines the
empirical strategy, which is based on propensity score matching esti-
mators. Section 9.4 presents the results. Section 9.5 finishes with con-
clusions and a discussion on the relationship between dropout rates
and test scores.
9.1 Methodological Issues and Theoretical Framework
9.1.1 General Framework

The basic equation for estimation is based on the idea that educational
outcomes depend on a production function based on certain inputs
(Hanushek 1986, 1996). Specifically, the literature considers that a mea-
sure of educational attainment, Yi; j; t of individual i in school j in a
given moment t, depends on the inputs of the school, Zj; t and on the
characteristics of the individual, Xj; t , such that:
Yi; j; t ¼ f ðXi; t ; Zj; t ; ei; t Þ ð9:1Þ
Included among the variables of school inputs, Zj; t , are the student-
teacher ratio, the educational attainment of the teachers, school infra-
structure, and school size. The characteristics of the individuals, Xi; t ,
control for household income, the educational level of the parents, the
number of siblings, and so on. In equation (9.1), ei; t captures the un-
observable characteristics of the individual, such as ability, skills, and
motivation that may influence test scores but cannot be measured. The
dependent variables, Yi; j; t , are the standardized test scores and the
dropout rate.
Equation (9.1) depends on several premises:
n
The quantity and quality of the school inputs are fundamental in the
production of better education, be they measured by higher standard-
ized test scores or lower dropout rates.
n
The characteristics of the individuals are also critical elements in the
production of better education. Individuals with better skills and better
nutrition and with an adequate home environment should also per-
form better in school.
n
These two groups of characteristics (school and individual/family)
interact to yield the third fundamental component in the education
production function.
In a meta-analysis of equation (9.1) Hanushek (1996) reaches several
conclusions. First, there is no clear relationship between several of the
school inputs and the quality of education. The pupil-teacher ratio, for
example, may have a concave relationship with the outcome measures.
A small number of pupils in the classroom may inhibit participation
and it is possible that the positive externalities of participation do not
occur. A large class size may also inhibit participation and may pro-
duce negative externalities such as noise. Second, the quality of the

teacher is a key element in the production of education. And third, the
role of the family is as important as the quality of the teacher.
The impact of the concession program on the quality of education
can be explained through several channels (for a general discussion,
see World Bank 2005). First, private participation assumes the applica-
tion of an already proven pedagogic model since the private schools
have already been operating for some time. Concession schools were,
in fact, handed over to private schools with the highest standardized
test scores. Likewise, concession schools are financially stable since the
State pays a fixed sum per student, ensuring the stability of the peda-
gogic model.
Second, the freedom to choose the teaching and administrative staff
may lead to a better quality of education than in the public schools,
where the teachers’ union makes it difficult to implement staff changes
(for instance, see Borgas and Acosta 2005; Duarte 1996). Third, conces-
sion schools have contracts that establish goals in terms of standard-
ized testing. Fourth, the infrastructure of these schools is superior to
that of public schools, providing the students better learning opportu-
nities. Concession schools were, in fact, built on better lots than was
the average public school, with better equipment and a complete set of
supplies for classrooms, laboratories, and libraries.
Finally, concession schools in general work actively with the pupils’
parents and the surrounding communities. The community work has
been one of the more discussed factors in visits to several concession
schools: school officials, pupils, parents and other community mem-
bers alike identify this community work as one the most important fea-
tures of the initiative.
9.1.2 Some Empirical Evidence

In the case of Colombia, Gaviria and Barrientos (2001), Barrera (2003),
and Sarmiento and others (2000) have estimated equation (9.1). These
studies reach similar conclusions as Hanushek (1996). They show that
school quality variables, such as the educational level of the teachers,
tend to have a positive impact on standardized test scores. Also, they
corroborate the importance of individual characteristics.
Early articles about concessions focused on the description of the
experience. Corpoeducación (2004) and Villa and Duarte (2005) de-
scribe the concession school initiative in great detail. Rodriguez (2005)
presents an overview of the concession schools program, explaining
the details of the program’s design and potential effects of the

program.
Sarmiento and others (2005) present an evaluation of concession
schools. They follow a different methodology than the one used here.
In short, they applied a very detailed questionnaire to 22 concession
schools and 10 public ones with similar characteristics. The ques-
tionnaire focused on internal processes and educational production.
Variables derived from the questionnaire were correlated with stan-
dardized test scores. They found that there are clear differences
between the concession and pure public schools in terms of adminis-
tration, autonomy of decisions, the capacity to adjust, and the impact
on the community. Overall, they found that concession schools are per-
forming better than the traditional public schools.
Charter schools in the United States offer additional insight. Despite
the differences across states in the implementation of charter school
programs in the United States, they are similar to the concession
schools in that they operate independently from the traditional public
school system, with (generally) a higher degree of autonomy than
the typical public school, and in some cases they explicitly target
low-income students (www.uscharterschools.org). Solmon and Gold-
schmidt (2004) found positive impacts of charter schools on standard-
ized test scores and other characteristics, such as the retention rate.1 In
contrast, Hanushek and others (2005) present evidence of a negative
impact of this type of school. Hoxby and Rockoff (2004) present strong
evidence in favor of charter schools, using the randomized character of
Chicago’s charter school program to determine the impact of charter
schools on standardized test scores. The randomized nature of the allo-
cation process allowed them to create a treatment group and a control
group that have the same observable and unobservable characteristics.
They found that students who attended an average of two years in
an elementary-level charter school in Chicago scored higher on both
math and reading tests. Finally, Hoxby and Murarka (2007), using the
randomized nature of placements in the charter system in New York
City, found that students in charter schools have higher test scores
than similar students in public schools.
9.2 Empirical Strategy
The basic methodology used to evaluate the impact of the program is

to compare the group of individuals that were admitted to a conces-
sion school (treatment group) to the group of students who attend

public school (comparison group) (see Heckman and others 1999). The
correct evaluation of a program requires the establishment of a coun-
terfactual for the treatment group; that is, what would have happened
had the treated individual not been treated. Of course, it is not possible
to observe the same individual in the two states and it is therefore nec-
essary to choose another individual, to serve as a comparison, who is
as similar as possible, both in observable and unobservable characteris-
tics, to the treated individual. Simple comparisons of mean test scores
(or dropout rates) between students in concession schools with stu-
dents in other types of schools may be biased measures of the true im-
pact because the two groups may be very different in their observable
and unobservable characteristics.
In mathematical terms,2 let Y 1 denote the standardized test score
or the dropout rate for a student who attended concession school
and Y 0 the individual who attended another type of school. Let T
denote the treatment condition, which is equal to one ðT ¼ 1Þ if the
person received treatment, and zero ðT ¼ 0Þ otherwise. The mean im-
pact of the program ðIPÞ will be given by IP ¼ EðY 1 =T ¼ 1Þ
EðY 0 =T ¼ 1Þ, for example, the difference in average test scores for the
same individual in two different states, with and without the inter-
vention. It is not possible, however, to observe the same individual in
both states. What is observable are two different individuals, one
attending a concession school and the other not, I~P ¼ EðY 1 =T ¼ 1Þ
EðY 0 =T ¼ 0Þ.
The estimator I~P can suffer from bias because of self-selection into
the program. In short, the two types of individuals may differ sys-
tematically in observable and unobservable characteristics such that a
typical control observation is not a good proxy of what would have
happened to a treated individual had he or she not received the treat-
ment. In the case of the concession school, the two groups, for example,
are not comparable in terms of income, thus leading to bias. That is,
EðY 0 =T ¼ 1Þ 0 EðY 0 =T ¼ 0Þ.
Ideally, a program like concession schools can be evaluated using
a simple randomization strategy: given that there exists excess de-
mand for concession schools, the government can perform a lottery
to allocate individuals between concessions and other public schools.
The randomization would ensure that the group of students that at-
tends a concession school (treatment group) and the one that attends
other public schools (control group) are very similar in observable
and unobservable characteristics. However, as already discussed, the

school placements were not done by a lottery.
A second approach3 would be to take as a control group students
who applied to a concession school but were assigned to another pub-
lic school. Still, evaluations based on the comparison of these two
groups may be prone to bias, mainly because the selection of students
into concession and other public schools is not random.4
In contrast, this chapter uses the whole sample of individuals who
attend public schools and uses matching estimators as the strategy to
‘‘reduce’’ selection bias. In short, the estimator will try to match each
treated individual with an individual who does not attend a conces-
sion school, based on the observable characteristics of the individuals
(a vector X). In this way, the estimator can be modified by I~P ¼
EðY 1 =X; T ¼ 1Þ EðY 0 =X; T ¼ 0Þ.
Moreover, instead of using the vector of characteristics x, it is possi-
ble to determine the probability of participation in the program (where
T is equal to 1 or 0), such that PðXÞ ¼ PrðT=XÞ ¼ f ðXÞ. This probability
is called the propensity score. In other words, the propensity score cap-
tures in a synthetic form the intention to participate in a program,
based on a broad vector of observable characteristics thought to in-
fluence the participation decision and the outcome measures. For
each individual (both in the treatment and control groups) a probabil-
ity of participating in the program is estimated using the observable
characteristics.
Therefore, the following estimation is used to calculate the impact of
the program:
~Þ; T ¼ 1Þ EðY 0 =PðX~Þ; T ¼ 0Þ

I~P ¼ EðY 1 =PðX ð9:2Þ
Clearly, the endogeneity problem can arise due to either observable

or unobservable characteristics. Equation (9.2) controls for the observ-
able characteristics, but not for the unobservable ones. Heckman and
others (1998) and Heckman, LaLonde, and Smith (1999) show that, in
fact, the bias of estimation may come from three margins, and the
most important one is the difference in the observable characteristics
among individuals. As stated previously, it is essential for rigorous
impact evaluation that the control and treatment groups have, on aver-
age, similar characteristics. Besides differences in the average treatment
and nontreated individual, there may be differences in the support of
the populations. For example, noneligible individuals cannot be part
of the control group since there are no treatment observations with the
characteristics of these individuals. The third potential difference be-
tween the groups arises from potential differences in the unobservable
characteristics. According to the estimates of Heckman and others
(1998), once the observable characteristics are similar between groups,
and the observations fall within the same support, the bias due to
unobservable characteristics is relatively small.
The evaluation of the concessions program in Bogota may suffer
from self-selection into the program from two sources. First, the loca-
tions of the schools were not randomly chosen. Authorities built the
concessions in areas where there was a high concentration of low-
income individuals and a scarce supply of public education. Second,
an unknown proportion of the students are not randomly assigned to
the schools. Individuals, when applying to public education, can state
their preferred school in their neighborhood. As previously discussed,
this chapter copes with the problem of endogeneity by using propen-
sity scores and matching estimators. In this way, I ‘‘minimize’’ the
endogeneity problem and can obtain estimates that approximate the
impact of the program. In any case, the predicted bias of the estimation
is downward since the pool of individuals that attend concessions
comes from extremely poor areas.
The chapter presents direct and indirect impacts of the concession
schools. In order to estimates these impacts, I separate the public
schools into three categories: (1) concession schools; (2) regular pub-
lic schools close to concession schools (that is, within the influence of
a concession school); and (3) public schools outside of the influence
of concession schools. The difference between students in concession
schools and the matched students who go to public schools outside
the influence of concession schools is the measure of direct impact.
In contrast, the indirect impact of concession schools is measured as
the difference between the students in public schools in the vicinity of
concession schools and the students in similar schools outside the con-
cessions’ area of influence. The area of influence is defined as being
within one kilometer of a concession school. Given that some conces-
sions are going to have more than one regular public school within
this definition of proximity, we also rank the proximity among the
nearby schools.5
The indirect impact is governed conceptually by the idea that nearby
schools may be forced to raise the quality of education to compete with
concession schools, as well as to respond to community pressure. This
idea is similar to the argument that vouchers will increase the quality
of education due to market and choice forces, as outlined by Friedman
(1955). Recent literature finds evidence in favor of this type of mecha-
nism. For instance, Hoxby (2002) shows that competition and choice
(in the form of vouchers and charter schools) increase productivity of
public schools significantly. In the case of Bogotá, not only do the con-
cession schools impact the nearby public schools through community
outreach but they also encourage better performance by regular pubic
schools due to the competition for resources. Since 2002, the central
government has transferred educational resources to the localities
based on the number of students enrolled in school. Presumably, con-
cessions attract students away from the regular public school system,
and the greater the difference in quality between regular and con-
cession schools, the greater the effect. In order to guarantee greater re-
sources, the nearby public schools need to match the performance of
the concession schools.
To estimate the indirect impact of the program, I use a two-step pro-
cess. I first identify the baseline. Given the identification of nearby
schools as discussed, I match them based on their common characteris-
tics with similar schools outside the influence area as of 1999. As stated
before, concession schools that started in 1999 were fully operational
by 2000. These control schools are determined using propensity scores.
Second, employing data for 2003, I estimate the indirect impact using
those students who attended schools close to a concession school as
treated individuals, and those students who attended the matched
schools found in the previous step as the control group. For the indi-
rect impact I only investigate effects on dropout rates under the as-
sumption that changes in standardized testing, via competition effects,
will tend to be observable only in the long term.
The baseline data can be used to control for differences in initial
characteristics. However, another problem that may persist is the dif-
ference in preexisting trends between the treated schools and the
matched schools. Unfortunately, data prior to 1999 does not exist to
test for this possibility. However, a priori, there is no theoretical
ground to believe that there exist systematic differences in trends be-
tween schools under the influence of concessions and schools outside
this area of influence.
In short, the estimation that I will present is based on matching esti-
mators using propensity scores to determine the treatment and control
groups. The estimation will present evidence of ‘‘direct impact’’ (drop-
out rates and test scores for concession versus matched non-concession
public schools outside the area of influence) and ‘‘indirect impact’’
(dropout rates for schools near the concession against matched schools
outside the area of influence of concessions).
9.3 Data and Results
9.3.1 Description of the Data

This chapter uses data from two sources: the Ministry of Education
(surveys C600 and C100) and Institute for the Development of Higher
Education (Instituto Colombiano para el Fomento de la Educación Su-
perior: ICFES), which administers standardized tests in Colombia. The
C600 and C100 surveys provide general data on an important array of
school characteristics and have an identification code that allows re-
searchers to link these datasets with the ICFES test scores. The Ministry
of Education collects the data directly via a questionnaire distributed to
all the schools in the country. Data presented are at the school and shift
level (the time period during which the school provides services).
The school-level data present information on the number of adminis-
trative personnel, the number of teachers and their level of education,
the number of teachers by subject areas, the number of physicians in
the school, the total number of students in the school by grade and by
age, the number of students who failed a grade, and the number of
students who dropped out. These data are available for both 1999 and
2003.
For 1999, there is also information on the physical characteristics of
schools. The main variables are: furniture in the school (chairs and
desks), support materials such as computers, the number of computers
exclusively for teaching purposes, the total number of laboratories, the
number of laboratories for specific subjects (physics, chemistry, biol-
ogy, construction, and farming), the total number of rooms, the num-
ber of classrooms, the number of libraries, the number of food facilities
and dorms, the number of other types of rooms, and the number of
sports facilities (soccer fields and basketball courts).
The surveys also break down the data at the shift level. There are
four main shifts during which students can attend school: in the morn-
ing (usually from 7 am to 12 pm), in the afternoon (from 12 pm to 4
pm), in the evening (from 2 pm to 7 pm); or they can attend a school
with a ‘‘complete’’ shift (from 8 am to 3 pm). The shift mechanism was
implemented throughout the country in the 1960s to maximize the use
of school infrastructure. In this way, some schools that previously

operated only during the morning were subsequently open for two
shifts (e.g., morning and afternoon), thus allowing them to enroll twice
the number of students. It is important to note that the data discrimi-
nate between the two shifts as if they were two different schools.
Second, data from ICFES provide test scores for individuals and
some student characteristics. The educational system in Colombia is
composed of three levels: primary spanning grades 1 to 5; basic sec-
ondary spanning grades 6 to 9; and middle secondary composed of
grades 10 and 11. The test scores used here are those of a general test
that is administered to all students finishing the eleventh grade. Al-
though the ICFES data provide test scores for various subjects, the
analysis here focuses on mathematics and reading scores. Student char-
acteristics such as gender and current enrollment status are also pro-
vided. Other variables linked to the individual students include: a
school code, the city where the test was taken, the semester (either A,
which is from February to November, or B, which runs from Septem-
ber to June), the type of secondary school that the individual attended
(mainly academic or technical), whether the school is public or private,
and the location of the school within the city.6 All ICFES data used in
this analysis are for 2003.
Unfortunately, ICFES data on individual characteristics are limited.
We use a fixed-effect model by location, which presumably can cap-
ture some of the socioeconomic characteristics of the individuals liv-
ing in the area. In fact, the data identified nine localities in the city that
are quite homogeneous in terms of income. In contrast, the informa-
tion on the schools from the Ministry of Education is very rich, and I
will exploit this as the source of variation in order to estimate the
impacts.
In order to estimate the direct effect of the concession schools, this
analysis uses the 2003 school and test-score data. A limitation of the
study is that, despite having individual observations, the effects are
calculated using variation across schools. In total I was able to get in-
formation for 17 concession schools (out of 25) versus 2,790 regular
public schools.
To estimate the indirect effect I use panel data at the school level
for 1999 and 2003. The sample size is 23 schools nearby the con-
cession schools and 416 schools outside the influence of a concession
school.
9.3.2 Descriptive Statistics

Table 9.1 presents data of the full sample of basic characteristics of
public schools, divided into concession (close to 1,050 individuals in
17 schools) and non-concession schools (36,000 individuals in 2,790
schools). Eligibility requirements preclude the use of data on private
schools, which may downplay the competition argument, though only
slightly since there is still competition for resources among public
schools.
The distribution of locations is quite similar between the two groups.
The majority of the schools are coed. In terms of the type of secondary
school, the majority of concession schools provide a classic academic
education, whereas 30 percent of the public schools provide technical
education. The distribution of shifts across public and concession
schools is quite different: concession schools use complete shifts,
whereas public schools are divided between morning, afternoon, and
evening shifts. Even though there are differences in the two groups,
the estimators presented as follows balance the two samples of stu-
dents in terms of observable characteristics. This is very important
since the estimators have to isolate the effect of concessions from differ-
ences in other variables. For example, new literature on the length of
the school day and the quality of education shows a positive relation-
ship between the two (Cerdan and Vermeersch 1996).
On average, public schools tend to be bigger than concession
schools. The number of students in grades lower than their age group
(‘‘overage’’) is lower in concession schools than in the rest of public
schools. The teacher-student ratio is quite similar (close to 28 students
per teacher), as is the average of teachers’ years of education (16).
The table also shows data on the two main impact variables, drop-
out rates and test scores. Dropout rates, measured by grade for grades
1 through 11, differ between the two types of schools. The dropout
rate is 18% in regular public schools (with a standard deviation
of 0.14), and 15% in concession schools (with a standard deviation of
0.09). Moreover, between grades five and six there is an increment in
the rate. This increment in the dropout rate between grades five and
six is also observed using national data (Barrera and Dominguez
2005). The rationale is that between those grades is the change between
two levels of education (primary and secondary), and the opportu-
nity cost for students starts increasing precisely after finishing primary
education.
Table 9.1
Basic Statistics: Public Schools
Non-concession Concession
Mean Std. Dev. Mean Std. Dev.

A. Characteristics
Localization
1 0.10 0.30 0.12 0.33
2 0.01 0.10 0.00 0.00
3 0.05 0.22 0.00 0.00
4 0.06 0.24 0.00 0.00
5 0.15 0.36 0.12 0.32
6 0.04 0.19 0.00 0.00
7 0.30 0.46 0.44 0.50
8 0.12 0.32 0.13 0.34
9 0.17 0.38 0.19 0.40
Type of school
Male 0.07 0.26 0.00 0.00
Female 0.03 0.16 0.00 0.00
Coed 0.90 0.30 1.00 0.00
Type of secondary
Academic 0.63 0.48 0.93 0.26
Technical 0.29 0.45 0.00 0.00
Both 0.07 0.26 0.07 0.26
Shift
Complete 0.03 0.18 1.00 0.00
Morning 0.40 0.49 0.00 0.00
Afternoon 0.14 0.35 0.00 0.00
Evening 0.43 0.49 0.00 0.00
Number of students
Total school 2954.60 1376.76 987.98 90.17
Taking exam 122.15 79.16 65.33 9.98
Old for grade 47.28 45.15 12.11 7.03
Repeating grade 103.30 73.72 47.56 35.93
Ratio teacher-stud. 28.44 6.20 27.20 3.48
Years of teacher education 16.07 2.28 16.02 0.99
Table 9.1
(continued)
Non-concession Concession
B. Outcomes
Dropout rates
Grade 1 0.040 0.05 0.020 0.02
Grade 2 0.031 0.04 0.023 0.02
Grade 3 0.026 0.04 0.019 0.02
Grade 4 0.026 0.03 0.015 0.02
Grade 5 0.020 0.33 0.018 0.02
Grade 6 0.071 0.09 0.021 0.02
Grade 7 0.040 0.04 0.025 0.02
Grade 8 0.038 0.04 0.032 0.03
Grade 9 0.039 0.05 0.036 0.03
Grade 10 0.033 0.04 0.038 0.04
Grade 11 0.012 0.02 0.021 0.05
Test scores
Mathematics 42.08 5.13 41.68 4.77
Reading 50.99 7.13 51.04 6.79
Number of observations
Students
Max. 36244 1056
Min. 34218 1013
Schools 2790 17
Figure 9.1 shows the dropout rate per grade. The dropout rates
in concession schools are lower and exhibit smooth behavior across
grades. In contrast, public schools mimic the behavior of the dropout
rates found in other datasets using national statistics (for example, Bar-
rera and Dominguez 2005).
Interestingly, dropout rates for concession schools increase signifi-
cantly in grade eight. This may indicate a change in the composition
of students in secondary education in the concession schools. Since
attending a concession school reduces the dropout rates for grade six,
more students, including those who would have dropped out in the
regular public system, are reaching higher grades. The population of
students in secondary education in the concession schools includes
those in the lower part of the income distribution, who have a higher
probability of dropping out of school.
Figure 9.1
Dropout Rates, Concessions versus Non-concessions
Figure 9.2 presents mean standardized test scores for regular public
and concession schools. It shows that there small differences between
regular public schools and concessions. On average, public schools
show slightly higher scores in mathematics and physics, and almost
identical results in reading.
Table 9.2 presents basic statistics for the public schools located with-
in a kilometer of a concession school, as well as public schools outside
the influence of concessions.
In 1999, regular public schools near concession schools had, on aver-
age, fewer resources than other public schools. For instance, the num-
ber of computers in schools near concession schools was only 17,
whereas it was 25 in schools in other parts of the city. Also, the ab-
solute numbers of teachers, administrative personnel, physicians, and
psychologists were lower in regular public schools near concession
schools. The educational attainment of teachers in 1999, measured as
the proportion of teachers with secondary studies to those with college
or graduate studies, was very similar across the two school types. Fi-
nally, regular public schools near concession schools had, on average,
24 students per teacher, compared to 26 students per teacher in the
rest of the public system. In sum, there is evidence that regular public
schools near concession schools were smaller and poorer than the rest
of the public system, although the quality of the teachers was similar.
Figure 9.2
Mean Test Scores
Given that concession schools are located in low-income areas, these

differences are not surprising and it is likely that any result based on
an unmatched sample will be biased.
In 2003, in general, the regular public schools near concession
schools were similar to the rest of the public schools, with some excep-
tions. The proportion of secondary schools with an academic focus was
higher in nearby schools than in public schools further away. Also, a
higher percentage of the nearby schools used the evening shift (0.61
versus 0.43, respectively). Furthermore, the nearby schools had lower
overage ratios.
Regular public schools near concession schools tended to have lower
rates than other public schools in 2003. The dropout rate increases from
2.1% in grade five to 5.1% in grade six, whereas the dropout rate for
other public schools rises from 2% in grade five to 7.1% in grade six.
However, the dropout rates in regular public schools near concession
schools tend to be higher than the ones in the rest of public schools at
higher grade levels. Again, these results are based on the unmatched
samples and thus the controls may not be a good counterfactual.
9.3.3 Results
Estimates of equation (9.2), the average treatment impact on the
treated, are presented as follows. There are several different estimators,
Table 9.2
Basic Statistics: Schools near Concession versus Public Schools
A. Characteristics of schools
Schools near a Schools far away

concession from a concession
1. ‘‘Before concessions’’ (1999) Mean Std. dev Mean Std. dev
Infrastructure
Number of computers 17.82 16.75 24.96 22.69
Number of classrooms 21.40 10.58 22.24 12.93
Number of bathrooms 1.45 0.78 1.67 0.87
Number of sport facilities 1.78 0.54 2.59 1.52
Teachers
Number of teachers 30.27 16.34 34.46 24.06
Number of administrative personnel 6.97 4.45 9.69 11.96
Number of physicians 0.06 0.24 0.45 1.10
Number of psychologists, etc. 1.52 1.29 2.19 1.85
% teachers with second education 0.07 0.18 0.06 0.12
% teachers with college education 0.60 0.23 0.60 0.62
% teachers with graduate education 0.25 0.17 0.34 0.34
Ratio students-teachers 24.92 3.24 26.16 28.78
Dropout rates 0.09 0.07 0.07 0.06
2. Follow-up (2003)
Localization
1 0.00 0.05 0.10 0.30
2 0.00 0.00 0.01 0.10
3 0.00 0.00 0.05 0.22
4 0.04 0.19 0.06 0.24
5 0.20 0.40 0.15 0.36
6 0.00 0.00 0.04 0.19
7 0.15 0.36 0.30 0.46
8 0.45 0.50 0.12 0.32
9 0.16 0.37 0.17 0.38
Gender of the school
Male 0.00 0.00 0.07 0.26
Female 0.00 0.00 0.03 0.16
Coed 1.00 0.00 0.90 0.30
Type of secondary
Academic 0.80 0.40 0.63 0.48
Technical 0.06 0.24 0.29 0.45
Both 0.14 0.35 0.07 0.26
Table 9.2
(continued)
Shift
Complete 0.04 0.20 0.03 0.18
Morning 0.28 0.45 0.40 0.49
Afternoon 0.07 0.25 0.14 0.35
Evening 0.61 0.49 0.43 0.49
Number of students
Total school 2544.18 786.25 2954.60 1376.76
Taking exam 136.97 69.02 122.15 79.16
Old for grade 35.52 35.30 47.28 45.15
Repeating grade 114.93 88.69 103.30 73.72
Years of teacher education 16.49 0.73 16.07 2.28
Dropout rates
Grade 1 0.035 0.04 0.040 0.05
Grade 2 0.030 0.03 0.031 0.04
Grade 3 0.030 0.03 0.026 0.04
Grade 4 0.018 0.02 0.026 0.03
Grade 5 0.021 0.02 0.020 0.33
Grade 6 0.054 0.06 0.071 0.09
Grade 7 0.029 0.03 0.040 0.04
Grade 8 0.051 0.04 0.038 0.04
Grade 9 0.046 0.03 0.039 0.05
Grade 10 0.040 0.03 0.033 0.04
Grade 11 0.012 0.02 0.012 0.02
Number of schools 23 416
ranging from the nearest neighbor to kernel estimators, that can be

used to determine the counterfactual. In this chapter the estimation is
based on the 10 nearest neighbors matching estimator, with a caliper
of 0.01; that is, the counterfactual is made of only those observations
with a propensity score within 0.01 of the propensity score of the
treated observation. All variables outside the common support were
excluded (see Vinha 2005). This analysis also checks the robustness of
the estimation. As a first approximation, results for the dropout rate
are quite stable and independent of the type of estimator; on the con-
trary, the results for the impact on test scores vary with the type of
estimation.
Table 9.3
Direct and Indirect Impact of Concessions: Effects over Dropout Rates
Direct impact over Indirect impact over
dropout rate dropout rate
Concession Schools near concession
Probit estimates Standard Standard
Depended variable Coefficient error Coefficient error
Total number of students 0.008 0.001 0.015 0.003
Dispersion in the test 0.366 0.080 0.708 0.178
Localization 4 13.472 8.162
Localization 5 0.024 0.168 9.129 8.556
Localization 7 0.212 0.138 13.929 8.089
Localization 8 0.052 0.163 12.411 8.234
Localization 9 0.048 0.155 12.455 8.311
Evening shift 0.243 0.173 1.966 0.430
Average no. of years of teacher 0.035 0.025 0.080 0.079
education
% of students repeating 1.723 0.351 1.407 0.502
% of students old for grade 3.327 0.719 5.040 1.709
Number of computers, 1999 0.027 0.020
Number of classrooms, 1999 0.053 0.038
Number of restrooms, 1999 4.686 0.719
Number of sport facilities, 1999 1.723 0.260
Number of teachers, 1999 0.476 0.278
Number of administrative 0.202 0.050
personnel 1999
Number of psychologist, 1999 0.069 0.187
Ratio teacher-stud., 1999 0.059 0.038
% teachers with second, 1999 35.714 10.066
% teachers with college, 1999 33.968 9.841
% teachers with graduate 25.912 8.898
studies, 1999
Log likelihood 529 231
Pseudo R 2 0.1391 0.5714
Number of observations 2334 1217
Table 9.3
(continued)
Direct impact over Indirect impact over
dropout rate dropout rate
Impact variable: Impact variable:
Matching dropout rate dropout rate
Variable: dropout rate
Difference treatment-control
Unmatched 0.0130 0.0069
ATT 0.0173 0.0082
Bootstrap statistics
Repetitions 1000 1000
Standard error 0.0039 0.0096317
Bias-corrected confidence 0.0244 0.0112 0.0341 0.0004
interval
The results for the direct and indirect effects on dropout rates are pre-
sented in Table 9.3. The unit of observation is the dropout rate by grade
by school. The first two columns present the results from the estimation
of equation (9.3) for concession schools, whereas columns three and
four present the results for the indirect impact on nearby regular public
schools. The estimation uses data at the school and grade level.
In general, the coefficients of the probit estimation to determine the
propensity scores are the move in the expected direction for both types
of impacts. I tested for the balance of characteristics across treated and
control schools. The groups are balanced in all variables except two
(number of students and proportion of students in grades lower than
their age). In order to improve the balance in the two samples, I omit-
ted observations with the extreme propensity scores from the analysis.
In any case, one variable remained unbalanced.
The lower part of table 9.3 presents the impact of the program on
dropout rates. The results show that concession schools (17 in the sam-
ple) have lower dropout rates (with a difference of 1.7 points) than sim-
ilar public schools (2,790 public schools). Given the range of dropout
rates (between 1.2% and 7.1%), this is an important and large effect. In
order to find the standard error for the estimator, a bootstrap proce-
dure was performed with 1,000 repetitions. The effect is statistically
different from zero at the 90% level of significance.
Columns three and four present evidence of the indirect impact.
That is, the probit is run with a dependent variable that equals one for
regular public schools near concession schools and zero for public
schools outside the influence of concessions. Again, in general, the pro-
bit estimates move in the expected direction and several of them are
statistically significant. The estimated indirect impact of concession
schools on nearby public school dropout rates is a reduction of 0.008
points. However, the standard error for the estimator (using a boot-
strap procedure with 1,000 repetitions) indicates that the 90% confi-
dence interval is between 0.03 and 0.0004. Thus, I cannot rule out
that there was no impact.
Table 9.4 presents estimations of the direct impact of concession
schools on the standardized test scores in mathematics and reading.
The data are at the individual level. As in the previous regressions, the
majority of the coefficients of the independent variables in the probit
estimation move in the expected direction, and several of them are
statistically significant, with one caveat: in contrast to the previous esti-
mation, there are several variables that are not balanced in the treat-
ment and control individuals (results are not shown). To address this,
I reduced the range of propensity scores in which the matching is per-
formed, with some improvements on the balance of the samples. The
implication is, again, that the treatment impact is for those individuals
within this range of propensity scores and not for the whole popula-
tion. The results hold for both estimations (math and reading).
The calculated impact is positive and statistically significant for both
reading and mathematics test scores. Mathematics scores for conces-
sion schools are almost one point higher than similar regular public
schools and the effect is significantly different from zero. The impact
on reading scores for concession schools is higher, with an estimated
impact of almost 2 points. Given that the average scores are 42.08 and
50.99, the results imply improvements of 2.4% and 4%, respectively.
It is important to note that a simple comparison between conces-
sion schools and other public schools shows that concession schools
have, on average, lower math tests. Indeed, the unmatched impact is
0.2099. However, once the estimation controls for the observable
characteristics, the impact of concessions on the score is positive and
significant.
9.4 Conclusions
In conclusion, there is strong evidence of a direct impact of concession

schools in reducing dropout rates. There is also some evidence of an
Table 9.4
Direct Impact of Concessions: Effects on Test Scores (Mathematics and Reading)
Math tests Reading tests
Concession Concession
Probit estimates Standard Standard
Depended variable Coefficient error Coefficient error
Total number of students 0.018 0.001 0.018 0.001

Dispersion in the test 0.351 0.033 0.197 0.029
Localization 5 0.441 0.079 0.542 0.080
Localization 7 0.149 0.063 0.140 0.064
Localization 8 0.073 0.075 0.079 0.076
Localization 9 0.150 0.071 0.181 0.072
Evening shift 0.102 0.072 0.068 0.073
Average years of teacher 0.089 0.012 0.095 0.011
education
% of students repeating 2.508 0.183 2.575 0.179
% of students old for grade 19.165 0.779 18.789 0.774
Constant 4.683 0.300 4.474 0.330
Log likelihood 2152.87 2846.556
Pseudo R 2 0.2768 0.268
Number of observations 18629 18630
Impact variable: Impact variable:
Matching mathematics reading
Variable: test scores
Difference treatment-control
Unmatched 0.2099 0.6234
ATT 0.9732 1.9364
Bootstrap statistics
Repetitions 100 100
Standard error 0.522 0.748
Bias-corrected confidence 1.2684 1.4655 1.5501 2.8472
interval
indirect impact of the concession schools on the dropout rates in

nearby regular public schools. Furthermore, there is evidence of a posi-
tive impact on test scores of students in concession schools when com-
pared with students in other public schools.
Moreover, as discussed previously, there is some evidence of down-
ward bias in the estimations. Indeed, OLS estimators are lower than
the matching estimators. This finding strengthens one important idea
of this chapter: given that propensity and matching estimators correct
only partially the sources of bias, presumably the effects of concessions
are larger than the ones presented here.
The results for dropout rates are especially important in light of the
current situation in the country. The enrollment rate in secondary edu-
cation in the cities is reaching levels of 85% but the dropout rates are
higher in the transition from primary to secondary education. Con-
cession schools seem to be a promising intervention for reducing
desertion from schools at this critical juncture. However, as public in-
vestment in each concession school is around $2.5 million, it is impor-
tant to perform a cost-benefit analysis.
The test score results are also promising, especially considering that
they are just the initial impacts. The concession schools program
started in 1999, four years before the test score data were collected.
Individuals who took the exam were in grade 11, and therefore, they
most likely transferred into the concession schools in grade 8. Presum-
ably, the impact on individuals who start in a concession school from
grade one onward will be higher.
Instruments for improving the quality of education are limited. In
fact, the educational sector in Colombia, and most likely in several
other countries, is subject to inflexible policies adopted in the past
under different conditions. Concession schools are an option that may
be able to generate consensus as a positive measure. However, the po-
tential scale of any such program may be limited. The program relies
on private, high-quality schools to manage public schools. Clearly,
there are only a limited number of such schools and, of those, even
fewer may participate in the program.
Notes
Formerly deputy director of Fedesarrollo (Colombia); now senior education economist,

HDNED, World Bank. The World Bank provided financial support for this project. The
secretaria de educación del distrito provided generous help and data. I would like to
thank the National Planning Department (Departamento Nacional de Planeaion, DNP)

for providing some of the data used in this article. I am indebted to Rajashri Chakrabarti,
Ronald Fryer, Brian Jacob, Harry Patrinos, Katja Vinha, and two anonymous referees for
very useful comments. The findings, interpretations, and conclusions expressed in this
chapter are entirely those of the author. They do not necessarily represent the views of
the International Bank for Reconstruction and Development/World Bank and its affili-
ated organizations, or those of the executive directors of the World Bank or the govern-
ments they represent.
1. Nelson and Hollenbeck (2001) present a critical view of Solmon and Goldschmidt.
2. The basic reference for the discussion of the problem is Heckman, LaLonde, and Smith
(1999). This section is based on chapter 2 in Vinha (2005), which presents a concise expla-
nation of the problem and the estimation using propensity scores and matching.
3. This approach was suggested by an anonymous referee.

4. Moreover, the information on who applies to a concession school is not available to
the public.
5. One area of future research will be to test for robustness of the results using another
radius for the area of influence.
6. The data identifies nine main geographical areas (subdivisions of location of the
school in the north, south, west, and east parts of the city).
References
Barrera, F. 2003. Decentralization and education: An empirical evaluation. Ph.D. disserta-

tion, Department of Economics, University of Maryland College Park.
Barrera, F., and C. Dominguez. 2005. Educación básica: Opciones futuras de polı́tica.
Work in process for the Misión para el Diseño de una Estrategia de Reducción de la
Pobreza y la Desigualdad, Fedesarrollo.
Borgas, G., and O. L. Acosta. 2005. Educational reform in Colombia. In Institutional
Reforms, ed. A. Alesina. Cambridge, MA: MIT Press.
Cerdan, P., and C. Vermeersch. 2006. More time is better: An evaluation of the full time
school program in Uruguay. Impact Evaluation Working Paper Series, World Bank,
Washington, DC.
Corpoeducación. 2004. Evaluation of concessions program. Unpublished document,
Bogotá.
Duarte, J. 1996. La debilidad del Ministerio de Educación y la politización de la educa-
ción: dos problemas a enfrentar en el plan decenal. Coyuntura Social 14: 145–167.
Friedman, M. 1955. Role of government in education. In Economics and the Public Interest,
ed. Robert Solo. New Brunswick, N.J.: Rutgers University Press.
Gaviria, A., and J. H. Barrientos. 2001. Caracterı́sticas del plantel y calidad de la educa-
ción en Bogotá. Coyuntura Social 25: 81–98.
Hanushek, E. A. 1986. The economics of schooling: Production and efficiency in public

schools. Journal of Economic Literature 24, no. 3: 1141–1177.
Hanushek, E. A. 1996. Measuring investment in education. Journal of Economic Perspectives

10, no. 4: 9–30.
Hanushek, E. A., J. F. Kain, S. G. Rivkin, and Gregory F. Branch. 2005. Charter school
quality and parental decision making with school choice. NBER Working Papers 11252,
National Bureau of Economic Research, Cambridge, Mass.
Heckman, J., H. Ichimura, J. Smith, and P. Todd. 1998. Characterizing selection bias using
experimental data. Econometrica 66, no. 5: 1017–1098.
Heckman, J., R. LaLonde, and J. Smith. 1999. The economics and econometrics of active
labor market programs. In Handbook of Labor Economics, ed. O. Ashenfelter and D. Card.
Amsterdam: North Holland.
Hoxby, C. 2002. School choice and school productivity (or, Could school choice be a tide
that lifts all boats?). NBER working paper no. 8873, Cambridge, MA.
Hoxby, C., and S. Murarka. 2007. New York City’s charter schools overall report. Cam-
bridge, MA: New York City Charter Schools Evaluation Project, June.
Hoxby, C., and J. Rockoff. 2004. The impact of charter schools on student achievement.
HIER working paper, Harvard University, Cambridge, MA.
Nelson, C., and K. Hollenbeck. 2001. Does charter school attendance improve test
scores?: Comments and reactions on the Arizona achievement study. Staff Working
Papers 01-70, W. E. Upjohn Institute for Employment Research, Kalamazoo, MI.
Rodrı́guez, A. 2005. Case Study: Public school concession model of Bogotá, Colombia.
Unpublished document, the World Bank.
Sarmiento, A., L. Becerra, and J. González. 2000. La incidencia del plantel en el logro edu-
cativo del alumno y su relación con el nivel socioeconómico. Mayo de 2004, borrador,
DNP.
Sarmiento, A., C. E. Alonso, G. Duncan, and C. A. Garzon. 2005. Evaluación de la gestión
de los colegios en concesión en Bogotá 2000–2003. Documento de trabajo, Departamento
Nacional de Planeaion, Bogotá.
Solmon, L., and P. Goldschmidt. 2004. Comparison of traditional public schools and
charter schools on retention, school switching and achievement growth. Policy report,
no. 192, Goldwater Institute.
Villa, L., and J. Duarte. 2005. Concessionary public schools in Bogotá: An innovation in
school management in Colombia. In Private Education and Public Policy in Latin America,
ed. L. Wolff, J. C. Navarro, and P. Gonzalez. Washington, DC: PREAL.
Vinha, K. 2005. The impact of the Washington Metro on development patterns, Ph.D. dis-
sertation, Department of Agricultural and Natural Resource Economics, University of
Maryland, College Park.
World Bank. 2005. Colombia: Contracting education services. Report no. 31841–CO. Colom-
bia and Mexico Country Management Unit, Education Unit, Human Development De-
partment, Latin America and the Caribbean Region. Washington, DC: World Bank.
10 Public and Private Schooling Initiatives in
England
While a lot of media and government policy attention has recently

been devoted to public-private initiatives in the English schooling mar-
ket, there is remarkably little research on the issue in the economics of
education field. This is perhaps surprising since the English schooling
system has, since the move to a quasi-market that began in the late
1980s (LeGrand 1991, 1993), been characterized by a lot of enthusiasm
directed towards more choice, more competition, and an increased
role for the market. In this chapter we address this question, first offer-
ing a general background discussion looking at the private provision
of schooling and how it has adapted over time, and second looking in
more detail at a specific, high-profile attempt to bring some aspects of
private sector ethos to pupils in the state sector.
There has been a long history of private education in England, but
over almost its entire history it has stayed largely independent of what
goes on in the state sector. Of course, it is not easy to define the charac-
teristics of schools in the private sector. In historical terms the classic
English ‘‘public schools’’ like Eton or Harrow generated the well-known
image of private education that Gordon, Aldrich, and Dean (1991) de-
scribe as ‘‘providing the sons of the commercial and entrepreneurial
classes with the manners of a gentleman. Definition, it appeared, came
from within: an acceptance, a recognition, an instinct of others who
had received a training which could be labelled ‘public school’’’ (p.
200). Over the twentieth century the private school (i.e., public schools
and other fee-charging schools) stayed firmly present in the English
education system. But they retained their independence: for example,
currently, private school pupils do not take the same examinations as
state school pupils, unless the school chooses that they do so.
Only relatively recently have there been serious attempts to bring
significant private sector involvement to the state sector in English
220 Stephen Machin and Joan Wilson
education. There are several reasons why this has become a popular
venture. First, the development of a quasi-market in education has po-
litical currency.1 Its link to incentives and to the perceptions of better
performance in the private sector lends itself naturally to the education
market. Second, in the last couple of decades there has been an up-
surge of public-private ventures in a range of industries in the United
Kingdom. Third, teacher unions have weakened, and traditionally they
would have opposed private finance initiatives and public-private
partnerships.
The chapter has two main parts. Section 10.1 discusses private edu-
cation in England. Section 10.2 looks at a particularly high-profile pri-
vate finance initiative that has been introduced to the English state
school sector recently, the city academies program. Section 10.3 offers
some conclusions.
10.1 Private Education in England
10.1.1 Brief History and Legislation

Private education has been important in England for a long time and
remains so today, despite the provision of state education for all facili-
tated through some of the significant education acts of the twentieth
century. The 1918 Education Act, for example, abolished any exemp-
tions to the compulsory minimum school-leaving age (then imposed at
age fourteen) and removed all fees in elementary schools. The 1944 Act
established a nationwide system of free, compulsory schooling from
age five to fifteen (with the compulsory minimum school-leaving age
raised from 14 to 15 in 1947), and organised the system into primary
and secondary schools.2
The 1988 Education Reform Act set the scene for the quasi-market
for schooling. The major provisions of the 1988 Act were to set up the
National Curriculum, to introduce testing and league tables, to offer
local management of schools, and to increase accountability (through a
regular inspection regime and by changing the nature of school gov-
erning bodies3). The act also set up grant-maintained (GM) schools
that were allowed to select up to 10 percent of their pupils on the basis
of ability or aptitude; and city technology colleges (CTCs), the first at-
tempt to bring the private sector into the state sector, as CTCs are
partially funded by private sector business.
The early acts, of course, had an impact on the extent of private
schooling in England and the 1988 Act clearly gave scope for the intro-
Public and Private Schooling Initiatives in England 221
Table 10.1
State and Private Schools in England, 2006
All1 Private Private Share
Number of schools 24091 2199 .091

Number of FTE pupils 7985349 595317 .075
Number of FTE teachers 459863 64503 .140
Pupil-teacher ratio 19.1 9.8 n/a
Notes: Authors’ own calculations using various data sources.
1. Includes primary, secondary, special, and private (independent) schools.
duction of the kinds of private finance initiatives and public-private

partnerships that have been a feature of English schools in the recent
past. What we consider next in this section is the extent of private
education in England and how outcomes for pupils attending private
schools differ from those in the state sector. This is important to estab-
lish to see why some advocate that elements of the private school
‘‘ethos’’—such as those that relate to financial and governance au-
tonomy as well as teaching practice—should be adopted in the state
sector.
10.1.2 The Extent of Private Education

According to the Department for Children, Schools and Families
(DCSF) statistics, just over 9 percent of schools are private (indepen-
dent) schools: table 10.1 reports that in 2006 there were 24,091 schools
in England, of which 2,199 were private. Figure 10.1 shows the evolu-
tion of the share of private schools through time. It was higher at
around 12.5 percent of schools in the early 1960s, dipping down to 8
percent by 1980 and then gradually rising through the 1990s and early
2000s to its current level of around 9.5 percent from 2004 onward.
Around 7 percent of pupils (around 600,000 out of 8 million) cur-
rently attend private schools, with the participation rate being higher
for secondary school pupils. Figure 10.2 shows changes over time in
the pupil share, which fell from 8 to 6 percent between the early 1960s
and mid-1970s, after which it climbed through the 1980s to around 7
percent by the early 1990s. Since then it has stabilized around this level
(7.6 percent in 2006).
There are significant regional variations in pupil participation and
the presence of private schools. Table 10.2 shows that only 3 percent of
schools in the North East are private, while in Inner London the share
Figure 10.1
Share of Private Schools in Total Schools, 1963 to 2006
Source: Calculated from various issues of Statistics of Education—Schools in England.
Figure 10.2
Share of Private School Pupils in Total School Pupils, 1963 to 2006
Source: Calculated from various issues of Statistics of Education—Schools in England.
Table 10.2
Regional Variations, 2006
Pupil-
Number Number teacher
Number of of pupils ratio in
of private Private Number in private Private private
schools schools share of pupils schools share schools
England 24091 2199 .091 7985349 595317 .075 9.8

North East 1235 38 .031 396799 16396 .041 12.3
North West 3450 225 .065 1106137 57661 .052 16.0
Yorkshire and 2432 135 .056 806575 32975 .041 13.8
Humberside
East Midlands 2225 142 .064 685052 37959 .055 9.4
West Midlands 2565 180 .070 883010 45945 .052 8.4
East of England 2800 215 .077 886343 63152 .071 8.4
Inner London 1127 237 .210 417383 65030 .156 9.0
Outer London 1729 244 .141 752668 70659 .094 9.7
South East 3930 547 .139 1285245 143891 .112 8.3
South West 2598 236 .091 766137 61649 .080 8.4
Note: Authors’ own calculations using various data sources.
is just above 20 percent. In terms of pupil participation, the range from

lowest to highest goes from 4 to about 16 percent. It is undoubtedly the
case that the private school sector is more heavily concentrated in the
wealthier parts of London and the South East of England.
10.1.3 Characteristics of Private Versus State Schools

Other than their fee-charging and charitable status,4 private schools
differ significantly from state schools along a number of dimensions.
The most obvious difference concerns teaching resources. There are far
more teachers for each pupil in private schools: the pupil-teacher ratio
is about 16.8 for state secondary schools and only 9.8 for private
schools (table 10.7 and table 10.1 [or 10.2] respectively).
Table 10.3 broadly summarizes differences in teacher characteristics
between the private and state sectors (drawing on the evidence of
Green et al. 2008). The first row restates the fact that pupil-teacher
ratios are markedly lower in private schools. The second, third, and
fourth rows consider teacher salaries, hours, and holidays. It is interest-
ing that salaries are, if anything, lower and hours are longer for private
school teachers. The most revealing observation may be that holidays
Table 10.3
Comparison of Teacher Characteristics in the Private Relative to the State Sector
Teacher characteristic Private against state Evidence
Pupil-teacher ratios Lower DCSF* statistics of education

Teacher salaries If anything, lower Labor Force Survey wage equations
(Green, Machin, Murphy, and Zhu 2008)
Teacher hours Higher Labor Force Survey hours equations
(Green, Machin, Murphy, and Zhu 2008)
Teacher holidays Longer Labor Force Survey paid holiday
entitlement equations (Green, Machin,
Murphy, and Zhu 2008)
Note: * Department for Children, Schools, and Families (DCSF), formerly known as the
Department for Education and Skills (DfES).
are longer, showing that teachers probably like the better work condi-
tions and holiday arrangements in private schools.
10.1.4 Outcomes for Private versus State School Students

Table 10.4 considers outcomes for private versus state school pupils.
On an observational level, children attending private schools do end
up with higher academic qualifications, are more likely to attend uni-
versity, and get paid more in the labor market than children from
the state sector. This has led to a widespread perception that private
schools do get better examination results for children who attend them.
It is also the case that, unlike most studies of state schools, some re-
search has identified a role for better resourcing to shape pupil out-
comes within the private sector. Graddy and Stevens (2005) note that
lower pupil-teacher ratios are an important determinant of fees in pri-
vate schools and are able to identify bigger resource effects in their
study of private schools in the U.K., showing that schools with lower
pupil-teacher ratios do in fact deliver better examination results.
Irrespective of any interpretation one might want to place on the
superior outcomes for private school children, it is probably the combi-
nation of the observation/perception that private school pupils ‘‘do
better’’ than state school pupils, together with the increased market-
ization of education, that has led to growing interest in transferring
private-sector-autonomous practices to the state sector. Compared to
other sectors in the U.K., where private sector financing and involve-
ment have been fairly substantial, this remains relatively new in the
education field. We consider some of the recent developments in the
next section of the chapter.
Table 10.4
Academic and Labor Market Outcomes for Private School Pupils Relative to State School
Pupils
Private
Outcome against state Evidence
GCSEs (exams taken Higher DCSF, Naylor, Smith, and McKnight
at age 15/16) (2002)
A Levels (exams taken
at age 17/18)
University attendance Higher In 2003, 13.2 percent of new first-year
students from independent schools
(Teaching Quality Information
National Student Survey)
Labor market earnings Higher Naylor, Smith, and McKnight (2002),
(among graduates) Green, Machin, Murphy, and Zhu
(2007)
10.2 Public-Private Initiatives—City Academies
10.2.1 Private Sector Involvement in the State Sector

In the area of education in England, there have been two main forms of
private sector involvement in the state sector: private finance initiatives
(PFIs) and public-private partnerships (PPPs). The former try to bring
aspects of private sector business more directly into the provision of
assets in the public sector. For example, PFI projects in schools may in-
volve the purchase of services from the private sector such as school
buildings, facilities such as sports halls, or specific services including
heating systems, information and communications technology (ICT),
or catering equipment. They are much more concerned with non-
academic operations (thus sometimes referred to by practitioners as
‘‘chore’’ not ‘‘core’’). In terms of functioning, private sector companies
bid for a PFI contract and, as for traditional procurement, this is usu-
ally a competitive tendering process to ensure value for money.
Public-private partnerships, on the other hand, encompass a wide
range of activities in which the public and private sectors work to-
gether, including joint ventures. The TeacherNet Web site describes
PPPs as being ‘‘about more than money. They are about improving the
services that the public sector provides. In terms of schools, that trans-
lates directly to raising educational standards.’’5
In the English schooling system, one of the most high-profile
partnerships with private sector co-funding has been the recent city
academies program. In some ways academies represent a mixture or

hybrid of both PFI and PPP in that they are a joint venture between
the state and their private sector sponsors, where the latter are directly
involved in the provision of education services and assets. We consider
the academies program next, both in terms of its nature and its capac-
ity to enhance pupil performance.
10.2.2 The City Academies Program

City academies are new independent state schools established on a
partnership or joint venture basis between central government and pri-
vate sector sponsors. So far the academies program has mainly been
applied to schools catering to the secondary school stage of education
(i.e., where pupils are beyond the age of 11 and 12). They are ‘‘new’’
schools in the sense of being newly classified as academies, and most
involve the development of a new school building. However, in
most cases an academy’s new physical facility is actually established
on the same land site as a school that the academy replaces (often a
failing urban school). According to the Academies Sponsor Prospectus
(2005), their ‘‘buildings and facilities—either new build or remodelling
of an existing school building—are financed and built in partnership
between the sponsor and the government.6 Their annual revenue fund-
ing comes entirely from the government at a level comparable to other
[local] schools. No fees are paid by parents.’’ The private sector spon-
sors can originate from a range of areas (e.g., business, faith groups,
individuals). Sponsors delegate management of the school to a largely
self-appointed board of governors or directors.7 In comparison with
state schools maintained by Local Education Authorities (LEA), ‘‘Aca-
demies governing bodies employ all Academy staff. The governing
body is responsible for agreeing [on] levels of pay, conditions of ser-
vice with its employees, as well as policies for staffing structure, career
development, discipline and performance management.’’8 The acad-
emy is expected to specialize in at least one subject area, chosen ini-
tially at the discretion of the sponsor, but with local community
interests considered. The most notable distinctions between academies
and other state schools are therefore their autonomy and the existence
and degree of involvement of their private sector sponsors.9 Table 10.5
summarizes public-private sector involvement in the delivery of serv-
ices and the financing of education in England. The table highlights
that academies are, in essence, a minority private/majority public
financed initiative.
Table 10.5
Delivery and Financing by Public and Private Actors in Education
Financing of initiative
Minority private/
Public majority public Private (capital)
Delivery of service
Public Management of Specialist schools1
(maintained) state
schools
Private Contracting-out Academies; Some Private finance
services LEA services initiatives
Public and private Supply teachers Education Action Contracting-out
(when needed) Zones2 services
Notes:
1. Specialist schools are maintained secondary schools that follow the National Curricu-
lum and specialize in a designated subject area. A small amount of financing must be
raised through private sector sponsors, while the majority of funding comes in the form
of additional government grants in excess of that received by nonspecialist state schools.
2. Education Action Zones involve a cluster of around 15 to 25 schools that adopt inno-
vative practices to raising education standards in disadvantaged areas. For this they re-
ceive around £1 million in additional funds (approx 75% of which comes from central
government and 25% is raised through private sector sponsors).
Source: Adapted from Gadkowski (2007, table 2).
There are some similarities, but also notable differences, between city
academies and U.S. charter schools. The latter are public (i.e., state)
schools set up from charters typically drawn up by groups of educa-
tors, parents, and community leaders (some are converted from exist-
ing public schools, although a small number were once private
schools). Like U.S. district public schools and U.K. city academies,
they receive public funding according to the number of students
attending (although, in a number of states, they do not receive the full
equivalent of their district counterparts). Unlike traditional district
schools, most charter schools do not receive funding to cover the cost
of premises and operating. Most rely on independent means to fund
their capital costs (even if some states may offer financial help with
startup costs). It is probably fair to say that, at least currently, acade-
mies remain more closely tied to the state system than do the majority
of U.S. charter schools.
The city academies program was introduced by David Blunkett, the
then secretary of state for education, in a speech in March 2000, and
the first projects were announced in September 2000. The first three
academies were established in September 2002. Nine more opened in
Table 10.6
City Academies Opened in September 2002 to September 2005
228
Academy Region Opening date Private sector involvement Specialization
The Business Academy, Bexley Outer London September 2002 Garrard Education Trust—£2.5 Business and enterprise
million
Greig City Academy, Haringey Inner London September 2002 Greig Trust and Church of Technology (especially
England £1.7 million ICT)
Unity City Academy, Middlesbrough North East September 2002 Amey plc—£1.9 million Applied enterprise
Capital City Academy, Brent Outer London September 2003 Sir Frank Lowe—£2 million Sports and art
The City Academy, Bristol South West September 2003 Bristol City Football Club and the Sports
University of the West of England
£1.7 million
The West London Academy, Ealing Outer London September 2003 Reed Executive plc £2 million Sports and enterprise
Manchester Academy, Manchester North West September 2003 United Learning Trust (The Business and enterprise,
Church Schools Company), and art
Manchester Science Park Ltd—
£1.3 million
The King’s Academy, Middlesbrough North East September 2003 Emmanuel Schools Foundation— Business and enterprise
£2 million
Djanogly City Academy, Nottingham East Midlands September 2003 Sir Harry Djanogly ICT
City of London Academy, Southwark Inner London September 2003 Corporation of London—£2 Business and enterprise
million and sports
The Academy at Peckham, Southwark Inner London September 2003 Lord Harris of Peckham—£3 Business and enterprise,
million and performing arts
Walsall City Academy West Midlands September 2003 Mercers’ Company and Thomas Technology
Telford Online—£2.5 million
London Academy, Barnet Outer London September 2004 SGI Limited—£1.5 million Business, enterprise, and
technology
Mossbourne Community Academy, Inner London September 2004 Sir Clive Bourne, Seabourne Technology
Hackney Group plc £1.6 million

Stockley Academy, Hillingdon Outer London September 2004 Insinger Townsley Stockbrokers Science and technology
and others—£1.5 million
Lambeth Academy, Lambeth Inner London September 2004 United Learning Trust—£1.5 Business and enterprise,
million and languages
Northampton Academy, Northampton East Midlands September 2004 United Learning Trust—£0.7 Sports, business, and
million enterprise
Trinity Academy, Doncaster Yorkshire and September 2005 Emmanuel Schools Foundation— Business and enterprise
The Humber £2 million
St Paul’s Academy, Greenwich Outer London September 2005 Roman Catholic Diocese of Sports and enterprise
Southwark—£0.2 million
Salford City Academy, Salford North West September 2005 United Learning Trust and Sports, business, and
Salford Diocese—£0.2 million enterprise
Marlowe Academy, Kent South East September 2005 Roger De Haan and Kent County Business and performing
Council—£2.2 million arts
The Harefield Academy, Hillingdon Outer London September 2005 David Meller and others—£0.4 Sports
million
Haberdashers Aske’s Knights Inner London September 2005 Haberdashers’ Livery Company ICT and sports science
Academy, Lewisham £0.3 million
Public and Private Schooling Initiatives in England
Haberdashers Aske’s Hatcham Inner London September 2005 Haberdashers’ Livery Company ICT and music
College Academy, Lewisham £0.7 million
Dixons City Academy, Bradford Yorkshire and September 2005 Dixons CTC Trust £0.6 million Performing arts and
The Humber product design
Academy of St. Francis Assisi, North West September 2005 Diocese of Liverpool and The Environment
Liverpool Roman Catholic Archdiocese of
Liverpool—£1 million
Macmillan Academy, West North East September 2005 Macmillan CTC £0.4 million Science and PE, and
Middlesbrough outdoor education
229
Note: See hhttp://www.standards.dfes.gov.uk/academies/projects/openacademies/?version=1i

Table 10.7
City Academies, 2006
All City City academies
secondary academies share
Number of schools 3358 27 .008
Number of FTE pupils 3298185 25196 .008
Number of FTE teachers 199131 1717 .009
Pupil-teacher ratio 16.8 14.8 n/a
Note: Authors’ own calculations using various data sources.
September 2003, five in September 2004, and a further ten in Septem-

ber 2005. Table 10.6 lists details of the twenty-seven academies set up
in those years.10 As can be seen from the table they cover quite a di-
verse range of specializations and sponsors, and are located in various
regions of the country. It is noteworthy that the places where they have
been introduced are, for the most part, characterized by social dis-
advantage and poor educational attainment. Table 10.7 also shows
that they are currently covering only a small (less than 1 percent) share
of schools and pupils, but many more are planned in the near future
(see note 10 at the end of this chapter).
The explicit aim from government is that city academies are being
introduced to raise educational standards. The presumption is that
standards will be raised through the innovative nature of the academy
culture and the skills brought to the academy by sponsors (e.g., spon-
sors may play a role in developing the curriculum). It is argued that
this will facilitate better management and governance, which in turn
will lead to improvements in educational attainment.
Some evaluation of the initial phases of the city academies program
has been conducted by PriceWaterhouseCoopers (PwC) (2007). (As
part of a five-year evaluation, PwC consultants have produced four an-
nual reports, in 2004, 2005, 2006, and July 2007). They discuss their
findings as early and ongoing work informing the progress of the
academies program. The 2007 report considers each of the twenty-
seven academies opening between September 2002 and September
2005. On the basis of their fieldwork they conclude that they are able
to identify positive endorsements of sponsors, positive feedback from
pupils, and evidence of innovative teaching and management styles
being implemented.
PwC has also carried out quantitative analysis of pupil performance,
by comparing improvement in final school-year exams (the General
Certificate of Secondary Education, or GCSE11) in academies with the

national average and with some selected groups of schools. They sum-
marize their evidence by stating that ‘‘academies’ progress in terms of
pupil achievement has generally exceeded corresponding improve-
ments at a national level and amongst other similar schools’’ (2007, ex-
ecutive summary, vii).
One serious point to note here is that the schools that were turned
into academies were typically located in socially deprived areas and
had very poor track records in terms of GCSE achievement. They were
among the worst performing schools in their respective LEAs, often
positioned right at the bottom of the stack. Comparing with the na-
tional average is not the right yardstick. Because of possible issues of
mean reversion (i.e., if they are at the bottom of the stack they are
likely to bounce back towards the mean), it is necessary to evaluate
their performance relative to comparable schools also characterized by
mean reversion. This is the approach we take in this chapter. Like PwC
we acknowledge that it is relatively early days yet in the academies
program and, given the very high profile of the academies, it is impor-
tant to get it right, but we take a first look at this in the next subsection.
10.2.3 A Start on Gauging the Impact of Academy Status on Pupil

Achievement
Conceptually, one could evaluate the impact of academy status on
pupil achievement if there were two identical schools (i.e., same levels
and trends in achievement prior to academy introduction) and one was
given academy status (and associated funding and autonomy) while
the other was not. Even if they are poorly performing schools but the
scope for mean reversion is the same, then a comparison of pupil
achievement before and after becoming an academy in the academy
school versus the non-academy school can provide an estimate of the
impact of becoming an academy on educational achievement.
Apart from one issue, to do with the pupil population changing (we
will return to this), a school-level analysis that compares schools that
become academies to a matched set of schools can provide a consistent
estimate. The issue is finding the matched schools. In our initial cut at
this, we have adopted two strategies. The first matches each academy
school with the nearest performing school via a one-to-one match on
pre-policy examination levels and trends in pupil achievement.12 In
the second, we also report results using all other secondary schools
in the academy’s LEA as a comparison group.
Table 10.8
School-Level Changes in GCSE Exam Performance in Schools Changing to Academy
Status and in Comparison Schools—% Getting 5 or More A*-C GCSEs
A. Academies Opening in September 2002
Post-academy,
Pre-academy, 2002–03 to
2000–01 2005–06 school Post-pre
school year years (averaged) change
Academies (3) 20.5 31.2 10.7
Matched schools (3) 21.7 36.2 14.5
D-i-D ¼
3:8 (10.1)
All other state schools in 37.2 48.4 11.2
LEA (23)
D-i-D ¼
0:5 (8.3)
National average 50.0 55.7 5.7
B. Academies Opening in September 2003
Post-academy,
Pre-academy, 2003–04 to
Academies (8) 23.4 37.5 14.1
D-i-D ¼
3.7 (9.5)
LEA (106)
D-i-D ¼
7.1 (7.2)
C. Academies Opening in September 2004

Post-academy,
Pre-academy, 2004–05 and
Academies (3) 24.0 37.7 13.7
D-i-D ¼
7.7 (13.8)
LEA (85)
D-i-D ¼
10.3 (8.9)
Table 10.8
(continued)
D. Academies Opening in September 2005
Pre-academy, Post-academy,
2003–04 2005–06 Post-pre
school year school year change
Academies (10) 46.0 55.6 9.6
D-i-D ¼
0.82 (16.9)
LEA (191)
D-i-D ¼
0.36 (13.1)
Note: Column 1: number of observations in parentheses. Three out of 27 academies are
new schools (one opening in 2003–04 and two opening in 2004–05). Three out of 27
academies each replace two predecessor schools (for one of the academies opening in
2002–03 and two opening in 2003–04). For one of the academy schools opening in 2002–
03 its two predecessor schools have the same match school. Column 4: D-i-D denotes
difference-in-difference. Robust standard errors in parentheses.
Table 10.8 carries out a descriptive analysis showing final-year

school examination performance in schools that become academies,
both as academies and in their predecessor school status. The analysis
is split into four panels, one for each of the newly opened cohorts of
academies for which we have examination data. The table shows a
pre-academy year of examination performance in column 1 (academic
year 2000–01 for the first cohort in the upper panel, through to 2003–
04 for the fourth cohort in the lower panel) and then subsequent per-
formance in the adjacent columns. The columns on the right show the
pre-post changes for schools that do become academies and for their
matched counterparts in each panel.
The table makes it evident that academies did improve their GCSE
performance after changing status. This is the case for all four cohorts,
with their improvement in GCSE performance rising between 9.6 per-
centage points (the fourth cohort) and 14.1 percentage points (the sec-
ond cohort).13
It is evident that, on average, these improvements look quite good
relative to changes in the national average. However, they look less im-
pressive when benchmarked against other poorly performing matched
state schools that did not become academies, but that were also prone
to mean reversion. This is because standards rose for the matched

schools as well, by between 6 percentage points (third cohort) up to
14.5 percentage points (first cohort). Therefore the difference-in-
differences (D-i-D) estimates in the table, which show the gap between
the pre-post change for academies relative to the pre-post change for
matched schools, are mostly positive but statistically indistinguishable
from one another. Qualitatively, the same pattern emerges if all state
schools in the academy’s Local Education Authority are used as the
comparator group. Overall there is some weak evidence of improved
performance in some of the cohorts, in numerical terms.
One important aspect that we have also considered is pre-academy
trends in the academy predecessor schools and the comparison
schools. These are shown in table 10.9. The (negative) gap between the
academy predecessor schools and the comparison schools was similar
and stable through time in the years prior to academy status, except
in the actual year before conversion where a dip in the gap seems
to occur. But this ‘‘permanent’’ longer-run difference is important to
consider.
Indeed, there is a serious need to carefully control for pre-policy
evolutions in GCSE scores for several years before academy status
occurred. We therefore report estimates of statistical difference-in-
difference models in table 10.10. The upper panel of the table shows
two sets of regression estimates for each of the four cohorts of academy
schools, using matched schools as comparators. The bottom panel
reports the same specifications for the broader set of control schools,
all state schools in the LEA. The two sets of regression estimates differ
in whether or not they also control for time-varying school characteris-
tics (log(school size), proportion eligible for free school meals, propor-
tion non-white).14
It is evident from the results in table 10.10 that benchmarking
against a longer pre-policy time period produces some positive acad-
emy effects. Given the small number of academies, statistical impreci-
sion is not so informative; although it is worth pointing out that the
magnitudes of the point estimates are fairly modest for three of the
four cohorts (especially once the control variables are included), but
there is clear evidence of significant positive effects for the third cohort
(the academies opening in September 2004). The pattern for most of the
cohorts is of no short-run effects of becoming an academy on GCSE
performance, but there is a positive effect for the one cohort.
Table 10.9
School-Level Year-on-Year Differences in the % of Pupils Attaining 5 or More A*-C GCSEs, 1995–96 to 2005–06
Academies opening in Academies opening in Academies opening in Academies opening in

September 2002 September 2003 September 2004 September 2005
Pre-academy– Pre-academy– Pre-academy– Pre-academy–
matched Pre-academy– matched Pre-academy– matched Pre-academy– matched Pre-academy–
schools and state schools schools and state schools schools and state schools schools and state schools
academy– in LEA and academy– in LEA and academy– in LEA and academy– in LEA and
matched academy–state matched academy–state matched academy–state matched academy–state
schools schools in LEA schools schools in LEA schools schools in LEA schools schools in LEA
differences differences differences differences differences differences differences differences
1995–96 5.7 (5.3) 22.3 (4.3) 2.8 (4.0) 14.9 (3.0) 8.7 (2.5) 24.6 (2.6) 2.4 (9.9) 0.2 (6.9)
1996–97 3.8 (3.9) 18.9 (2.8) 3.3 (4.1) 15.8 (2.8) 20.7 (3.1) 28.7 (2.0) 1.3 (10.7) 1.4 (7.9)
1997–98 7.7 (6.9) 22.4 (5.3) 4.6 (5.4) 16.1 (4.3) 4.0 (6.4) 26.0 (4.5) 4.4 (11.0) 2.9 (7.8)
1998–99 9.6 (0.8) 20.9 (3.0) 5.0 (5.0) 18.1 (4.0) 2.7 (6.4) 28.3 (4.5) 0.2 (11.0) 0.1 (7.4)
1999–00 7.9 (5.9) 21.8 (4.9) 7.2 (5.3) 17.1 (3.8) 4.7 (2.7) 26.2 (2.2) 2.7 (11.9) 1.5 (8.6)
2000–01 2.7 (5.4) 16.4 (4.6) 6.1 (5.2) 16.9 (4.0) 6.7 (3.4) 28.6 (2.1) 2.9 (12.1) 1.5 (8.9)
2001–02 Becomes academy 5.6 (5.6) 15.0 (4.5) 8.7 (8.6) 28.1 (6.0) 4.5 (12.8) 1.5 (9.5)
Public and Private Schooling Initiatives in England
2002–03 9.7 (7.5) 18.0 (5.5) Becomes academy 5.3 (10.8) 27.5 (6.6) 7.1 (13.6) 3.5 (10.5)
2003–04 6.7 (7.6) 20.0 (5.0) 3.4 (7.2) 11.7 (5.4) Becomes academy 3.4 (12.6) 4.0 (9.8)
2004–05 3.3 (12.4) 14.1 (10.1) 1.2 (8.7) 7.5 (6.4) 1.7 (10.0) 20.5 (7.0) Becomes academy
2005–06 0.3 (10.3) 10.8 (8.6) 3.0 (7.8) 7.7 (4.4) 3.0 (7.3) 15.0 (5.0) 4.3 (11.2) 4.4 (8.7)
Note: Robust standard errors in parentheses.
235
Table 10.10
School-Level Difference-in-Difference Estimates of City Academy Status on GCSE Performance, 1995–96 to 2005–06
236
A. Comparison with matched schools

(1) (2) (3) (4) (5) (6) (7) (8)
Becomes academy 0.49 1.57 3.80 0.01 10.00 8.95 1.86 0.15
(3.56) (3.88) (3.11) (3.42) (4.18) (5.01) (3.73) (3.85)
School fixed effects Yes (6) Yes (6) Yes (18) Yes (18) Yes (6) Yes (6) Yes (19) Yes (19)
Year dummies Yes (9) Yes (9) Yes (9) Yes (9) Yes (9) Yes (9) Yes (9) Yes (9)
Time varying controls No Yes No Yes No Yes No Yes
R-squared .80 .85 .68 .81 .64 .66 .95 .94
Number of schools 7 7 21 17 6 6 19 17
B. Comparison with all other state schools in LEA
(9) (10) (11) (12) (13) (14) (15) (16)
Becomes academy 3.02 3.58 8.45 4.12 8.27 4.58 3.01 2.86
(3.01) (3.44) (2.52) (2.71) (2.67) (3.36) (2.24) (2.14)
School fixed effects Yes (27) Yes (27) Yes (116) Yes (116) Yes (89) Yes (89) Yes (201) Yes (201)
Year dummies Yes (9) Yes (9) Yes (9) Yes (9) Yes (9) Yes (9) Yes (9) Yes (9)
Time varying controls No Yes No Yes No Yes No Yes
R-squared .84 .85 .84 .87 .87 .89 .87 .88
Number of schools 26 26 110 94 81 81 186 120
Note: Coefficient estimates (robust standard errors); Control variables are time-varying school characteristics—log(school size), proportion eligible
for free school meals, proportion nonwhite. GCSE performance is measured by the percentage of pupils getting 5 or more A*-C GCSE grades.
As noted, it is possible that the pupil mix may have changed as the
school changed status to become an academy. It turns out that pupil
mobility among students in predecessor schools is low and most who
were there at key stage 3 (aged thirteen/fourteen and in school year
nine) also take their GCSEs two years later in the academy. We have
therefore considered value added-models at the pupil level that are
able to address this possible issue of changing pupil mix. They are
reported in table 10.11. One difference is that, because of our availabil-
ity of pupil-level data, only three cohorts can be considered. Nonethe-
less, in the second and third cohorts (respectively those opening in
September 2003 and 2004) there are positive (small but mostly statisti-
cally significant) impacts on pupil value-added. Thus while standards
tended to not rise so much on average, there is some evidence that
value-added improved significantly in some of the schools that were
granted academy status.
10.3 Concluding Remarks
In this chapter we have considered some of the issues to do with

public-private partnerships in English education. We began by exam-
ining the role of the private sector in English schools and the increased
appetite for private sector ethos, autonomy, and attitudes to be
brought to state sector schools in the context of the quasi-market for
schooling. We presented some early empirical evidence on a high-
profile private-public joint venture, the city academies program. It is
premature for this joint venture to be appraised, and we find no evi-
dence of general positive effects on academic attainment from academy
status. However, there is substantial variation and there is some spe-
cific evidence of positive effects in some of the cohorts of new acade-
mies. That there are some positive short-run effects for some academies
is interesting in that, in the longer term, the pupil profile of schools
may change as well.
The academies program is evolving rapidly and it is likely that chil-
dren may need more exposure to the academies for substantial benefi-
cial effects on achievement to occur. It is evident that a very important
future research exercise on the role of private sector collaboration in
the state school sector will be to evaluate the impact of their more
widespread introduction on the academic performance of English
pupils.
Table 10.11
238
Pupil-Level Value-Added Models of Progression from Key Stage 3 to Key Stage 4 (GCSE ), 2001–02 to 2004–05
Dependent variable: Pr(5 or more A*-C GCSEs)
Academies opening in Academies opening in Academies opening in
September 2002 September 2003 September 2004
One-on-one Other state One-on-one Other state One-on-one Other state
matched schools matched schools matched schools
schools in LEA schools in LEA schools in LEA
(1) (2) (3) (4) (5) (6)
Becomes academy 0.49 0.22 0.24 0.23 0.78 0.48
(0.12) (0.12) (0.17) (0.12) (0.26) (0.23)
Controls for key stage 3 achievement Yes Yes Yes Yes Yes Yes
School fixed effects Yes (6) Yes (27) Yes (18) Yes (114) Yes (6) Yes (86)
Year dummies Yes (3) Yes (3) Yes (3) Yes (3) Yes (3) Yes (3)
R-squared .77 .37 .31 .16 .30 .13
Number of pupils 4706 19862 11383 81862 3932 60902
Number of schools 7 26 21 108 6 81
Note: Coefficient estimates (robust standard errors); both 100 to maintain comparability with tables 10.8, 10.9, and 10.10.
Notes
We would like to thank Rajashri Chakrabarti, Susan Dynarski, Tom Kane, Sandra
McNally, Paul Peterson, and an anonymous referee for a number of helpful comments
on an earlier draft.
1. See Le Grand (1991, 1993) and the more recent discussion of the quasi-market in
Machin and Vignoles (2005).
2. The compulsory minimum school-leaving age was raised again in 1973 to sixteen, the
level at which it currently stands. It is expected that it will be raised to eighteen in
England by 2015 (see BBC News, November 6, 2007: hhttp://news.bbc.co.uk/1/hi/
education/7080699.stmi).
3. The 1988 Act established parental representation on school governing bodies and
required Local Education Authorities (LEAs) to delegate staff recruitment and financial
management to the board of school governors (Machin and Vignoles 2005).
4. Private schools have charitable status and so their fee income is not taxed.
5. Source: hhttp://www.teachernet.gov.uk/management/resourcesfinanceandbuilding/
funding/schoolsprivatefinanceinitiative/faqs/i.
6. Specifically, sponsors ‘‘make a charitable donation of 10% of the building costs, up to
a total of £2 million (or £1.5 m for an Academy based on remodelled rather than
completely new buildings) towards the initial capital cost of their Academy.’’ Capital
contributions of the sponsor are expected to be made over the length of the building proj-
ect (two to three years). Source: hhttp://www.standards.dfes.gov.uk/academies/pdf/
AcademySponsorProspectus2005.pdf?version=1i. These contributions may be replaced
with sponsor-provided long-term endowment funds, of equivalent financial value,
for academies opening in the future (NAO 2007). Following further changes made to
the academies programme in 2007, the 2 million funding requirement is waived when the
sponsor is a university, a fee-charging private school, or a high-achieving state school
(Sibieta et al. 2008).
7. An academy usually has around thirteen governors, typically seven of whom are
appointed by the sponsor. Of the remaining six, one must be a local authority representa-
tive, and at least one elected parent representative is required.
8. Staff usually transfer directly from the predecessor school to the academy school, and
their existing employment terms and conditions are legally protected. Source: hhttp://
www.standards.dfes.gov.uk/academies/what_are_academies/organisation/?version=1i.
Table 10.5 shows that more than half of the academy school set up by September 2005
were in the inner and outer London regions (14 of the 27 listed). This reflects government
targets requiring 60 of the first 200 academy schools to be established in London (NAO
2007).
9. The autonomy of academies relative to other state schools seems to extend to their pol-
icy on pupil expulsions. In the academic year 2005–06, academy schools permanently
excluded 5.5 pupils in every 1,000. The comparative figure for other state secondary
schools in England was 2.4 in every 1,000 (See BBC News story July 13, 2007, hhttp://
news.bbc.co.uk/1/hi/education/6897429.stmi).
10. We list only those academy schools set up in September 2002 through September 2005
because to date we do not have pupil examination performance data covering academies
opening in subsequent years. However, as of August 2007, there were forty-six acade-
mies open (the twenty-seven listed in table 10.5, plus nineteen that opened in September
2006) and many more are intended to follow (the current U.K. government has unveiled
plans to establish 400 academies, with at least 200 of these open or planned to be opened
by 2010).
11. GCSEs are the examinations taken at the end of the final year of compulsory school-
ing (in year eleven at age fifteen/sixteen). GCSEs were first introduced in 1986, to GCE
O-Level and the Certificate of Secondary Education (CSE) qualifications. The first GCSE
exams were sat in 1988 and were originally graded from A to G. Nowadays pupils take
a number of GCSEs, which differ across schools, in different subjects, and these are
assigned pass grades ranging from A through G. The A grade was first implemented
in 1994 as a means of classification for outstanding examination achievement. The cur-
rent key headline figure at school level is the percentage of children achieving five or
more GCSEs with grades A through C. All Academies open in September of a given
year, and comparison of final school year exams in the Academy schools relative to the
national average is based on the results of GCSE exams taken in the summer of the fol-
lowing year.
12. Specifically, our pre-policy matching is based on school-level data from the academic
year 1995–96 to the academic year prior to the one in which the school became an acad-
emy (since we have school-level data available from 1995–96 onward). As a third crite-
rion we also look for similarities between the match school and the pre-academy school
using their school type (in the maintained sector other school types are community, vol-
untary aided, voluntary controlled, foundation, or city technology college). Finally, we
also considered the nearest school in geographical terms as a spatially matched control
group, with similar findings to those reported.
13. These numbers also show that, in terms of GCSE performance, the schools that be-
came academies were right near the bottom of the pile in national terms before becoming
academies (for example, in the 2001–02, 2002–03, 2003–04, and 2004–05 academic years
the national averages were 51.5, 52.9, 53.7, and 57.1 percent, respectively).
14. School fixed effects involve controlling for time-invariant characteristics of schools.
References
Gadkowski, L. B. 2007. Britain’s city academies: Can corporate citizenship strengthen

state-supported education? Unpublished dissertation, London School of Economics.
Gordon, P., R. Aldrich, and D. Dean. 1991. Education and Policy in England in the Twentieth
Century. London: Woburn Press.
Graddy, K., and M. Stevens. 2005. The impact of school resources on student perfor-
mance: A study of private schools in the united kingdom. Industrial and Labor Relations
Review 58: 435–451.
Green, F., S. Machin, R. Murphy, and Y. Zhu. 2007. The changing private returns to inde-
pendent education in Britain. Unfinished draft.
Green, F., S. Machin, R. Murphy, and Y. Zhu. 2008. Competition for private and state
school teachers. Centre for the Economics of Education Discussion Paper No. 94. London:
London School of Economics.
Le Grand, J. 1991. Equity and Choice. London: Harper Collins.

Le Grand, J. 1993. Quasi-Markets and Social Policy. London: Macmillan.
Machin, S., and A. Vignoles. 2005. What’s the Good of Education? The Economics of Educa-
tion in the United Kingdom. Princeton: Princeton University Press.
National Audit Office (NAO). 2007. The Academies Programme. London: The Stationery
Office.
Naylor, R., J. Smith, and A. McKnight. 2002. Why is there a graduate earnings premium
for students from independent schools. Bulletin of Economic Research 54: 307–378.
PriceWaterhouseCoopers. 2007. Academies evaluation, 4th annual report. London: De-

partment for Education and Skills.
Sibieta, L., H. Chowdry, and A. Muriel. 2008. Level Playing Field? The Implications of School
Funding. Berkshire, UK: CfBT Education Trust.
IV Conclusions
11 Education Contracting: Scope of Future Research
Harry A. Patrinos
School choice is often promoted as a means of increasing competition

in the school system (Friedman 1955). It is believed that competition
will lead to efficiency gains as public and private schools compete for
students and try to improve quality while reducing expenses (Neal
2002). By encouraging more private schools, vouchers will allow
schools to become innovative and bring improvements to the learning
process. Public schools, in order to attract the resources that come with
students, will improve. Opponents claim that under a voucher system
private providers will be unaccountable to taxpayers. Claims of effi-
ciency gains are also questioned. Further, critics sometimes claim that
choice will lead to privatization, less public control of education, and
increased segregation (Ladd 2002). Others argue that increased choice
may lead to social breakdown and civil conflict, and that without
institutional controls, private schools may teach racism and religious
intolerance, thus exacerbating existing social tensions. However, except
for the United States, there are few rigorous—even fewer random—
impact evaluations (Kremer 2003).
In most countries the state is both the major financer and provider of
education. However, government efforts to expand schooling have not
reached all members of society. While governments have an interest in
promoting and financing education, it does not necessarily follow that
the public sector has a role in providing it. There are often other pro-
viders, such as churches, home schools, and private schools, both for-
profit and not-for-profit. By extending financing to these providers,
governments would give all parents the opportunity to participate
more fully in their child’s education by choosing the school that is right
for them.
246 Harry A. Patrinos
Forms of Contracts
While there are many ways to incorporate elements of choice in school

systems, as well as many means of including the private sector in the
provision of services at the compulsory level, here the concern is with
forms of contracting for service delivery. Contracting refers to the pro-
cess whereby a government procures education or education-related
services, of a defined volume and quality, at an agreed price, from a
specific provider for a specified period, with the provisions between fi-
nancier and service provider recorded in a contract.
Many forms of contracting are used in education, depending on
which services are bought from the private sector. Norman LaRocque
in chapter 4 of this volume summarizes the forms of contracting. I use
the same categorization and present a few examples and the strength
of the evidence based in table 11.1. There are many cases around the
world of school places being bought from the private sector by govern-
ment, including Australia, Japan, and the United Kingdom. Contract-
ing with schools to enroll publicly funded students is used extensively
and has proven to be a good strategy for rapidly expanding access to
education while avoiding large capital costs. Also, contracting for sup-
port services (meals, maintenance) is widely used. Contracting for
management services is one of the most important issues, but it is very
difficult to implement, not least because of the challenge of identifying
Table 11.1
Selected Education Contracting Examples and Evidence Base
Private finance
Voucher-type programs Charter-like schools initiative
OECD countries
Denmark* United States**** Australia
Italy* United Kingdom** Canada
Netherlands** Germany
Sweden*** Ireland
United States**** United Kingdom
Developing/transition countries
Chile** Colombia*** Colombia
Colombia**** Fe y Alegria** Mexico
Czech Republic**
Notes: Strength of evidence: * weak; ** strong; *** very strong; **** excellent (randomized)
Source: Author’s compilation.
Education Contracting 247
measurable performance criteria. Contracting a private actor to operate

a public school has proven controversial in some countries, regardless
of results. A few countries are experimenting with contracts for private
financing and construction of schools. An extension of the private fi-
nance model is contracting for private actors to run schools, as well as
finance and build them, a model that has not yet been tried in educa-
tion. Contracting for professional services (such as curriculum devel-
opment) is also fairly easy to specify and monitor.
Evidence
In general, the best evaluations use designed experiments that ran-

domly assign benefits and include a true control group. In the case in
which the allocation of vouchers is not random, the type of children
leaving the worst schools are presumably the more able students, and
the students remaining in the school are, therefore, the less able ones.
In the absence of a randomized design, or some form of natural experi-
ment, rigorous techniques such as propensity-score matching, local
average-treatment effects, regression discontinuities, and so on, are
used. Outcomes to evaluate could be cost savings, increases in enroll-
ments, and improvements in learning, among others. The many bene-
fits of contracting also make such projects highly amenable to proper
impact evaluations.
From the United States, Hoxby (2003a) shows that the public schools
are threatened by competition, leading them to reform and improve
so as not to lose students. Thus, competition has an impact on public
schools, especially where there was previous popular dissatisfaction
with them. Hoxby (2003b) reviews the evidence from the United States
relying on policy experiments. She finds that public schools do in fact
respond to competition by raising their achievement and productivity.
Student achievement rises when students attend voucher or charter
schools. Further, she argues that voucher and charter school programs
do not cream-skim; they disproportionately attract students who were
performing badly in their regular public schools. Howell and Peterson
(2002) show evidence that vouchers benefit African-American students
more than participants from other ethnic groups in their examination
of test scores, parental satisfaction, parent-school communications, and
political tolerance among students and parents participating in pro-
grams in New York; Dayton, Ohio; Washington, D.C.; and San Anto-
nio, Texas. West and Peterson (2006) analyze Florida’s accountability
law, the Aþ Plan, where the average test performances of students in

grades three through ten must be reported annually for each school,
and in which students at twice-failed schools are given the opportunity
to attend another school. They show positive impacts on student
performance.
Milwaukee’s experiment with school vouchers receives a lot of atten-
tion. The program started in 1990 and gives selected children from
low-income families taxpayer-funded vouchers to allow them to attend
private schools. Rouse (1998) compares voucher students with control
groups, using various statistical methods to remove biases. Her con-
clusion is that in standardized reading tests there appears to be no dif-
ference between voucher students and their public school peers. But
standardized math test scores rose significantly more rapidly for stu-
dents who used vouchers to attend private schools than for their coun-
terparts in public schools. This suggests that providing vouchers to
low-income students to attend private schools could help increase the
mathematical achievement of those students who participate.
A good deal of research has evaluated charter schools in the United
States (Hoxby 2004; Bettinger 2005). A few studies have found signifi-
cant improvement, but several have found either no impact or dete-
rioration in school performance. Hoxby’s (2003a) analysis of charter
schools in Michigan and Arizona shows positive evidence.
Vouchers in Other Countries
There are a few systematic evaluations of the effects of vouchers worth

mentioning. An experiment with vouchers is going on now in Italy. An
immediate outcome was that the cost of private schooling declined.
Using a longitudinal study of voucher recipients, Brunello and Checchi
(2005) find that private schools may be selected for different reasons
than quality considerations. Also, by exploiting individual data on
voucher applicants, they present evidence that the percentage of
voucher applicants is higher when the average quality of private
schools is higher. However, their study does not establish causation,
and the authors rightly point to the need for further research with bet-
ter data.
Large-scale, universal programs exist in a few European countries.
Some are so old and well established that it is almost impossible to
evaluate them in the usual sense. These would include Denmark and
the Netherlands. In Denmark, any group of parents can claim public
funding by declaring themselves a private school if they have at least

twenty-eight students. The state funds all schools that meet minimum
requirements. All schools receive grants according to the number of
students enrolled. About 12 percent of children attend a private school,
and this percentage has been increasing in recent years. Andersen and
Serritzlew (2006) attempt to analyze the impact of private school com-
petition on Danish language and math test scores. They use a logit
model and OLS to investigate the effect of market share on test scores,
controlling for a variety of school factors, socioeconomic background,
teachers, and local taxes. They use cross-section data for 2002. They
conclude that competition does not improve the achievement of public
school students, but does increase public school expenditures per stu-
dent. This study, however, suffers from the lack of over-time data and
any identifying variable to control for selection.
The Dutch education system has been decentralized and demand-
driven since 1917 (Patrinos 2002). Almost 70 percent of schools in the
Netherlands are administered and governed by private school boards.
Public and private schools are funded by government on an equal
footing, and most parents have a choice of several schools near their
homes. While schools are free to determine what is taught and how,
the Ministry of Education does impose a number of statutory quality
standards. In recent years, there has been a trend towards greater au-
tonomy and decentralization. Many central government powers have
been transferred to the level of the individual school. There is evidence
that Dutch students are among the best educated in the world (accord-
ing to TIMSS and PISA results),1 and that the system produces very
low levels of learning inequality (Patrinos 2004). More recently, in a
working paper Himmler (2007) uses cross-section data and instru-
ments for the number of Catholics living in the local education market
and estimated the impact of competition on public school outcomes.
He uses a variety of competition measures, including geographic prox-
imity and the Herfindahl index (a commonly accepted measure of mar-
ket concentration) to control for the possible endogeneity of Catholic
competition to public school quality. He finds that competition leads
to improved secondary school grades, and that competition does not
increase per-student spending or grade inflation. However, the study
completely discounts the presence of sorting by ability.
More recent universal choice systems are more easily subject to
evaluation. The sweeping changes that ended communism in East-
ern Europe had a major impact on education. Immediately after the
transition, private schools became legal and soon began receiving

public funds. In the Czech Republic, 4 percent of students attend pri-
vate schools overall, but as much as 14 percent at the upper secondary
level, up from 1 and 3 percent in 1991. A direct analysis of the impact
of the voucher on private and public schools was undertaken, finding
that private schools have arisen in response to market incentives (Filer
and Munich 2002). Private schools are more common in fields where
public school inertia has resulted in an undersupply of places. Also,
private schools are more common where the public schools are per-
forming worse academically. Public secondary schools facing signifi-
cant private competition in 1995 improved their relative success in
obtaining university admissions for their graduates by over four posi-
tions (out of 77) between 1996 and 1998. One problem with the study
is that data on university access come after the establishment of private
schools in the Czech Republic. Still, that would mean that the positive
effect is underestimated. More serious might be the lack of controls on
selection effects, since the reform was not a randomized experiment.
But private schools tend to attract weaker students than the students
remaining in the public system.
In the early 1980s, less than 1 percent of Swedish children attended
private schools, even though half of the private schools did not charge
fees and citizens felt the centralized public system was unresponsive.
A series of reforms in the 1990s introduced greater parental influence
through the devolution of funding and management. New rules al-
lowed money to follow students, and municipalities were required to
fund private schools. As a result, enrollments in private schools con-
tinue to grow, to more than 10 percent in 2003. Evaluations generally
show positive results. Sandström and Bergström (2005) find that
results improve for public school students when the degree of competi-
tion from private schools increases. They instrument using the share
of students attending independent schools and who have no failing
grades to control for endogeneity of private school selection, and use a
sample selection model. They control for socioeconomic background,
immigrant status, distance, and school factors. They use panel data for
the period 1992–97. They specifically test for the impact of greater
competition in the municipality on improving the performance of pub-
lic school students. They find that ninth-grade math test scores increase
by one-fifth of a standard deviation when the local private school en-
rollment share increases by 10 percent. They also find that the share of
students in private schools is larger if academic achievement of public

school students was low prior to the enactment of the reform. Ahlin
(2002) also finds positive effects of local school competition in Sweden.
The analysis improves upon Sandström and Bergström (2005) by in-
cluding a value-added specification on test scores. Ahlin incorporates
previous student performance into her equations and finds that a 10
percent increase in the local share of private school enrollment leads to
one-fifth of a standard deviation increase in ninth-grade math perfor-
mance, but no significant effects on English or Swedish. Ahlin controls
for student characteristics, family background and municipal charac-
teristics, and uses data over time (1991–97). She also estimates quantile
regressions and reports homogeneous effects for both low- and high-
performing students.
Chile is the only developing country to have a universal voucher
program. Subsidized private schools have proven to be slightly more
cost-effective than municipal schools. Research results are mixed. Most
studies, however, are subject to criticism, not least because of the reli-
ance on post-voucher plan data and no preprogram trends (Hoxby
2003b). The identification problem is serious since the reform did not
include any arbitrary exclusions. Hsieh and Urquiola (2006) use an in-
strumental variable approach and over-time data with a fixed effects
model. Presumably, the fixed effects model will control for observable
and unobservable time-invariant effects. This study finds no evidence
that choice improved average educational outcomes as measured by
test scores, repetition rates, and years of schooling. However, the
authors find evidence that the voucher program led to increased sort-
ing, as the ‘‘best’’ public school students left for the private sector. In a
careful examination of the evidence, Cristian Bellei (chapter 8 in this
volume) finds that different studies yield very different conclusions
due to methodological differences; data limitations; and differences in
abilities to control for various biases such as selection, inability to in-
clude appropriate controls, interaction effects, and so on. When correct-
ing for such biases no evidence is found that private voucher schools
are more effective than public schools in Chile.
In the 1990s Colombia experimented with a targeted voucher pro-
gram. The program was oversubscribed so a lottery was put in place
to randomly assign students to private schools with excess capacity.
Taking advantage of the random design, several rigorous evaluations
have been undertaken. The program led to considerable enrollment
increases, especially for the disadvantaged, at a low cost to govern-

ment. The Colombia voucher program grew substantially, incorporat-
ing more than 125,000 students from poor families, was efficient in
terms of lower unit cost per beneficiary student, and provided educa-
tion of at least comparable quality to that provided in public schools.
Taking advantage of the randomized design, Angrist and others
(2002) and Angrist, Bettinger, and Kremer (2006) found that voucher
beneficiaries had higher educational attainment: they were 10 percent-
age points more likely to finish the eighth grade three years after they
won the vouchers. They were also 5 to 6 percentage points less likely
to repeat a grade, scored 0.2 standard deviations higher on achieve-
ment tests than non-voucher students, and were 20 percent more likely
to take the college entrance exam than students who had not won a
voucher in the lottery. They were also 0.6 to 1.0 percentage points less
likely to be married and 2.5 to 3.0 percentage points less likely to be
working. Angrist and others (2002) also estimated that voucher benefi-
ciaries were likely to earn U.S.$36 more in wages each year, compared
to the U.S.$24 per beneficiary that it cost the government to provide
vouchers instead of places in public schools. In a study of the pro-
gram’s longer term effects, Angrist, Bettinger, and Kremer (2006) found
that the program improves scores for both average students and those
over the 90th percentile.
Charter Schools and Variants in Other Countries
The most extensive example of contracting in Latin America is the Fe

y Alegrı́a (FyA) school network, which is described by Norman La-
Rocque in chapter 4. FyA has a long history in Venezuela, where it
was founded in 1955. FyA schools are believed to outperform tradi-
tional public schools. Allcott and Ortega (2007) estimate the average
treatment effect (ATE), the difference in test scores between students
in treated and untreated schools, or how a student would have per-
formed in Fe y Alegrı́a as opposed to in a public school. They estimate
this effect using ordinary least squares and propensity score matching,
after assuming that there is no omitted variables bias and that the
treatment effect is homogeneous. Their outcome variables are math
and verbal scores. They control for nationality, socioeconomic status,
household factors, and school factors. While they cannot fully rule out
bias due to unobservables, they do argue that their results do not suffer
from lack of overlap and differing economic environments. They find
an average treatment effect of 0.1 standard deviations. They argue that

the better performance of the Fe y Alegrı́a school comes from its labor
contract flexibility and decentralized administrative structure.
The city of Bogotá, Colombia, has introduced the concession schools
program, under which twenty-five newly built public schools in poor
neighborhoods were turned over to high-quality private institutions.
The program began in 2000 and serves 26,000 students. Schools are
paid U.S.$506 per student (below the average annual cost of a student
who attends a half-day public school), must provide educational ser-
vices to poor children, and must accept all students. The provider has
full autonomy over school management and is assessed on results. A
recent impact evaluation using propensity score matching by Felipe
Barrera-Osorio (chapter 9) in this volume finds that dropout rates
were lower in concession schools and competition from concession
schools led to a decline in dropout rates in nearby public schools.
There are few similar programs outside the United States. A similar
version may be the United Kingdom’s City Academies program, which
in the first two to three years of its existence does not show any
improved pupil performance relative to state schools (see Stephen
Machin and Joan Wilson, chapter 10 in this volume).
Private Finance Initiative (PFI)
The United Kingdom, starting in 1997, pioneered initiatives to contract

with private consortiums to provide school facilities. Canada, Austra-
lia, and other countries have implemented similar PFIs to expand
private involvement in financing and providing infrastructure. Devel-
oping countries are following suit now, including Mexico and Colom-
bia. There is little evidence on the impact of contracting for the
delivery of education services, although some studies have assessed
PFIs on the price and timeliness of delivery of infrastructure.
Conclusions and a Call for Further Research
We know more about certain types of contract options. Namely, we

know a lot about voucher programs and charter schools in the United
States. We know much less about other contracting forms from other
countries. Therefore, impact evaluations are needed to increase the in-
formation base by which policy makers can make informed decisions.
While the assessment literature for innovative education contracting
programs is increasing rapidly over time, most of the research is for the
United States. Meanwhile, there are some very interesting programs in
developing and transition countries that are not receiving much atten-
tion. More research could help us uncover whether cross-country dif-
ferences in the operation and funding of schools matter for student
achievement.
It would appear that competition matters. Most studies show im-
provements in public schools over time that are subject to competition
from private schools in the local education market. This is the case
in more developed countries, such as the United States, Sweden, the
Czech Republic, and the Netherlands; but not in Denmark, and proba-
bly not in Chile. While all nonrandomized studies are problematic to
some extent, the lack of over-time data (Netherlands, Denmark), lack
of preprogram data (Chile), and difficulties controlling for selection or
sorting (Netherlands, Czech Republic, Denmark) make some of the
studies suspect. At a minimum, evaluations of such programs should
be done over time, with data covering several years before the imple-
mentation and at least a few years after program launch.
The fundamental problem in estimating the impact of voucher pro-
grams is that students and schools often self-select into the program
(the selection problem). In this case a comparison between students
who participate and those who do not participate confounds the effects
of the program with the initial differences in characteristics between
participants and nonparticipants. It is likely that better informed
households are more likely to apply to voucher programs. In this case,
students from better informed and more active households may per-
form differently than students who did not apply to the program.
Therefore, any observed final educational outcomes not only comprise
the results from the voucher program but also the inherent differences
in characteristics of the families or students. One possible solution in
order to evaluate outcomes in established school choice programs
might be to experiment with the motivation to participate. That is, pol-
icy makers could offer more information on the benefits of vouchers,
for example by publishing test scores, to a targeted group of families
in disadvantaged economic conditions.
Many researchers rely on the instrumental variables (IV) approach,
which uses a variable with two characteristics: it can explain participa-
tion in the program but is uncorrelated with the outcome measures of
interest. Clearly, the main problem of the IV approach is validity. The
majority of available variables that are correlated with participation
are correlated with the outcome as well. Even when one finds a vari-
able correlated with participation, it is impossible to test whether it is
uncorrelated with the unobservable part of the outcome variable. The
other potential methods—Heckman correction models, difference-in-
difference estimators, and matching estimators—are based on strong
assumptions.
The need for more evaluations using strong designs is clear. Voucher
programs are ideal settings in which to design experimental setups.
The demand for education is increasing, and the resources are scarce.
In this case, allocation of benefits by lottery will ensure horizontal eq-
uity and transparency to the process. In the event that the voucher is
targeted using a proxy index, then the program can be evaluated by re-
gression discontinuity design (RDD), given unbiased estimators. The
RDD estimation is possible if the program identified its beneficiaries
using an assignment variable. For example, if selection was based on a
proxy-mean index, then regression discontinuity analyses could poten-
tially be used in such cases, since they make use of the assignment
variable and the observations with scores close to the cutoff point for
program eligibility. If all those with a score below a certain cutoff are
beneficiaries and those with a score above the cutoff are denied access
to the program, then students with scores just below the cutoff point
(beneficiaries) are very similar to students above the cutoff point (com-
parison group). Thus, it is possible to compare the outcome variables
for those two groups and attribute the differences to the program.
Social issues and vouchers must be on the research agenda. Vouch-
ers and other forms of private participation should not only be as-
sessed in terms of efficiency in improving test scores, but also in terms
of their impact on social tensions. It could be that without institutional
control, private schools may teach racism and religious intolerance,
thus exacerbating social tensions. These are important empirical ques-
tions that must be included in future research. For example, in the case
of Pakistan, recent findings on the enrollment numbers differ by an
order of magnitude from those reported by and in the media. The reli-
gious school sector (madrasa) is small compared to educational options
such as public and private schooling, accounting for less than 1 percent
of overall enrollment in the country (Andrabi and others 2006).
The evaluation of programs is fundamental for the selection of suc-
cessful public policies. In effect, a positive evaluation may lead to the
realization of the importance of investing part of the national budget in
a program with proven results. A negative evaluation, on the contrary,
may imply considerable savings for the public treasury since it is pos-
sible to cancel programs that, without the evaluation, would have con-
tinued and maybe even been extended.
There is a need for multi-country, -institutional, and -year research,
with various sources of funding, and a common methodological frame-
work, under a common organizing framework. The research could go
beyond analyzing what works and explain why it works, how, and
under what circumstances.
Notes
World Bank, 1818 H Street NW, Washington, D.C., 20433 ([email protected]).

The views expressed here are those of the author and should not be attributed to the
World Bank.
1. Trends in International Mathematics and Science Study (TIMSS) and Program for In-
ternational Student Assessment (PISA).
References
Ahlin, A. 2002. Does school competition matter? Effects of large scale school choice re-
form on student performance. Working paper, Department of Economics, Uppsala Uni-
versity, Sweden.
Allcott, H., and D. Ortega. 2007. The performance of decentralized school systems: Evi-
dence from Fe y Alegrı́a in Venezuela. Working paper, Instituto de Estudios Superiores
de Administración, Caracas, Venezuela.
Andersen, S. C., and S. Serritzlew. 2006. The unintended effects of private school compe-
tition. Journal of Public Administration Research and Theory 17: 335–356.
Andrabi, T., J. Das, A. Ijaz Khwaja, and T. Zajonc. 2006. Religious school enrollment in
Pakistan: A look at the data. Comparative Education Review 50, no. 3: 446–477.
Angrist, J., E. Bettinger, E. Bloom, E. King, and M. Kremer. 2002. Vouchers for private
schooling in Colombia: Evidence from randomized natural experiments. American Eco-
nomic Review 92, no. 5: 1535–1558.
Angrist, J., E. Bettinger, and M. Kremer. 2006. Long-term consequences of secondary

school vouchers: Evidence from administrative records in colombia. American Economic
Review 96, no. 3: 847–862.
Bettinger, E. P. 2005. The effect of charter schools on charter students and public schools.
Economics of Education Review 24: 133–147.
Brunello, G., and D. Checchi. 2005. School vouchers Italian style. IZA Discussion Paper
Series 1475, Bonn, Germany.
Filer, R. K., and D. Munich. 2002. Responses of private and public schools to voucher
funding. Working paper, Center for Economic Research and Graduate Education of
Charles University and Economics Institute of the Academy of Sciences of the Czech Re-
public, Prague.
Friedman, M. 1955. Role of government in education. In Economics and the Public Interest,
ed. Robert Solo. New Brunswick, N.J.: Rutgers University Press.
Himmler, O. 2007. The effects of school choice on academic achievement in the Nether-
lands. Working paper, Georg-August-Universität Göttingen, Germany.
Howell, W. G., and P. E. Peterson. 2002. The Education Gap: Vouchers and Urban Schools.
Washington, D.C.: Brookings Institution.
Hoxby, C. M. 2003a. The Economics of School Choice. Chicago: University of Chicago Press.
Hoxby, C. M. 2003b. School choice and school competition: Evidence from the United
States. Swedish Economic Policy Review 10: 9–65.
Hoxby, C. M. 2004. A straightforward comparison of charter schools and regular public
schools in the United States. Working paper, Harvard Institute of Economic Research.
Hsieh, C.-T., and M. Urquiola. 2006. The effects of generalized school choice on achieve-
ment and stratification: Evidence from Chile’s voucher program. Journal of Public Econom-
ics 90, no. 8–9: 1477–1503.
Kremer, M. 2003. Randomized evaluations of educational programs in developing coun-
tries: Some lessons. American Economic Review Papers and Proceedings 93, no. 2: 102–106.
Ladd, H. F. 2002. School vouchers: A critical view. Journal of Economic Perspectives 16, no.
4: 3–24.
Neal, D. 2002. How vouchers could change the market for education. Journal of Economic
Perspectives 16, no. 4: 25–44.
Patrinos, H. A. 2002. Private education provision and public finance: The Netherlands as
a possible model. Occasional Paper No. 59, NCSPE, Columbia University.
Patrinos, H. A. 2004. Private education provision and public finance: The Netherlands as
a possible model. Paper presented at the 2004 AEA meetings, San Diego, CA.
Rouse, C. E. 1998. Private school vouchers and student achievement: An evaluation of the
Milwaukee Parental Choice Program. Quarterly Journal of Economics 113, no. 2: 553–602.
Sandström, F. M., and F. Bergström. 2005. School vouchers in practice: Competition will
not hurt you. Journal of Public Economics 89, no. 2–3: 351–380.
West, M. R., and P. E. Peterson. 2006. The efficacy of choice threats within school account-
ability systems: Results from legislatively induced experiments. Economic Journal 116, no.
510: C46–C62.
Contributors
World Bank
Cristian Bellei
Harvard University
Eric P. Bettinger
Case Western Reserve University
Rajashri Chakrabarti
Federal Reserve Bank
Geeta G. Kingdon
University of Oxford
Michael Kremer
Harvard University
Norman LaRocque
Education Forum
Stephen Machin
University College London
Karthik Muralidharan
University of California–San Diego
Thomas J. Nechyba
Duke University
Harry Anthony Patrinos
World Bank
Paul E. Peterson
Harvard University
260 Contributors
Joan Wilson
London School of Economics
Ludger Wößmann
University of Munich
Index
Aþ Plan, 247–248 Catholic Church, 4, 48

Ability tracking, 58–59 Chile and, 168, 170
Academies Sponsor Prospectus, 226 competition effect and, 249
Aedo, Cristian, 169 Jesuits and, 80
All India Education Survey, 111 Caucutt, E., 57
Andersen, S. C., 249 Cerdan, P., 205
Angrist, Joshua, 144, 149, 152, 252 Certification, 230–231
Arizona, 248 Chakrabarti, Rajashri, 3–11
Army schools, 111 Charter schools, 4, 198
ASER survey, 115, 136n4 accountability and, 75, 78
Australia, 5, 83–84, 246 city academies and, 226–227
Average treatment effect (ATE), 252 contracting and, 75, 78
Fe y Alegrı́a (FyA) and, 80, 252–253
Banerjee, A. S., 112 government red tape and, 78
Barrera-Osorio, Felipe, 8–9, 193–218 legal issues and, 78
Barrientos, J. H., 197 United States and, 8, 75, 78, 198, 226–227
Bashir, Sajitha, 118, 126–127 Chaudhury, Nazmul, 92, 96
Bayer, P., 49 Checchi, D., 248
Belgium, 19 Chicago Public Schools (CPS), 75
Bellei, Cristian, 7–9, 165–192, 251 Chile, 4, 8–10, 251, 254
Bergström, F., 250–251 competition effect and, 168, 189
Bettinger, Eric, 7, 9, 143–164, 248, 252 cream skimming and, 167
Bishop, John H., 18, 23 enrollment and, 167
Bloom, Erik, 144 institutional design in, 166–167
Blunkett, David, 227 language and, 167–168
Book banks, 99 market-oriented initiatives and, 165–168,
Bravo, David, 169 188
Brazil, 4–5 mathematics and, 167–168
funding patterns and, 19 parents’ education level and, 178–179
private schools and, 47–48 peer effects and, 180
public-private partnerships and, 14, 19 public-private partnerships and, 165–191
Bribery, 93 regression analysis of, 167–188
Brunello, G., 248 school types in, 166–167
Brunner, E., 64 SES index and, 168, 172–177, 180–182,
Build-operate-transfer (BOT), 73 191n7
socioeconomic status and, 179–181
Calderon, Alberto, 146 student selection and, 172–178
Carnoy, Martin, 169 systemic effects and, 167–169
262 Index
Chile (cont.) future policy research and, 247–249, 254

vouchers and, 7, 132, 166–167, 170–181, India and, 101–103
251 Concession schools, 10n1
Chubb, John E., 23–24, 49 autonomy and, 198
Church schools, 18 Colombia and, 8, 78–80, 193–217, 253
Catholic, 4, 48, 80, 168, 170, 249 direct/indirect impact of, 201, 209–216
competition effect and, 249 dropout rates and, 195, 202–203, 205, 207,
City academies, 9 209, 213
Academies Sponsor Prospectus and, 226 economic issues and, 193, 197, 216, 253
Blunkett and, 227 enrollment and, 193
certification and, 230–231 infrastructure of, 203–204
charter schools and, 226–227 Instituto Colombiano Para El Fomento
General Certificate of Secondary De La Educación Superior (ICFES) and,
Education (GCSE) and, 230–234, 237, 203–204
239n11 open-door policy of, 194
Local Education Authorities (LEA) and, performance and, 193–195, 198–216
226, 231, 234 poor families and, 194
performance of, 230–237 public school differences and, 193–194
private finance initiatives (PFIs) and, pupil-teacher ratio and, 196–197
225–226 quality of, 193–194
program of, 226–231 regression analysis of, 196–216
specialization and, 230 resources of, 203, 208–209
sponsors and, 230 self-selection and, 201
TeacherNet and, 225 test scores and, 194–195, 199, 208
City technology colleges (CTCs), 220 Contracting
Classroom size, 83 charter schools and, 75, 78, 252–253
Colombia and, 145, 196–197 concession schools and, 78–80
India and, 91, 95–96, 101–103 defined, 71
United Kingdom and, 223–224 Education Management Organizations
Cleveland Scholarship and Tutoring (EMOs) and, 74–75
program, 6 forms of, 246–247
Cohen-Zada, D., 54 government use of, 71
Colombia, 4–6, 9 infrastructure public-private partnership
dropout rates and, 195, 202–203, 205, 207, (PPP) model and, 73–74, 82–84
209, 213 international examples of, 74–84
enrollment and, 143, 193 management contract model and, 72–73
Instituto Colombiano Para El Fomento model topologies and, 72–74
De La Educación Superior (ICFES) and, operational contract model and, 72
144–145, 150–162, 203 performance and, 247–248
Ministry of Education and, 79, 203 policy design and, 84–86
Plan de Ampliación de Cobertura de la potential benefits of, 72
Educación Secundaria (PACES) and, private finance initiatives (PFIs) and, 83–
143–148, 155–156, 159 84, 225–226, 253
public-private partnerships and, 143– private management and, 74–80
163 public schools and, 74–80
pupil-teacher ratio and, 145, 196–197 service delivery model and, 72–73
regression analysis of, 143–163, 196–216 service procurement and, 71
vouchers and, 7, 132, 143–163, 251–252 transparency and, 72, 85
Common Core of Basic Competence, 150 United States and, 74–78
Competition effect vouchers and, 248–252 (see also Vouchers)
Chile and, 167–168, 189 Contreras, Dante, 169
Colombia and, 202 Cooley, J., 49
Index 263
Corpoeducatión, 197 per pupil spending and, 68n5

Côte d’Ivoire, 5, 81 Plan de Ampliación de Cobertura de la
Cream skimming, 47, 56–59, 65–66, 167 Educación Secundaria (PACES) and,
Credit constraints, 24–29, 35, 40 146
Czech Republic, 250, 254 poor families and, 24, 116, 131–133, 145,
152–153, 194
De, Anuradha, 91, 99 private finance initiatives (PFIs) and,
Denmark, 248–249, 254 225–226, 253
Department for Children, Schools and property values and, 67
Families (DCSF), 221 public-private partnerships and, 14, 18–
Department of Education and Training 23 (see also Public-private partnerships
(DET), 83–84 [PPPs])
Direct Payment Agreement, 127 quasi-public school markets and, 60–64
District Information System for Education rents and, 49, 54
(DISE), 113 rural private schools and, 99–105
Dixon, Pauline, 91, 93, 104–105, 116, 118 Salary Distribution Act and, 127
Dominguez, C., 205 socioeconomic status and, 179–181, 250–
Drèze, J., 111, 115, 118, 125–126 251
Dropout rates, 195, 202–203, 205, 207, 209, subsidies and, 52–53, 81, 127, 130–131,
213 134, 152, 165–171, 178–188, 191n7, 251
Duarte, J., 193 teacher salaries and, 91, 101, 125–127,
Duflo, Esther, 112 130–131, 223
transparency and, 72, 85
Economic agents, 48–49 unions and, 127–129
Economic issues vouchers and, 3–4, 6, 143–163 (see also
bribery and, 93 Vouchers)
concession schools and, 79, 193, 197, 216, Edison Schools, 75
253 Education Act, 220
contracting and, 71–86 Education guarantee schools, 111
cream skimming and, 57–58 Education Management Organizations
credit constraints, 24–29, 35, 40 (EMOs), 74–75
different cost types and, 55 Education Reform Act, 220
Direct Payment Agreement and, 127 Education Service Commission, 112, 128
Educational Service Contracting (ESC) Education Service Contracting (ESC), 81
and, 81 Education Voucher system (EVS), 81–82
education policy and, 48–49 Efficiency, 245, 252, 255
equilibrium and, 48–59 Chile and, 165
free riders and, 51–52 Colombia and, 156
Fund for Assistance to Private Education India and, 97, 104, 107, 125, 130–131
(FAPE), 81 public-private relationships and, 5, 9, 17,
funding effects and, 26–35 23–24, 41, 72, 86
grants-in-aid system and, 129–130 relative effectiveness and, 118–122
Herfindahl index and, 249 resource, 47, 51–54
housing choices and, 47, 60–64 United States and, 47, 51–59, 63, 66
human capital and, 3 Egypt, 84
income heterogeneity and, 53 Epple, D., 57–59
Indian schools and, 112–113, 122–126 Equations
introducing politics and, 53–54 achievement score, 119–120
lack of resources and, 132 performance analysis, 25–30
niche markets and, 47 production function, 196
operation interactions and, 36–38 program impact, 200
performance analysis and, 25–40 voucher effect, 152
264 Index
Equilibrium Grant-maintained (GM) schools, 220

cream skimming and, 56–59 Grants-in-aid system, 129–130
education policy and, 48–50 Greece, 4, 14, 19
peer effects and, 54–55 Green, F., 223
policy and, 52–57 Greene, William H., 26
Equity issues, 131–133 Gutierrez, Marybell, 145
Examination boards, 111
Hanna, Rema, 112
Ferreira, F., 49 Hanushek, Eric, 23, 49, 132, 196–198
Ferreyra, M., 54, 58 Harris, J., 49
Fe y Alegrı́a (FyA), 80, 252–253 Heckman, J., 199–201
Filer, R. K., 250 Herfindahl index, 249
Finland, 14, 16, 19 Himmler, O., 249
Florida Horizontal differentiation, 54–56, 59
Aþ Plan and, 247–250 Housing choices, 47
Opportunity Scholarship program, 6–7 geographical constraints and, 59
Supreme Court and, 6–7 public schools and, 60
France, 4 quasi-public schools and, 60–63
Free riders, 51–52 residential segregation and, 60–61
Friedman, M., 245 vouchers and, 62–65
Fuchs, Thomas, 18 Howell, W. G., 247
Fund for Assistance to Private Education Hoxby, C., 49, 59, 198, 247–248
(FAPE), 81 Hsieh, Chang-Tai, 132, 167–168, 251
Human capital, 3
Gallego, Francisco, 168, 170 Human Development Network, 13
Galor, Oded, 24
Gauri, Varum, 167–168, 171–172 Iceland, 14, 19
Gaviria, A., 197 India, 6
General Certificate of Secondary Educa- absenteeism and, 91–92, 95, 97, 103–105,
tion (GCSE), 230–234, 237, 239n11 128
Germany, 5 achievement scores and, 119–121
Glinskaya, Elena, 112 aided schools and, 92–93, 107, 108n6, 112,
Goldschmidt, P., 198 115–123, 126–134, 136n4
Government, 92–93, 112–113. See also bribery and, 93
Policy conditions in, 99, 101
charter schools and, 75, 78 decentralizing reform and, 132
city academies and, 225–237 Direct Payment Agreement and, 127
contracting and, 71–86, 246–247 District Information System for Educa-
grant-maintained (GM) schools and, tion (DISE) and, 113
220 Education Service Commission and, 112,
private school sponsorship and, 81 128
Right to Education Bill and, 107 enrollment and, 93, 111–118, 135
subsidies and, 52–53, 81, 127, 130–131, examination boards and, 111
134, 152, 165–171, 178–188, 191n7, 251 government recognition and, 113
vouchers and, 143–163 (see also Vouchers) grants-in-aid system and, 129–130
Government Assistance to Students and infrastructure and, 99, 101
Teachers in Private Education internal efficiency and, 118–126
(GASTPE), 81 lack of reliable data for, 111–112
Govinda, R., 118 lack of resources and, 132
Goyal, M., 112 language and, 119–123, 125
Graddy, K., 224 Madhya Pradesh, 118–119, 133
Gradstein, Mark, 23–24 mathematics and, 108n7, 118–125
Index 265
Ministry of Human Resource Develop- mathematics and, 150–152, 157

ment (MHRD) and, 126–127 median scores on, 145
National Commission on Teachers and, outcomes analysis of, 153, 156–162
128–129 popularity of, 150
official/household data and, 113–116 regression analysis of, 150–152, 155–161
parent characteristics and, 102, 104–105 subsidies and, 152
performance in, 104–105, 111–112, 119– valid ID and, 150
122 voucher effect and, 155–162
per-pupil expenditures in, 122–123 Instrumental variables (IV) approach, 112,
private schools and, 91–103 254–255
public-private partnerships and, 106–107, Ireland, 19
126–134 Italy, 4, 14
public school failure and, 97
pupil-teacher ratio (PTR) and, 96, 101– Jalan, Iyotsna, 112
103 Japan, 246
recognized schools and, 111 Jesuits, 80
relative sizes of school sectors in, 112–118 Jiminez, Emmanuel, 24, 155
Right to Education Bill and, 107 Justman, Moshe, 23–24, 54
rural/urban growth and, 116–118
Salary Distribution Act and, 127 King, Elizabeth, 143–145
sampling methodology for, 92–94 Kingdon, Geeta G., 6, 9, 91, 93, 111–139
school costs in, 122–126 Korea, 16, 19
Tamil Nadu, 118, 133 Kremer, Michael, 5, 9
teacher characteristics and, 91–92, 103– Colombia and, 144, 146
105, 127–129, 132 future research and, 245, 252
teacher salaries and, 91, 125–126 India and, 91–109, 115, 119, 128
transfer certificates (TCs) and, 113 Krueger, Alan, 49, 152
typology of school types in, 112–113
unions and, 127–129, 132 Ladd, Helen F., 132, 245
Uttar Pradesh, 113–114, 118–119, 122– LaLonde, R., 200
123, 125, 127, 129, 134, 136n1 Language, 55, 249
vouchers and, 106, 132–133 Chile and, 167–168, 170–174, 186
Information and communications Colombia and, 150, 156–159
technology (ICT), 225 India and, 108n7, 118–121, 123, 125,
Infrastructure public-private partnership 136n1
(PPP) model, 73–74, 82–84 LaRocque, Norman, 5, 71–87, 126, 132,
Initiatives, 8–9 246, 252
city academies and, 225–237 Larrañaga, Osvaldo, 169
equity issues and, 131–133 Latvia, 14, 19
further research for, 253–256 Learning disabilities, 67
market-oriented, 165–168, 188 Leases, 73, 82
mobilization policy and, 47–67 Lebanon, 5
Plan de Ampliación de Cobertura de la Legal issues
Educación Secundaria (PACES) and, charter schools and, 78
143–144, 155–156, 159 Florida’s accountability law and, 247–
private finance initiatives (PFIs) and, 83– 248
84, 225–226, 253 vouchers and, 6–7
teacher response to, 49 Levine, D., 112
Instituto Colombiano Para El Fomento De Libraries, 99, 197, 203
La Educación Superior (ICFES), 144 Literacy. See Reading
concession schools and, 203–204 Local Education Authorities (LEA), 226,
language and, 150–152 231, 234
266 Index
Lotteries Moe, Terry M., 23–24, 49

Colombia and, 143–156, 159, 162 Morales-Cobo, Patricia, 145
PACES and, 145–148 Munich, D., 250
Luxembourg, 19 Muralidharan, Karthik, 5, 9, 91–109, 112,
115, 119
Machin, Stephen, 8, 219–240 Murarka, S., 198
Management, 3–6. See also Policy Muzammil, Mohammad, 122, 126–127
charter schools and, 75, 78
concession schools and, 78–80 National Commission on Teachers (NCT),
contracting and, 71–86 128–129
service delivery and, 80–82 National Sample Survey, 95
Mathematics Neal, D., 245
Chile and, 167–179, 182–187, 191n8 Nechyba, Thomas J., 5, 9, 47–69, 126, 132–
Colombia and, 150–161, 198, 204, 207– 133
208, 214 Netherlands, 4, 5, 19, 248–249, 254
India and, 108n7, 118–125 New Jersey, 65
Instituto Colombiano Para El Fomento Newlon, E., 58
De La Educación Superior (ICFES) and, New Schools Project, 83–84
157 New York, 198, 247
Plan de Ampliación de Cobertura de la New Zealand, 4–5, 14, 132
Educación Secundaria (PACES) and, 146 Nonformal education centers, 94
Program for International Student Assess- Nongovernmental organizations (NGOs),
ment (PISA) and, 17, 32 18, 80
public-private partnerships analysis and, Noronha, Claire, 91, 99
13, 17, 30, 32–35, 40, 43n12, 248–252 Norway, 14, 19
Trends in International Mathematics and Nutrition, 194
Science Study (TIMSS) and, 249
Mayer, Daniel, 148 Oates, Wallace, 49
McEwan, Patrick, 169–170 Ohio, 6, 247
McHugh, C., 58 Operational contract model, 72
McMillan, R., 49 Orazem, Peter, 143, 145
Mehta, Arun, 99, 111 Ordinary least squares (OLS) analysis, 95,
Meier, Volker, 24 136n7, 170, 188, 215, 249
Members of the Legislative Assembly Organisation for Economic Co-Operation
(MLAs), 128 and Development (OECD), 17–18, 32
Members of the Legislative Council Ostoic, Carolina, 170
(MLCs), 128
Mendez, Jairo, 143 Pandey, Priyanka, 112
Mexico, 4, 14, 19 Paqueo, Vicente, 24
Michigan, 248 Pardo, Carlos, 145
Micro Impacts at Macroeconomic Policies Parents, 6, 247–248
(MIMAP), 116 education level of, 25, 178–179
Milwaukee, 6, 248 housing choices and, 47, 59–65
Milwaukee Parental Choice Program income levels of, 67
(MPCP), 6 India and, 102, 104–105
Ministry of Education (Colombia), 203 policy and, 49
Ministry of Education (Netherlands), 249 valuation of education by, 25–27
Ministry of Human Resource Develop- Patrinos, Harry A., 9, 13, 245–257
ment (MHRD), 126–127 Pedagogical targeting, 47
Minneapolis-St. Paul, 8 cost differences and, 55
Minnesota, 78 equilibrium and, 54–55
Mizala, Alejandra, 169–170 horizontal differentiation and, 54–56
Index 267
peer effects and, 54–55 Peterson, Paul E., 3–11, 13, 247–248
vertical differentiation and, 56–59 Philadelphia, 75
Peer effects Philippines, 5, 81–82
cross-type, 55–56 Plan de Ampliación de Cobertura de la
equilibrium and, 54–55 Educación Secundaria (PACES)
horizontal differentiation and, 54–56 background of, 145–148
positive, 47 lotteries and, 143–156, 159, 162
socioeconomic status and, 180 poor families and, 145
vertical differentiation and, 56–59 pupil-teacher ratios and, 145
Penn World Tables, 126 regression analysis of, 145–163
Performance. See also Regression analysis subsidies and, 152–153
Aþ Plan and, 247–248 valid ID and, 146–148
achievement scores and, 119–121 voucher effects and, 150–163
alternative empirical specifications and, Policy
27–30 ability tracking and, 58–59
Chile and, 167–172 Chile and, 165–191
city academies and, 230–237 city academies and, 225–237
Colombia and, 143–163, 193–216 Colombia and, 143–163, 193–217
competition effect and, 101–103, 168, 189, complex challenges for, 48–51
245, 247–249, 254 concession schools and, 194
concession schools and, 193–195, 198–216 constraint definition and, 48–49
examination boards and, 111 contracting and, 71–86
funding effect and, 26–27 cream skimming and, 47, 56–59, 65–66
further research for, 253–256 design lessons for, 84–86
General Certificate of Secondary Educa- Direct Payment Agreement and, 127
tion (GCSE) and, 230–234, 237, 239n11 economic issues and, 48–49
Herfindahl index and, 249 Education Act and, 220
India and, 104–105, 111–112, 119–122 education contracts and, 245–256
instrumental variable methods and, 112, Education Reform Act and, 220
254–255 equilibrium and, 48–50, 52, 54–57
mathematics and, 13, 17, 30, 32–35 (see Florida’s accountability law and, 247–248
also Mathematics) free riders and, 51–52
policy design and, 85 future, 245–256
Program for International Student government red tape and, 78
Assessment (PISA) and, 5, 13, 16–18, 28, horizontal differentiation and, 54–56
32 housing choices and, 47, 59–65
propensity-score matching and, 112 income heterogeneity and, 53
pupil fixed-effects approaches and, 112 India and, 91–108
quality and, 70, 80 (see also Quality) introducing politics and, 53–54
reading and, 13, 17, 33, 38, 41, 121–125, new form of public-private partnerships,
155, 161–162, 198, 204, 208, 214, 248 130–133
relative effectiveness and, 118–122 parents and, 49
science and, 13, 17, 33, 41, 191n8, 249, pedagogical targeting and, 47, 54–56
256n1 Plan de Ampliación de Cobertura de la
selection bias and, 25–27, 35–36 Educación Secundaria (PACES) and,
study results on, 30–40 143–144, 155–156, 159
systemic effects and, 167–169 private sector mobilization and, 47–67
teacher promotions and, 122–123 real world outcomes and, 49
treatment effect models and, 112 resource efficiency and, 47, 51–54
Trends in International Mathematics and Right to Education Bill and, 9, 107, 130–
Science Study (TIMSS) and, 249, 256n1 132, 134
vouchers and, 252 (see also Vouchers) Salary Distribution Act and, 127
268 Index
Policy (cont.) official/household data and, 113–116

United Kingdom and, 219–240 pedagogical targeting and, 47, 54–56
vertical differentiation and, 56–59 performance analysis and, 23–30, 104–
voice and, 51–52 105 (see also Performance)
voucher targeting, 62–67 Plan de Ampliación de Cobertura de la
Poor families, 24 Educación Secundaria (PACES) and,
concession schools and, 194 145–148
disadvantaged children and, 131–132 positive peer effects and, 47
equity issues and, 131–133 public-private partnerships and, 3–11 (see
India and, 116 also Public-private partnerships [PPPs])
nutrition and, 194 pupil participation rates and, 221, 223
Plan de Ampliación de Cobertura de la quasi-public school markets and, 61–64
Educación Secundaria (PACES) and, 145 relative cost of, 122–126
subsidies and, 152–153 resource efficiency and, 47, 51–54
Pradesh, Andhra, 101 rural, 91–107
PriceWaterhouseCoopers (PwC), 230–231 school infrastructure and, 99, 101
Primary sampling units (PSUs), 92 sector size effects and, 112–118
Pritechett, Lant, 23 service delivery and, 80–82
Private finance initiatives (PFIs), 83–84, socioeconomic status and, 179–181, 250–
225–226, 253 251
Private schools state schools and, 223–224, 226
bribery and, 93 student assessment and, 13–14
Catholic, 4, 48, 80, 168, 170, 249 student selection and, 172–178
Chile and, 165–191 subsidies and, 52–53, 81, 127, 130–131,
classification of, 18 134, 152, 165–171, 178–188, 191n7, 251
classroom size and, 83, 91, 95–96, 101– superior results in, 23
103, 145, 196–197, 223–224 teacher absenteeism and, 5–6, 91–92, 95,
Colombia and, 143–163, 193–216 97, 103–105, 128
competition effect and, 101–103, 168, 189, unaided, 112–113
245, 247–249, 254 unrecognized, 111
contracting and, 80–82 vertical differentiation and, 56–59
cream skimming and, 47, 56–59, 65–66, vouchers and, 6–8, 62–64, 143–163, 251–
167 252
cross-country involvement analysis and, Probability proportional to size (PPS), 94–
17–23 95
effectiveness of, 118–122, 169–170 PROBE Team, 91, 116, 121–122
enrollment and, 113–115, 221 Program for International Student Assess-
equilibrium without peer effects and, 54– ment (PISA), 5
55 background questionnaires and, 18
free riders and, 51–52 database of, 17–18
funding patterns and, 18–23 literacy and, 17–18
growth in, 116–118 OECD and, 17–18
horizontal differentiation and, 54–56 public-private partnerships and, 13, 16–
housing choices and, 47, 59–64 18, 28, 32
India and, 91–103 Propensity-score matching methods, 112
internal efficiency of, 118–126 Psacharopolous, George, 155
introducing politics and, 53–54 Public-private partnerships (PPPs)
mobilization policies and, 47–67 accountability and, 23–24
niche markets and, 47 charter schools and, 4, 8, 75, 78, 80, 198,
nongovernmental organizations (NGOs) 226–227, 252–253
and, 18, 80 Chile and, 165–191
nonsubsidized, 122–123, 166–167, 170 city academies and, 225–237
Index 269
Colombia and, 143–163, 193–217 internal efficiency of, 118–126

concession schools and, 193–217 introducing politics and, 53–54
contracting and, 71–86 outcomes of, 224
debate over, 13 performance analysis and, 24–30 (see also
empirical models for, 24–30 Performance)
funding and, 14, 16, 18–23, 26–35, 129– price of, 49–50
130 private management of, 74–80
grants-in-aid system and, 129–130 quasi, 60–63, 65
historical perspective on, 126–130 relative cost of, 122–126
importance of cross-country differences residential segregation and, 60–61
in, 23–24 sector size effects and, 112–118
India and, 106–107, 126–134 student assessment and, 13–14
infrastructure model and, 73–74, 82–84 student selection and, 172–178
management and, 3–6, 71–86 teacher absenteeism and, 5–6
New Schools Project and, 83–84 voice and, 51–52
operation interactions and, 36–38 Pupil fixed-effects approaches, 112
performance analysis and, 24–40 (see also Pupil-teacher ratio (PTR)
Performance) Colombia and, 145, 196–197
policy design and, 84–86 India and, 91, 95–96, 101–103
poor families and, 24 United Kingdom and, 223–224
Program for International Student Assess-
ment (PISA) and, 13, 16–18, 28, 32 Quadrant-dummy specification, 28–29
as project-based endeavors, 13 Quality, 78, 80
proposed new form of, 130–133 ability tracking and, 58–59
regression analysis and, 25–40 (see also Chile and, 165–168, 171, 176, 185, 189–
Regression analysis) 190
resources and, 3–4 Colombia and, 155, 159, 161–162, 193–
systemic effects and, 13–14 197, 201–202, 208, 216
TeacherNet and, 225 concession schools and, 193–194, 197
teacher unions and, 127–129 cream skimming and, 47, 56–59
United Kingdom and, 19, 127, 219–240 future policy and, 245–249, 252–253
vouchers and, 38 (see also Vouchers) India and, 106–107, 129–131, 134
Public schools pupil-teacher ratio and, 96, 101–103, 196–
ability tracking and, 58–59 197
Chile and, 165–191 U.S. private sector mobilization and, 51–
classification of, 4, 18 68
Colombia and, 143, 162 Quasi-public schools, 60–63, 65
competition effect and, 247–247
concession school differences and, 193– Randomization strategy, 199–200
194 Rawlings, Laura, 145
contracting and, 74–80 Reading, 248
cross-country involvement analysis and, Colombia and, 155, 161–162, 198, 204,
17–23 208, 214
effectiveness of, 118–122, 169–170 India and, 121–125
enrollment of, 75 Program for International Student
equilibrium without peer effects and, 54– Assessment (PISA) and, 17–18
55 student achievement and, 13, 17, 33, 38,
failure of, 132 41
free riders and, 51–52 Regression analysis
full-fledged, 67 bootstrap procedure and, 214
funding patterns and, 18–23 Chile and, 167–188
India and, 97 Colombia and, 143–163, 196–216
270 Index
Regression analysis (cont.) Sarmiento, A., 198

concession schools and, 196–216 Schmid, Juan Pedro, 112
cross-country variation and, 25–38 School choice. See Policy
funding effects and, 26–35 Science, 13, 17, 33, 41, 191n8, 249, 256n1
India and, 104–105, 111–112, 119–122 Selection bias, 25–27, 35–36
Instituto Colombiano Para El Fomento Sen, Amartya, 118
De La Educación Superior (ICFES) and, Serritzlew, S., 249
150–152, 155–161 Service delivery model, 72–73
instrumental variables (IV) approach Shleifer, Andrei, 23
and, 254–255 Smith, J., 200
operation interactions and, 36–38 Solmon, L., 198
ordinary least squares (OLS), 95, 136n7, Sonstelie, J., 64
170, 188, 215, 249 State schools, 223–224
performance studies and, 25–30 city academies and, 226–237
Plan de Ampliación de Cobertura de la Local Education Authorities (LEA) and,
Educación Secundaria (PACES) and, 226
145–163 Stevens, M., 224
quadrant-dummy specification and, 28– Students
29 ability tracking and, 58–59
randomization strategy and, 199–200 cream skimming and, 47, 56–59, 65–66,
voucher effects and, 150–163 167
within-country variations and, 38–40 cross-type peer effects and, 55–56
Regression discontinuity design (RDD), 255 dropout rates and, 195, 202–203, 205, 207,
Rents, 49, 54 209, 213
Residential segregation, 60–61 fixed-effects approaches and, 112
Resources, 47, 51–52 horizontal differentiation and, 54–56
cross-type peer effects and, 55–56 Instituto Colombiano Para El Fomento
equilibrium and, 54–55 De La Educación Superior (ICFES) and,
horizontal differentiation and, 54–56 144–145, 150–162
income heterogeneity and, 53 learning disabilities and, 67
introducing politics and, 53–54 parents’ education level and, 25, 178–179
pedagogical targeting and, 54–56 participation rates and, 221, 223
vertical differentiation and, 56–59 peer effects and, 47, 54–59, 180
‘‘Right to Education’’ Bill, 9, 107, 130–132, per-pupil expenditures and, 68n5, 122–
134 123
Rockoff, J., 198 poor nutrition and, 194
Rodriguez, A., 197–198 Program for International Student
Rodriguez, Jorge, 169 Assessment (PISA) and, 13, 16–18, 32
Romaguera, Pilar, 169–170 pupil-teacher ratio (PTR) and, 91, 95–96,
Romano, R., 57–59 101–103, 145, 196–197, 223–224
Rouse, C. E., 248 selection issues and, 172–178
Russia, 19 socioeconomic status and, 179–181, 250–
251
Saavedra, Juan, 146 transfer certificates and, 93
Salary Distribution Act, 127 vertical differentiation and, 56–59
Samson, Meera, 91, 99 Sundararaman, V., 112
San Antonio, 247 Sweden, 14, 19, 254
Sanchez, Fabio, 143 Switzerland, 19
Sandström, F. M., 250–251 Synagogues, 4
Sanhueza, Claudia, 169 Systemic effects, 13–14, 167–169
Sapelli, Claudio, 169–170 System of Measurement of Educational
Saran, M., 125–126 Quality (SIMCE), 166, 171, 175, 189–190
Index 271
Tan, J., 155 TeacherNet and, 225

TeacherNet, 225 vouchers and, 132
Teachers United States, 5
absenteeism and, 5–6, 91–92, 95, 97, 103– charter schools and, 8, 75, 78, 198, 226–
105, 128 227
accountability and, 121–122, 128 competition effect and, 247–248
block grants and, 130–131 contracting and, 74–78
degrees and, 92 Florida’s accountability law and, 247–248
Direct Payment Agreement and, 127 further research for, 253–254
enrollment reporting and, 115 private sector mobilization and, 47–67
idleness and, 121, 128 public/private funding patterns and, 19
National Commission on Teachers and, vouchers and, 3, 6–7, 132
128–129 University Grants Council, 127
pedagogical targeting and, 47, 54–56 Urquiola, Mighel, 132, 167–168, 251
pupil-teacher ratio (PTR) and, 91, 95–96,
101–103, 145, 196–197, 223–224 Varghese, N. V., 118
salaries of, 91, 101, 125–127, 130–131, Vegas, Emiliana, 169
223 Venezuela, 5, 80
Salary Distribution Act and, 127 Vermeersch, C., 205
socioeconomic status and, 179–181 Vertical differentiation, 56–59
unions and, 127–129, 132 Vial, Bernardita, 170
Teal, Francis, 112 Villa, L., 193
Texas, 247 Voice, 51–52
Thayer, M., 64 Vouchers, 3–4, 9, 67, 245, 253
Tooley, James, 91, 93, 104–105, 116, 118 Chile and, 7, 132, 166–167, 170–181, 251
Torres, Carlos, 145 college exams and, 144
Trade unions, 18, 127–129, 132 Colombia and, 7, 132, 143–163, 251–252
Transfer certificates (TCs), 93, 113 cream skimming and, 65–66
Transparency, 72, 85 Czech Republic and, 250
Treatment effect models, 112 Denmark and, 248–249
Trends in International Mathematics and federally funded, 6–7
Science Study (TIMSS), 249, 256n1 further research for, 254–255
geographical targeting and, 63
United Kingdom, 4–5, 246 housing choices and, 62–65
charter schools and, 8 India and, 106, 132–133
city academies and, 8–9, 225–237 Instituto Colombiano Para El Fomento
city technology colleges (CTCs) and, 220 De La Educación Superior (ICFES) and,
Department for Children, Schools and 155–162
Families (DCSF) and, 221 legal issues and, 6–7
Education Act and, 220 lotteries and, 143–156, 159, 162
Education Reform Act and, 220 Netherlands and, 248–249
enrollment and, 221 per pupil spending and, 68n5
grant-maintained (GM) schools and, 220 Philippines and, 81–82
Local Education Authorities (LEA) and, Plan de Ampliación de Cobertura de la
226, 231, 234 Educación Secundaria (PACES) and,
private finance initiatives (PFIs) and, 83, 143–163
225–226, 253 political economy of, 63–64
private schools and, 220–224 private schools and, 6–8, 62–64, 143–163,
public-private partnerships and, 19, 127, 251–252
219–240 quasi-public schools and, 62–65
pupil participation rates and, 221, 223 student selection and, 172–178
state schools and, 223–224, 226 Sweden and, 250–251
272 Index
Vouchers (cont.)
targeting policies and, 62–67
teacher unions and, 132
United States and, 3, 6–7, 132
Washington, D.C., 7, 75, 78, 247

Weiner, Myron, 122
West, M. R., 247–248
Wilson, Joan, 8, 219–240
Wohlgemuth, Darin, 143, 145
World Bank, 13, 143, 194, 197
Wößmann, Ludger, 3, 5, 9, 13–45
Zeira, Joseph, 24

School Choice International

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

School Choice International

Uploaded by

Copyright:

Available Formats

School Choice International

School Choice International

Exploring Public-Private Partnerships

edited by Rajashri Chakrabarti and Paul E. Peterson

The MIT Press

For information about special quantity discounts, please email special_sales@mitpress

Library of Congress Cataloging-in-Publication Data

School choice international : exploring public-private partnerships / edited by Rajashri

I School Choice: Worldwide Perspectives

1 Perspectives in Public-Private Partnerships in Education 3

2 Public-Private Partnerships and Student Achievement:

3 Mobilizing the Private Sector in the United States:

4 The Practice of Public-Private Partnerships 71

II Traditional Forms of School Choice

5 Public-Private Schools in Rural India 91

6 School-Sector Effects on Student Achievement in India 111

III Public Funding of Privately Managed Schools

7 School Vouchers in Colombia 143

8 The Public-Private School Controversy in Chile 165

9 The Concession Schools of Bogotá, Colombia 193

10 Public and Private Schooling Initiatives in England 219

11 Education Contracting: Scope of Future Research 245

List of Contributors 259

The importance of school choice in education cannot be overempha-

Ferguson, Chester Finn, Ariel Fiszbein, Roland Fryer, Edward Glaeser,

Rajashri Chakrabarti and Paul E. Peterson

The undeniable importance of human capital for economic growth has

Chile, and Colombia; charter schools in the United States; concession

versus privately ﬁnanced and managed schools. Chapters in part III

possible answer to this apparent puzzle is that head teachers in private

voucher program in the United States was established in Washington,

importance of using correct empirical methodology and high-quality

Chakrabarti was a postdoctoral fellow at Harvard University in the John F. Kennedy

The issue of public-private partnership (PPP) is a much-debated topic,

possible differences in the effects of PPPs when they exist in different

lic funding of schools, 20 out of 29 countries have more than 87%

2.1 Public and Private Involvement in Schooling around the World

Before we describe the models and econometric evidence, this section

2.1.1 The PISA Database

(e.g., Bishop 1992). The speciﬁc student-level database used in this

2.1.2 Differing Patterns of Public and Private Funding and

Test Performance and Sample Size

2.2 Why Should the Public-Private Division Matter?

Do these cross-country differences in public versus private involve-

2.3 The Empirical Models

This section describes the different speciﬁcations of the empirical

2.3.1 Observables, Unobservables, and the Possibility of Selection

where Ti is the achievement test score of student i, Os is a dummy

Consider ﬁrst the observable background features B. Following the

EðgÞ ¼ b 2 þ ½ðFs Fs Þ1 Fs0 Ui b4 : ð2:2Þ

Assuming for the moment only one unobserved variable Ui , we can

2.3.2 Alternative Empirical Speciﬁcations

Ti ¼ a þ b1 TLc þ b2 BLc þ b3 BRc þ Bi b4 þ nc þ hs þ ei : ð2:5Þ

This quadrant-dummy speciﬁcation has the same advantages as the

Note that Os is now a dummy representing whether the student’s

where ac is a country-speciﬁc intercept implemented by adding a full

same country. The previous considerations suggest that the relative

By doing so, this speciﬁcation can depict whether the within-country

2.4 Empirical Results on Public-Private Division and Student

This section presents the empirical results on the association between

2.4.1 Public vs. Private Funding and Operation Measured at the

Country-level public-private measures School-level public-private measures