8/12/2019 16. Elacqua

1/37

The politics of education reform in Chile: When

ideology trumps evidence

Gregory ElacquaPrepared for the Research Seminar in Comparative Politics

February 10, 2008

1

8/12/2019 16. Elacqua

2/37

1 Introduction

Since the return of democracy in Chile in 1990, education reforms have focused on improving

quality and equity through curricular reform, increased investment in teachers salaries,

school construction, longer school days, and the provision of computers and Internet in all

schools (Cox, 2003). Most experts agree that these investments increased coverage, especially

for low-income children, improved the quality of school facilities, provided many children with

the opportunity to spend more hours a day at school (OECD, 2004), and increased parent

satisfaction with the quality of their childrens schools (e.g. Fundacion Futuro, 2005).

Despite these positive outcomes, and a fourfold increase in spending in inflation adjusted

terms since 1990, there has not been significant improvement in the average quality of learn-

ing. Student achievement in Chile is among the highest in Latin America, but still lags

significantly behind a number of emerging countries in Asia and Eastern Europe, as demon-

strated by the poor results achieved on the Trends in International Mathematics and Science

Study (TIMSS), the Programme for International Student Assessment (PISA), and the In-

ternational Civic Education Study (CIVIC). National test scores have also been stagnant

since 1997, and large test score gaps persist both among socioeconomic groups and between

public and private schools. Schools are also stratified by socioeconomic status. Students

attending private schools, on average, come from families that have much higher incomes

and that are headed by parents with substantially more schooling than students enrolled in

public schools. 1

These factors converged to motivate one of the largest protests in Chilean history that is

widely known as the march of the penguins - in reference to the protesters school uniforms.

The protest began in May 2006, less than three months after President Michelle Bachelet

took office. More than 600,000 students walked out of class and occupied hundreds of schools

all over Chile, demanding the overhaul of the education system. The student movement had

widespread popular support among university students, the teachers union, the workers

union, and average citizens.2

1For an analysis of the quality and equity of Chiles education system see Contreras and Elacqua (2005)and Mizala and Romaguera (2006).

2Public support for the protests was nearly universal with almost 90 percent of Chileans polled sayingthat they supported the student movement (El Mercurio, 2006a).

1

8/12/2019 16. Elacqua

3/37

The student demands included more teachers and improved school construction, the elim-

ination of fees for the national college entrance exam, and free student public transportation

fares. With prices of copper, Chiles chief export, at record highs, and government reserves

with several years of budget surpluses, the students maintained that the government could

afford to invest more in education.3

President Bachelet responded to the students demands by offering an emergency spend-

ing bill of close to USD 135 million a year that covered free student transportation passes,

a waiver of university exam fees for the poorest students, and the renovation of over 1,200

public schools. The offer was rejected by students, who, in addition to the additional re-

sources, demanded the L.O.C.Es (Ley Organica Constitucional de Ensenanza) reform. The

L.O.C.E was a Pinochet-era constitutional education law that decentralized schooling in

Chile, making it possible for almost anyone to open a school4 and receive government fund-

ing without having to conform to any standard of quality. The students main objection was

that for-profit schools were allowed to compete with non-profit schools and public schools

for students (El Mercurio, 2006b).

Education in Chile occurs in a mixed market with 53 percent of students enrolled in

public schools, 25 percent in for-profit voucher schools, 15 percent in non-profit (religious

and secular) voucher schools, and 7 percent in private non-voucher schools. The position

taken by the students and others opposed to funding for-profits is the belief that for-profit

providers cannot be trusted to place the interests of children ahead of profitability (OEI,

2007). Skeptics have countered that for-profit schools have stronger incentives to reduce

costs, and more importantly, to innovate, leading to both higher quality and greater efficiency

in education (Tironi, 2006). Neither of these arguments, however, is based on any data in

Chile on the quality of education provided by the different types of private schools (for-profit

or non-profit).

There is also a persistent scholarly debate on this topic. Researchers have developed

a number of theories as to why non-profit organizations might outperform for-profit firms

3A common slogan on student banners read Copper sky high and education in the gutter (Rohter,2006).

4The only formal requirement to open a school in Chile is to have a high school diploma (Montt et al.,2006).

2

8/12/2019 16. Elacqua

4/37

in mixed industries. One view is that for-profit firms have incentives to take advantage of

customers by providing inferior services buyers cannot evaluate (Hausmann, 1987). Under

these circumstances, economic theory predicts that non-profit providers will come into exis-

tence to provide high quality services to poorly informed customers who seek a trustworthy

organization (Weisbrod, 1988). A second viewpoint is that non-profits are better positioned

to provide under-satisfied demands, such as the provision of goods to disadvantaged popula-

tions, than for-profit firms because they may rely on donations of money or volunteer time

to finance the provision of these goods, while for-profits must satisfy a market survival test

(Rose-Ackerman, 1996).

These claimed advantages have not gone unchallenged. Critics argue that the ambition

of profit fosters efficient decision making by for-profit firms. In contrast, non-profit organi-

zations are insulated from competitive pressures and thus have little incentive to manage

their firms efficiently (Glaeser and Schleifer, 2001). For instance, because non-profits cannot

distribute profits to owners, critics assert that their managers have less incentive to minimize

costs and may, for example, pay themselves excessive salaries.

Other critics suggest that there are likely no systematic differences in the objectives of

for-profit and nonprofit suppliers. Non-profits may engage in profit making activities and,

conversely, for-profit firms may have a deep commitment for the services they produce. Mis-

sion driven firms may find the constraints placed on non-profit organizations too restrictive,

and profit maximizing firms may find it more advantageous to choose non-profit forms, due

to tax exemptions, for example (Weisbrod, 1998).

Empirical studies generally corroborate the theoretical predictions of higher quality in

the non-profit sector. For instance, Lukesetich et al. (2000) show that non-profit nursing

homes spend more per-patient on nursing care and less on administrative expenses than for-

profit homes. Ford and Kasserman (2000) find that non-profit kidney dialysis clinics provide

significantly longer treatment than for-profit dialysis clinics. Similarly, non-profit hospitals

provide more uncompensated care than for-profits (Schlesinger et al., 1987). In studies

of prisons, Hart et al. (1997) find that for-profit prisons hire lower quality prison guards

than non-profits. The empirical studies of day care centers also show systematic quality

differences between non-profit and for-profit centers. Non-profits rank higher along input

3

8/12/2019 16. Elacqua

5/37

measures such as child-staff ratios and staff experience, while for-profits generally provide

lower quality services for similar fees (Morris and Helburn, 2000).

Much of the existing empirical research in education treats private schools as an aggre-

gate category and no studies in Chile, and very few studies in other countries, have examined

whether performance differs across for-profit and non-profit schools. The evidence on this

point is limited because there are so few schooling systems that provide public funding to

private schools. While different combinations of private and public provision (funding and

management) are observed in many countries, most schools continue to be funded and oper-

ated primarily by the government (OECD, 2006), and non-profit status is usually required

for private educational institutions (James, 1993).5 This suggests that the positions taken

by proponents and critics of the proposal to cease funding for-profit voucher schools have

not been informed by any evidence in Chile or elsewhere on the relative performance of these

schools.

In the midst of a fierce debate in Congress and the media on the merits of the students

proposal, Bachelet ordered the creation of a presidential advisory panel with 81 members

to seek a national consensus on education reform. During the six months of talk, the body

evaluated the Chilean educational system and sought to reach a consensus on the structural

change in the system. But the talks collapsed in December 2006 when members of the social

block, which comprised of students, teachers, and parents, rejected the Advisory Councils

final report. Bachelet announced that her government would study the report and announce

a proposal.

Four months later, in March 2007, when Bachelet announced a reform of the L.O.C.E.

it was apparent that the social blocks demands were being heard. On Monday April 9,

the President met with the Minister of Education, a delegation of students, and various

congressman and senators to sign the legislation that would put an end to the L.O.C.E. and

create a new General Law of Education (Ministry of Education, 2007).

5Some researchers in the United States have made finer distinctions between private non-profit charterschools. For instance, recent research distinguishes between market-oriented and mission-oriented non-profit charter schools (Brown et al., 2005). Although they find differences between market-oriented schoolsand mission-oriented schools across different dimensions, the theoretical typology they use for distinguishingbetween non-profit charter schools limits their ability to disentangle the inter-institutional differences drivenby incentives and legal constraints placed on non-profit and for-profit schools.

4

8/12/2019 16. Elacqua

6/37

The key points of the law deal with reducing discrimination and selection in public and

private voucher schools, which is currently a widespread practice. Under the new law, no

school will be able to select students. The law also proposes to eliminate for-profit voucher

schools, so that only non-profit and public schools will be able to operate an educational

organization.

The goal of this paper is to evaluate the merits of this proposal by examining whether or

not for-profit voucher schools are less effective, all else equal, than non-profit voucher schools

in Chile. Using a highly-detailed unique data set Ive constructed from the administrative

records of the Ministry of Education, I compare the achievement of eighth-graders in public

schools, for-profit and non-profit (religious and secular) voucher schools, and non-voucher

schools.

This is not the first paper to examine private voucher school performance in Chiles

national voucher program. Earlier work used aggregated school level data (Mizala and

Romaguera, 2000). More recently, researchers used student-level data and attempted to

control for selection bias (Anand et al., 2006; Sapelli and Vial, 2002; McEwan, 2001). Most

of these studies show a private school advantage over public schools, although the differences

are usually small.

This paper differs from earlier work by examining achievement across private voucher

schools according to their ownership type. I consider for-profit (independent and franchise)

voucher schools, non-profit (religious and secular) voucher schools, and private non-voucher

schools; prior researchers have used a single category to describe all private vouchers schools,

with the exception of McEwan (2001) who studied Catholic school achievement and Elacqua

et al. (forthcoming) who examined achievement across private voucher schools according to

their network size. The findings presented in this study demonstrate that, while differential

quality is found across private voucher school types (for-profit and non-profit), the differences

do not always comport with the positions on either side of the policy debate in Chile.

The remainder of the paper is organized as follows. The second section reviews some

background on Chiles school system and describes the school ownership types that will

serve as the key analytical categories. The next section sets forth the empirical strategy that

will be used to compare student achievement in public and private school types, and describes

5

8/12/2019 16. Elacqua

7/37

the data that will be used to implement it. The fourth section presents and interprets in the

empirical analysis. The final section concludes and discusses policy implications.

2 Background on Chile

During the 1980s, the school system in Chile experienced a sweeping reform program enacted

by the military government (1973-1990). First, the government decentralized the administra-

tion of schools, transferring responsibility for public school management from the Ministry

of Education to local municipalities. Second, the administration altered the financing of

public and most private schools. Public schools continued to be funded centrally, but mu-

nicipalities started to receive a per-pupil payment for every child attending their schools.

As a result, enrollment losses came to have a direct effect on their education budgets. Most

importantly, private schools that did not charge tuition began receiving the same per-pupil

payment as public schools.6 Tuition-charging private schools continued to operate without

public funding.

The essential features of this system have remained in place for over a quarter-century.

The only significant modification was in 1994, when the Ministry instituted a financing

scheme that allowed all private voucher schools to charge limited tuition (Montt et al.,

2006).The voucher and shared financing reforms sparked a massive redistribution of students

across private and public schools, as well as the creation of many new private schools. In

1981, 15 percent of Chilean K-12 students attended private schools that received some public

subsidy, and another 7 percent attended more elite, unsubsidized private schools. By 1990,

32.4 percent of students attended private voucher schools. By 2004, enrollment in such

schools had reached almost 40 percent of total enrollment. Most of these gains were at the

expense of public school enrollments. Adding in the 7.6 percent of students in elite privatenon-voucher schools, leaves only a slight majority of Chilean students in public schools (see

Figure 1).

Figure 1 Here

6Chiles voucher formula includes adjustments for rural schools and high schools, but is flat with respectto student socioeconomic characteristics.

6

8/12/2019 16. Elacqua

8/37

Most researchers generally use a single category to describe private voucher schools in

Chile. However, there is a great deal of variability in the private voucher sector. Prior to

the voucher reforms in 1981, most subsidized private schools were non-profit (Aedo, 2000).

When private subsidized schools began to receive the same per-pupil payment as the public

schools, a flood of mostly for-profit voucher schools entered the market. Table 1 shows how

primary and secondary school students are distributed across school types. Public schools

account for the majority (62 percent) of schools and enrollments (53 percent). For-profit

voucher schools account for 23 percent of schools and 25 percent of enrollments. Non-profit

voucher schools account for about 9 percent of schools and close to 15 percent of enrollments.

Non-voucher schools represent less than 8 percent of schools and enrollments.

Table 1 HereOver the period 1990 and 2004, the total number of for-profit schools increased by 36

percent, and total enrollment in for-profit schools increased by 62 percent (see Table 2).

For-profit schools have far outpaced growth rates of other school types. The total number

of public schools decreased by 4 percent and total enrollments in public schools increased

by only 3 percent. Non-profit voucher have consolidated of the years yielding a moderate

growth in the number of schools and enrollments (see table 2).

Table 2 Here

Table 3 provides basic descriptive information about the 4 school types. Non-profit

schools have, on average, more students per school than other schools. The data show that

public schools are more likely to serve low-income and rural students than all categories of

private schools. Table 3 also shows that for-profit voucher schools enroll a slightly higher

proportion of disadvantaged students than non-profit voucher schools and that students that

attend private non-voucher schools come from the most highly educated families. The data

also show that about half of private voucher schools (for-profit and non-profit) charge tuition.

Table 3 Here

For-profit and non-profit voucher schools in Chile are diverse in membership. Table 4

describes the management and financing of public and non-voucher schools and six types

of private voucher schools (for-profit franchise, for-profit independent, Catholic, Protestant,

7

8/12/2019 16. Elacqua

9/37

and non-sectarian). Table 5 shows how primary and secondary school students are dis-

tributed across these seven school categories. For-profit franchises, which represent about 20

percent of all for-profit schools, are probably those that best fit the description of educational

privatization proponents (Chubb, 2001). Most of these schools were founded in the last 5

years. They are often controlled by a group of off-site owners, in some cases with private

shareholders, often have ties to other industries, and are characterized by networks of cam-

puses. These for-profit schools, which account for over 4 percent of schools and 7 percent

of enrollments, stand in varying degrees of contrast to for-profit independent schools, which

account for about 80 percent of all for-profit schools and 19 percent of total schools and 18

percent of enrollments. For-profit independent schools are especially small in size and scale,

suggesting that when these groups set out to establish a school, they are probably looking

to create a school that only provides services to children in the community. Many of these

schools were founded during the first decade of the reform, and anecdotal evidence suggests

that many of the owners are former public school teachers who were expelled by the military

government.7

Table 4 Here

Table 5 Here

Non-profit voucher schools, including Catholic, Protestant,8

and non-sectarian organiza-

tions,9 are more likely to be characterized by an academic and/or religious mission rather

than profit maximization. These schools, which are often subsidized by the Church or local

businesses, often have access to donated facilities and teachers willing to work for below-

market salaries, and thus are able to provide a range of services to students whose costs

exceed the voucher and tuition payments. Most non-profit schools are also characterized

by networks of campuses that are affiliated through religious congregations or foundations.

Catholic schools account for about 75 percent of non-profit schools and 7 percent of all7The National Private Voucher School Association (CONACEP) provided me with the descriptive in-

formation about for-profit voucher schools. I have also conducted a number of focus groups with a diversegroup of for-profit school owners.

8There are 4 private voucher schools of other religious orientations. These schools were dropped from thesample.

9Most of the non-sectarian non-profit schools are branches of foundations that were created for otherspecific tasks, such as the Aid Corporation for Children with Cancer. Some foundations were created bycommunity development groups such as the Rural Social Development Corporation.

8

8/12/2019 16. Elacqua

10/37

schools and 12 percent of total enrollments. Only about 3 percent of students in Chile are

enrolled in Protestant and non-sectarian voucher schools.10

3 Empirical Strategy

In this section, I describe an empirical strategy for comparing public and private for-profit

and non-profit schools student achievement that will correct for selection bias.11 I hypothe-

size that student achievement, measured as student performance on standardized tests, can

be modeled as a function of student socioeconomic characteristics (family background, home

resources, and peer groups).12

Formally, I posit that linear models of the following form can explain student achievement:

Aij =Xijj+ ij (1)

where (Aij) is the test score of the ith student in the jth school type is a function of

independent variables that describe student and the students peer group demographics (Xij),

as well as a disturbance term (ij). In this analysis, I have one public school category, one

private non-voucher school category and 2 categories of private voucher schools (for-profit

and nonprofit).13 The sample is divided among school categories, as I estimate separate

regression coefficients for each sub-sample.

Using the estimates of, one can predict the achievement of a typical student in each

school category. The choice of this students characteristics is arbitrary, but I will use the

mean characteristics of for-profit school students (denoted as X) because I am interested in

comparing for-profit voucher and non-profit voucher school outcomes. Thus, the predicted

10The National Federation of Private Schools (FIDE) provided me with the descriptive information aboutnon-profit voucher schools. I have also conducted an interview with the National Catholic School Organiza-tion.

11Much of the debate around differences between public and private schools has revolved around statisticaltechniques that purport to control for student background characteristics and for potential selection onunobserved variables. See Vandenberghe and Robin (2004) for a critical review of different methods.

12I include peer group controls because a body of literature has documented the positive spillover effectsof having high-ability peers and the negative effects of being surrounded by disadvantaged students (Zimmerand Toma, 2000).

13For ease of exposition, I am using the two aggregate private voucher school categories to explain theempirical strategy. In the empirical analysis, I will also subdivide for-profit schools in two categories (franchiseand independent) and non-profits in three categories (Catholic, Protestant and nonsectarian).

9

8/12/2019 16. Elacqua

11/37

achievement of the average for-profit voucher school student in the jth school category is:

A= Xj (2)

To measure the difference in achievement between two school categories, I subtract one

prediction from another. The corresponding standard error can also be calculated.14 For

example, I may estimate the corrected difference between non-profit voucher schools (j= 2)

and for-profit voucher schools (j = 1). This provides an approximation of the expected

increase (or decrease) in test scores for the average for-profit voucher school student if she

were to attend a non-profit voucher school.

If the independent variables perfectly account for student and peer demographics, then

the strategy outlined above yields unbiased estimates. More likely is that some variables

are imperfectly measured or omitted from the regressions. For instance, non-profit voucher

schools may be able to select more qualified students, on average, than their for-profit voucher

school counterparts (school choice bias). Similarly, the average student attending a non-

profit voucher school may be more likely to have other attributes (such as having parents who

place a higher value on education) than the average student attending a for-profit voucher

school (parental choice bias).

For these reasons, a simple comparison of student outcomes in for-profit and non-profit

voucher schools is unlikely to give unbiased estimates of the impact of non-profit schools

on student achievement. Prior research has often applied variants of two-stage procedures

developed by Heckman (1979) in an attempt to diminish the parental choice selection bias.

15 This analysis usually consists of a single equation model in which the dependent variable

is the probability of choosing a school (e.g. for-profit or non-profit) and the independent

variables are factors that are believed to influence the choice. These methods presume that

a choice is made between only two schooling alternatives.

In the Chilean context, there are four school categories available to students.16 Lee (1983)

14For the standard errors, I used the following formula X(Vj +Vforprof it)X)1

2 where Vforprofit is theestimate variance covariance matrix from the for-profit regression andV is the comparison groups regression(Murnane et al., 1985).

15In order be able to control for school choice bias, information on school selection practices would berequired. This information is currently not available in Chile.

16In the second part of the empirical analysis I subdivide voucher schools into 5 categories, so there willbe seven school categories available to parents in that selection model.

10

8/12/2019 16. Elacqua

12/37

has developed a two-stage selection bias procedure for cases where choice is among several

alternatives.

Consider the following model:

Mij =Tijj+ uij(j = 1, 2, 3, 4) (3)

where Mij is a latent variable and Tij is a vector of variables determining school choice for

student i in school type j. Let Mbe a polychotomous variable that can take values 1 to 4

(M=j if the jth school type is chosen). A student attends the jth school type(M=j) iff:

Mj > MaxM

p (p= 1, 2, 3, 4, p=j ) (4)

Given assumptions about the error term uij , equation (3) can be estimated as a multi-

nomial logit (Maddala, 1983). Lee (1983) demonstrates how estimates from the multinomial

logit can be used to construct a selectivity term for each observation ( ij), which then be-

comes an explanatory variable in the student achievement regressions:

Aij =Xijj+ ijj+ ij (5)

wherej is an additional parameter to be estimated. The variable (ij) is equivalent to the

inverse Mills ratio in the common two-step correction proposed by Heckman (1979). It is

defined as:

ij = (1(ij))

ij(6)

Where (.) is the standard normal density, (.) is the normal distribution function, and

ij is the estimated probability (derived from the multinomial logit) that the ith student

chooses the jth school type.

In general, the independent variables that influence student achievement (Xij) in equation

(1) are quite similar to those which influence school choice (Tij) in equation (3). Demographicmeasures, among others, belong in both equations. In the subsequent empirical analysis,

however, it is necessary that one or more variables be included inTij that are excluded from

Xij, in order to identify the model. The key empirical problem in implementing a two-stage

model is in distinguishing the non-profit school effect (or the for-profit school effect) from

the effect of other variables that are not observed. A variable (or variables) is needed that

11

8/12/2019 16. Elacqua

13/37

affects the probability of attending a non-profit voucher school and that is not correlated

with the error term in the outcomes equation.

In related studies in the United States researchers have assumed that the supply of

Catholic schools is a determinant of Catholic school choice, but is not correlated with stu-

dent achievement (Neal, 1997). Following Neal (1997), I hypothesize that an individuals

probability of choosing a given school type is affected by the number of schools per square

kilometer of each type in her municipality.17 All else equal, students are more likely to

choose schooling alternatives that are more densely concentrated in their municipalities.18

It is assumed, however, that school densities are not correlated with student achievement.19

3.1 Data

The previous models are estimated with student data from Chiles national standardized

test, (Sistema de Medicion de la Calidad de la Educacion-SIMCE), which assesses students

in grades 4, 8, and 10 in language, mathematics, history and geography, and natural sciences

in odd years.20 In 2004, SIMCE evaluated 251,642 eighth graders. Students test scores are

complemented with parent and teacher questionnaires, which include socioeconomic and en-

vironmental information regarding the students, their families, their peers, and their schools.

Table 6 provides definitions of the dependent and independent variables used in the analy-

sis. The dependent variables SPANISH and MATH were originally expressed as the number

of items correct on the test, although I standardized these variables to a mean of 0 and a

17Municipalities are recognized neighborhoods in Chile around which many municipal services are orga-nized.

18Municipalities are recognized neighborhoods in Chile around which many municipal services are orga-nized. Municipalities are important in how people think about neighborhoods and how municipal servicesare organizedthat is, they have both a social reality in terms of defining neighborhoods and a political realityin terms of defining public services (e.g. Valenzuela, 1997). Over 80 percent of primary school students go toschool in their home municipality. Thus, the density measure provides a good proxy for local neighborhoodschooling options.

19

I attempt to corroborate this assumption in the subsequent analysis. See appendix 2 and 3. A reader ofa previous draft of this paper pointed out that if student achievement is low in these schools, demand maydecline, which could eventually affect supply. While there is no evidence on the supply of schools to supportor refute this concern, previous research on the demand side in Chile does suggest that parental decisionsare more influenced by student demographics than actual school quality (Elacqua et al., 2006).

20For additional information on the SIMCE test, see www.simce.cl. SIMCE employs an Item ResponseTheory Methodology.

12

8/12/2019 16. Elacqua

14/37

standard deviation of 1.21

Several independent variables characterize student demographics. These include the

students gender (FEMALE), years of parental schooling (MTHSCH and FTHSCH), self-

reported household income (INCOME), the number of non-school related books in the stu-

dents home (BOOKS1-BOOKS6, expressed as a series of dummy variables). I imputed

missing parent education information using student peer characteristics. A set of dummy

variables (MTHMISS and FTHMISS) is included to identify those observations with imputed

data.

I calculated student peer information by averaging individual student information over

all the students in a given classroom. AVMTHSCH and AVFTHSCH provide measures of

the average parental schooling, while AVINCOME is the average household income in each

classroom.

I also introduce a variable to indicate the relative isolation of the school (RURAL) and the

total number of students enrolled at the school (SCHOOLSIZE). Although not reported in

the subsequent analysis, I also included regional dummy variables - relative to the Metropoli-

tan Region - in the regressions to account for differences across regions. To approximate the

number of neighborhood schooling options a family confronts, I include a measure of the

number of schools in each category per square kilometer in each municipality (SCHOOL-

SKM2).

Table 6 Here

Table 7 provides descriptive statistics for the 251,642 students that comprise the sample,

divided by school category. The distribution across school categories in the sample is similar

to the universe of enrollments (see tables 1 and 5). According to table 7, 50 percent of

students attend public schools, 28 percent of students are enrolled in for-profit voucher

schools, 16 percent attend non-profit voucher schools, and about 7 percent of families send21There is no unambiguous definition of what constitutes small or large differences across school types. The

standard deviation is the most commonly used measure of variability in education research. In a standardnormal distribution, approximately 68 percent of the cases lie between -1 and +1 standard deviations aboveor below the mean. Two standard deviations above and below the mean contain roughly 95 percent. Thus,if a student begins with a test score that is extremely low (relative to most individuals), an increase of fourstandard deviations would allow her to outperform the majority of students. Of course, in most studieseffects are much smaller. For instance, in this study, effects are some fraction of a standard deviation.

13

8/12/2019 16. Elacqua

15/37

their children to non-voucher schools. The data presented in Table 7 also shows that most of

the for-profit school students attend independent schools that do not belong to a franchise

and most of the non-profit students are enrolled in Catholic schools.

On average, most schools are urban, though a larger proportion of public schools (20

percent) are rural compared to all categories of private schools (less than 6 percent). For-

profit franchise and Catholic schools have the largest average number of students per school.

In general, private school parents have higher levels of education, books in the household,

and income. Parents in non-voucher schools followed by Catholic schools have the highest

levels of these indicators.

Table 7 Here

4 Empirical Results

A summary of the results for private and public school coefficients is provided in Table

8.22 Table 8 presents the results when a broad set of control variables and corrections for

selection bias are made. The table is divided into two panels. The top panel summarizes

the results for Spanish, while the bottom presents the results for mathematics. The first

row displays the unadjusted difference in test scores between non-profit, public, non-voucher

and for-profit voucher schools, which is the omitted reference category. The subsequent rows

present the differences after accounting for individual and peer attributes and selection bias.

The first column displays the for-profit-non-profit school test score gap. The subsequent

columns present the for-profit-public school achievement gap and the for-profit-non-voucher

school test score gap.

The simple uncorrected estimates show that the Spanish and mathematics achievement

of students that attend for-profit schools is substantially lower, on average, than that of

non-profit school students and higher than that of public school students. The first row also

indicates a large non-voucher-for-profit test score gap.

After controlling for student and peer attributes and selection bias,23 I still find a sig-

22See appendix 1 for the multinomial logit model and the achievement regressions with the selectivitycorrection. In the interest of space, Ive only included the results for the first model. The regression outputfor the second model that I present below are available upon request.

23Point estimates of the selectivity coefficients (lambda), while not precisely estimated, are generally

14

8/12/2019 16. Elacqua

16/37

nificant and positive, but small, non-profit school Spanish (0.104 standard deviations) and

mathematics (.094 standard deviations) effect. However, the corrected test score estimates

indicate that there remains only a very small and significant difference in Spanish achieve-

ment between for-profit and public and for-profit and non-voucher schools. There is no sig-

nificant difference in mathematic achievement between for-profit and public and non-voucher

schools.

Table 8 Here

These results provide evidence of the effectiveness of non-profit voucher schools, but no

evidence on the difference in quality between public and for-profit schools. However, for-

profit schools and non-profit schools, as I discussed above, are a heterogeneous lot. The

data presented in table 5 show that 80 percent of for-profit schools are independent andmany are run by former teachers. In contrast, franchise schools, which account for about

20 percent of this sector, are often controlled by a group of off-site entrepreneurs that and

are characterized by networks of campuses. In addition, non-profit schools are composed of

Catholic, Protestant and non-sectarian schools with very different educational missions. It

is essential to separate Catholic schools from other schools because previous research has

demonstrated that Catholic schools, all else equal, usually outperform public schools and

other private schools (McEwan, 2001; Bryk et al., 1993). By doing so, we avoid confoundingthe effect of attending a non-profit school with the effect of a Catholic school.

Here I examine whether some types of for-profit and non-profit schools are more effective

than others. Table 9 summarizes the results separating for-profit and non-profit schools

by ownership type. As in the prior analysis, unadjusted estimates suggest that students in

for-profit independent schools have higher Spanish and mathematics achievement than pub-

lic school students and significantly lower achievement than non-voucher school students.

suggestive of negative selection, although it is only statistically significant for non-voucher schools. Seeappendix 2 and 3. I find similar results (available upon request) in the model that subdivides for-profit andnon-profit schools by ownership type. Even so, the results do not provide enough evidence to convincinglyreject the null hypothesis of selection bias, especially in the case of Catholic and non-voucher schools. Apossible solution would be to find a more complete set of instrumental variables, that is covariates that arecorrelated with school choice and uncorrelated with test scores, to diminish the amount of collinearity thelambda introduces to the achievement regressions. This would allow one to estimate coefficients of Catholicand non-voucher schools with greater precision. Unfortunately, instrumental variables with these propertiesare not easy to identify in Chile.

15

8/12/2019 16. Elacqua

17/37

For-profit students that attend schools that belong to a franchise score, on average, 0.1 stan-

dard deviation higher than for-profit independent students. Raw differences are even higher

between Catholic and for-profit independent students (.29 standard deviations) and non-

sectarian students (.21 standard deviations). There is no significant unadjusted difference in

Spanish or mathematic achievement between Protestant and for-profit independent schools.

After controlling for student and peer attributes and selection bias, we still find a sig-

nificant and even more substantial positive for-profit franchise school (0.22 standard devi-

ations), Catholic school (0.19 standard deviations), and non-sectarian school (.84 standard

deviations) Spanish achievement effect.24 However, the corrected test score estimates in-

dicate that there is no significant difference in Spanish achievement between Protestant,

public, non-voucher and for-profit independent schools. The results in table 9 also demon-

strate that for-profit franchise schools and Catholic schools have a considerable advantage

in mathematics (0.17 standard deviations) over for-profit independent schools, once student

and peer attributes and selection bias are controlled for. In addition, there is a substantial

non-sectarian school mathematics effect. There is no significant difference in mathematic

achievement between for-profit independent, public, and non-voucher schools. It appears

that Protestant schools produce slightly lower mathematics test scores than for-profit inde-

pendent schools after accounting for confounding independent variables.

Table 9 Here

Are the magnitudes of these Catholic, non-sectarian, and for-profit franchise effects sub-

stantial? Research in the United States has found that Catholic schools have an effect size

of around .10 standard deviations (Neal, 1997), which some have argued is not noteworthy

(Levin, 1998). I find that Catholic schools and for-profit schools that belong to a franchise

have even more substantial effect sizes, over .17 of a standard deviation.25 Finally, I find

a lack of any consistent difference between student achievement in public, Protestant and24With the exception of the non-sectarian schools, accounting for selection reduces or does not change a

school sectors advantage or disadvantage relative to for-profit independent schools. As indicated in table8, the effect of non-sectarian schools becomes very large after controlling for selection bias. It is likely thatcollinearity induced by the selectivity term, together with the fairly small sample size, led to movements inthe coefficient estimates.

25Non-sectarian non-profit voucher schools have even larger effect sizes, nearly one standard deviation.However, as I mentioned in the previous footnote, these effect sizes need to interpreted with caution.

16

8/12/2019 16. Elacqua

18/37

for-profit independent schools. In some cases it also appears that Protestant schools produce

slightly lower achievement than for-profit independent schools.

5 Conclusion

There has been a vigorous policy debate in Chile on the performance of for-profit and non-

profit schools. Some argue that for-profit schools cannot be trusted to place the interest of

students over profitability. Buried in this position is the belief that for-profits would cut

quality in the process of cutting costs (Levin, 2002). Skeptics have countered that for-profit

schools have incentives to reduce costs and to innovate, leading to both higher quality and

greater efficiency in education (Tironi, 2006; Hoxby, 2003). Neither of these arguments,

however, is based on any empirical evidence on the differential performance across school

types.

To gain insight into this debate, Ive examined the Chilean school system where vouchers

have been implemented on a large scale and where for-profit and non-profit school supply has

increased. This paper compares the academic achievement of eighth-grade students across

for-profit, non-profit, public, and non-voucher schools. I have also subdivided for-profit and

non-profit schools by ownership type: for-profit independent, for-profit franchise, Catholic,

Protestant, and non-sectarian voucher schools. What I find is a mixed story. Controllingfor individual and peer characteristics and selection bias, the results suggest that a for-profit

school student achieves slightly lower than a comparable student in a non-profit school.

There is no consistent difference in student achievement in for-profit and public schools.

However, an average student in a for-profit independent school performs significantly lower

than a similar student enrolled in a Catholic, non-sectarian, and for-profit franchise school.

The results also show that there is no important difference in achievement between public

and for-profit independent schools. There is also not a statistically significant difference inSpanish test scores between for-profit independent and Protestant schools. In mathematics,

it appears that Protestant voucher schools have slightly lower achievement than for-profit

independent schools.

The Catholic school effect is consistent with previous research in the United States (Neal,

1997). Researchers have argued that Catholic schools foster an environment in which rigorous

17

8/12/2019 16. Elacqua

19/37

academic work is pursued within a supportive and caring environment (Bryk et al., 1993).

The positive for-profit franchise effect is also consistent with previous research in Chile

(Elacqua et al., forthcoming). Some of the reasons that may explain the for-profit franchise

advantage include the benefits of scale of educational inputs (Chubb, 2001). In addition,

some researchers have argued that being embedded in a larger organization reduces and

facilitates the flow of information (such as research on best practices) between the schools

in a franchise (McMeekin, 2003).26 The evidence on low-quality Protestant schools is also

consistent with some research in the United States, which has found that these schools

dedicate more of their resources for preparing its students for the Kingdom of Heaven than

on academic activities (Peshkin, 1986).

From a policy perspective, perhaps the most interesting finding of this research is the

variation of student achievement within both the for-profit and non-profit sectors. These

findings suggest that policies oriented to eliminate for-profit schools and continue funding

non-profit (religious and non-sectarian) schools, as proposed in the General Education Act

President Bachelet signed and sent to Congress, is unlikely to improve educational outcomes.

It is highly unlikely that these mixed research findings will resolve the education policy

debate in Chile. As some scholars and practitioners have pointed out, nuanced research

findings are usually much harder to interpret and less likely to influence policy and public

understanding than straightforward ideological positions (Belfield and Levin, 2005). At a

recent conference that explored the reasons why research rarely influences policy, David

Driscoll, the Commissioner of Education of the state of Massachusetts, summarized the

terms of this debate: The exact science of ideology always trumps the inexact science of

research (Driscoll, 2007).

26It may also be the case that high achieving for-profit schools may be more likely to establish franchises (orto join a franchise) than low quality for-profit schools. In a competitive schooling environment, low qualityfor-profit schools may be unable to attract students and additional resources needed to expand operations.Data on the characteristics of for-profit school owners would improve our understanding of the complexdecisions involved in establishing a for-profit voucher school and expanding operations.

18

8/12/2019 16. Elacqua

20/37

References

Aedo, R. (2000). La educacion privada en Chile: Un estudio historico-analitico desde el

periodo colonial hasta 1990. Santiago, Chile: RIL editores.

Anand, P. A. Mizala and A. Repetto (2006). Using school scholarships to estimate the effect

of government subsidized private education on academic achievement in Chile. National

Center for the Study of Privatization in Education Occasional Paper 120. Teachers College.

Columbia University.

Belfield C. and H.M. Levin (2005). Privatizing Educational Choice: Consequences for Par-

ents, Schools, and Public Policy. Boulder and London: Paradigm Publishers

Byrk, A. S., Lee, V. E., and P. B. Holly. (1993). Catholic schools and the common good.

Cambridge, MA: Harvard University Press.

Brown, H., J. Henig, T. Holyoke, and N. Lacireno-Paquet. (2005). The Influence of Founder

Type on Charter School Structures and Operations. American Journal of Education, vol-

ume 111, pp. 487-588.

Chubb, J. E. (2001). The profit motive. The private can be public. Spring. Education Next.

Contreras, D. and Elacqua, G. (2005). El desafio de la calidad y equidad en la educacion

chilena, En foco-Expansiva, Santiago, Chile.

Cox, C. (2003).Politicas educacionales en el cambio del siglo: La reforma del sistema escolar

chileno. Santiago, Chile: Editorial Universitaria

Driscoll, D. (2007). The politics of knowledge: why research does (or does not) influence

education policy. American Enterprise Institute, Washington D.C. May Available at:http://www.aei.org/events/filter.all,eventID.1455/summary.asp

Elacqua, G. D. Contreras and F. Salazar (forthcoming). The effectiveness of private school

franchises in Chiles national voucher program. Education Next.

19

8/12/2019 16. Elacqua

21/37

Elacqua, G., M. Schneider, and J. Buckley (2006). School choice in Chile: Is it class or the

classroom? Journal of Policy Analysis and ManagementVol. 25, Issue 3.

El Mercurio (2006a, May 30). Negociacion bajo presion. Editorial. El Mercurio.

El Mercurio (2006b, June 1). Libertad de ensenanza en riesgo. Editorial. El Mercurio.

Ford J. and D. Kasserman. (2000). Ownership structure and the quality of medical care:

Evidence from the dialysis industry. Journal of Economic Behavior and Organization. 43,

279-93.

Fundacin Futuro (2005). Educacion: la opinion de los padres. Santiago, Chile: Fundacion

Futuro.

Glaeser, E. L. and Shleifer, A., (2001). Not-for-profit entrepreneurs, Journal of Public Eco-

nomics, Elsevier, vol. 81(1), pages 99-115, July.

Hart, O., Shleifer, A. and Vishny, R. W. (1997). The proper scope of government: Theory

and an application to prisons, Quarterly Journal of Economics, vol. 112, pp. 1126-61.

Hausmann, H. (1987). Economic theories of nonprofit organization. InThe nonprofit sector:

A research handbook, ed. W. Powell, 27-42. Yale University Press: New Haven.

Heckman, J. (1979). Sample selection bias as a specification error, Econometrica, 47, pp.

153-161

Hoxby, C. (2003). School choice and school productivity (or could school choice be a tide that

lifts all boats?) In C. Hoxby (Ed.) The Economics of school choice. Chicago: University

of Chicago and NBER press

James, E. (1993). Why do different countries choose a different public-private mix of educa-

tional services? The Journal of Human Resources. 28(3), 571-592.

Lee, L.-F. (1983). Generalized econometric models with selectivity, Econometrica, 51, pp.

507-512

20

8/12/2019 16. Elacqua

22/37

Levin, H.M. (1998). Educational vouchers: effectiveness, choice, and costs. Journal of Policy

Analysis and Management, 17, pp.373-391.

Levin, H. M. (2002). Potential of for-profit schools for educational reform. National Cen-

ter for the Study of Privatization in Education. Teachers College, Columbia University.

Occasional paper. No. 47.

Lukesetich, W., M. Edwards, and T. Caroll. (2000). Organizational form and nursing home

behavior. Nonprofit and Voluntary Sector Quarterly, 29, 255-279.

Maddala, G.S. (1983). Limited and dependent qualitative variables in econometrics. Cam-

bridge: Cambridge University Press.

McEwan, P. (2001). The effectiveness of Public, Catholic, and Non-Religious Private Schools

in Chiles Voucher System, Education Economics, Vol. 9, No2, pp. 103-128.

McMeekin, R. (2003). Networks of schools. Education Policy Analysis Archives, 11(16).

Ministry of Education (2007). Presidenta de La Republica firma proyecto que reemplaza la

LOCE. Santiago, Chile: Ministerio de Educacin

Mizala, A. and P. Romaguera (2005). Calidad de la educacion chilena: una asignatura pen-diente. Unpublished manuscript. CIEPLAN

Mizala, A., and P. Romaguera (2000). School performance and choice: The Chilean experi-

ence. Winter, Journal of Human Resources.

Montt, P., G. Elacqua, G., P. Gonzalez, P. Pacheco, and D. Razynski (2006). Hacia un

sistema decentralizado solido y fuerte. Santiago, Chile: Ministerio de Educacin. Serie

Bicentenario.

Morris, J. and S. Helburn. (2000). Child care center quality differences: The role of profit

status, client preferences, and trust. Nonprofit and Voluntary Sector Quarterly, vol.29,

no.3.

21

8/12/2019 16. Elacqua

23/37

Murnane, R.J. Newstead, S. and Olsen, R.J. (1985). Comparing public and private schools:

the puzzling note of selection bias.Journal of Business and Economic Statistics, 3, pp.23-

35.

Neal, D. (1997). The effects of Catholic secondary schooling on educational achievement.

Journal of Labor Economics, 15, pp. 98-123.

OECD (2004). Revision de politicas nacionales de educacion: Chile.Paris: OECD.

OECD (2006). Education at a glance. Paris: Organization for Economic Co-operation and

Development.

OEI (2007, April 24). Chile: Lucro y calidad en la educacin.Madrid, Spain: OEI.

Peshkin, A. (1986). Gods choice: The total world of a fundamentalist Christian school.

Chicago: University of Chicago Press.

Rohter, L. (2006). Chilean promised a New Deal: Now striking youth demand it. New York

Times. June 5

Rose-Ackerman, S. (1996). Altruism, nonprofits, and economic theory. Journal of Economic

Literature, 34(2), 701-728.

Sapelli, C., and B. Vial (2002). The performance of private and public schools in the Chilean

voucher system. Cuadernos de Economa, 118, 423-54.

Schlesinger, M., Bentkover, J., and D. Blumenthal. (1987). The privatization of health care

and physicians perceptions of access to hospital services. Millbank Quarterly, 65, 25-58.

Tironi, E. (2006, July 3). Sobre lucro en la educacion. Opinion, El Mercurio.

Valenzuela, J. P. (1997). Descentralizacion Fiscal: Los ingresos municipales y regionales en

Chile. Politica Fiscal Series, ECLAC.

Vandenberghe and Robin (2004). Evaluating the effectiveness of private education across

countries: a comparison of methods. Labour economics11 pp. 487-506.

22

8/12/2019 16. Elacqua

24/37

Weisbrod, B. (1988). The nonprofit economy. Cambridge, MA: Harvard University Press.

Weisbrod, B. (1998). Institutional form and organizational behavior. In W. Powell and E.S.

Clemens (Eds.),Private action and the public good. New Haven, CT: Yale University Press.

Zimmer, R. W., and Toma, E. F. (2000). Peer effects in private and public schools across

countries.Journal of Policy Analysis and Management, 19(1), 75-92.

23

8/12/2019 16. Elacqua

25/37

Figure 1. Enrollment share in public and private schools, 19792004.

59.7%

52.6%

15.1%

43.1%39.9%

6.9% 6.7% 7.6%

78%

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

1981 1990 2004

Percentageoftotalprimaryand

secondaryenrollments

Public

Private voucher

Private non-voucher

24

8/12/2019 16. Elacqua

26/37

Table 1: Distribution of primary and secondary school students across school types, 2004Percent of schools Percent of enrollment

Public 62 52.6

For-profit voucher 23 24.9

Non-profit voucher 9.1 15

Private non-voucher 6 7.6

Total 100 100

Number of schools or students 9,427 3,215,405

25

8/12/2019 16. Elacqua

27/37

Table 2 The growth of private schooling in Chile, 1990-2004

Schooltype

School numbers Student enrollments

1990 2004Percentchange

1990 2004Percentchange

Public 6,072 5,843 -3.8% 1,642,414 1,689,720 2.9%

For-profit 1,592 2,167 36.1% 494,843 799,806 61.6%

Non-profit 700 856 22.3% 343,755 482,560 40.4%

Non-voucher

521 561 7.7% 198,602 243,319 22.5%

Total 8,885 9,427 6.1% 2,679,614 3,215,405 20%

26

8/12/2019 16. Elacqua

28/37

Table 3: Private schools by ownership type: Descriptive summary

School type NRural(%)

Averageschool

size

ChargeTuition

(%)

Mothers'years of

education

Vulnerablestudents

(%)**

Public 5,843 22% 290 18%* 8.1 39%

For-profit 2,167 6% 369 53% 10.1 18%

Non-profit 856 5% 564 50% 10.8 14%

Non-voucher 561 2% 434 100% 15 0%

*Only public high schools can charge fees. Elementary schools are required by law to be free.

**Only elementary school students (k-8)

27

8/12/2019 16. Elacqua

29/37

Table 4: A taxonomy of p ublic and pri vate schools in Chile

School Type Management Financing

*Contributions include services of personnel and monetary and in-kind donations.

Public Schools Departamento de Administracin dela Educacin Municipal; part ofmunicipal bureaucracy

National vouchers; municipalcontributors; Regional DevelopmentFund (infrastructure)

Municipal Corporation; quasi-autonomous from municipalbureaucracy

For-profitindependentvoucher

Individual entrepreneurs. National vouchers; tuition payments;parent-center fees

For-profit franchisevoucher

For-profit partnerships and firms(with private shareholders) thatbelong to a chain of schools and areoperated by the same legal schoolowner

National vouchers; tuition payments;parent-center fees

Catholic voucher Branches of Catholic church,including religious orders, parishes,archdiocese and religiousfoundations.

National vouchers; tuition payments;parent-center fees; churchcontributions*

Protestant voucher Protestant churches, includingMethodist, Baptist, Seventh-DayAdventist, Anglican, Lutheran, andPresbyterian churches.

National vouchers; tuition payments;parent-center fees; churchcontributions*

Nonsectarianvoucher

Foundations, universities,Community development groups(NGOs)

National vouchers; tuition payments;parent-center fees; foundationcontributions*

Private non-voucher

Catholic church; Protestantchurches; for-profit individuals andfirms; non-profit foundations.

Tuition payments; parent-centerfees; church and foundationcontributions*

28

8/12/2019 16. Elacqua

30/37

Table 5: Distribution of primary school students across 7 school categories, 2004Percent of schools Percent of enrollment

Public 62 52.6

For-profit franchise 4.2 6.7

For-profit independent 18.7 18.1

Non-profit Catholic 6.9 12

Non-profit Protestant 1.0 1.4

Non-profit nonsectarian 1.1 1.6

Private non-voucher 6 7.6

Total 100 100

Number of schools or students 9,424 3,215,006

29

8/12/2019 16. Elacqua

31/37

Table 6 Variable definitions

Variable Description

SPANISH Student score on the eighth-grade Spanish test (standardized to a mean of 0

and a standard deviation of 1)

MATH Student score on the eighth-grade mathematics test (standardized to a mean

of 0 and a standard deviation of 1)

FEMALE Dummy variable indicating whether student is female

MTHSCH Years of schooling of students mother

MTHMISS Dummy variable indicating whether MTHSCH is missing

FTHSCH Years of schooling of students father

FTHMISS Dummy variable indicating whether FTHSCH is missing

INCOME Monthly family income, divided by 100,000

BOOKS1-BOOKS6 Six dummy variables indicating the number of books in the family home,

ranging from 1 (5 or less) to 6 (more than 200). BOOKS2 is omitted in

regressions.

AVMTHSCH Average schooling of student mothers in classroom

AVFTHSCH Average schooling of student fathers in classroom

AVINCOME Average monthly household income of students in classroom

SCHOOLSIZE Total enrollment in school

SECPROPORTION Proportion of secondary students over total enrollment

RURAL Dummy variable indicating whether school is rural

30

8/12/2019 16. Elacqua

32/37

Table 7 Sample descriptive statistics

Note: Standard deviations are in brackets.

Sample Public

For-profit

independent

For-profit

franchise Catholic Protestant Nonsectarian Nonvoucher

SPANISH 0.000 -0.234 0.037 0.137 0.324 0.041 0.242 0.836

[1.000] [.964] [.98] [.983] [.924] [.982] [.985] [.796]

MATH 0.000 -233 0.034 0.15 0.28 0.028 0.275 0.896

[1.000] [.964] [.979] [.977] [.929] [.959] [.995] [.772]

Public 0.501

For-profitindependent

0.212

For-profit franchise 0.064

Catholic 0.126

Protestant 0.016

Nonsectarian 0.015

Nonvoucher 0.065

FEMALE 0.49 0.489 0.47 0.473 0.55 0.472 0.472 0.476

MTHSCH 10.756 9.44 11.34 11.064 12.117 11.365 11.4877 15.42

[3.424] [3.12] [3.1] [2.974] [2.913] [3.037] [3.11] [2.002]

MTHMISS 0.106 0.1033 0.112 0.104 0.088 0.112 0.099 0.132

FTHSCH 10.824 9.466 11.433 11.113 12.039 11.5 11.424 16

[3.548] [3.189] [3.172] [3.041] [3.036] [3.067] [3.357] [2.216]

FTHMISS 0.144 0.15 0.145 0.136 0.118 0.15 0.131 0.145

INCOME 3.42 2.2 3.546 3.147 3.91 3.288 3.854 11.597

3.259 1.632 2.83 2.337 3.005 2.372 3.242 3.418

BOOKS1 0.103 0.152 0.071 0.072 0.049 0.068 0.069 0.003

BOOKS2 0.443 0.523 0.411 0.419 0.375 0.451 0.386 0.097

BOOKS3 0.3 0.243 0.351 0.353 0.38 0.346 0.357 0.326

BOOKS4 0.091 0.054 0.106 0.101 0.121 0.085 0.112 0.239

BOOKS5 0.032 0.015 0.034 0.031 0.041 0.03 0.043 0.142

BOOKS6 0.031 0.013 0.027 0.024 0.033 0.021 0.034 0.193

RURAL 0.123 0.196 0.06 0.06 0.05 0.091 0.038 0.02

AVMTHSCH (peer) 10.765 9.437 11.349 11.05 12.15 11.385 11.482 15.623

[2.363] [1.59] [2.013] [1.775] [1.684] [1.775] [2.095] [.837]

AVFTHSCH (peer) 10.842 9.473 11.452 11.101 12.073 11.54 11.405 16.252

[2.496] [1.692] [2.05] [1.79] [1.792] [1.727] [2.284] [.943]

AVINCOME (peer) 3.411 2.191 3.533 3.142 3.9 3.281 3.833 11.6

[2.619] [.799[ [1.801] [1.285] [1.799] [1.17] [2.362] [1.573]

SCHOOLSIZE 763.331 673.369 696.26 1071.306 1053.257 688.28 866.29 828.341

[667.358] [436.452] [662.267] [1020.878] [1056.541] [446.32] [575.81] [515.759]

SECPROPORTION 0.111 0.036 0.153 0.133 0.202 0.141 0.184 0.321

N(students 251,642 126,092 53,365 16,166 31,783 3,931 3,738 16,567

31

8/12/2019 16. Elacqua

33/37

Table 8 Differences between non-profit voucher, public, and private non-voucher schools, with average characteristicsof for-profit voucher school students

Non-profitvoucher

Public PrivateNon-voucher

SPANISH

Unadjusted Difference 0.228 -0.294 0.776

[0.006] [0.005] [0.009]

Difference adjusted for:

Individual SES 0.138 -0.180 0.325

[0.028] [0.022] [0.050]

Individu al/peer SES 0.091 -0.056 0.076

[0.027] [0.023] [0.091]

Individual/peer SES/selectivity 0.104 -0.061 0.161

[0.037] [0.026] [0.100]

N 39,096 124,757 16,396

MATH

Unadjusted Difference 0.193 -0.294 0.835

[0.006] [0.005] [0.009]

Difference adjusted for:

Individual SES 0.119 -0.160 0.289

[0.030] [0.023] [0.053]

Individu al/peer SES 0.071 -0.034 -0.074

[0.029] [0.024] [0.097]

Individual/peer SES/selectivity 0.094 -0.050 0.036

[0.042] [0.028] [0.106]

N 39,107 124,533 16,395

Note: Standard errors in brackets.

32

8/12/2019 16. Elacqua

34/37

Table 9: Differences between for-profit franchise, Catholic, Protestant, Non-sectarian, Public, and non-voucherschools, with average characteristics of for-profit independent school students

For-profitfranchise

Catholic Protestant Non-sectarian

Public Non-voucher

SPANISH

Unadjusted Difference 0.099 0.287 0.004 0.206 -0.271 0.800

[0.009] [0.007] [0.017] [0.017] [0.005] [0.009]

Difference adjusted for :

Individual SES 0.210 0.219 0.007 0.217 -0.127 0.368

[0.041] [0.029] [0.090] [0.091] [0.022] [0.049

Individual/peer SES 0.200 0.159 0.013 0.282 0.006 0.122

[0.036] [0.028] [0.076] [0.072] [0.023] [0.089]

Individual/peer SES/selectivity 0.221 0.185 0.017 0.836 -0.004 0.194

[0.060] [0.036] [0.201] [0.156] [0.028] [0.094]

N 15,994 31,537 3,873 3,686 124,757 16,396

MATH

Unadjusted Difference 0.115 0.245 -0.006 0.242 -0.267 0.862

[0.009] [0.007] [0.017] [0.017] [0.005] [0.009]

Difference adjusted f or:

Individual SES 0.197 0.184 -0.066 0.280 -0.109 0.332

[0.044] [0.032] [0.087] [0.088] [0.023] [0.051]

Individual/peer SES 0.191 0.122 -0.061 0.343 0.026 -0.024

[0.039] [0.032] [0.081] [0.069] [0.025] [0.094]

Individual/peer SES/selectivity 0.173 0.164 -0.086 0.986 0.010 0.069

[0.062] [0.041] [0.272] [0.146] [0.030] [0.101]

N 15,992 31,548 3,866 3,693 124,533 16,395

Note: Standard errors in brackets.

33

8/12/2019 16. Elacqua

35/37

Appendix 1 A multinomial logit model of school choice

Public Non-profitNon-

voucher

FEMALE 0.06 0.26*** -0.01

[-1.59] [-5.18] [-0.07]

MTHSCH -0.01* 0.07*** 0.31***

[-2.32] [-11.13] [-18.41]

MTHMISS 0.02 -0.04 0.31

[-0.45] [-0.74] [-1.82]

FTHSCH -0.01* 0.02*** 0.25***

[-2.41] [-3.89] [-16.24]

FTHMISS 0.09** 0.01 0.25*

[-2.59] [-0.44] [-2.05]

INCOME -0.07*** 0.02* 0.45***

[-3.92] [-2.21] [-21.73]

BOOKS1 0.05 -0.11* -0.86***

[-1.65] [-2.51] [-3.80]

BOOKS2 -0.09*** 0.06** 0.19**

[-3.68] [-2.88] [-2.63]

BOOKS3 -0.12** 0.05 0.36***

[-3.11] [-1.54] [-4.44]

BOOKS4 0.02 0.08 0.31**

[-0.34] [-1.87] [-2.96]

BOOKS5 0.05 0.01 0.26

[-0.62] [-0.19] [-1.9]

RURAL 0.93*** 0.05 -1.89***

[-4.63] [-0.22] [-3.96]

SCHOOLSIZE 0.00*** 0.00* -0.00***

[-4.83] [-2.23] [-3.93]

SECPROPORTION -3.43*** 0.93* 8.99***

[-6.18] [-2.47] [-13.27]

DENSITY1 0.03 -0.04 0.02

[-1.2] [-1.87] [-0.45]

DENSITY1(2) 0 0 0

[-1.57] [-1.83] [-0.99]

DENSITY2 -0.02 0 0.01

[-1.10] [-0.25] [-0.24]

DENSITY2(2) 0 0 0

[-0.7] [-0.94] [-0.06]

DENSITY3 0 0.02 0.01

[-0.03] [-0.94] [-0.13]

DENSITY3(2) 0 0 0

[-1] [-0.69] [-0.08]

DENSITY4 -0.03 -0.02 0.01

[-1.17] [-0.88] [-0.3]

DENSITY4(2) 0 0 0

[-0.21] [-1.19] [-0.56]

CONSTANT 1.16*** -2.27*** -11.17***

[-3.55] [-8.17] [-16.62]

Pseudo R-squared 0.54 Number of observations : 205172Robust standard errors in brackets*significant at 10%; ** significant at 5%; significant at 1%

34

8/12/2019 16. Elacqua

36/37

Appendix 2 Spanish achievement regressions with school densityPublic For-profit Non-

profitNon-

voucher

FEMALE 0.1288*** 0.1161*** 0.1087*** 0.1509***

[0.0064] [0.0094] [0.0128] [0.0158]

MTHSCH 0.0342*** 0.0303*** 0.0319*** 0.0324***

[0.0012] [0.0017] [0.0024] [0.0041]

MTHMISS -0.0859*** -0.0770*** -0.0233 -0.1066**

[0.0174] [0.0221] [0.0269] [0.0462]

FTHSCH 0.0288*** 0.0248*** 0.0229*** 0.0168***

[0.0011] [0.0017] [0.0022] [0.0036]

FTHMISS -0.0460*** -0.0801*** -0.0728*** -0.0233

[0.0114] [0.0164] [0.0203] [0.0334]

INCOME 0.0128*** 0.0061*** 0.0042* -0.0065*

[0.0029] [0.0018] [0.0022] [0.0033]

BOOK1 -0.0768*** -0.0523*** -0.0625*** 0.014

[0.0088] [0.0173] [0.0222] [0.1371]

BOOK2 0.1224*** 0.0986*** 0.1130*** 0.0838***

[0.0076] [0.0096] [0.0122] [0.0230]

BOOK3 0.1733*** 0.1667*** 0.1695*** 0.1589***

[0.0137] [0.0129] [0.0166] [0.0255]

BOOK4 0.2184*** 0.2333*** 0.2232*** 0.1716***

[0.0249] [0.0211] [0.0248] [0.0271]

BOOK5 0.2439*** 0.2523*** 0.1787*** 0.1992***

[0.0265] [0.0231] [0.0274] [0.0269]

RURAL 0.1924*** 0.0944* -0.0083 0.094

[0.0212] [0.0508] [0.0464] [0.0655]

SCHOOLSIZE 0 0.0001*** -0.0000* 0.0002***

[0.0000] [0.0000] [0.0000] [0.0000]

SECPROPORTION 0.0503 -0.1076* 0.0757 -0.0291

[0.0780] [0.0647] [0.0746] [0.1341]

AVMTHSCH 0.0682*** 0.0711*** 0.0774*** 0.0152

[0.0076] [0.0115] [0.0174] [0.0234]

AVFTHSCHO 0.0136* 0.0442*** 0.0320* 0.0403*

[0.0075] [0.0120] [0.0172] [0.0221]

AVINCOME 0.0334*** -0.0017 0.0045 0.003

[0.0128] [0.0127] [0.0127] [0.0100]

LAMBDA -0.0477 -0.026 -0.0077 -0.0796**

[0.0442] [0.0313] [0.0275] [0.0365]

CONSTANT -1.8667*** -1.9906*** -1.7850*** -1.1152***

[0.0554] [0.0903] [0.1506] [0.3104]

Number of Observations 100376 56690 33350 13697

R-squared 0.1133 0.1442 0.1399 0.0725

Robust standard errors in brackets*significant at 10%; ** significant at 5%; significant at 1%Regional dummies were also included in the regressions.

35

8/12/2019 16. Elacqua

37/37

Appendix 3 Mathematics achievement regressions with school densityPublic For-profit Non-profit Non-

voucher

FEMALE -0.0628*** -0.0839*** -0.0905*** -0.0267*

[0.0067] [0.0101] [0.0136] [0.0160]

MTHSCH 0.0350*** 0.0269*** 0.0297*** 0.0312***

[0.0012] [0.0017] [0.0023] [0.0043]

MTHMISS -0.1129*** -0.0927*** -0.044 -0.1497***

[0.0169] [0.0217] [0.0282] [0.0454]

FTHSCH 0.0270*** 0.0248*** 0.0199*** 0.0212***

[0.0011] [0.0017] [0.0023] [0.0037]

FTHMISS -0.0684*** -0.0609*** -0.1098*** -0.049

[0.0113] [0.0156] [0.0216] [0.0368]

INCOME 0.0134*** 0.0063*** 0.0082*** -0.004

[0.0029] [0.0018] [0.0021] [0.0030]

BOOK1 -0.0757*** -0.0636*** -0.0492** 0.080

[0.0085] [0.0173] [0.0240] [0.1474]

BOOK2 0.1202*** 0.1073*** 0.0972*** 0.0789***

[0.0077] [0.0093] [0.0114] [0.0220]

BOOK3 0.1705*** 0.1513*** 0.1488*** 0.1475***

[0.0136] [0.0131] [0.0165] [0.0242]

BOOK4 0.2009*** 0.2147*** 0.2180*** 0.1550***

[0.0240] [0.0215] [0.0256] [0.0257]

BOOK5 0.1843*** 0.2234*** 0.1655*** 0.1741***

[0.0284] [0.0233] [0.0276] [0.0257]

RURAL 0.1666*** 0.0900* -0.043 0.037

[0.0228] [0.0483] [0.0458] [0.1131]

SCHOOLSIZE 0.000 0.0001*** 0.000 0.0002***

[0.0000] [0.0000] [0.0000] [0.0000]

SECPROPORTION 0.057 -0.1272* 0.139 -0.108

[0.0835] [0.0696] [0.0869] [0.1301]

AVMTHSCH 0.0681*** 0.0620*** 0.0794*** 0.010

[0.0082] [0.0128] [0.0196] [0.0242]

AVFTHSCHO 0.0173** 0.0497*** 0.019 0.0764***

[0.0083] [0.0126] [0.0192] [0.0235]

AVINCOME 0.0310** 0.012 0.016 0.001

[0.0139] [0.0135] [0.0138] [0.0116]

LAMBDA -0.075 -0.006 -0.023 -0.1055***

[0.0458] [0.0374] [0.0326] [0.0356]

CONSTANT -1.7689*** -1.8887*** -1.5496*** -1.4845***

[0.0577] [0.0970] [0.1643] [0.3507]

Number of Observations 100580.000 56881.000 33428.000 13728.000

R-squared 0.108 0.150 0.139 0.086

Robust standard errors in brackets*significant at 10%; ** significant at 5%; significant at 1%Regional dummies were also included in the regressions.