Professional Documents
Culture Documents
School Vouchers: A Survey of The Economics Literature: Dennis Epple, Richard E. Romano, and Miguel Urquiola
School Vouchers: A Survey of The Economics Literature: Dennis Epple, Richard E. Romano, and Miguel Urquiola
School Vouchers: A Survey of The Economics Literature: Dennis Epple, Richard E. Romano, and Miguel Urquiola
https://doi.org/10.1257/jel.20150679
441
442 Journal of Economic Literature, Vol. LV (June 2017)
who is eligible to receive them, their source 5. What political-economy factors deter-
of funding, and the criteria for private-school mine the existence and design of
participation. For example, vouchers are fre- voucher programs?
quently “targeted” to low-income students,
are sometimes funded privately rather than Research frequently focuses on more spe-
through tax proceeds, and religious schools cific questions that help to get at these
may or may not be eligible to participate.2 fundamentals.
The research that has analyzed these Our review begins by describing the issues
programs seeks to answer five fundamental and controversies that frame the research on
questions: these questions (section 2). We set out the
“for” and “against” positions typically (and
1. What effects do vouchers have on the at times informally) cited on vouchers. The
students who use them? brief discussion is sufficient to show that the
2. Do vouchers induce nonrandom migra- answers to questions 1–5 can depend on both
tion of students from public to private the characteristics of the program analyzed
schools, possibly lowering the achieve- and the context into which it is introduced.
ment of students that remain in the We then summarize the features of
public sector via peer effects or other voucher programs that have been imple-
channels? mented throughout the world (section 3).
3. Do voucher programs pressure pub- We make a distinction between two program
lic schools to become more efficient, types. First, by small-scale programs, we
increasing the achievement of students mean those that place significant restrictions
that remain in the public sector? on who can receive vouchers. The most com-
4. What is the net effect of vouchers on mon restrictions involve income or geogra-
aggregate educational performance? phy; for instance, vouchers may be made
available only to low-income children in a
given municipality within a country. Second,
2 This program heterogeneity poses a challenge in deter- by large-scale programs we mean those in
mining which programs to classify as voucher programs. which vouchers are distributed countrywide
From the perspective of this review, the most difficult deci-
sion regards whether to include charter schools under the with minimal restrictions on the type of chil-
voucher umbrella. Most charter-school advocates would dren who can use them.
strenuously resist classifying charter schools as voucher We then present a brief synopsis of the
schools, pointing to state requirements that the former be
chartered as public schools, subjected to the oversight of theoretical literature (section 4). It reveals
the state-designated charter authorizer, and bound by con- that even in a qualitative sense, the answers
straints on admissions and funding. The c ounterargument to questions 1 –5 depend on voucher design.
is that, in some states, charter schools can be privately
owned and operate under constraints on admission and The main exception surrounds question 2,
financing not markedly different from those imposed by where most models suggest that voucher sys-
some of the more restrictive voucher programs. While tems will display a tendency towards strati-
there is merit to both perspectives, we view the distinction
between charters and vouchers to be meaningful despite fication by ability and/or income—although
the fuzzy boundary between the two, in part because all of this too can be mitigated by design and
the voucher programs we review permit use of the voucher depends on context.
in private schools. Moreover, there is tremendous hetero-
geneity in charter school characteristics, both across and Finally, we turn to reviewing the empirical
within states, making inclusion of charters in this review work—the focus of this survey (section 5). In
unwieldy if not unmanageable. Hence, we have chosen terms of question 1, the empirical research
not to include research on charter schools in this review.
See Epple, Romano, and Zimmer (2017) for a review of does not suggest that awarding students a
research on charter schools. voucher is a systematically reliable way to
Epple, Romano, and Urquiola: School Vouchers 443
improve their educational outcomes. A per- For instance, the accumulated research has
haps surprisingly large proportion of the begun to provide guidance regarding how
most rigorous studies suggest that being voucher programs may be formulated to
awarded a voucher has an effect that is sta- limit adverse effects related to sorting while
tistically indistinguishable from zero. At the preserving achievement-enhancing effects
same time, there is also evidence that in related to competition. Third, there is evi-
some settings, or for some subgroups or spe- dence that the returns to a well-functioning
cific outcomes, vouchers can have a substan- education system can be large, with the asso-
tial positive effect on those who use them. ciated implication that a good understanding
In addition, however, some recent evidence of voucher design could be very useful.
points to some discouragingly large negative A final note is that given the evidence we
test score effects. In terms of question 3, have reviewed, our sense is that work orig-
the literature generally suggests that com- inating in a single case (e.g., a given coun-
petition from vouchers leads public schools try) or in a single research approach (e.g.,
to improve. That said, it also makes clear randomized-control trials) is unlikely to fully
that it is very difficult to isolate the effect of answer questions 1–5. The work on vouchers
competition on public-sector value added suggests that educational markets are com-
(the object of interest); this reflects that as plex, and that therefore fairly wide ranging
implied by the answer to question 2, vouchers empirical and theoretical work will be neces-
typically lead to sorting and can thus affect sary to make progress.
public schools through channels other than
productivity enhancements. Taken together,
2. The Issues
these findings point to an ambiguous answer
to question 4 regarding the net effect of To provide perspective for our review, we
vouchers. Finally, empirical work finds sup- begin with an overview of the types of issues
port for theoretical predictions regarding the theoretical and empirical research on vouch-
political economy of voucher adoption. ers must address. These issues are complex,
Our “bottom line” assessment is that those in part because the effect of a voucher pro-
hoping for definitive answers to questions gram depends on both its design and the
1–5 will not find them in the research to institutional and economic setting in which
date. In our view, the available answers to it is introduced. For instance, the effects
these questions are insufficient to warrant may depend on the size of the program and
recommending that vouchers be adopted also on the alternative: What educational-
on a widespread basis. In that respect, the provision regime would prevail without the
effects of vouchers have been disappointing, voucher program? The “effects” that are of
relative to early views on their promise. interest are themselves a fundamental issue.
That said, our view is also that the record What is the social objective?
definitely warrants continued explora- To introduce these issues, we list some
tion. This assessment reflects three factors. classic claims that frame the voucher con-
First and as stated, there is evidence that troversy—and virtually everything about
in some cases vouchers can have signifi- vouchers remains controversial—with-
cant positive effects on educational perfor- out discussing any literature. This listing
mance, or at least produce substantial cost will begin to illustrate the challenges that
savings. Second, there is some indication research on vouchers faces, and our hope is
that the prevalence of such results might be that the subsequent review will help to c larify
increased with improved voucher design. what issues remain the most unsettled. We
444 Journal of Economic Literature, Vol. LV (June 2017)
begin simply by stating some arguments for wherein quality rises with socioeco-
and against vouchers, essentially as they are nomic status would be reduced. In short,
commonly expressed. vouchers would provide both efficiency
The arguments in favor of vouchers and equity gains.5
include:
The arguments against vouchers include:
• Vouchers would lead to market or
quasi-market provision of education,
• Vouchers would lead to the sorting of
with competition among providers and students across schools along charac-
choice by students inducing efficient teristics like income and ability (such
provision.3 The alternative of public pro- sorting is often referred to as stratifica-
vision is characterized by weak incen- tion). For example, the private sector
tives, both because public providers are might “cream skim” the highest income
politically managed and monopolized, or most motivated children away from
and because the exercise of student public schools. Teachers would sort as
choice is limited. Thus, both static and well—the most advantaged students
dynamic efficiency would be promoted would be taught by the best teachers
by vouchers, with gains coming both and the least advantaged by the worst.
from private-school advantages and a • Such sorting would have negative con-
public-school response. sequences due to peer effects. These
• Market provision would lead to educa- would arise directly, for example, if the
tional variety, better matching prefer- ability to interact with higher-achiev-
ences to supply. Diversity would increase ing peers helps students to learn or to
with respect to aspects like curricula and acquire useful networks. Peer effects
teaching methodology, an improvement could also reflect indirect mechanisms,
over the excessive homogeneity associ- e.g., if school oversight of wealthier par-
ated with monopolized public provision. ents disciplines school administrators
• While there might be concerns regard- and teachers.
ing externalities from educational attain- • Even if peer effects do not exist, sort-
ment (for example, a modern democratic ing would adversely affect less-advan-
society requires citizens to be literate taged students through informational
in a common language), restrictions on channels. For example, being at a “bad”
private providers could address these. school could stigmatize students in the
Similarly, the level of the voucher would labor market, affecting their incentives
address capital market failures affecting to study. Further, sorting might be det-
educational investment.4 rimental if the mixing of students along
• By decoupling residence and school categories such as race and religion
choice, vouchers would increase access
to quality education, especially for stu-
dents at the low end of the s ocioeconomic 5 Additionally, a more philosophical argument for
ladder. Stratified educational provision vouchers rests on the substitution of student choice for
public choice. As noted, the traditional economic version
of this argument emphasizes better matching of prefer-
ences to educational supply. The noneconomic version of
3 In our discussion, we will refer to the student and the the argument places value on freedom of choice per se,
student’s household as just the “student.” while rejecting the paternalistic alternative. Since the liter-
4 If necessary, this could be supplemented with policies ature we review considers mainly economic outcomes and
supporting educational loans. considerations, we abstract from such issues.
Epple, Romano, and Urquiola: School Vouchers 445
Milwaukee Racine Florida Cleveland Douglas, CO Ohio Indiana New Orleans Washington, DC
Targeting Below 300% Below 300% Failing school Open to all, All students Underperforming Below 150% Below 250% of Below 300%
poverty poverty 2013– priority to low public poverty or failing poverty poverty or on
income school food stamps
Admission to Lottery Lottery Lottery Lottery Lottery Pvt school Pvt school Pvt school Pvt school
oversubscribed admission admission admission admission
pvt schools criteria criteria criteria criteria
Years of program 1990– 2011– 1999–2006 1995– 2013– 2005– 2011– 2008– 2004–
Enrollment 24,027 in 2012 520 in 2012 3,649 in 2012 6,300 in 2009 494 in 2011 12,685 in 2009 3,919 in 2011 8,000 2013 3,105 in 2012
(cap 500)
Funding per $6,442 in $6,442 in K–8: $4,250; K–8: $4,250; $4,600 in K–8: $4,250; $4,700 for K–8, $7,617 maximum K–8: $8,256;
student 2012/13 2013/14 HS: $5,000 HS: $5,000 2011/12 HS: $5,000 HS slightly in 2011/12 HS: $12,385 in
higher 2013
Transportation Some, not Some, not Yes if voucher Yes Yes Up to school No Voucher can be
provided routinely routinely school in used to pay for
student’s district metro/bus
Number of 112 in 2012 13 in 2013 32 in 2012 35 in 2012 23 in 2013 305 in 2009 280 in 2013 about 130 52 in 2012
participating
private schools
Can tuition K–8: No. HS: K–8: No. HS: K–8: No. HS: K–8: No. HS: Yes Yes, above 200% Yes No Yes
supplement be Yes above 220% Yes above 220% Yes above 200% Yes above 200% poverty
required? poverty poverty poverty poverty
Epple, Romano, and Urquiola: School Vouchers
Can vouchers Yes since 1998 Yes Yes Yes Yes Yes Yes Yes Yes
be used at
religious
schools?
Same Yes beginning Yes Yes Yes Yes Yes Yes Yes Yes
achievement 2006
exams as in
public schools?
447
448 Journal of Economic Literature, Vol. LV (June 2017)
voucher programs in the United States.9 All in 1998, with students enrolled in religious
of these programs apply for grades K–12, schools retaining the right to opt out of
and almost all are targeted to students in religious instruction. The voucher pays the
low-income households or schools desig-
lesser of tuition at a private school and the
nated as underperforming. Programs vary standard district allocation, $6,442 in 2010.
in age, ranging from Milwaukee, which Initially, schools could not charge additional
started in 1990, to New Orleans, which tuition. Beginning in 2011, high schools were
started in 2008. Programs also vary substan- permitted to charge additional tuition to eli-
tially in the number of students receiving gible students above 220 percent of the pov-
vouchers. Several programs require that erty line. Transportation is provided by the
oversubscribed voucher schools choose district if the student is within a set atten-
students by lottery. Others permit private dance area. Participating private schools set
schools to apply the same admission crite- the number of available slots for voucher stu-
ria for voucher as for non-voucher students. dents, and must accept all students, conduct-
Funding per student varies across programs, ing a lottery if oversubscribed. They must
but most provide sufficient resources to also be accredited by one of several agencies.
attract participation of a substantial number Private schools must also meet at least one of
of private schools. Vouchers may be used in the following four performance standards:10
religious schools in all of these programs. (1) at least 70 percent of voucher-supported
All now require that voucher recipients take students must advance a grade level, (2) fre-
the same standardized examinations as pub- quency of attendance by voucher stu-
lic school students. While not detailed in dents must be at least 90 percent, (3) at
the table, all programs require participating least 80 percent of program students must
schools to meet curricular and other criteria. demonstrate significant academic progress,
Some programs (e.g., Milwaukee) require or (4) at least 70 percent of v oucher-student
schools to be pre-accredited by an approved families must meet parental involvement cri-
national agency; some (e.g., Ohio) require teria set by the school.
that schools obtain a state charter, and oth- The income threshold for eligibility has
ers require annual reporting to an oversight been on an upward trend. It was raised from
body. the initial 175 percent of the federal poverty
Milwaukee is in many respects the most level to 220 percent in 2005, and 300 percent
important voucher program in the United more recently. This allowed the program’s
States and has served as a model for others; coverage to grow. In 2004 it distributed
it therefore merits additional discussion. The about 24,000 vouchers, accounting for about
Milwaukee Parental Choice Program was 23 percent of total district enrollments. By
introduced in 1990 in the Milwaukee school 2002, 102 private schools were participating,
district, targeting K
–12 students with house- including 26 schools reported as entering as
hold income not exceeding 175 percent of a result of the voucher program (Chudgar,
the federal poverty level. At its inception, Adamson, and Carnoy 2007).
the program was not available to students
attending religious schools. That changed Tax-Credit-Funded Programs.—We next
turn to tax-credit-funded voucher programs,
9 Though we discuss programs in the present tense, the
Florida program described in table 1, known as the Florida
Opportunity Scholarship Program (FOSP), was declared 10 Wisconsin Administrative Code, chapter PI 35,
unconstitutional and suspended in 2006. p. 117.
Epple, Romano, and Urquiola: School Vouchers 449
summarized in table 2.11 The operation and sponsoring organization is the CSF, and a
funding of one of the earliest, the Florida brief description of its operation illustrates
Corporate Income Tax Credit Scholarship key characteristics of this type of program.
Program (FTC), illustrates similarities and The CSF received a founding contribution
differences between these programs and from the Walton Family Foundation and
the tax-funded programs summarized in has provided vouchers to low-income stu-
table 1. The FTC was established in 2001 dents in numerous municipalities includ-
and is financed by corporate contributions, ing New York City, Charlotte, Dayton,
for which donors get a 100 percent corporate Baltimore, and Washington, DC. In 1999,
income tax credit for contributions that do it received 1.25 million applicants for
not exceed 75 percent of their tax liability. 40,000 vouchers.
Total contributions are capped at 88 million Its Baltimore program is typical. It tar-
dollars, currently. Vouchers are for free- or gets low-income students in grades K –8. In
reduced-lunch students and the program is 2008, it distributed vouchers to 490 students
administered by approved nonprofit agen- attending 70 private schools, 64 of which were
cies. In 2012, the FTC awarded about 51,000 religious. The average voucher was $1,759
vouchers to students attending about 1,300 and the maximum was $2,000. Families are
schools. This makes it the largest voucher required to pay at least $500 themselves,
program in the United States—roughly twice and their average payment was $2,711. All
the size of Milwaukee, although relative to a privately funded programs of which we are
much larger population. aware are similarly targeted, typically by
As shown in table 2, the average voucher per income, and some to racial/ethnic groups.
student among tax-credit-funded programs Some also target students that are identified
was $4,335 in the 2012 school year. Private as having high academic potential but lim-
schools may impose their own admission poli- ited means.
cies, restricted only by antidiscrimination stat-
3.1.2 Colombia13
utes, and can charge tuition in addition to the
voucher, so long as this is their normal policy. Small-scale voucher programs are much
less common outside the United States, but
Privately Funded Programs.—Roughly Colombia provided a salient, if short-lived,
fifty privately funded voucher programs example. Specifically, in 1992 it began oper-
also exist in the United States.12 The largest ating the PACES secondary-school voucher
program.14 This initiative was launched to
increase secondary-school enrollment—the
11 The tax-credit-funded programs detailed in table 2
intent was for private participation to help
are to be distinguished from state income tax credit and
deduction programs available to households for educa- ease public-sector capacity constraints. The
tional expenses that currently exist in Arizona, Minnesota, vouchers, which were renewable contingent
Illinois, and Louisiana. Given restrictions on amounts, on grade completion, were targeted at
eligibility, and state income taxation, these programs have
limited effects. For example, the most recently passed pro- entering students who were: (1) residing in
gram in Louisiana in 2008 allows taxpayers to deduct from
the state income tax 50 percent of educational expenses,
up to the minimum of $5,000 per child or the total tax-
able income of the individual. With the maximum marginal 13 For further reference see King et al. (1997), King
income tax of 6 percent in Louisiana, the maximum sub- et al. (1998), and Angrist et al. (2002), on which this dis-
sidy to educational expenditure is below $300. cussion is based.
12 This calculation counts separately programs in dif- 14 PACES stands for Programa de Ampliación de
ferent municipalities that are administered by the same Cobertura de la Educación Secundaria—program for
organization. increasing secondary school enrollment.
450
Table 2
Tax-Credit-Financed Voucher Programs
Targeting Below 230% of All public-school Below 200% of Below 300% of Below 300% of Below Below 250% of
poverty level students poverty level or in poverty poverty level or in $75,000 and in poverty level.
underperforming underperforming u nderperforming
public school. public school. public school.
Enrollment 51,023 in 2 012–13 13,285 in 2011–12 2,890 in 2012–13 10,600 in 2 011–12 N/A 45,100 in 2011–12 382 in 2012–13
Funding per $4,335 average in $3,388 average in $880 average in $1,031 in 2 011–12 Up to $5,000 $990 average in $2,690 average in
student 2012–13 2011–12 2012–13 2011–12 2012–13
Number of 1,330 in 2 012–13 N/A N/A 154 in 2011–12 N/A 400 in 2012–13 54 in 2012–13
participating
private schools
Notes: Tax credits to businesses are generally limited as is the statewide total credit and amount available to provide vouchers to eligible students. Lotteries are
then generally used to provide vouchers to applicants. The poverty level for eligibility is generally measured by the federal poverty level.
Epple, Romano, and Urquiola: School Vouchers 451
TABLE 3
International Voucher Programs
Enrollments in private
or independent For-profit Selective Religious Significant
Years in voucher-funded operation admissions affiliation tuition charges
Country operation schools allowed allowed allowed allowed
as municipal schools. These changes resulted rotests. The current Bachelet administra-
p
in substantial private-school entry. By 2009, tion has submitted to congress a proposal
about 57 percent of all students attended with three main ingredients: (1) the elimi-
private schools, with voucher schools alone nation of tuition top-ups, (2) the end of the
accounting for about 50 percent.20 ability of private voucher schools to operate
Chile’s scheme imposes few restrictions on for profit, and (3) the introduction of a sig-
private schools. These can receive voucher nificant reduction of the ability of private
subsidies regardless of their religious status schools to select students. The proposals
and operate for profit. They are allowed to are still under discussion, and the details of
implement admissions policies subject to implementation remain to be seen.
few restrictions and, as of 1994, can charge
3.2.2 Denmark24
tuition add-ons.21 The latter are capped at
about four times the voucher payment, but Denmark has a long tradition of subsi-
this constraint rarely binds.22 Public schools dizing independent schools. Since the Free
operate under more restrictions. They are School Act of 1855, it has allowed parents
not allowed to turn away students unless and organizations to set up independent
oversubscribed, and cannot charge tuition schools to which any child can apply, and
at the primary level. All schools must imple- which are allowed to have religious affilia-
ment elements of a national curriculum and tions. Historically, these schools were funded
participate in annual standardized exams, through a scheme by which the State reim-
the results of which have been public since bursed a large portion of their expenses.
the 1990s.23 In 1992, this system was replaced with one
Recent years have seen further reforms. that provides independent schools with a
Since 1997, schools charging tuition a dd-ons grant based on the number of pupils enrolled
are required to provide exemptions on these by a certain date. Public schools continue to
for a percentage of low-income students. be financed by a combination of national and
In 2008, the flat voucher became differen- local government allocations; they do not
tiated: it was increased for low-income stu- receive per-student payments.
dents. However, not all schools receive these The voucher-type payment for indepen-
additional subsidies, as they have to comply dent schools is indexed to expenditures in
with a number of conditions to receive them. public schools and varies with two factors:
Even further significant reforms to the school size (with higher payments for smaller
voucher system are under active discus- schools, to account for economies of scale)
sion, in part in reaction to persistent student and the age distribution of students and
teachers. These payments cover only about
20 The “elite” unsubsidized private schools continued to 80 percent of average educational costs, and
account for about 6 percent of enrollments. independent schools are therefore allowed
21 Over the years, e ducation-related legislation often to charge tuition (low-income households
mentions that private schools should not select students.
The anecdotal evidence indicates that this rarely binds— can apply for waivers) or seek external grants
for instance no admissions lotteries are required. We to cover the remainder. Despite this, total
return to this issue below.
22 Most of the “elite” unsubsidized private schools
per-pupil expenditures are slightly lower in
could take vouchers but choose not to; see Urquiola and
Verhoogen (2009).
23 These tests have been used for purposes of account-
ability and targeting. For instance, Chay, McEwan, and
Urquiola (2005) consider a program that targeted the 900 24 For further discussion see Justesen (2002), on which
worst-performing schools in the country. this discussion is based.
454 Journal of Economic Literature, Vol. LV (June 2017)
the independent-school sector.25 By 2005 admission for various reasons, including reli-
independent schools accounted for about gious affiliation.
12 percent of enrollments. They operate
3.2.4 New Zealand 27
subject to nationwide collective agree-
ments with teachers, and to basic curricu- In 1989, New Zealand implemented a
lar requirements that leave them relatively decentralization initiative transferring con-
broad pedagogical autonomy. trol of each public school from a national
department of education to a “Board of
3.2.3 Holland26
Trustees”—largely consisting of parents—
The Dutch 1848 constitution allows for elected locally. In 1991, this system was
churches, foundations, and parental associ- extended by granting per-pupil funding to
ations to set up independent school boards all schools, including independent and “inte-
that operate schools to which any child can grated” institutions. The latter are schools
apply. The 1917 constitution further includes which, while being institutionally indepen-
commitments of equal financial support for dent, had been affiliated with the public sys-
public and independent schools. While both tem since the 1970s; most, though not all,
types of schools receive funds for infrastruc- have a religious affiliation. At present, the
ture, a substantial part of schools’ support is enrollment shares of public, integrated, and
in the form of a per-pupil grant, with greater independent schools are 85, 11, and 4 per-
payments when they enroll children of low cent respectively.
socioeconomic status. These arrangements put in place a key
While this system was initially set up to ingredient of a voucher system—schools that
allow for transfers to Catholic and Protestant attract more students receive greater fund-
schools, at present it also covers schools with ing. That said, they depart from the textbook
other religious affiliations. Nevertheless, a voucher in some ways. First, not all schools
majority of independent schools still identify receive the same p er-student funding. Public
as Protestant or Catholic, with enrollment schools receive subsidies for teacher sala-
shares of 27 and 29 percent, respectively. ries, operational costs, and capital expenses;
The public sector’s share is 35 percent, integrated schools are only compensated for
with the remaining 9 percent of children in teacher salaries and operational costs; and
schools of other types. Independent schools independent schools receive only a portion
must be run on a n ot-for-profit basis and of the p er-student payments awarded to
“top-up” tuition charges are not allowed. integrated schools (the percentage has fluc-
In addition, these schools must implement tuated around 30 percent over the years).
at least parts of a core national curriculum, Second, public and integrated schools do not
participate in national standardized exams, have control over teacher pay; pay scales are
and comply with regulations regarding centrally determined for all but the indepen-
aspects like class size, teacher qualifications, dent schools.
and minimum enrollments. Private schools In addition, while public schools may
implement selection policies and may deny supplement their central subsidies via fund-
raising activities and donations, they are not
25 Justesen (2002) indicates that in the aggregate, 77 allowed to charge mandatory fees. Integrated
percent of independent schools’ resources come from schools are allowed to collect donations and
voucher-type payments, 18 percent from user fees, and the
remaining 5 percent from other external sources.
26 For further reference, see Justesen (2002), Patrinos 27 This description is based on Ladd and Fiske (2001);
(2002), and Levin (2004), on which this discussion is based. Adams (2009); and Lubienski, Lee, and Gordon (2013).
Epple, Romano, and Urquiola: School Vouchers 455
Bohlmark and Lindahl (2012) point (p. 89). Implicit in this government role
out that there was relatively little voucher would be assurance that voucher funds be
school entry through 1998; at that point, the spent on education. Friedman argued that
independent-school share began to grow. By competition for students would induce
2004, the independent-school enrollment schools to operate efficiently and reward
rate reached 10 percent for high schools and quality teaching, with effective schools
6 percent for primary and lower secondary. establishing good reputations. The poor
By 2009, independent schools accounted for would have educational choices not bound
roughly 10 percent of all students. The rea- by the residence restrictions embodied in
sons for the relative lack of voucher school neighborhood public school systems. In
entry for the first several years are unclear. Friedman’s view, the education environment
These could include anything from a lack was not sufficiently different from other
of information or risk aversion on the part market settings to interfere significantly with
of parents, to the fact that the 1996 relax- effective functioning of such a marketplace
ation of centralized wage setting might have for education.
allowed independent schools to compete More recent research has modeled edu-
more effectively. cational vouchers taking account of distinc-
tive features of the education environment.
Table 4 provides a summary of the charac-
4. Theory
teristics of models that we discuss in this
In this review, our primary emphasis is section. The delineation of model charac-
on empirical research (a comprehensive teristics in table 4 is imperfect and does not,
review of theoretical and computational of course, fully describe differences across
research is provided by Epple and Romano papers. For example, the table indicates
2012). Nonetheless, this section provides a whether the vouchers that are studied are
brief summary of the theoretical literature targeted, but does not indicate the type of
with an emphasis on empirical and policy targeting, which varies in important ways
implications. across studies. Likewise, there are import-
The case for a market-based educational ant differences in what makes public schools
voucher was laid out by Milton Friedman heterogeneous in the models that have such
(1962), who provided a vision for voucher differentiation. In the discussion below,
design and an enumeration of the benefits these modelling differences are highlighted.
that he foresaw from voucher adoption. He In reviewing this recent literature, we make
supported public funding of education on reference to how it helps to address the five
the grounds that such funding was warranted fundamental questions on vouchers that we
by the social externalities flowing from an set out in section 1.
educated population and due to borrowing A central theme that emerges is that
constraints, but argued that public fund- the answers to these questions depend on
ing need not imply public provision. He voucher design. Regarding question 2, vir-
envisioned a system in which parents could tually all theoretical analyses predict that
choose a school for their child with public a laissez-faire design will induce “cream
funding going to the chosen school. The role skimming,” with the associated implication
of government would be to provide funding for question 1 that some students will gain
while “… insuring that schools meet certain more than others; and some will be made
minimum standards, such as inclusion of worse off unless the effects on public-school
minimum common content in programs…” performance (question 3) are substantial.
TABLE 4
Characteristics of Theoretical Models of Vouchers
Public schools Student differentiation Vouchers Public choice
Private
Peers affect Homoge- Heteroge- Rent Imperfectly No topping Limit price Housing Public
Voucher models quality neous neous seeking Income Aptitude observed Universal Targeted up admissions discriminate market expenditure Voucher
Manski (1991) X X X X X X
Epple and Romano X X X X X X
(1998)
Nechyba (1999) X X X X X X X
Nechyba (2000) X X X X X X X
Nechyba (2003) X X X X X X X X
McMillan (2005) indirectly X X X X X
Ferrerya (2007) X X X X X X
Epple and Romano X X X X X X X X
(2008)
Macleod and Urquiola X X X X X
(2009)
Ferrerya and Liang X X X X X X X
(2012)
Chakrabarti (2013c) X X X X
Nielson (2013) X X X X X
Chakrabarti (2013d) X X X X X X
Public-choice models
Ireland (1990) X X X
Epple and Romano X X X X
(1996)
Epple, Romano, and Urquiola: School Vouchers
Theoretical models are often paired with for educational quality rising with income
a computational counterpart to quantify and motivation; a positive peer externality
magnitudes, distributional effects, and, from highly motivated students; educational
with respect to question 4, net impacts. As quality determined by expenditure per stu-
problematic implications of the laissez-faire dent and peer quality; analysis of alternative
design have become better understood, public-sector objectives including rent seek-
research has increasingly emphasized ways ing; and zero-profit private schools that set
in which benefits from voucher-induced tuition to maximize enrollments with tuition
competition can be obtained without adverse and a voucher spent on educational inputs.31
distributional effects. While the natural Manski uses this setup to assess if vouchers
focus with regard to question 1 is on educa- would induce changes that equalized edu-
tional outcomes, theory has also developed cational opportunity. The simulations and
interesting implications regarding impacts outcomes he considers are numerous, but
on residential choice and housing values, and overall the conclusion is that vouchers are
the connection to voucher design. Regarding not a “panacea.” A key prediction is that, as
question 3, theoretical research has iden- the voucher level rises, the fraction of highly
tified potential sources of efficiency gains motivated students in the public schools
from educational competition, as well as tends to fall, especially in poor communities.
ways that public-school performance might He states that “even in the most favorable
be adversely affected. Failures of voucher case, a systemic choice system would not
proposals at the ballot box have motivated come close to equalizing educational oppor-
research addressing question 5. tunity across income groups.” Thus, Manski’s
analysis predicts that cream skimming of the
4.1 The Effects of Vouchers
sort raised in question 2 will adversely affect
The theoretical and computational litera- less-motivated students.
ture typically begins from a characterization Epple and Romano (1998) study private-
of the educational environment, while taking and public-school competition when stu-
the existence and characteristics of vouch- dents vary in ability and household income,
ers as exogenous. The question then is how and school quality increases with peer abil-
the introduction of a given voucher program ity. Private schools maximize profits and
into a school “market” affects school effec- can price discriminate, i.e., charge tuition
tiveness, the distribution of outcomes and that varies with student ability and income.
welfare across students, the distribution of Schools have fixed costs, as well as variable
students across schools, tax revenues, public- costs that are an increasing convex function of
school expenditures, residences, and prop- enrollment, i.e., cost per student is U-shaped
erty values. In our discussion below, we draw in enrollment. The model gives rise to the
out the predictions of theoretical models, following predictions. First, because school
while also noting those that have not yet quality increases with peer ability, private
received empirical testing. schools charge lower tuitions (i.e., provide
Manski (1992) pioneered this type of
approach developing a theoretical and com- 31 While Manski describes students as varying in motiva-
putational model that captures features of the tion, we label such variation as “aptitude” in table 4. Many
educational environment including: public authors describe students as varying in ability, and we have
and private sectors between which students just chosen one term in the table to describe student vari-
ation along these lines. Note, too, that Manski considers
can choose; students differing by house- rent-seeking public schools as we indicate in table 4, but
hold income and motivation, with demand also other objectives.
Epple, Romano, and Urquiola: School Vouchers 459
more financial aid) for high-ability students. serving the relatively more able and afflu-
Second, higher-income households with ent, lower-achieving voucher schools serving
low-ability students pay a tuition premium to the relatively less able and affluent, and a
enable their children to attend high-quality public-school sector with lower achievement
schools. Thus, the school system will have still. Epple and Romano (2008) extend this
a general tendency towards stratification setup to show that these properties persist
in two respects. There will be stratification when school quality depends on expenditure
across schools within the private-school sec- per student in addition to peer quality.33
tor and there will be stratification between Could all students nonetheless have
the public and private sectors. Moreover, improved educational outcomes with the
private schools will be differentiated in voucher? If private schools have an educa-
quality and will exhibit “diagonal stratifi- tional approach that is sufficiently superior to
cation,” with each private school having a that of the public schools they supplant, and
student body that is a “diagonal slice” in the if the remaining public schools are induced
income–ability plane. The
lowest-income by competition to adopt a superior delivery
and lowest-ability students will attend public approach, all students might have higher
schools. Thus, the model predicts that there achievement than in the no-voucher equilib-
will be a higher correlation between income rium. Computational analysis calibrated to
and ability in public than in private schools.32 the US context suggests, however, that some
The model predicts that introduction of students will benefit from the voucher—the
a universal ( flat-rate) voucher will induce comparatively more able and affluent—while
additional private schools to enter, with each others—the comparatively less able and
entering school being of lower quality than affluent—will be hurt. Regarding question 4,
the preceding entrant, and each exhibiting the effect on normed aggregate achievement
the diagonal-stratification pattern. As the (equal to future earnings) may be positive or
amount of the voucher increases, average negative depending on the extent to which
peer quality in the public schools is pre- private-school education delivery is more
dicted to decline as private entrants “cream effective than preexisting public schools, and
skim” higher-income and higher-ability the extent to which public schools upgrade
students from the public schools. If a com- delivery in response to competition. In
paratively low voucher is introduced, those summary, the model yields unambiguous pre-
switching to private schools will attend a dictions about stratification, private-school
school with higher peer quality than the pricing, and relative achievement across the
public-sector school they depart. As the
predicted school hierarchy, while predicted
voucher level is increased, however, stu- aggregate effects depend on the impact of
dents who are induced to switch to private vouchers on educational effectiveness. It
school exit a public-school sector whose
peer quality has been diminished by cream 33 One additional finding is that low-quality “bottom
skimming to a private school with compar- feeder” schools may enter when vouchers are available,
atively low peer quality. Thus, regarding providing financial aid “kickbacks” to induce low-income
questions 1 and 3, the model predicts that households to choose low-quality schools. It is shown that
this can be prevented by a mandate that the voucher be
there will be high-achieving voucher schools spent on education. Theoretical models have generally
assumed that kickbacks are not allowed. The incentive to
kickback monies to poor students raises the related ques-
32 These and other predictions of the model are tested tion as to whether vouchers would lead schools to pro-
in Epple, Figlio, and Romano (2004) and are found to be vide noneducational goods to students as a way around a
supported by the data. requirement to spend all of a voucher on education.
460 Journal of Economic Literature, Vol. LV (June 2017)
bears emphasis that these predictions are Relative to the design of Epple and Romano
for a universal (flat-rate) voucher design, the (2008), the absence of enhanced voucher
Chilean voucher (at least in roughly its first funding for low-ability students reduces
two and a half decades) being perhaps the incentives for schools to seek out and retain
closest observed counterpart. less able students.
Epple and Romano (2008) also investigate Nechyba (1999, 2000, 2003) analyzes the
the implications of voucher design for cream effects of voucher programs in multi-district
skimming, showing that an ability-targeted local economies. He develops a rich theoret-
and tuition-constrained voucher can preserve ical and computational model to investigate
efficiency benefits from competition while the effect of several voucher programs under
eliminating cream skimming and providing alternative public-school financing schemes.
relatively uniform benefits across the distri- He demonstrates the importance of house-
bution of student income and ability types. hold mobility and general equilibrium effects
The tuition constraint disallows “top-ups” in predicting outcomes from large-scale
and “kickbacks.” Such an ability-targeted voucher programs. In his 1999 framework,
design has not been implemented in prac- there are multiple local school districts, a
tice.34 Voucher designs requiring that fixed stock of heterogeneous housing units,
oversubscribed schools select by lottery
neighborhoods within districts differentiated
and that all school funds be spent on edu- by housing quality, district-wide homoge-
cation, coupled with a prohibition against neous public schools, perceived education
topping up, may be the nearest operational quality that varies with expenditure per stu-
counterpart. Chakrabarti (2013b) provides a dent and average peer quality, and peer qual-
model of such a voucher and tests the sort- ity that is correlated with household income.
ing predictions, as discussed in section 5.3.35 Tuition varies across private schools, but, in
contrast to Epple and Romano (1998, 2008),
34 In related work, Eden (1994) examines efficient
price discrimination is not permitted, imply-
voucher policy in a model with peer effects within schools ing each private school is specialized to serve
and an achievement externality to society. Education is one student type. This and the willingness
a pure investment good and capital markets are (implic- of higher-income households to pay a pre-
itly) perfect. He shows that an achievement subsidy
aligns school and social incentives, and combined with a mium for quality results in stratification by
type-dependent voucher equal to the efficient expenditure peer ability and income in the private sector.
plus the student’s peer externality cost induces an efficient Households simultaneously choose where to
(zero-profit) equilibrium in which students pay nothing out
of pocket. live (district and neighborhood), whether to
35 Chakrabarti (2013b) assumes students differ continu- send their child to public or private school,
ously in income and ability, with demand for school quality and vote for a d istrict-wide property tax
increasing in both. School quality is equated to expendi-
ture per student, with a maximum quality. She considers a used to finance public schools. Nechyba
voucher that, for simplicity, covers the highest-quality cost conducts policy analysis in a computational
of education, effectively implying no topping up. Neither
can private schools kick back any of the voucher. Private
schools have no incentive to base admissions on ability types value quality by more and there are utility costs of
due to an absence of peer effects, as with a voucher that applying. The latter is because tuition is fully covered, util-
requires equal probability of admission. Private-school ity costs of applying are independent of income, and the
slots are, however, limited. While the voucher covers all potential monetary cost may not be enough to deter appli-
tuition costs, students face utility costs of applying for a cation. In contrast, in the enrollment stage, there will likely
voucher that they may not get, and may face a monetary not be ability sorting but there will be sorting by income.
cost of attendance if, for example, transportation costs The former is because ability sorting has already occurred,
are not covered. Chakrabarti shows that there is sorting and the latter is because monetary-attendance costs that
by ability at the application stage, but there may not be arise for some will deter take up by some lower-income
sorting by income. The former is because higher-ability students.
Epple, Romano, and Urquiola: School Vouchers 461
c ounterpart calibrated to data for New Jersey. Most households taking such vouchers would
His model predicts that private schools will reside in or move to low-income districts,
emerge in poor communities as high-income whether or not targeted to these districts.
households take vouchers and relocate to More generally, for realistic values, vouchers
occupy higher-quality housing in such com- targeted to the poor district will have small
munities. Hence, stratification of income effects. Similarly, income-targeted vouchers
and property values across communities is will have modest effects unless school qual-
reduced. Poorer households do not experi- ity depends largely on child ability. In that
ence improved peer quality in their (pub- case, low-income parents of h igh-ability chil-
lic) schools, however, because incumbent dren will choose private schools in districts
somewhat higher-income households opt for with poor-quality public schools. Milwaukee
private schools or relocate to communities would appear to be fertile ground for empir-
with better public schools. Expenditure per ical investigation of Nechyba’s predictions of
student rises in public schools as long as the the effects of vouchers on household location,
voucher is not high enough to induce more but such testing has not been undertaken.
than half the population to attend private Ferreyra (2007) builds on Nechyba’s
schools. Hence, public-school quality could model introducing both preferences for reli-
increase if this increased spending offsets gious schools and idiosyncratic (randomly
the decline in public-school peer quality. By drawn) preferences for school types (pub-
allowing mobility and expenditure effects in lic, private, secular, religious) and location.
public schools, Nechyba’s analysis predicts She estimates the parameters of the model
more favorable effects of universal vouch- using data from seven metropolitan areas.
ers on poor students, relative to Epple and She then uses these estimates to simulate
Romano (1998, 2008). On the other hand, the effect of several voucher programs. In
by not allowing price discrimination, bene- particular, Ferreyra examines the differential
fits from vouchers to high-ability students, effects of vouchers depending on whether
whether rich or poor, are curtailed. these can be used at religious schools.37 She
Nechyba (2000, 2003) extends this frame- finds that both types of voucher programs
work by studying vouchers targeted to poor increase private-school enrollment and give
individuals and poor districts, as compared rise to mobility effects of the type identified
to universal vouchers. This is of particular by Nechyba. She also finds that a prohibition
relevance for the US context.36 He concludes on the use of vouchers at religious schools
that a small n on-means-tested voucher tar- results in less private-school enrollment and
geted to residents of low-income districts is can shrink religious enrollments as some stu-
largely equivalent to a universal voucher that dents take a voucher and switch from reli-
is not targeted, due to household mobility. gious to secular private schools. Milwaukee’s
1998 shift to allowing the use of vouchers
36 As reported in tables 1 and 2, targeting vouchers
at religious schools provides a potentially
to the poor usually characterizes US voucher programs. promising environment for testing these pre-
Targeting to poor districts would be similar to the prac- dictions, although, to date such testing has
tice of targeting to poorly performing schools if households not been undertaken.
need only reside where their designated public school
is so labeled to get a voucher, though prior attendance
requirements limit this. In addition to differences in the
underlying models, the ability targeting analyzed in Epple 37 We describe the vouchers in Ferreyra (2007) as
and Romano (2008) has a normative focus, while targeting universal and n ontargeted in table 4, but the variation in
forms studied in Nechyba (2000, 2003) are better moti- whether vouchers can be used at religious schools or not is
vated empirically. central to her paper.
462 Journal of Economic Literature, Vol. LV (June 2017)
Neilson (2013) develops and tests a model schools serving low-income students charge
with geographically differentiated schools lower tuition and provide lower effort than
that compete for vouchers that are higher for private schools serving high-income stu-
poorer households. The empirical applica- dents. The rent-seeking public-school sec-
tion is to Chile, as we discuss in section 5.3. tor will provide one of two quality levels, a
Private schools are differentiated by their high level sufficient to attract both high-
endogenously chosen quality, as well as by and low-income students, or a low level
their location. There are no peer effects. that attracts only low-income students.
Households differ demographically (e.g., If the former, high-income students pre-
in income) and in residence, with idiosyn- fer public schools over paying tuition to
cratic preferences over schools, as well as attend private school; hence the public sec-
(estimated) systematic differences in prefer- tor attracts all students. McMillan consid-
ences. Locational differences among private ers the effect of a universal voucher in the
schools and idiosyncratic preferences imply high p ublic-school effort case. The voucher
market power.38 Profit-maximizing private lowers the cost of private-school education,
schools choose quality below the competitive and may induce high-income households to
(zero-profit) level, with the quality reduction switch to private schools. If this happens,
increasing in their market power. The quality public schools choose to lower effort to the
markdown is greater in poorer areas, where level required to retain only low income stu-
households are estimated to be more p rice dents. A voucher could, alternatively, induce
sensitive. As such, vouchers that are higher public schools to increase effort to retain
for poorer households have a greater posi- high-income students. Hence, McMillan
tive effect on quality. This paper speaks to provides a mechanism such that, instead
question 1, with gains to voucher students of improving public-school effectiveness,
coming largely from reduction in market voucher-induced private-school competi-
power among private providers; but it is tion, and associated income stratification,
also relevant to question 3 on public-sector may have an adverse effect on public-school
responses. effectiveness. McMillan’s framework thus
In contrast to the research discussed so captures an endogenous peer effect associ-
far, McMillan (2004) is squarely focused ated with variation in how parents of differ-
on question 3. This paper endogenizes how ent income levels are able to induce school
public schools adjust their effectiveness in effectiveness.
response to competition from more effec- Building on McMillan (2005), Ferreyra
tive private schools. In McMillan’s frame- and Liang (2012) model imperfect paren-
work, households are of two income types, tal and policy maker monitoring of schools’
with high-income households willing to effort choices. Households vary in ability
pay more than low-income households for and income, and higher-ability households
school quality. Schools exert effort, which are more efficient at monitoring their chil-
raises their effectiveness but comes at a cost dren’s learning. Competitive private schools
to them. Competition constrains private are sufficiently small that no free-rider prob-
schools to provide efficient effort. Private lem arises in parental monitoring, while free
riding prevails in the public sector. They
demonstrate that combining vouchers with
38 Nielson’s schools are then differentiated “vertically”
increased public monitoring of the public
by quality and “horizontally” by location and idiosyncratic
appeal. Epple et al. (2013) also provide a model of vertical sector has the potential to increase every-
and horizontal school differentiation, applied to colleges. one’s achievement and aggregate welfare.
Epple, Romano, and Urquiola: School Vouchers 463
Chakrabarti (2013c) also develops a model her innate ability, her effort, and her school’s
where vouchers could increase or decrease value added. In the second period, the indi-
effort of rent-seeking public schools. In vidual is employed in a competitive labor
her model, students differ continuously in market. MacLeod and Urquiola assume
income and ability, with demand for qual- that innate ability and effort are not directly
ity increasing in both and school quality observed; employers infer ability from all the
depending on school effort and mean ability. available information. A key assumption is
Private schools cannot price discriminate as that individual innate ability is more accu-
in Nechyba’s (1999) model. Universal vouch- rately assessed by schools than by employers.
ers induce higher-ability students to exit the For example, schools might be better able
public sector, implying students at the margin to administer admissions exams or conduct
of attending public school are of lower ability parental interviews. As a result, employers
in the voucher regime. Increasing effort and rationally use an individual’s school of ori-
public-school quality has a smaller marginal gin as a signal of her skill. In turn, students
effect on increasing their attendance. If this seek to attend schools with good reputations,
is the dominant effect, then public schools where a school’s reputation is the expected
worsen as in McMillan (2004). skill of its graduates.
Motivated by voucher programs like the Two key sets of empirical implications
former Florida program (FOSP), targeted emerge. First, “laissez-faire” school sys-
to failing schools, Chakrabarti then contrasts tems have a tendency towards stratification
such a voucher with a policy that awards by ability. Students in n onselective schools
vouchers only if the public school fails to (e.g., the public sector) will be hurt by this
meet a quality standard. She shows that stratification; their low ability is revealed to
with appropriate setting of the quality stan- employers by their failure to gain admission
dard, such a program will induce increased to a selective school. Second, the effects of
public-school effort and quality improve- school competition induced by vouchers will
ment. This is because public schools have depend on design. For example, schemes
a stronger incentive to improve to meet the that restrict schools’ ability to select students
standard and avert the voucher and loss of will maximize effort on the part of students
students, while students at the margin of and their willingness to choose schools with
attending private schools would always exit the highest value added. Schools, in turn,
with a universal voucher. She goes on to test will be forced to build their reputations on
the model as discussed in section 5.3. their advantage in value added, as opposed
In exploring why sorting might adversely to just their ability to select high-ability stu-
affect students left in the public sector, dents. In contrast, systems that facilitate
the above models focus on peer effects.39 selection will tend to lower students’ study
MacLeod and Urquiola (2009, 2012) depart effort and raise the probability that they
from this by studying informational mecha- choose schools based on peer quality rather
nisms instead. Specifically, they model the than value added.
combination of educational and labor mar- To summarize, MacLeod and Urquiola
kets. In a first period, each individual attends (2009) show that even in the absence of peer
school and accumulates skill as a function of effects, the reputational mechanisms empha-
sized by Friedman (1962) do not ensure that
39 See Sacerdote (2011) and Epple and Romano (2011)
vouchers will increase the production of
for recent surveys of the literature on educational peer skill. The intuition behind this result is two-
effects. fold. First, the fact that school membership
464 Journal of Economic Literature, Vol. LV (June 2017)
allows students to convey their innate ability voucher). Invoking the public-sector bud-
reduces the incentive they face to work hard get constraint eliminates one of these vari-
and do well in school. Second, under some ables, leaving a two-dimensional choice
conditions, rational parents will not always set and the accompanying challenges for
prefer the highest value-added schools, analyzing majority rule set forth by Plott
and rational schools will not always choose (1967). Second, even if one variable, say the
to compete on value added. These implica- voucher, is exogenous, preferences over the
tions are consistent, for example, with the tax rate are not single peaked. Epple and
well-identified empirical evidence that selec- Romano (1996) investigate the second of
tive schools only sometimes produce higher these issues, considering voting over the tax
learning (e.g., Clark 2010; Abdulkadiroglu, rate that funds public educational expen-
Angrist, and Pathak 2014; Pop-Eleches and diture and the voucher, taking the voucher
Urquiola 2013). amount as given.40 They show that, despite
the non-single-peaked preferences, equilib-
4.2 Vouchers and Political Economy
rium under majority rule is likely to exist for
The research reviewed thus far studies realistic parameter values. The equilibrium
implications of vouchers but does not analyze is of an ends-against-the-middle form, with
the endogenous public choice of voucher a coalition of poor and wealthy households,
policy, a subject of obvious importance given comprising half the population, favoring a
the poor performance of voucher proposals reduction in the tax rate and m iddle-income
in referenda in California and Michigan. households, comprising the other half, pre-
Ireland (1990) provided the foundation for ferring an increase.41
research on this issue. In his framework, Work to endogenize voucher choice has
households obtain utility from the education followed two avenues. One is to consider vot-
of their children and from the consump- ing one issue at a time. The other is to limit
tion of other goods. Households’ demand the choice set in other ways, such as requiring
for educational expenditure is increasing in that the voucher equal public expenditure
income. Expenditures on public schools and or by considering voucher-only economies.
on a voucher, if any, are funded by a propor- Hoyt and Lee (1998) endogenize both the
tional income tax. Ireland investigates how voucher and tax rate by considering sequen-
public-school spending is impacted by the tial voting with the voucher determined first
introduction of a universal voucher smaller and then the tax rate second. Employing
than per-student public-school expendi- information on the income distribution in
ture. The effect may be either an increase each state, they find that there are some states
or decrease, depending on whether the in which introduction of a $1,000 voucher
reduction in outlay for students who switch would permit lowering the tax rate without
from public to private school in response to lowering public expenditure per student.
the voucher is sufficient to offset the cost of
providing a voucher to students who would
attend private school anyway. 40 See also Glomm and Ravikumar (1998).
41 In a very similar model, assuming existence of majority-
Ireland treats the voucher and tax rate
choice equilibrium, Rangazas (1995) identified the trade-
as exogenous; subsequent work has sought offs in the public choice of expenditure in the public school
to model these as chosen by majority rule. for a given voucher. His computational analysis predicts
This effort encounters two challenges. First, that a voucher equal to 1.25 percent of a teacher’s annual
salary would cause per-student public expenditure to
the policy vector has three variables (tax increase. Investigation of “ends-against” voting is under-
rate, public-school expenditure per pupil, taken by Brunner and Ross (2010).
Epple, Romano, and Urquiola: School Vouchers 465
Chen and West (2000) take the voucher Romano (2014). They analyze simultaneous
as equal to public-school expenditure per voting over the tax rate, public-school expen-
student in their examination of the political diture per student, and the voucher, exploit-
economy of income targeting. If vouchers ing the citizen-candidate model of Besley
produce some efficiency gains, they find that and Coate (1997). They provide necessary
the targeted regime is likely to be majority and sufficient conditions for equilibrium,
preferred both to the no-voucher status quo and show computationally that equilibrium
and to a n ontargeted voucher regime. exists for realistic parameter values. They
Another approach, voucher-only econ- also show that a voucher is likely to gar-
omies, is employed by Fernandez and ner greater political support when income
Rogerson (2003) to study vouchers in a inequality is low. Intuitively, when inequality
dynamic setting in which education spend- is low, a relatively small number of house-
ing impacts adult earnings. They consider holds choose private school. A marginal
three alternative voucher programs: a uni- increase in the voucher induces a relatively
versal flat-rate voucher, a means-tested large exodus from the public schools, permit-
voucher, and a “means-equalizing” voucher ting an increase in public-school expenditure
that depends on household income and the per student with a lower tax rate, a change
amount of income devoted to education. that garners unanimous support.
They find that all three alternatives increase Epple, Romano, and Sarpca (2014)
utilitarian welfare substantially, relative to extend this model to include income target-
the purely private system, and all tend to cor- ing via simultaneous voting over the tax rate,
rect the inefficiency from low investment on expenditure per student in public schools,
the part of poor households.42 the voucher amount, and the maximum
Bearse et al. (2013) continue the study income of households eligible for vouchers.
of income-targeted programs by consider- They find that income targeting increases
ing a voucher that is positive for the lowest- political support for vouchers by limiting
income household and declines linearly with their use by h igh-income households that
income to zero. A sequential voting equilib- would use private school even in absence
rium, with the tax rate chosen first, followed of vouchers, and that a targeted voucher
by the parameters of the voucher program, always garners political support. The pref-
is shown to exist. Their computational model erence for targeted vouchers conforms to
shows that, compared to the n o-voucher pub- observation, but the finding that a targeted
lic equilibrium, the means-tested voucher voucher would always garner political sup-
chosen by majority rule benefits the poor via port does not. This latter finding brings
higher education spending and a lower tax to the fore limitations of the workhorse
rate, while also benefitting wealthy house- Ireland (1990) framework, particularly the
holds who prefer private schooling coupled assumptions that all households have the
with a low tax rate. same preference function and differ only
An alternative approach to overcome the in income. Evidence on voting by legisla-
Plott existence issue is adopted by Epple and tors for voucher proposals discussed in sec-
tion 5, along with differences in opinions
42 Fernandez and Rogerson (2003) note that their three evoked by vouchers, point to consideration
voucher systems can be viewed as analogous to three dif- of ideological differences in preferences as
ferent systems of state grants to local districts (foundation, an avenue for extending the Ireland frame-
means tested, and power equalizing). Hence, their analysis
can alternatively be viewed as informing the political econ- work. This is being pursued in ongoing
omy of public-school finance. research of Epple, Romano, and Sarpca.
466 Journal of Economic Literature, Vol. LV (June 2017)
To summarize the research on the polit- ublicly and privately funded voucher pro-
p
ical economy of vouchers, two prominent grams have been explicitly designed as exper-
themes emerge. One is that the majority iments. In other instances, nonexperimental
of voters, those who do not choose private programs are oversubscribed, and random
school in the absence of vouchers, will ben- assignment arises from the use of lotteries to
efit by targeting vouchers so as to prevent allocate scarce slots. These cases, at least in
take-up by those who would attend private principle, establish a clear counterfactual—
school anyway. The other is that a voucher on average the observed and unobserved
offering less than p
er-student expenditure in characteristics of treated and untreated
public school will generally be preferred by groups should be identical, and therefore
those who would continue to attend public simple comparisons of their achievement
school. Such a partial voucher induces some can reveal the causal effect of vouchers.
households to switch to private school, and One aspect to bear in mind is that in all
this yields a net tax savings to those attend- the programs we discuss, those who are
ing public school equal to the differential offered a voucher are not required to use
between per-student public spending and it. Hence, a distinction arises between the
the voucher. effects calculated by focusing the compar-
ison on those who have been offered the
voucher and those who actually take it up.
5. The Empirical Evidence
A comparison of the average outcomes of
This section reviews the empirical evi- those offered and not offered the voucher
dence on each of the five questions raised yields an “Intent to Treat” (ITT) estimate.
above. For each, we highlight the method- A “Treatment on the Treated” (TOT) esti-
ological challenges that arise and we focus mate adjusts for the proportion of students
the review on the papers that have most suc- who take up the voucher—thus providing
cessfully dealt with these challenges. This an approximation to the effect of the treat-
implies that we discuss some voucher pro- ment on those who received it. Both types
grams more than others, depending on the of estimates have analytical advantages. For
question at hand. instance, the ITT estimate might provide a
reasonable approximation to the effect of
5.1 Question 1: What effects do vouchers implementing a small-scale voucher scheme,
have on the students who use them?—The since it is always the case that the proportion
key challenge in answering this question is of students who take up the voucher is less
establishing credible counterfactuals; e.g., than one.
what would the outcomes of voucher win- Finally, in some cases we also review
ners have been had they not received a the results of papers that, facing a lack of a
voucher? While different types of research (quasi-) experimental counterfactual, seek
attempt to do this, the papers on small-scale to establish one by using matching tech-
voucher programs are the most focused on it niques or otherwise attempting to control
and have tackled it with the highest degree for observable characteristics. In such cases,
of credibility. one must bear in mind that estimates can still
This reflects that, in many cases, their be biased if unobserved student or parental
setup at least emulates a randomized con- characteristics are correlated with treatment.
trolled trial in which subjects are randomly To preview the bottom line on question 1,
assigned to treatment (voucher) or control the evidence does not suggest that awarding
(
no-voucher) groups. Specifically, some students a voucher is a systematically reliable
Epple, Romano, and Urquiola: School Vouchers 467
way to improve their educational outcomes. emerges for reading after three years, but
A perhaps surprisingly large proportion of none for math). Overall, there is little evi-
the best-identified studies suggest that win- dence that the Washington DC Opportunity
ning a voucher has an effect on achieve- Scholarship Program resulted in a sustained
ment that is statistically indistinguishable improvement on test scores.
from zero. Moreover, three recent studies In contrast, Wolf et al. (2010a) report
find large negative effects on test scores of that the program had a large and statisti-
voucher recipients. This is contrary to what cally significant impact on graduation rates.
one would expect, for example, if private After four years, students who were offered
or independent schools had systematically a voucher (ITT) were 12 percentage points
higher value added. There is, however, recent more likely to graduate than those who were
evidence from two randomized-control stud- not, with a corresponding TOT effect of
ies that point to more favorable effects on 21 percentage points. Exploring heterogene-
attainment. There is also evidence that in ity in impacts, Wolf et al. find similar effects
some settings, or for some subgroups or spe- among students originally in schools desig-
cific outcomes, vouchers can have substan- nated as “in need of improvement.”
tial positive effects on those who use them. The School Choice Scholarship Foundation
A question is, therefore, what accounts for created three voucher programs—New
the variation in estimated impacts? The liter- York City, Dayton, and Washington, DC—
ature offers some tentative and useful clues, that also conform to experimental design.
but no definitive guidance. This reflects two The most intensively studied of these is
aspects we will be explicit about. First, the the one in New York, where in 1997 a lot-
best evidence on question 1 comes from very tery was conducted among approximately
different settings—in this section we review 11,000 applicants (Peterson et al. 2003).
studies on the United States, Colombia, and None of these experiments yield significant
India—and while all these provide useful test score effects for non–African American
evidence, extrapolating is difficult, as these students. Nonetheless, Mayer et al. (2002)
settings vary along multiple dimensions. find the program increased the test scores
Second, the experimental studies can provide of African American students in New York
clear counterfactuals and credibly answer by about 6 percentile points (ITT). Krueger
question 1, but they deliver a “reduced- and Zhu (2004) show, however, that this find-
form” answer that does not fully reveal what ing is sensitive to how ethnicity is coded, as
mechanisms account for the effects—a fur- well as to how one handles students with
ther reason for why extrapolation is difficult. missing baseline scores.43 In Washington,
Peterson et al. (2003) find significant effects
5.1.1 The United States
for African American students at the end of
Wolf et al. (2010a, 2010b) report on the two years of treatment, but these vanish by
Washington DC Opportunity Scholarship the third. In Dayton, they find that African
Program, which used an experimental design. Americans had a 4 percentile point advan-
Their sample consists of roughly 2,300 stu- tage at the end of two years (significant at
dents, of whom about 60 percent were the 10 percent level). Overall, this group of
offered a voucher, with the rest serving as a experiments suggests some—albeit not very
control group. Of those offered a voucher, robust—indication of test score effects for
77 percent made use of it. The authors find
no significant impact on test scores after 43 On this issue see also Barnard et al. (2003), Krueger
one, two, or four years (a significant effect and Zhu (2003), and Peterson and Howell (2004a, 2004b).
468 Journal of Economic Literature, Vol. LV (June 2017)
African American students, and none for the sample of public-school students as the com-
rest. parison group, obtaining similar findings.
As in the DC Opportunity Scholarship Witte et al. (2008, 2009, 2010, 2011, 2012)
Program, however, these experiments pro- conduct a fi ve-year study of the effects on
duced better results in terms of graduation voucher recipients (TOT) using a match-
rates, at least for African American students. ing strategy. They find no significant effects
Specifically, Chingos and Peterson (2015) on test scores in the first, second, and third
study college enrollment as another outcome year of the program. A statistically significant
in New York. At the time of the experiment fourth-year gain of 0.15 standard deviations
(late 1990s), participants were in grades K–5. emerges in reading; gains in mathematics
Chingos and Peterson were able to obtain are also evident, but are significant only for
follow-up information on college enrollment students in grade seven. Overall, this analy-
for a remarkable 99 percent of the roughly sis suggests that during this post-expansion
2,700 students in the original study. They phase, the Milwaukee voucher program had
find no significant differences between the little, if any, effect on test scores. In con-
treatment and control groups as a whole, but trast Cowen et al. (2011) and Cowen et al.
a significant difference for African American (2013) use a matching strategy and find pos-
students—for those offered a voucher (ITT), itive impacts on years of schooling, although
the estimated increase in part- and full-time the results are not entirely robust to differ-
enrollment is 7 percentage points, a 20 per- ent specifications. Beyond potential biases
cent increase. from unobserved characteristics, a concern
The above papers focus on voucher pro- with these longer-term m atching-based esti-
grams that were designed with an explicit mates emerges from the possible existence
experimental setup in mind. This was not the of contemporaneous non-voucher-related
case in Milwaukee, which displays variation policies. For instance, starting in 2002 the
in its ability to deliver clear counterfactuals public reporting of schools’ test scores began
over time. Specifically, in its early years the to be required, and this might have affected
Milwaukee program featured randomization schools’ performance quite aside from
due to oversubscription, but starting in 1994, vouchers.
increases in the income cap and the incor- While the studies discussed above find
poration of Catholic schools had the effect little systematic evidence of positive effects
of making vouchers generally available to of vouchers on achievement, they do not
most eligible students without recourse to find significant negative effects. By contrast,
lotteries. Abdulkadiroglu, Pathak, and Walters (2015)
Rouse (1998) analyzes impacts during (APW henceforth) find large negative test
both periods.44 First she exploits random- score effects for the Louisiana Scholarship
ization, finding little evidence of effects for Program (LSP). APW study test scores in
reading. Her estimates (both ITT and TOT) 2012–13, the first year following statewide
for mathematics are substantial—statistically adoption of the Louisiana voucher program,
significant effects of 0.3 to 0.5 standard devi- a program that previously had been available
ations over a four-year period. She also fol- only in New Orleans.
lows Witte (1997) in considering a random APW exploit lottery selection of students
to oversubscribed private schools for LSP
voucher applicants in third through eighth
44 See also Witte, Sterr, and Thorn (1995) and Greene, grades, studying effects on scores in stan-
Peterson, and Du (1996). dardized tests of math, science, social studies,
Epple, Romano, and Urquiola: School Vouchers 469
and English. APW find that 1,412 students performing, but must also be accepted by
who applied to oversubscribed schools were a participating private school. Mathematics
selected by lottery, with subsequent test scores decline each year by around −0.5σ
scores available for upwards of 85 percent and reading by around −0.3σ, estimates that
of these students. Causal local average treat- are very robust to alternative specifications.
ment effects on math, science, and social Because participating private schools can
studies are −0.413σ, −0.263σ, and −0.331σ select students, propensity score matching
respectively, all significant at the 1 percent is employed for identification. Specifically,
level. A negative but insignificant effect is voucher takers in barely eligible schools are
found for English. matched on observables to ineligible stu-
APW conduct extensive analyses to inves- dents in public schools that barely exceeded
tigate robustness and possible differences in low performance. As the authors discuss,
impacts of the voucher across student income this sharp identification limits the treated
groups and geographic areas. Their findings students to 445, implying that effects on
of negative impacts of vouchers are robust, voucher takers in the worst public schools
e.g., with estimation methods accounting for might differ.
sample attrition, and are found for all income Why students in these studies elect to
groups and geographic areas. Participating attend private school where they perform
private schools are found to have falling relatively poorly on tests is an important and
enrollments prior to their participation in open research question. Students may have
LSP, suggesting that the voucher program is alternative goals, e.g., religious study, or may
attracting schools that are struggling. It will be making mistakes. It is notable that expen-
be important to investigate whether these diture per student in the attended private
negative effects persist for subsequent years schools is generally lower than in their public
of the LSP, but as APW conclude, “These alternatives.
results suggest caution in the design of Overall, the evidence on the United States
voucher systems aimed at expanding school finds not very robust effects on test scores,
choice for disadvantaged students.” most frequently nonexistent, some positive
A first indication regarding persistence effects on blacks, but also the just-discussed
of negative effects of the LSP emerges large negative effects. More robust evidence
from work by Mills and Wolf (2016). These has accumulated regarding positive impacts
authors use a slightly larger sample of LSP on graduation probabilities, particularly for
students, and are able to expand the analysis black students.
to a second year. Mills and Wolf state that
5.1.2 Colombia
their first-year results largely match those
in APW. For the second year, they still find Colombia yields perhaps the most positive
negative effects, although somewhat smaller evidence that emerges from (quasi-)exper-
in magnitude. imental work on question 1. Specifically,
Figlio and Karbownik (2016) also find Angrist et al. (2002) exploit voucher lot-
large negative and significant effects on teries implemented in the cities of Bogotá
both math and reading scores of voucher and Cali. In terms of test scores, their key
recipients in the statewide Ohio “EdChoice finding is that three years after the alloca-
Scholarship Program,” notably persisting tion, voucher winners scored 0.2 standard
for students through the three years of the deviations higher on achievement tests.
study. Voucher-eligible students have their Using a similar design, Angrist, Bettinger,
designated public school deemed to be low and Kremer (2006) find that, correcting for
470 Journal of Economic Literature, Vol. LV (June 2017)
differences in test taking between lottery schools? This is a possibility, since winners
winners and losers, the program increased on average used the voucher to “upgrade”
high-stakes college admissions test scores
to more expensive institutions—Angrist
also by 0.2 standard deviations. They fur- et al. (2002) report that vouchers “crowded
ther find an effect on longer-term outcomes: in” educational expenditure to some extent.
voucher winners were 15–20 percent more There is less data on the resource changes
likely to complete secondary school, less that vouchers imply in the United States.
likely to work while in school, and less likely A final possible channel relates to student
to marry or cohabit as teenagers. incentives. In Colombia the vouchers were
In short, the experimental results from renewable contingent on grade completion,
Colombia are more positive than those from and thus the program included an incentive
the United States in every measured dimen- component—voucher winners faced a stron-
sion. A relevant question is: why are voucher ger reward for doing well at school. Thus,
winners in Bogotá and Cali benefiting more superior performance could have been due
consistently than those in New York City or to incentives, rather than to the voucher pro-
Washington DC? None of the experimental vision itself.45
studies reveal the precise channels through To summarize, the Colombian experiment
which its effects work—the clear identifica- suggests that vouchers had a positive effect
tion comes at this cost, to some extent—so it on tests scores and several other outcomes,
is not possible to answer this question defin- but the difficulty in attributing effects to a
itively. Nevertheless, the institutional differ- precise channel makes it difficult to draw
ences between the US and Colombia voucher precise policy implications.
experiments render some differences more
5.1.3 India
or less likely as potential candidates.
Is it possible that public schools in Turning to India, Muralidharan and
Colombia are much weaker than in the Sundararaman (2015) analyze an experiment
United States, and so the opportunity to use in villages of the state of Andhra Pradesh, as
a private school has a large effect? This may described in section 3.1.3. They find that four
play a role but is unlikely to provide a full years after treatment, lottery winners did not
explanation, as—somewhat unusually—both have higher test scores than losers in Telugu
voucher winners and losers in Colombia (the local language), Math, English, Science,
tended to enroll in private schools. For and Social Studies; in contrast, there was a
example, Angrist et al. (2002) point out that positive and significant effect in Hindi. The
while about 94 percent of lottery winners authors interpret this as an overall positive
attended private school in the first year, so effect, since private schools achieve higher
did 88 percent of the losers. This partially Hindi results with no disadvantage in the
reflects (section 3.1.2) that a requirement for other tests.
application was to have been accepted at a
private school (the stated goals of the pro-
gram were related to raising enrollment rates 45 Digging even deeper into mechanisms, the gains in
by increasing private participation). Thus, Colombia could have also reflected peer effects. However,
Bettinger, Kremer, and Saavedra (2010) suggest that
one might reasonably expect the program to at least for a subgroup of PACES beneficiaries, at least
be more attractive to students strongly inter- observable peer quality does not seem to account for the
ested in private school anyway. results. Specifically, they show that the advantage found in
Angrist et al. (2002) persists, even when one considers stu-
Is it possible that the positive findings dents who chose to attend vocational schools with typically
reflect greater resources at the receiving lower observable peer quality.
Epple, Romano, and Urquiola: School Vouchers 471
Interpreting the effect in Hindi is com- Hanna, and Ryan (2012) and Muralidharan
plicated by the fact that public schools in and Sundararaman (2011).
Andhra Pradesh essentially do not teach Muralidharan and Sundararaman (2015)
Hindi at all. More specifically, Muralidharan emphasize that, coupled with the above
and Sundararaman use interesting comple- findings on achievement, the fact that the
mentary data to show that the private schools private schools spend so much less implies
spend much more time teaching Hindi a substantial private productivity advantage.
(essentially relative to a base of zero in public They also show that the crux of this cost dif-
schools) and substantially less on the remain- ference is in teacher salaries—private-school
ing subjects, except for English.46 teachers are younger, typically less trained,
The setting this paper explores also more often female, and on average make
raises interesting contrasts with the US and only one-sixth the salary of their public-
Colombia cases described above. Importantly, sector counterparts. In short, private-school
in contrast to Colombia, where voucher win- students have instructors who are paid much
ners benefited from greater expenditure, less. On the other hand, in some dimensions
Muralidharan and Sundararaman (2015) private-school students have access to more
point out that private schools in the villages educational resources. For instance, they
analyzed have expenditures that are only enjoy lower class sizes, and the probability
about one-third of those in public schools. that they are in multi-grade classrooms is
This reflects that many private schools in lower by about 50 percentage points.47
India operate without subsidies, even as they In short, private and public schools in
cater to very low-income individuals. This is Andhra Pradesh are organized very differ-
observed in other settings, including Pakistan ently, but on net have no testing performance
and parts of Africa (e.g., Andrabi, Das, and differences except in Hindi, which the pub-
Khwaja 2013); in these areas, private expan- lic schools do not teach. It is again the case
sion has been observed on a magnitude that that the myriad of differences in setting pro-
in middle- or high-income countries would vide interesting implications, but complicate
seem to require significant public subsidies. extrapolation. Just to cite one example, the
This might reflect parental willingness to voucher schools in Andhra Pradesh, India,
escape a deeply dysfunctional public sector on average have lower class sizes than public
in which there is evidence of rampant absen- schools; the opposite is the case in present-
teeism by teachers; see, for example, Duflo, day Chile.
To summarize, question 1 is perhaps the
most straightforward among those we ask,
46 The authors further note that not all private schools
and yet the above review makes clear that
use Telugu or Hindu as the language of instruction, with the answer to it is complex. The results are
some using English instead. While acknowledging that this not clear-cut—in many cases and for many
is an endogenous choice (and hence not a feature of the outcomes transferring students to private
experimental design), they present suggestive evidence
that the choice of English as a medium of instruction dis- school does little to their achievement; in
rupts learning. If parents do not realize this is the case, others, it improves or lowers it substantially.
then further regulation of private schools may be war- In addition, the most rigorous research on
ranted. But another possibility is that parents are aware
of this but willing to make the sacrifice if, for example, this question t ypically delivers r educed-form
English has high labor-market returns. A broader point the
findings around Hindi and English illustrate is that choice
is likely to produce more of what parents want that may 47 The research in Duflo, Dupas, and Kremer (2011)
or may not be skills along the precise dimensions policy suggests this might have a large favorable impact on
makers prefer. performance.
472 Journal of Economic Literature, Vol. LV (June 2017)
results. These do not reveal the mechanisms and of course randomizing at the country
that account for the differences, making it level is not feasible.
hard to draw clear implications. It is relevant
5.2.1 Large-Scale Voucher Programs:
to point out that the evidence—particularly
Chile and Sweden
that from the United States—is consistent
with vouchers improving some types of skills The original design of Chile’s voucher
more than others. For example, it is possible scheme allowed private schools ease in set-
that the positive effects on graduation rates ting up admissions policies. These could
in the United States stem from improve- include features such as admissions exams
ments in n oncognitive skills, while the lack and parental and student interviews.49 In
of a consistent impact on test scores points addition, since the m id-1990s, schools have
to weaker impacts in a cognitive dimension.48 been able to charge tuition add-ons. Under
such circumstances, models like Epple and
5.2 Question 2: Do vouchers induce Romano (1998) and MacLeod and Urquiola
onrandom migration from public to private
n (2009, 2015) suggest that the introduction
schools, possibly lowering the achievement of of vouchers would lead to cream skimming
students that remain in the public sector?— from the public sector, and stratification by
Another fundamental question on vouchers income and/or ability within the private
is whether these lead to sorting. As discussed sector.
above, this is the common prediction of Hsieh and Urquiola (2003, 2006) analyze
theoretical work. However, that work also this by looking at the growth of the private
makes clear that the type of sorting observed sector across municipalities. They essen-
will be a function of the rules governing the tially implement a difference-in-differences
voucher system. analysis that asks if stratification measures
We begin by considering the evidence increased more in markets with greater
on large-scale voucher programs. These are growth in the private-voucher sector. Again,
analytically suited to addressing question 2 this is not equivalent to a randomized exper-
in that they provide a chance to study situa- iment and biases could arise, for instance,
tions in which large numbers of students of from preexisting trends. Nevertheless,
all types are given a chance to exercise school Hsieh and Urquiola find evidence that the
choice, and in which schools get a chance to voucher-induced growth in the private sec-
enter and exit the market in response. tor was associated with a “middle class”
Yet the evidence from large-scale pro- exodus from public schools consistent with
grams has disadvantages too. Most impor- cream skimming; this is robust to the use of
tantly, almost by definition, very clear candidate instrumental variables for private
identification is difficult to obtain from these growth.
programs. They involve the distribution of McEwan, Urquiola, and Vegas (2008)
vouchers to anyone who wishes to use them, additionally present a descriptive analysis
and are typically implemented c ountrywide. of sorting in small markets. The idea—one
Thus, it is very difficult to establish clear related to an approach used in industrial
counterfactuals regarding what would have
happened in the absence of these programs,
49 Over the years, more restrictions on selection by
private schools have been written into law. In our under-
standing, these include generalized statements regarding
prohibitions on selection rather than specific, legally bind-
48 We thank an anonymous referee for this observation. ing mechanisms (e.g., centralized lotteries).
Epple, Romano, and Urquiola: School Vouchers 473
o rganization—is that fixed costs determine independent schools must partially apply a
that private voucher schools must be of at first-come-first-served criterion if oversub-
least a minimum size in order to break even. scribed, and tuition add-ons are not allowed.
This implies that very small markets—say Nevertheless, research points to some sort-
very small towns—will tend be served by ing effects. For instance, Sandstrom and
public schools only. Larger towns will have Bergstrom (2005) report that indepen-
private participation. McEwan, Urquiola, dent-school students are more likely to be
and Vegas (2008) focus on the range of immigrants and to have parents with higher
towns (by population size) where private income and education (see also Bjorklund
entry is first observed, comparing the degree et al. 2006). Further, Bohlmark and Lindahl
of sorting observed just below the approxi- (2007) find that public schools tend to lose
mate size threshold that determines private students who are second-generation immi-
entry to that observed just above. This is akin grants or whose parents are more highly
to a “fuzzy” regression discontinuity design, educated, but find no evidence of sorting by
albeit one with limited statistical power. income.
They find that private entry is indeed related In more recent work that considers effects
to stratification, consistent with the first pri- through 2009, Bohlmark, Holmlund, and
vate-school cream skimming the highest Lindahl (2015) find that after accounting
ability/income kids from the public sector. for residential sorting, school segregation
Finally, a large number of studies point to increased more in municipalities in which
cross-sectional evidence of high stratification school choice became more prevalent (as
in Chile. For instance, Valenzuela, Bellei, measured by the number of independent
and de los Rios (2010) suggest that Chile dis- schools in operation). Specifically, segrega-
plays one of the highest levels of s chool-level tion in such areas has grown between immi-
stratification by socioeconomic status in the grants and natives, and between children of
OECD. In addition, Mizala, Romaguera, parents with high/low levels of education.
and Urquiola (2007) suggest stratification is Nevertheless, they conclude that the magni-
particularly extensive in the private sector.50 tude of the effect of choice on segregation
In closing the discussion on Chile, it is is relatively low, and that Sweden still ranks
worth mentioning that there is widespread as a country with relatively low levels of
consensus among observers and policy mak- across-school sorting along such dimensions.
ers there around the claim that vouchers
5.2.2 Small-Scale Voucher Programs
have facilitated sorting. In addition and as we
discuss below, the high degree of stratifica- The literature on small-scale programs has
tion in the school sector seems to be a signif- also explored question 2. These programs
icant contributing factor to student protests can be expected to have a different impact
that have persisted over a number of years. on sorting (relative to large-scale programs)
Stratification has been less of a focus on the due to the fact that they tend to display
research in Sweden. Speculating, this might four features: (1) targeting of low-income
reflect that Sweden’s voucher design likely students, (2) lotteries in cases of oversub-
promotes sorting less than Chile’s. As stated, scription, (3) requirements that voucher pro-
ceeds be used for educational expenses (i.e.,
not to provide financial aid), and (4) rules
50 For other examples of large-scale school market lib-
against tuition “ top-ups.” In addition, in
eralization leading to stratification, see Lucas and Mbiti
(2012) for the case of Kenya. For related evidence in the the case of the United States the potential
United States, see Urquiola (2005). for voucher-induced sorting is set against a
474 Journal of Economic Literature, Vol. LV (June 2017)
c onsistent with this, suggesting that voucher ability/income children leaving to pri-
participants tend to come from schools vate schools, and
with worse academic performance and
higher rates of violence. In addition, they (3) A peer effect—for instance, if there is
are more likely to exit public schools when nonrandom sorting the performance
they encounter more convenient (e.g., by of those “left behind” may be affected
distance) and varied private-school options. by no longer interacting with departing
Finally, they are more likely to exit schools students.
with high concentrations of African American
students, and the latter holds regardless of While (1) is the object of interest in this case,
the student’s own race. the presence of (2) and (3) make it very dif-
To summarize, the evidence suggests ficult to isolate this effect. Specifically, in
that, perhaps not surprisingly, vouchers can order to isolate (1) it would be necessary to
result in the nonrandom reallocation of stu- control for children’s characteristics, some of
dents across and within sectors. That said, which may be unobserved. Further, even if
the details of program design clearly matter. one observed all student characteristics, ade-
For one example from among large-scale quately controlling for peer effects is difficult,
programs, Chile’s design generally facilitates given that the literature has not produced a
sorting more than Sweden’s. consensus on the functional form of such
effects, or even on whether a stable functional
5.3 Question 3: Do voucher programs form exists (Carrell, Sacerdote, and West
pressure public schools to become more effi- 2013). A clear prior on the direction of at least
cient?—A key reason to introduce competi- some of the effects in (1)–(3) could provide
tion into any industry is the possibility that analytical leverage to empirically get a sense
it will lead to productivity improvements. of the direction of the others. However, as
Question 3 asks whether vouchers induce noted in section 4, theory does not provide
these in the public sector, perhaps as its unambiguous guidance on any of these.
schools attempt to fend off gains on the part Hsieh and Urquiola (2003, 2006) present
of private competitors. evidence of these difficulties for the case of
To illustrate the methodological chal- Chile. For example, they show that public
lenges that this question raises, Hsieh and performance worsened in municipalities that
Urquiola (2003) point out that a first-pass experienced greater private growth after the
answer to it is provided by calculating the introduction of vouchers. They point out that
difference in average public-school perfor- this could be driven purely by composition
mance before and after the introduction of effects—(2) in the listing above. They there-
vouchers. They point out however, that one fore call for a focus on how the introduction
would ideally want to decompose this change of vouchers affects aggregate performance.
into three effects: This does not solve the challenge posed by
(3), but it does control for (2) and poten-
(1)
The public sector’s change in value tially comes closer to identifying the effect
added—essentially the object of inter- of vouchers on overall school productivity.
est in question 3 We will return to that issue in the context
of discussing question 4 (which focuses on
(2)
A composition effect—for example, aggregate, net effects); for now we review
vouchers may worsen public-sector the evidence on question 3, keeping in mind
performance if they result in higher the above challenges.
476 Journal of Economic Literature, Vol. LV (June 2017)
5.3.1 The United States the poverty line. Their logic is that the com-
petitive impact of voucher availability on
Hoxby (2003a) studies question 3 in the public schools will be greatest when public
context of Milwaukee. She measures the schools have nearby private competitors.
intensity with which public schools face They create four measures of private-school
competition by the proportion of their stu- proximity, and find that public-school
dents that are eligible for vouchers. In achievement is related to each of these mea-
Milwaukee, this measure varies because sures significantly, albeit modestly: a one
eligibility for a voucher is dependent on stu- standard-deviation increase in the number
dents’ income. In this approach, schools with of nearby private schools raises achievement
a low proportion of poor students serve as by 0.02 to 0.03 standard deviations.
the control group. As an additional control, A special feature of the (now canceled)
Hoxby chooses a set of low-income public FOSP facilitates identification of program
schools outside of Milwaukee. She finds that effects, but potentially confounds their inter-
more intensively treated public schools have pretation for purposes of analyses related
higher rates of productivity growth measured to vouchers. Specifically, in this program
as achievement on a standardized exam, rela- Florida schools were graded based on stu-
tive to expenditure. dent performance on a series of standardized
Chakrabarti (2008) extends Hoxby’s exams. If a school received an F grade twice
approach, considering two periods in the within four years, its students became eligi-
Milwaukee voucher system. She designates ble for a voucher. Schools that received one
1990–97 as Phase I; Phase II consists of years F faced a threat of a voucher, and the perfor-
after 1998, roughly when the program expe- mance of students in these schools can then,
rienced changes including the expansion of for example, be compared to that of their
the proportion of eligible students and reli- peers in schools that received a D grade.
gious school eligibility. She uses school-level Figlio and Rouse (2006) examine test-
data, with the proportion of students on free score gains in schools receiving one F grade
or reduced lunch (FRL) as a measure of the using several strategies including regression
intensity of treatment. As controls, she uses discontinuity, with D schools serving as a con-
the thirty-three Wisconsin public schools trol group. They find effects on high-stakes
outside Milwaukee that had at least 25 per- mathematics tests (i.e., tests relevant to grad-
cent of their population eligible for FRL, ing of the schools) of about 0.2σ, larger than
had a black population of at least 15 percent, effects on low-stakes exams, suggesting non-
and were in locales similar to Milwaukee in trivial gains but also some strategic focusing
1990. Chakrabarti finds little evidence that of resources. However, they emphasize and
the voucher program had effects on public present some evidence suggesting that the
schools in Phase I. By contrast, for Phase II, gains may be due to the stigma associated
she finds significant gains on the order of with receiving another F grade, rather than
0.1–0.15 standard deviations; these are sta- to the threat of loss of funding associated
tistically significant in reading and language with a voucher. In short, the program may
arts, but mostly not in mathematics and confound the effects of vouchers with those
science. of accountability.
Figlio and Hart (2014) study the effect Chakrabarti (2013a, 2013d) continues
of the FTC on public-school performance. the study of the FOSP program. Using
This program is available for students from difference-in-differences and regression
families with incomes below 185 percent of discontinuity designs, Chakrabarti (2013a)
Epple, Romano, and Urquiola: School Vouchers 477
provides evidence that schools receiving voucher eligibility threshold, they find large,
one F focused resources on improving the significant three-year achievement impacts
scores of students predicted to be near the of voucher eligibility ranging between 0.05σ
boundary of the threshold of failure, and to 0.1σ, and four-year impacts of similar
on preparing for the writing exam, where magnitude. These estimates are robust to
performance is believed to be more eas- including demographic controls and allow-
ily improved. She provides some evidence ing different slopes on the two sides of the
that the voucher threat was important, for eligibility threshold. Analysis of impacts by
example, finding larger effects of receiv- demographic groups suggests that these
ing an F grade on schools that faced more estimated gains accrued primarily to white,
private-school competition, as in Figlio and non-disadvantaged students. Insignificant
Hart (2014). Chakrabarti (2013a) focuses on effects are found with polynomial functions.
pure gaming effects of the program by inves- Hence, while the results with linear func-
tigating whether schools classify students tions point to positive competitive impacts,
with an eye to preventing their scores from lack of robustness to more flexible functional
counting toward the school grade. This anal- forms argues for caution.
ysis again compares F to D schools, and finds
5.3.2 Canada
significantly increased classification of stu-
dents as having limited English proficiency. Chan and McMillan (2009) study the
While F schools might have tried to do the effect on public-school performance from
same by classifying more students as special a private-school tax credit in Ontario. This
education, their test scores also excluded was passed into law on short notice in June
from the school’s grade, the gains from this 2001, with the credit becoming available
strategic reclassification would come at the in January 2002. The plan provided for a
cost of these students then becoming eligi- private-school tax credit that was scheduled
ble for another Florida voucher program. to grow in increments over five years, but the
These schools did not increase such clas- program was canceled in December 2003,
sification. It bears repeating that studies retroactively to January 2003. Using the
of the FOSP program provide interesting 2002–03 private-school enrollment share in a
avenues for identification of the program public-school attendance zone as their mea-
effects, but the challenge of disentangling sure of private-school competition, Chan and
the accountability and voucher effects weak- McMillan find that a 1 standard-deviation
ens implications that can be drawn for the increase in competition is associated with a
more common accountability-independent statistically significant 0.1 standard-deviation
voucher programs. increase in the percentage of public-school
Figlio and Karbownik (2016, FK hence- students achieving the provincial perfor-
forth) study the competitive impact of the mance standard for grade 3.
Ohio voucher program, EdChoice, using a
5.3.3 Sweden
regression discontinuity analysis. The idea
is to compare performance of students Sandstrom and Bergstrom (2005) consider
assigned to public schools barely eligible whether students in Swedish public schools
for vouchers to ineligible students in pub- perform better if they live in municipali-
lic schools that barely avoided their stu- ties that have a larger share of independent
dents being voucher eligible. The estimated schools. As stated, such an evaluation con-
impacts are sensitive to functional form. fronts difficulties that originate in the non-
With linear functions on either side of the random sorting that follows private entry, as
478 Journal of Economic Literature, Vol. LV (June 2017)
well as the endogeneity of this entry. The first stated above, some studies focus not on
of these might be somewhat mitigated, since questions 1–3, but rather on the net effect
Swedish independent schools are not allowed of vouchers—question 4. The reason to do
to select on ability. Nevertheless, applying to this can be stated by summarizing some of
an independent school is still endogenous. the difficulties that arise in answering ques-
Sandstrom and Bergstrom (2005) deal with tions 1–3, and why even ideal answers to
this by way of a Heckman correction. In order these three questions may only give a partial
to address the endogeneity of private entry, sense of the overall effect of vouchers.
they use variables approximating whether Specifically, as stated, question 1 can be
local authorities are “hostile to independent credibly addressed with experiments, but
schools.” Specifically, they proxy for these atti- doing so provides only a partial assessment of
tudes using measures of the extent to which the impact of vouchers. For instance, to the
municipalities contract out responsibilities to extent that a private advantage is at least par-
the private sector. The identifying assumption tially due to a peer effect, then this gain will
is that this attitude will only affect educational not be independent of the size of the private
outcomes through the channel that munici- sector and/or the sorting its growth induces.
palities with less hostility will be less likely In other words, the advantage conferred by
to block independent-school entry. The key transferring to a private school may dissipate
finding is that the presence of independent as the private sector grows and incorporates
schools results in better public performance weaker children. In some scenarios—e.g.,
in a GPA-type measure, as well as in stan- if private schools are not more productive
dardized mathematics exams and an indicator and peer effects are linear in means, pri-
for whether students passed all three exams vate expansion may be zero sum (Hsieh and
necessary for high-school admission. Urquiola 2003).
To summarize, several studies of Further, even a solid answer to ques-
public-school response to voucher-school
tion 1 does not provide an assessment of
competition have measured intensity of the c onsequences on the children not using
competition either by the proportion of a vouchers. Studying questions 2 and 3 begins
public school’s students who are potentially to provide a sense of this, but immediately
eligible for vouchers, or by the proximity of raises significant empirical challenges. For
private-school alternatives. Virtually all of
instance, if vouchers induce sorting, then it
these studies find that public-school achieve- is very hard to empirically isolate their effect
ment increases with the intensity of treat- on public-school value added.
ment. That said, most of these analyses do One alternative in the face of these dif-
not have an iron-clad strategy to deal with ficulties—particularly when looking at
potential biases from composition effects large-scale programs—is to simply analyze
(which section 5.2 suggests could be signif- market-level net effects (Hsieh and Urquiola
icant) or with potentially confounding poli- 2003). This does not allow one to isolate the
cies such as accountability.53 specific channels through which vouchers
work, but can provide a sense of their aggre-
5.4 Question 4: What is the net effect of gate effect.
vouchers on educational outcomes?—As
5.4.1 Chile
53 Figlio
One of the more common approaches to
and Hart (2014) study the effects of Florida’s
FTC program before implementation, and therefore avoid addressing question 4 has been to use panel
sorting-related concerns. data for multiple local school markets. The
Epple, Romano, and Urquiola: School Vouchers 479
changes in the private sector’s reach within scores increased substantially—by twenty-
these are then compared to the change nine and forty-eight points respectively
in average performance. This controls for from 2003 to 2011. Hanushek, Peterson,
market-specific fixed characteristics. In this and Woessmann (2012) estimate that Chile
spirit, Hsieh and Urquiola (2006) apply had the second-highest growth rate among
a difference-in-differences approach to forty-nine countries they studied for this
municipal-level data for 1982 to 1996, sug- period. Looking at national tests, Neilson
gesting that while municipalities with greater (2013) shows evidence of improvement
private growth display clear signs of greater among voucher schools in fourth-grade test
sorting, they display no relative advantage in scores during 2008–12—this contrasts with
terms of the evolution of achievement on stan- stagnant performance between 2003 and
dardized tests and years of schooling. As with 2007. At the same time, this progress seems
the analyses of the effects of large-scale pro- to have significantly decelerated, depending
grams on sorting (question 2), the key source on the subject tested, as measured by the last
of concern with these estimates—despite the PISA tests in 2013.
use of some candidate instrumental variables The literature features another approach
(e.g., population density)—is that private to look at the net aggregate effects of choice:
entry into school markets is endogenous. For structural estimation. Papers that take this
instance, if outcomes had been declining in route often achieve interesting analytical
areas where the private sector grew more, richness, but also require strong assump-
these effects would underestimate the salu- tions. Beginning with test scores as an
tary effects of competition. outcome, Neilson (2013) argues that the
Other work implements a c ross-sectional 2008–12 improvement cited in the previ-
variant of this idea by looking for an ous paragraph is due to a 2008 reform that
instrument for the prevalence of voucher
increased the voucher for low-income stu-
schools and evaluating its effect on a ggregate dents. Specifically, the reform increased
performance. Here again, the challenge is the voucher by about 40 percent for the
finding credible instrumental variables for poorest 40 percent of the population, with
private enrollment. Auguste and Valenzuela schools having higher concentrations of
(2006) use distance to a nearby city and find low-income students receiving even higher
evidence of cream skimming and significant payments. In exchange for this higher sub-
positive effects on achievement. Gallego sidy, voucher schools were required to elim-
(2006) uses the density of priests per dio- inate tuition top-ups for these students, and
cese and finds substantial effects on average to refrain from selective admissions. About
achievement. three-fourths of voucher schools eventually
Another way to consider aggregate chose to participate.
effects—the one that takes this logic the fur- The paper uses rich data (that includes dis-
thest—is to simply look at aggregate country tance to school) to estimate school-specific
performance, particularly in international quality measures and a random utility model
tests. Hsieh and Urquiola (2003) point out of school choice by heterogeneous house-
that Chile’s performance did not improve holds. The results suggest that the targeted
in the first twenty years after the voucher voucher: (1) increased average school qual-
reform. Recent experience in this area has ity by 0.21σ, (2) increased average vouch-
been more favorable. After dropping by five er-school quality by 0.16σ, and (3) did not
and seven points respectively from 1999 to affect quality at the non-voucher (elite) pri-
2003, Chile’s 8th grade math and science vate schools. These estimates are then used
480 Journal of Economic Literature, Vol. LV (June 2017)
to parcel achievement gains between those family background characteristics. The iden-
that result as low-income students switch to tification originates in that individuals vary
higher-quality schools (as tuition drops to in the number of years they were exposed
zero) and to increases in school quality. The to the voucher regime. The parameters of
counterfactual with the targeted voucher the structural model are estimated using
introduced but with no change in school about 100,000 person–year observations.
qualities is simulated to estimate the former The results are then used to simulate choices
effect. The analysis suggests that one-third and outcomes for those fully exposed to the
of the achievement gain is from changes in voucher regime and to compare these to the
school choice, and two-thirds is due to school counterfactual with the voucher program
quality improvements. Further research shut down.
here might consider further whether gains The results suggest that the introduction
can be explained “simply” by increased of vouchers led to an increase in earnings
school expenditure from the voucher. from years attending municipal and subsi-
This approach addresses very interesting dized private primary schools, but a decrease
issues, but also requires strong assumptions. from years attending secondary schools.
For instance, p eriod-specific school quality The authors note that the secondary-school
is measured by regressing test scores on a effects might be explained by entry of less-
school fixed effect and a number of controls. efficient schools induced by the voucher,
This assumes there is no selection on unob- and by reduced per-student expenditures
servable characteristics; e.g., parental tastes noted in section 3. Educational attainment
for educational achievement or religious is estimated to be substantially increased by
instruction.54 In addition, the framework the voucher regime. For example, full expo-
assumes that students can attend any school sure to the regime is estimated to increase
they are willing to travel to and pay for. While high-school graduation by 3.6 percent, and
consistent with the Chilean legislation, these completion of at least two years of college
assumptions would seem at odds with the by 2.6 percent. Average lifetime earnings are
extensive stratification and heated current estimated to be unchanged by the voucher.
discussion over the implementation of more The increased attainment is offset by the
binding mechanisms to stop selection, such lower return to secondary education and
as centralized lotteries. delayed entry into the labor force.55 Earnings
Finally, Bravo, Mukhopadhyay, and Todd variation is reduced by the voucher, as those
(2010) present an evaluation of the attain- at the bottom end of the earnings distribu-
ment and labor market—as opposed to test tion benefit from improved primary educa-
score—effects of the Chilean voucher regime tion, while those at the top end suffer from
based on estimation of a structural model of weakened secondary schools. The authors
dynamic school and labor-market choices. find substantial increases in average dis-
They use the 2002 and 2004 waves of a large counted lifetime utility, approximately
survey of w orking-aged individuals that 10 percent. These gains arise from the utility
includes educational and work history, and of time spent attending school and not work-
ing. Gains are found at all percentiles, with
larger increases at lower percentiles than Specifically, measures (1)–(3) are found to
higher percentiles. increase with the i ndependent-school enroll-
The paper is ambitious and the findings of ment share. For instance, a 10 percentage-
large attainment effects, no (average) earn- point increase in this share is associated with
ings effects, and substantial lifetime utility a 0.08σ increase in language and math scores
gains add new findings on the net effects of at the end of ninth grade, and a 0.04σ rise in
vouchers. Identifying the effects of vouchers the fraction of individuals completing at least
at different school levels and on different one semester of university. As stated, these
school types is also notable, relative to other estimates are obtained using data at the
approaches. Again, however, the structural municipality level. When observations are
approach comes at a cost. First, the identi- aggregated further the results, with respect
fication challenges inherent in a large-scale to test score and grade gains, are robust,
program are still present. Second, to make while those with respect to college atten-
the model tractable, individuals are of “just” dance and years of schooling are somewhat
three types. Related to this, the model has less so.
limited potential to provide insight into sort- Bohlmark and Lindahl (2012) conclude
ing effects on schooling and labor-market by discussing the contrast of findings for
outcomes. For example, might the finding Sweden relative to Hsieh and Urquiola’s
of lower earnings from attending second- (2006) findings for Chile. They observe
ary school be a result of sorting (with both that the more favorable results for Sweden
municipal and subsidized schools having with respect both to cream skimming and
worse students on average in the voucher educational outcomes are consistent with
regime), whether through composition or the predictions of the reputational model
peer effects? of MacLeod and Urquiola (2009). Namely,
the fact that it is harder for independent
5.4.2 Sweden schools to cream skim may imply that they
may seek to build reputations for quality on
Bohlmark and Lindahl (2008) present value added, rather than peer composition;
an analysis analogous to that of Hsieh and parental school choice may in turn be driven
Urquiola (2006)—they ask if outcomes by value added as well.
improved by more in municipalities that expe- However, recent work suggests a poten-
rienced more extensive independent-school tial source of caution with respect to the
entry. They focus on three types of student test-score-related results from Sweden.
outcomes: (1) GPA for the ninth and twelfth Specifically, while as stated the content of
grades, (2) participation in higher education the tests is nationally standardized, they are
(a dummy for having completed at least one graded at each school. A concern that had
year of education within six years of leav- been mentioned in Sweden is that indepen-
ing compulsory schooling), and (3) years of dent schools might grade more leniently;
schooling eight years after compulsory edu- indeed this is something that Bohlmark
cation. On outcome (1), the results point to a and Lindahl (2008) themselves mention.
small positive effect on average ninth-grade This was recently analyzed in a regrading
GPA that does not persist until grade 12; on exercise described by Tyrefors, Hinnerich,
(2) and (3) there is no evidence of an effect. and Vlachos (2013). Independent graders
Bohlmark and Lindahl (2012) extend this reexamined exams from different schools.
analysis to several additional cohorts, fi
nding The authors point out that independent
significantly more positive conclusions. schools were more likely to have their grades
482 Journal of Economic Literature, Vol. LV (June 2017)
towns/markets. Specifically, first some towns be 8 percent lower than in a district with a
were selected for distribution of vouch- housing price premium 15 percent below
ers; second, within the towns selected for the average.
treatment, some children were randomly Brunner and Sonstelie (2003) analyze a
selected to receive the vouchers. This allows 2000 voucher proposal in California. They
Muralidharan and Sundararaman (2015) to study opinions on the proposal using data
go beyond the usual comparison (lottery win- from a statewide opinion poll conducted
ners versus lottery losers) and address poten- three months prior to the vote. The poll asked
tial externalities on children who remain in respondents whether they supported the
public school. For example, by comparing voucher, their perception of quality at their
non-applicants in towns that did not receive local public school, and their income and
vouchers to non-applicants in towns that did, demographic characteristics. The authors
they can get a sense of negative effects on investigate responses among three groups of
children “left behind” in the public sector. home owners: those without children, those
The authors find little if any evidence of such with children in public school, and those
externalities. In short, the gains in perfor- with children in private school. For owners
mance found in the tests for Hindi may rep- without children, support for the voucher
resent true a ggregate-level gains. was inversely related to local public-school
quality. This is in keeping with a concern for
5.5 Question 5: What political-economy property values. Households with children in
factors determine the existence and design private schools are much more supportive of
of voucher programs?—We close our review the voucher, but also at the margin have a
of the empirical literature by looking at evi- larger reduction in support as public-school
dence related to political economy. Brunner, quality increases. For those with children in
Sonstelie, and Thayer (2001) study voting on public school, voucher support was positively
the 1993 California voucher proposal using related to public-school quality, though sig-
data from 3,786 precincts in Los Angeles nificant at only the 10 percent level. Brunner
County. They take Nechyba (1999) as their and Sonstelie note that this finding might
point of departure, noting that this model emerge if (as Nechyba’s analysis predicts)
predicts that a voucher would cause house some households with children in public
prices to fall (rise) in neighborhoods with school expect to move to a neighborhood
high (low) quality schools. The authors use a with low-quality public schools (low housing
hedonic equation to estimate the relationship prices), take up the voucher, and send their
of housing prices and school quality—mea- children to private school. Households with
suring the latter by scores on standardized children already in private school who choose
tests—to estimate the housing price pre- nonetheless to live in a neighborhood with
mium for schooling in each district. This a high-quality public school would presum-
price premium is then included in a regres- ably not relocate in response to the voucher.
sion in which the dependent variable is the Hence, results that at first glance seem con-
proportion of voters supporting the voucher. tradictory can in principle be reconciled.
They find strong support for predictions with Brunner and Imazeki (2008) extend the
respect to housing owners. Holding other analysis of California voting, arguing that
variables constant, their estimates imply that higher-income voters’ support for a voucher
the difference in the vote favoring vouch- depends on the extent of choice in local
ers between districts with a housing pre- education markets. In markets with many
mium 15 percent above the average would districts, higher-income households can be
484 Journal of Economic Literature, Vol. LV (June 2017)
expected to have paid a substantial housing in some way, provide support for private
price premium to locate in a high-quality dis- schools. These include proposals for trans-
trict. For such households, a private-school portation for private-school students, in
voucher program could adversely affect addition to a variety of voucher models. All
home values. By contrast, in low-choice ten referenda were defeated. Kenny then
markets, high-income households would turns to a descriptive assessment of factors
likely benefit from being able to use vouch- that influence whether a voucher proposal is
ers to pay for private schooling. They use considered by a state legislature, a number
block-level voting data from the 2000 ballot of which have been passed or at least sup-
initiative (Proposition 38) that would have ported by one chamber, and factors that raise
offered a flat voucher of $4,000 per student its probability of success. He concludes that
in California. They create an index for choice ideology plays a central role, noting that vir-
among public schools and estimate regres- tually all proposals were in states where the
sions in which the dependent variable is a Republican Party controlled the legislature,
logistic transformation of the fraction of yes and among those, the successful proposals
votes. The regressions include income, an were in states with relatively more conserva-
interaction of income and the choice index, tive Republicans. Kenny also finds evidence
and demographic controls. They predict that that voucher proposals focused on big-city
income will have a positive coefficient, the districts tend to garner more support.
interaction of income and the choice index Kenny (2010) investigates voting on two
will have a negative coefficient, and the sum voucher proposals that have come before the
of the coefficients will be negative. Their pre- US Congress. One, an amendment to the No
dictions are s upported. Moreover, the effects Child Left Behind (NCLB) Act, would have
are quite large and are robust to a variety of allowed federal funds to help children in
specification checks, providing evidence that poorly performing or unsafe public schools
the extent of Tiebout choice impacts voucher attend private schools. The other was to
support. authorize a voucher for Washington, DC.
Brunner, Imazeki, and Ross (2010) Only four Democrats, found by Kenny to be
exploit the idea that votes on the California highly conservative, voted for either. Hence,
voucher initiative may signal intent to use Kenny again focuses on an empirical descrip-
the voucher. They find that support for the tion of the determinants of votes by House
voucher by white households with children Republicans. He considers the degree of
increased with the proportion of nonwhite conservatism of the legislator—based on
students in their children’s schools. No com- Americans for Democratic Action (ADA)
parable phenomenon was present either scores—and characteristics of the legislator’s
for nonwhite households or for households district: percent urban, percent low income,
without children. They provide some evi- percent black, percent teachers unionized,
dence that households may be responding to and percent of private-school attendees. Of
correlates of race/ethnicity rather than race/ these, the ADA score is highly significant
ethnicity per se. For example, voucher sup- in the expected direction. The variation in
port among n onwhite households with chil- percent teachers unionized is negatively
dren increased with the share of n onwhite correlated with voting on the NCLB amend-
students with limited English proficiency. ment, but essentially uncorrelated with the
Kenny (2005) studies support for vouch- position on the Washington, DC vote. Kenny
ers, both in referenda and state legislatures. notes this is consistent with teachers’ unions
He identifies ten referenda initiatives that, exerting more effort to defeat the NCLB
Epple, Romano, and Urquiola: School Vouchers 485
also suggests that work originating in a sin- Catholic Schools.” Journal of Political Economy 113
gle country or in a single research approach (1): 151–84.
Andrabi, Tahir, Jishnu Das, and Asim Ijaz Khwaja.
is unlikely to completely answer questions 2013. “Students Today, Teachers Tomorrow: Iden-
regarding vouchers. S mall-scale experiments tifying Constraints on the Provision of Education.”
are appealing in providing strong statisti- Journal of Public Economics 100: 1–14.
Angrist, Joshua D., Eric Bettinger, Erik Bloom,
cal identification, but do not always isolate Elizabeth King, and Michael Kremer. 2002. “Vouch-
mechanisms (e.g., peer effects, differences ers for Private Schooling in Colombia: Evidence
in expenditure per pupil) and leave open from a Randomized Natural Experiment.” American
Economic Review 92 (5): 1535–58.
the issue of scalability. L
arge-scale programs Angrist, Joshua D. Eric Bettinger, and Michael Kremer.
provide scope for assessment of the effects 2006. “Long-Term Educational Consequences of
of vouchers in practice, but identification is Secondary School Vouchers: Evidence from Admin-
istrative Records in Colombia.” American Economic
a greater challenge due to potential selec- Review 96 (3): 847–62.
tion effects and associated differential peer Auguste, S., and J. Valenzuela. 2006. “Is It Just Cream
effects (and, sometimes, to potential con- Skimming? School Vouchers in Chile.” Unpublished.
Barnard, John, Constantine E. Frangakis, Jennifer L.
founding effects of contemporaneous policy Hill, and Donald B. Rubin. 2003. “Principal Strati-
changes). For research, the ongoing tasks fication Approach to Broken Randomized Experi-
include continuing refinement of identifi- ments: A Case Study of School Choice Vouchers in
New York City.” Journal of the American Statistical
cation strategies, investigating longer-term Association 98 (462): 299–323.
impacts, providing a better understanding Bearse, Peter, Buly A. Cardak, Gerhard Glomm, and
of why effects emerge or fail to emerge, and B. Ravikumar. 2013. “Why Do Education Vouchers
Fail at the Ballot Box?” European Journal of Political
marshaling theory and evidence to improve Economy 32: 26–37.
voucher design. Belfield, Clive R. 2006. “The Evidence on Education
Vouchers: An Application to the Cleveland Scholar-
References ship and Tutoring Program.” Unpublished.
Besley, Timothy, and Stephen Coate. 1997. “An Eco-
Abdulkadiroglu, Atila, Joshua D. Angrist, Susan M. nomic Model of Representative Democracy.” Quar-
Dynarski, Thomas J. Kane, and Parag A. Pathak. terly Journal of Economics 112 (1): 85–114.
2011. “Accountability and Flexibility in Public Bettinger, Eric, Michael Kremer, and Juan E. Saavedra.
Schools: Evidence from Boston’s Charters and Pilots.” 2010. “Are Educational Vouchers Only Redistribu-
Quarterly Journal of Economics 126 (2): 699–748. tive?” Economic Journal 120 (546): F204–28.
Abdulkadiroglu, Atila, Joshua D. Angrist, and Parag Bjorklund, Anders, Melissa A. Clark, Per-Anders Edin,
A. Pathak. 2014. “The Elite Illusion: Achievement Peter Fredriksson, and Alan B. Krueger. 2006. The
Effects at Boston and New York Exam Schools.” Market Comes to Education in Sweden: An Evalu-
Econometrica 82 (1): 137–96. ation of Sweden’s Surprising School Reforms. New
Abdulkadiroglu, Atila, Parag A. Pathak, and Christo- York: Russell Sage Foundation.
pher R. Walters. 2015. “School Vouchers and Student Bohlmark, Anders, Helena Holmlund, and Mikael
Achievement: First-Year Evidence from the Louisi- Lindahl. 2015. “School Choice and Segregation:
ana Scholarship Program.” School Effectiveness and Evidence from Sweden.” Institute for Evaluation of
Inequality Initiative Working Paper 2015.06. Labour Market and Education Policy Working Paper
Adams, Mark. 2009. “Tomorrow’s Schools Today: New 2015:8.
Zealand’s Experiment 20 Years On.” George Mason Bohlmark, Anders, and Mikael Lindahl. 2007. “The
University Mercatus Center Working Paper 09-01. Impact of School Choice on Pupil Achievement, Seg-
Aedo, Cristian, and Osvaldo Larrañaga. 1994. “The regation and Costs: Swedish Evidence.” Institute for
Chilean Experience.” In Social Service Delivery Sys- the Study of Labor Discussion Paper 2786.
tems: An Agenda for Reform, edited by Cristian Aedo Bohlmark, Anders, and Mikael Lindahl. 2008. “Does
and Osvaldo Larrañaga, 13–49. Washington, DC: School Privatization Improve Educational Achieve-
Inter-American Development Bank. ment? Evidence from Sweden’s Voucher Reform.”
Ahlin, Asa. 2003. “Does School Competition Matter? Institute for the Study of Labor Discussion Paper
Effects of a Large-Scale School Choice Reform on 3691.
Student Performance.” Uppsala University Depart- Bohlmark, Anders, and Mikael Lindahl. 2012. “Inde-
ment of Economics Working Paper 2003:2. pendent Schools and Long-Run Educational Out-
Altonji, Joseph G., Todd E. Elder, and Christopher comes: Evidence from Sweden’s Large Scale Voucher
R. Taber. 2005. “Selection on Observed and Unob- Reform.” CESifo Working Paper 3866.
served Variables: Assessing the Effectiveness of Bositis, David A. 2008. “National Survey of African
488 Journal of Economic Literature, Vol. LV (June 2017)
American Families’ Views on Education.” http:// Chakrabarti, Rajashri. 2013d. “Vouchers, Public School
jointcenter.org/sites/default/files/Education_0.pdf. Response, and the Role of Incentives: Evidence from
Bravo, David, D. Contreras, and C. Sanhueza. 1999. Florida.” Economic Inquiry 51 (1): 500–526.
“Educational Achievement, Inequalities and Private/ Chan, Ping Ching Winnie, and Robert McMillan.
Public Gap, Chile 1982–1997.” University of Chile 2009. “School Choice and Public School Perfor-
Department of Economics Working Paper 163. mance: Evidence from Ontario’s Tuition Tax Credit.”
Bravo, David, Sankar Mukhopadhyay, and Petra E. Unpublished.
Todd. 2010. “Effects of School Reform on Educa- Chay, Kenneth Y., Patrick J. McEwan, and Miguel
tion and Labor Market Performance: Evidence from Urquiola. 2005. “The Central Role of Noise in
Chile’s Universal Voucher System.” Quantitative Evaluating Interventions That Use Test Scores to
Economics 1 (1): 47–95. Rank Schools.” American Economic Review 95 (4):
Brunner, Eric J., and Jennifer Imazeki. 2008. “Tiebout 1237–58.
Choice and Universal School Vouchers.” Journal of Chen, Zhiqi, and Edwin G. West. 2000. “Selective
Urban Economics 63 (1): 253–79. versus Universal Vouchers: Modeling Median Voter
Brunner, Eric J., Jennifer Imazeki, and Stephen L. Preferences in Education.” American Economic
Ross. 2010.” Universal Vouchers and Racial and Eth- Review 90 (5): 1520–34.
nic Segregation.” Review of Economics and Statistics Chetty, Raj, John N. Friedman, and Jonah E. Rockoff.
92 (4): 912–27. 2014a. “Measuring the Impacts of Teachers I: Eval-
Brunner, Eric J., and Stephen L. Ross. 2010. “Is uating Bias in Teacher Value-Added Estimates.”
the Median Voter Decisive? Evidence from Refer- American Economic Review 104 (9): 2593–632.
enda Voting Patterns.” Journal of Public Economics Chetty, Raj, John N. Friedman, and Jonah E. Rockoff.
94 (11–12): 898–910. 2014b. “Measuring the Impacts of Teachers II:
Brunner, Eric J., and Jon Sonstelie. 2003. “Homeown- Teacher Value-Added and Student Outcomes in
ers, Property Values, and the Political Economy of Adulthood.” American Economic Review 104 (9):
the School Voucher.” Journal of Urban Economics 54 2633–79.
(2): 239–57. Chingos, Matthew M., and Paul E. Peterson. 2015.
Brunner, Eric J., Jon Sonstelie, and Mark Thayer. 2001. “Experimentally Estimated Impacts of School
“Capitalization and the Voucher: An Analysis of Pre- Vouchers on College Enrollment and Degree Attain-
cinct Returns from California’s Proposition 174.” ment.” Journal of Public Economics 122: 1–12.
Journal of Urban Economics 50 (3): 517–36. Chubb, John E., and Terry M. Moe. 1990. Politics,
Card, David, Martin D. Dooley, and A. Abigail Payne. Markets, and America’s Schools. Washington, DC:
2010. “School Competition and Efficiency with Pub- Brookings Institution Press.
licly Funded Catholic Schools.” American Economic Chudgar, Amita, Frank Adamson, and Martin
Journal: Applied Economics 2 (4): 150–76. Carnoy. 2007. “Vouchers and Public School Perfor-
Carnoy, Martin. 1997. “Is Privatization through Edu- mance: A Case Study of the Milwaukee Parental
cation Vouchers Really the Answer? A Comment Choice Program.” http://www.epi.org/publication/
on West.” World Bank Research Observer 12 (1): book_vouchers/.
105–16. Clark, Damon. 2009. “The Performance and Competi-
Carrell, Scott E., Bruce I. Sacerdote, and James E. tive Effects of School Autonomy.” Journal of Political
West. 2013. “From Natural Variation to Optimal Economy 117 (4): 745–83.
Policy? The Importance of Endogenous Peer Group Clark, Damon. 2010. “Selective Schools and Academic
Formation.” Econometrica 81 (3): 855–82. Achievement.” BE Journal of Economic Analysis and
Chakrabarti, Rajashri. 2008. “Can Increasing Private Policy 10 (1).
School Participation and Monetary Loss in a Voucher Coons, John E., and Stephen D. Sugarman. 1978.
Program Affect Public School Performance? Evi- Education by Choice: The Case for Family Control.
dence from Milwaukee.” Journal of Public Econom- Berkeley and Los Angeles: University of California
ics 92 (5–6): 1371–93. Press.
Chakrabarti, Rajashri. 2013a. “Accountability with Costrell, Robert M. 2010. “The Fiscal Impact of the
Voucher Threats, Responses, and the Test-Taking Milwaukee Parental Choice Program: 2010–2011
Population: Regression Discontinuity Evidence Update and Policy Options.” School Choice Demon-
from Florida.” Education Finance and Policy 8 (2): stration Project Milwaukee Evaluation Report 22.
121–67. Cowen, Joshua M., David J. Fleming, John F. Witte,
Chakrabarti, Rajashri. 2013b. “Do Vouchers Lead to and Patrick J. Wolf. 2011. “Student Attainment and
Sorting under Random Private School Selection? the Milwaukee Parental Choice Program.” School
Evidence from the Milwaukee Voucher Program.” Choice Demonstration Project Milwaukee Evalua-
Economics of Education Review 34: 191–218. tion Report 24.
Chakrabarti, Rajashri. 2013c. “Impact of Voucher Cowen, Joshua M., David J. Fleming, John F. Witte,
Design on Public School Performance: Evidence Patrick J. Wolf, and Brian Kisida. 2013. “School
from Florida and Milwaukee Voucher Programs.” Vouchers and Student Attainment: Evidence from
BE Journal of Economic Analysis and Policy 13 (1): a State-Mandated Study of Milwaukee’s Parental
349–94. Choice Program.” Policy Studies Journal 41 (1):
Epple, Romano, and Urquiola: School Vouchers 489
Economic Journal: Applied Economics 4 (4): 226–53. vidence from a Two-Stage Experiment in India.”
E
MacLeod, W. Bentley, and Miguel Urquiola. 2009. Quarterly Journal of Economics 130 (3): 1011–66.
“Anti-lemons: School Reputation and Educational Neal, Derek. 2002. “How Vouchers Could Change the
Quality.” National Bureau of Economic Research Market for Education.” Journal of Economic Per-
Working Paper 15112. spectives 16 (4): 25–44.
MacLeod, W. Bentley, and Miguel Urquiola. 2012. Nechyba, Thomas J. 1999. “School Finance Induced
“Anti-lemons: School Reputation, Relative Diversity, Migration and Stratification Patterns: The Impact
and Educational Quality.” Institute for the Study of of Private School Vouchers.” Journal of Public Eco-
Labor Discussion Paper 6805. nomic Theory 1 (1): 5–50.
MacLeod, W. Bentley, and Miguel Urquiola. 2015. Nechyba, Thomas J. 2000. “Mobility, Targeting, and
“Reputation and School Competition.” American Private-School Vouchers.” American Economic
Economic Review 105 (11): 3471–88. Review 90 (1): 130–46.
Manski, Charles F. 1992. “Educational Choice (Vouch- Nechyba, Thomas J. 2003. “Introducing School Choice
ers) and Social Mobility.” Economics of Education into Multidistrict Public School Systems.” In The
Review 11 (4): 351–69. Economics of School Choice, edited by Caroline M.
Mayer, Daniel P., Paul E. Peterson, David E. Myers, Hoxby, 145–94. Chicago and London: University of
Christina Clark Tuttle, and William G. Howell. 2002. Chicago Press.
“School Choice in New York City after Three Years: Neilson, Christopher. 2013. “Targeted Vouchers, Com-
An Evaluation of the School Choice Scholarships petition among Schools, and the Academic Achieve-
Program.” Mathematica Policy Research Reference ment of Poor Students.” Unpublished.
No. 8404-045. Patrinos, Harry Anthony. 2002. “Private Education
McEwan, Patrick J. 2001. “The Effectiveness of Pub- Provision and Public Finance: The Netherlands as a
lic, Catholic, and Non-religious Private Schools in Possible Model.” Columbia University National Cen-
Chile’s Voucher System.” Education Economics 9 (2): ter for the Study of Privatization in Education Occa-
103–28. sional Paper 59.
McEwan, Patrick J. 2004. “The Potential Impact of Peterson, Paul E., William G. Howell, Patrick J. Wolf,
Vouchers.” Peabody Journal of Education 79 (3): and David E. Campbell. 2003. “School Vouchers:
57–80. Results from Randomized Experiments.” In The
McEwan, Patrick J., and Martin Carnoy. 2000. “The Economics of School Choice, edited by Caroline M.
Effectiveness and Efficiency of Private Schools in Hoxby, 107–44. Chicago and London: University of
Chile’s Voucher System.” Educational Evaluation Chicago Press.
and Policy Analysis 22 (3): 213–39. Peterson, Paul E., and William G. Howell. 2004a. “Effi-
McEwan, Patrick J., Miguel Urquiola, and Emiliana ciency, Bias, and Classification Schemes: A Response
Vegas. 2008. “School Choice, Stratification, and to Alan B. Krueger and Pei Zhu.” American Behav-
Information on School Performance: Lessons from ioral Scientist 47 (5): 699–717.
Chile.” Economía 8 (2): 1–42. Peterson, Paul E., and William G. Howell. 2004b.
McMillan, Robert. 2004. “Competition, Incentives, and “Voucher Research Controversy: New Looks at the
Public School Productivity.” Journal of Public Eco- New York City Evaluation.” Education Next 4 (2):
nomics 88 (9–10): 1871–92. 73–78.
Mills, Jonathan N., and Patrick J. Wolf. 2016. “The Plott, Charles R. 1967. “A Notion of Equilibrium and
Effects of the Louisiana Scholarship Program on Its Possibility under Majority Rule.” American Eco-
Student Achievement After Two Years.” Louisiana nomic Review 57 (4): 787–806.
Scholarship Program Evaluation Report Number 1. Plucker, Jonathan, Patricia Muller, John Hansen, Russ
Mizala, Alejandra, and Pilar Romaguera. 2000. “School Ravert, and Matthew Makel. 2006. “Evaluation of
Performance and Choice: The Chilean Experience.” the Cleveland Scholarship and Tutoring Program:
Journal of Human Resources 35 (2): 392–417. Technical Report 1998–2004.” Bloomington, IN:
Mizala, Alejandra, Pilar Romaguera, and Miguel Center for Evaluation and Education Policy.
Urquiola. 2007. “Socioeconomic Status or Noise?
Pop-Eleches, Cristian, and Miguel Urquiola. 2013.
Tradeoffs in the Generation of School Quality Infor- “Going to a Better School: Effects and Behavioral
mation.” Journal of Development Economics 84 (1): Responses.” American Economic Review 103 (4):
61–75. 1289–324.
Mizala, Alejandra, and Miguel Urquiola. 2013. “School Rangazas, Peter. 1995. “Vouchers and Voting: An Initial
Markets: The Impact of Information Approximat- Estimate Based on the Median Voter Model.” Public
ing Schools’ Effectiveness.” Journal of Development Choice 82 (3–4): 261–79.
Economics 103: 313–35. Rouse, Cecilia Elena. 1998. “Private School Vouchers
Muralidharan, Karthik, and Venkatesh Sundararaman. and Student Achievement: An Evaluation of the Mil-
2011. “Teacher Performance Pay: Experimental Evi- waukee Parental Choice Program.” Quarterly Jour-
dence from India.” Journal of Political Economy 119 nal of Economics 113 (2): 553–602.
(1): 39–77. Rouse, Cecilia Elena, and Barrow, Lisa. 2009. “School
Muralidharan, Karthik, and Venkatesh Sundararaman. Vouchers and Student Achievement: Recent Evi-
2015. “The Aggregate Effect of School Choice: dence and Remaining Questions.” Annual Review of
492 Journal of Economic Literature, Vol. LV (June 2017)