Cross-National Differences in Educational Achievement Inequality

Cross-national Differences in Educational Achievement Inequality
Author(s): Guillermo Montt

Source: Sociology of Education , JANUARY 2011, Vol. 84, No. 1 (JANUARY 2011), pp. 49-
68
Published by: American Sociological Association
Stable URL: https://www.jstor.org/stable/23057035
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
American Sociological Association is collaborating with JSTOR to digitize, preserve and extend
access to Sociology of Education
This content downloaded from

168.176.5.118 on Fri, 07 May 2021 22:29:48 UTC
All use subject to https://about.jstor.org/terms
A. A
AMERICAN SOCIOLOGICAL ASSOCIATION
Sociology of Education
84(1) 49-68
Cross-national Differences in © American Sociological Association 201 I
DOI: 10.1177/0038040710392717
Educational Achievement http://soe.sagepub.com
USAGE
Inequality
Guillermo Montt1
Abstract
School systems are called not only to instruct and socialize students but also to differentiate among them
Although much research has investigated inequalities in educational outcomes associated with student
family background and other ascriptive traits, little research has examined cross-national differences in
the total amount of differentiation that school systems produce, the total achievement inequality. Th
article evaluates whether two dimensions of educational systems—variations in opportunities to learn
and intensity of schooling—are associated with achievement inequality independent of family backgroun
It draws data from the Programme for International Student Assessment for more than 50 school system
and models the variance in achievement. Findings suggest that decreasing the variability in opportunities t
learn—in the form of greater homogeneity in teacher quality and the absence of tracking—within th
school system might reduce achievement inequality. More intense schooling is also related to lower
achievement inequality to the extent that this intensity is homogeneously distributed within the school
system, particularly in the form of a more highly qualified teacher workforce.
Keywords
comparative education, educational achievement, achievement inequality, PISA, variance regression
Under a meritocratic ideal, scholastic achievement In this article, I examine country-to-country dif
is a function of talent, ambition, and effort in ferences in total achievement inequality, and
school. Schools are called to provide equal oppor more specifically, I assess the extent to which
tunities to learn and identify differences among two important conceptual dimensions of school
their students so that students are properly allo ing, opportunities to learn and the intensity of
cated into the labor market; schools are gatekeep schooling, help account for cross-national varia
ers in a sorting process. Following this framework, tion in achievement inequality.
sociologists who study educational inequality An educational system is expected, among
have been concerned with whether scholastic other things, to differentiate and sort its students
achievement and attainment are independent by
of identifying or creating inequalities among
them. Because educational experience and edu
ascriptive background factors (e.g., race/ethnicity,
socioeconomic background) and hence whether cational achievement in early, primary, and
schools reproduce or reduce social inequality.
Yet inequality due to family background or other
'University of Notre Dame, IN, USA
ascriptive traits represents only a portion of the
total inequality in educational outcomes. The Corresponding Author:
overall variation in achievement, or total achieve
Guillermo Montt, University of Notre Dame, 810
Flanner Hall, Notre Dame, IN 46556
ment inequality, provides an alternative metric for
assessing equality within an educational system.Email: gmonttar@nd.edu

168.176.5.118 on Fri, 07 May 2021 22:29:48 UTC
50 Sociology of Education 84( I)
secondary schooling are important

To identify determin
the contribution of schools to edu
of higher education, cational inequality, market,
labor it is critical to first account
and for adul
outcomes (Featherman, Hauser, and Sewell the substantial amount of inequality produced by
1976; Kerckhoff 1993; Meyers et al. 2004; Pallas family and home environment.
2000), the total achievement inequality in second
ary school translates into inequalities in access to
higher education. These may, subsequently,
Family Background and Home
become disparities in income, wealth, and occupa
Environment
tional status. Also, total achievement inequality
may have other, more immediate outcomes; at the At the individual level, socioeconomic back
national level, it is an important predictor of school ground is the most robust and consistent predictor
violence (Baker and LeTendre 2005). of student achievement both in the United States
Although inequalities in opportunity, access, and abroad (Baker, Goesling, and LeTendre
and quality of education have been historically 2002; Buchmann and Hannum 2001; Heyneman
reduced (Baker and LeTendre 2005), leading to and Loxley 1983). As a result, students from
an increased equality of human capital production high-income families score higher on standardized
both within and between countries (Castello and tests and are more likely to go to college than stu
Domenech 2002), inequality in achievement is dents from lower income families. Socioeconomic
inherent to all educational systems and is unlikely background is also important in explaining
to be eliminated. The relevant question regarding achievement inequality among schools. In the
inequality in educational outcomes is, then, not United States and abroad, for example, student
about the existence of inequality but the amount body composition is one of the strongest predic
of inequality that is produced by school systems. tors of school-to-school differences in achieve
This study examines total inequality in student ment (Coleman 1990; OECD 2007a; Scheerens
learning in an international sample of schools and and Bosker 1997).
students. It explores the characteristics of educa Given the robust relationship between socioeco
tional systems that exacerbate or reduce inequality nomic background and achievement, we should
in achievement within nations. In addition to the expect that two school systems with identical insti
influence of family background on achievement, tutional characteristics at the school and country
I hypothesize that the remaining inequality is level will have different amounts of inequality if
related to two primary dimensions of a nation's one school system has a more uneven distribution
school system: variation in opportunities to learn of student background than the other. There is con
and the overall intensity of schooling. In particu siderable variation in the amount of socioeconomic
lar, does variation in opportunities to learn and diversity of student populations across countries.
in the overall intensity of schooling correspond Spain, Greece, and the United States, for example,
to the total amount of inequality in a school sys have a more diverse student population than Japan,
tem independent of the distribution and effects Norway, and Australia (Arnett 2007).
of family background? It is important to consider that certain social in
Using achievement data from the Organization stitutions may constrain socioeconomic diversity
for Economic Co-Operation and Development's through the reduction of inequalities in living con
(OECD's) Programme for International Student ditions, thereby reducing inequalities in opportu
Assessment (PISA) for more than 50 countries, I nity. These social institutions have been at the
model the variance in achievement as a measure center of the reduction in inequality of opportunity
of the dispersion in student learning within coun in the Netherlands, Sweden, and other countries
tries as a function of indicators of family back (Breen and Jonsson 2005; Shavit and Blossfeld
ground, opportunities to learn, and intensity of 1993) and include social welfare programs and
schooling. I find that both variations in opportuni welfare states. For example, social insurance and
ties to learn, particularly in the form of equality in risk management institutions are crucial for redis
the distribution of teachers across schools and the tributing outcomes (Korpi and Palme 1998), reduc
absence of tracking, and greater intensity of ing poverty (Brady 2005), and reducing the
schooling, in the form of better teachers across variability in family background, all of which af
the board, are related to a reduction in total fects students' opportunities to learn (Valenzuela,
achievement inequality. Tironi Barrios, and Scully 2006).

168.176.5.118 on Fri, 07 May 2021 22:29:48 UTC
Montt
Although informative, studies of educational achievement and thus in generating inequality.

inequality that focus on the relationship between Student learning is sensitive to teacher attributes
family background and achievement generally such as teacher training and experience
miss the total amount of inequality that a school sys (Greenwald, Hedges, and Laine 1996;
tem produces. It is possible—at least theoretically— Konstantopoulos 2006; Nye, Konstantopoulos,
that a school system that reduces the relationship and Hedges 2004; Rivkin, Hanushek, and Kain
between academic achievement and socioeco 2005), so homogeneity in teacher attributes across
nomic background could still produce high schools
levels should reduce inequality. Differences in
of inequality in educational outcomes. Such is
material resources available to students, by con
true of Taiwan, whose tracking system tends tohave had less success in explaining differen
trast,
reduce the effect of socioeconomic status on ces in student achievement (Buchmann and
achievement while still producing considerable
Hannum 2001; Gamoran, Secada, and Marrett
levels of dispersion in achievement among differ 2000; Greenwald et al. 1996; Heyneman and
ently tracked students (Broaded 1997). Loxley 1983). This is particularly true in industri
alized societies. In less developed countries,
For this reason, studies of inequality in student
learning ought to go beyond examining a few wherespe some schools may lack basic material re
cific ascriptive factors to consider the school and
sources that are crucial for instruction (e.g., text
instructional practices that are associated withbooks), school-to-school differences in resources
may affect achievement growth (Baker et al.
total achievement inequality. Only a limited num
ber of studies analyze the total inequality
2002;
in Fuller 1987). Smaller class sizes lead to
achievement, and these generally focus on aanar
modest increase in student performance (Ma
and Klinger 2000; Nye, Hedges, and
row set of characteristics of school systems,
Konstantopoulos 2002) by allowing a closer rela
most notably differentiation and standardization
tionship between the teacher and both students
(see Van de Werfhorst and Mijs 2010 for a com
and parents, increasing student engagement,
parative review). The distribution of opportunities
reducing
to learn in schooling systems and the intensity of the amount of time dedicated to disci
schooling provide a broader conceptual frame
plining students, and enabling teachers to use
work for understanding the amount of achieve new pedagogical practices that are possible or
ment inequality that school systems produce. even work better in smaller classes (Ehrenberg
et al. 2001). Because teacher attributes, and to
a lesser extent school resources and class size,
are related to student achievement, greater vari
Variability/Standardization of
ability in these resources within a school system
Opportunities to Learn will be related to increased levels of total achieve
Sorensen and Hallinan (1977) introduce a three ment inequality.
part model of student learning, wherein achieve Data from PISA 2006 suggest that educational
systems have differing amounts of school-to
ment is a function of ability, effort, and opportuni
school variation in teacher credentials, school re
ties to learn. According to this model, differences
in opportunities to learn induce variation insources,
stu and class size. The interquartile range sta
dent achievement growth even among students tistic, which measures the difference between the
scores of cases falling at the top and bottom
who exhibit similar effort and ability. Elements
25th percentile, provides a simple metric for
of opportunities to learn include teacher quality,
describing the variation in measures of opportuni
school resource quality, curriculum organization,
ties to learn within different countries. In
and class size. Greater standardization in opportu
Colombia and Mexico, for example, schools
nities to learn in the school system should provide
more homogeneous school experiences forvary stu widely in terms of class size; The interquar
tile range in class size is 25 students. The compa
dents and reduce the total inequality in achieve
ment within a school system. rable range for the Netherlands and Finland is
only 5 students. In terms of teacher credentials,
A more equal distribution of schools' human
and material resources reduces variability in
schools in Portugal, Switzerland, and Austria are
opportunities to learn, yet some elements heterogeneous
of (the interquartile range in the pro
opportunities to learn have been shown toportion
be of teachers with International Standard
more important than others in predicting Classification of Education [ISCED] 5a diplomas

168.176.5.118 on Fri, 07 May 2021 22:29:48 UTC
students across
in the school is greater thansectors (Carbonaro
0.75), and Covay whereas
in Australia, Japan,2010)
andand total achievement
Canada inequality.are
There is, homoge
(the interquartile range is
however, the possibility that close to
a larger private sector 0). Fin
may foster
PISA's index of school equality by allowing schools
resources hasto be an inte
more flexible.
tile range greater than 1.5 Control over hiringindex
(the decisions, cur ranges
to 2.1 at the school ricular offerings, and in
level) budget allocations
Qatar, may Argen
allow private-sector
Kyrgyzstan, and Brazil, schools the flexibility to
signaling heterogen
in school resources, meet
butthe particular
anneeds of low-achieving stu
interquartile ran
only 0.75 in Norway, dents inLatvia,
their local context, potentially
andreducing Croatia, in
dispersion in achievement
ing greater homogeneity in how (Bryk, Lee, and
school reso
are distributed. Holland 1993; Chubb and Moe 1990; Coleman
The curricular organization of school systems and Hoffer 1987; Fuchs and Wo(3mann 2007).
can also produce variations in opportunities to School systems vary in the magnitude of the
learn and increased inequality in achievement private sector and their allowance of school auton
if different students are exposed to different omy. For example, all schools in Hungary,
instructional content. This can occur in school sys Iceland, Latvia, the Netherlands, New Zealand,
tems that implement tracking and in school and Sweden have the freedom to set the starting
systems that do not have a standardized curricu salaries for their teachers, whereas a quarter or
lum. In the context of increasing equality in qual less of the schools in France, Germany, Italy,
ity of instruction, these differentiations are often Luxembourg, and Portugal have this liberty
remnants of nineteenth-century educational sys (Fuchs and Wopmann 2007). Private schools are
tems that explicitly incorporated inequalities fairly common in Ireland and Belgium, for exam
within the school system (Baker and LeTendre ple, where more than 60 percent of the students
2005). Tracking can occur between schools in attend privately managed schools, but are rela
terms of school type (as in the German system), tively rare in Austria, Brazil, and Mexico, where
course of study (e.g., vocational schools in less than 15 percent of the students attend pri
Japan), tracks or streams within comprehensive vately managed schools (Vandenberghe and
secondary schools (e.g., high schools in the Robin 2004).
United States), ability grouping within classrooms In sum, independent of student ability and
(e.g., reading instruction in elementary schools ineffort, variations in opportunities to learn should
the United States), and according to geographic lead to variations in student learning and greater
location (LeTendre, Hofer, and Shimizu 2003). total achievement inequality. These features that
Countries that provide different curricula for dif vary within countries, however, fail to tell the
ferent types of students generate larger amounts entire story about achievement inequality. Two
of inequality than those that adopt a comprehen school systems with equal distributions of oppor
sive schooling system (Hanushek and WopSmann tunities to learn will have different levels of total
2005). By contrast, a school system with a highly achievement inequality if their intensity of school
standardized curriculum should produce less total ing differs.
achievement inequality by ensuring greater equal
ity in the content and coverage of material across
schools and classrooms (Stevenson and Baker Intensity of Schooling
1991). Coleman's call for increased equality of opportu
Certain institutional arrangements may gener nity is a call for intense schooling that is indepen
ate greater variability in opportunities to learn dent of a child's social environment (Coleman
across schools, which is a nontrivial element of 1990), such that greater intensity reduces inequal
the total inequality in an educational system. A ities that exist prior to children's entrance into pri
larger private sector and greater school autonomy mary schools and which are reinforced by their
will allow for greater variability in curriculum out-of-school experiences. To the extent that this
organization, resources, and teacher quality across intensity is equally distributed, achievement
schools. In this scenario, better resources and inequality will be reduced. Features of intense
higher quality teachers will tend to be concen schooling include the total amount of time dedi
trated in schools that enjoy higher budgets, cated to instruction, the overall quality of teachers,
increasing the differences in achievement between the overall class size of the schools, the overall

168.176.5.118 on Fri, 07 May 2021 22:29:48 UTC
Montt 53
resource (OECD 2009a). Per-pupil expenditures on queduca

education. tion exceeded $10,000 in Austria, Luxembourg,
An earlier entrance into the school system, Norway, Switzerland, and the United States yet
a lengthier school year, and a lengthier school were less than $3,000 in Chile, Brazil, Mexico,
day are all signs of increased intensity of school Slovakia, and Poland (OECD 2009b).
ing because students have greater exposure to Cross-national research on achievement has
structured and homogeneous learning environ practically ignored total achievement inequality
ments (Arnett 2007; Entwisle, Alexander, and as an outcome for educational systems. Except
Olson 2000; Meyers et al. 2004). Because vari for studies of tracking and total inequality (see
ability in student learning is lower when school Hanushek and Wopmann 2005; Huang 2009; or
is in session (Downey, von Hippel, and Broh Gamoran 2009; and Van de Werfhorst and Mijs
2004), systems with higher intensity in terms of 2010 for reviews), studies that take a cross
length of exposure will reduce total achievement national approach to education have generally
inequality. The overall teacher quality of the looked at the effectiveness of institutional and
schooling system is another dimension of intensity schooling practices as a source for generalizability
of schooling that is likely to affect achievement of findings at a national level (Baker and
inequality. Better qualified teachers are more LeTendre 2005; Ramirez 2006). This article con
able to adapt curricular material, subject knowl tributes to research on educational inequality in
edge, and pedagogical techniques to the needs of two ways. First, by focusing on total achievement
their students, thereby providing an enhanced inequality, it expands the study of educational
schooling experience for all students and affecting inequality by going beyond the few discrete
student achievement (Gamoran 1993; Rowan, ascriptive sources of inequality typically exam
Correnti, and Miller 2002). The presence of better ined. Second, it broadens the sources of achieve
teachers is especially beneficial to students from ment inequality by taking into account factors
disadvantaged backgrounds (Nye et al. 2004), related to opportunities to learn and the intensity
reducing total achievement inequality. Smaller of schooling that occur at the school and country
overall class sizes allow for a closer relationship levels.
between teachers and both students and parents,
increase student engagement, and reduce the DATA AND METHOD
amount of time dedicated to disciplining students
(Ehrenberg et al. 2001), also reducing total Data for this study are drawn from OECD's PISA.
achievement inequality. In 2006, PISA surveyed more than 400,000 15
The overall level of school resource quality year-old students in more than 50 countries, both
and public expenditure on education should also OECD members and partner economies, focusing
affect achievement inequality. Better resources on the practical applications of knowledge in the
in schools may reduce the likelihood that schools domains of reading, mathematics, and science.
would lack crucial instructional resources. This approach to measuring achievement is less
sensitive to the curricular organization of subjects
Resources, however, may not be spent in a targeted
in each country and has a stronger link to the abil
fashion to reduce educational inequality (Condron
ities required later in life than achievement tests
and Roscigno 2003) and may in fact be unequally
that have a curricular base. PISA also asks stu
distributed to the benefit of advantaged students
(Moser and Rubenstein 2002). dents about family background, learning habits,
engagement, and motivation in school. PISA
These features of the intensity of schooling
vary across countries. For example, the also surveys school principals, who provide con
total
textual12
yearly hours of instruction for students aged information about the students' schools
and an assessment of the quality of the learning
to 14 is more than 1,000 hours in Italy, France,
Chile, and Mexico, but less than 800 hoursenvironment
in of the school (OECD 2007a).
Sweden, Slovenia, and Luxembourg (OECD
For each participating country, PISA adopted a
stratified
2009b). Teachers have high levels of formal edu clustered sampling strategy. Depending
cation in Poland and Slovakia, where more onthan
the needs of each country, a random sample
90 percent have attained master's degrees.of
Inschools
con was selected (within strata according
to regions, school types, or other categories rele
trast, less than 8 percent of teachers have obtained
vant for each nation). Within each school,
MAs in Brazil, Slovenia, Turkey, and Iceland

168.176.5.118 on Fri, 07 May 2021 22:29:48 UTC
a random sample of ratios—are largely functions of the

approximately mean and var
thirty-fiv
year-old students was drawn.
iance and To ensure
are thus not reported here. the
parability of results Following
across Western and Bloome (2009),
countries, the the t
population of PISAvariance
isin 15-year-old
mathematics achievement is modeled
individ
enrolled in a school,through
without regard
the contribution to (i)the
of each individual in
level school (/') to his or
they attend: Different her country's (k) variance:
countries do not
the same grade-level the squared deviation of individual
definitions, scores from
rendering
parisons across grade levels
the country mean (cr2^).difficult (OEC
Generalized linear models
2007a). estimate the expected value of this deviation from
It is important to note that only 15-year-olds a combination of individual-, school-, and country
who attended school on the day of the survey level variables related to socioeconomic background
were sampled, so the results in this study are not (ses), opportunities to learn (ptl), intensity of school
necessarily generalizable to the 15-year-old popu ing (int), and a set of control variables (cont). For
lation. Generalizability to the 15-year-old popula specification as a generalized linear model, this
tion can be biased if enrollment rates are not full. model uses a log link function and a gamma distri
Also, comparisons across countries can be biased bution with one degree of freedom:
if selection into schooling among 15-year-olds
varies across countries. Consequently, results
from this study pertain to the relationships
log (<4*) =a° + K*+~K*
between school systems and total achievement + M4£ + P*r + e**
inequality among 15-year-olds who are enrolled
where a2jk = (yijk -yi)2
in school. Nonetheless, to assess the possibility
of bias in generalizing to the 15-year-old popula The expected value of the mean squared devi
tion, I estimated complementary models that ation is the average mean squared deviation, or the
include gross enrollment rates as a control, models variance. Thus, results from this model can be
that restrict the sample of countries to OECD mem used to estimate the group-level inequality (e.g.,
bers, and models that restrict the sample to coun the country-level variance in achievement) based
tries with gross enrollment rates greater than 95. on individual-, school-, and country-level attrib
The results from these models do not alter substan utes. Like studies of income inequality that use in
tively the results presented here.1 Also, PISA is dividuals' incomes to estimate group-level
a cross-sectional study. As such, it is not well inequality, this study uses students' test scores to
suited for assessing causality between variables or measure group-level inequality (see Western,
assessing historical trends in its measures. Bloome, and Percheski 2008 for a study of
This study focuses on the total achievement group-level income inequality using variance
inequality in mathematics across nations. function regressions). The country-level variance
Achievement inequality is defined as the variance in achievement can be estimated by aggregating
of student math test scores—or equivalently, the individuals within a country. Also, the effects of
standard deviation—such that countries with the distribution of school- and individual-level at
more variance in their test scores are moretributes on country-level inequality can be esti
unequal than others. The variance measures mated
the by comparing the distributions of these
dispersion of a distribution, thus indicating
attributes across countries to produce a snapshot
the probability that scores concentrate around
of inequality and its sources at the individual,
the mean or spread apart from it. Moreover,
school, and country levels.
because the location and scale of PISA test scores SAS's procedure for the estimation of general
ized linear models (PROC GENMOD) is used to
are arbitrary, the value of 0 lacks meaning; other
commonly used measures of inequality such as
estimate these models accounting for the nested
the Gini coefficient, the Theil Index, andstructure
the of the data (students within schools
Variance of the Logs are not inappropriatewithin
for countries) and uses empirical standard er
use with the PISA test scores (Allison 1978).
rors to calculate the statistical significance of the
Also, PISA scores are transformed to a normal
estimates.2 Because models have a logarithmic
distribution, so achievement distributionslink
forfunction, interpretation of the effect of an
each country approximate normality. As a result,
independent variable ((3^) on the expected value
other measures of inequality—such as percentile
of the mean squared deviation is of a factor of ePl.

168.176.5.118 on Fri, 07 May 2021 22:29:48 UTC
Monti
The independent variables included in the on Education. The curricular policies of each
models were obtained from four sources. All stu country were reviewed, and countries were identi
dent* and school-level variables are obtained fied as those in which the central government de
directly from the PISA data set. Most country termines the curriculum (2), countries in which
level variables were also obtained from the regional or local agencies have some ability to
PISA data set by aggregating student- or adapt the centrally mandated curriculum (1), or
school-level variables (OECD 2007b, 2007c). countries in which there is no central government
Other country-level variables came from intervention in designing the curriculum (0).
the World Development Indicators data set I incorporate several measures for the intensity
(World Bank 2008) or a codification of of schooling, all of which are measured at the
United Nations Educational, Scientific and country level. The length and duration of the
Cultural Organization's World Data on lower secondary school year and the starting age
Education (Amadio 2007). of compulsory education are both obtained from
Family background and home environment are the World Data on Education. The public per
measured by the deviation of the students' pa expenditure in secondary education (in con
pupil
rents' occupational status from the national stant 2000 dollars) comes from the World
mean, mother's educational attainment in ISCED Development Indicators. Average resource quality
is measured through PISA as the country-level
levels, and at the school level, parent's average
average index of school resources. Similarly,
occupational status. At the country level, inequal
both country-level average class size and teacher
ities in living conditions are captured by a measure
of income inequality (Gini coefficients) and outcertification are aggregates of each school princi
of-pocket health expenditures. Out-of-pocket pal's reported class size and the proportion of
health expenditures in a country are determined teachers with ISCED 5a degrees.
by the distribution of resources within a countryFinally, models control for each country's
and a government's commitment to its popula average math achievement, an aggregate of stu
tion's health (Arnett 2007; Quadagno 1987)dents' as scores in PISA's math test, and each coun
part of the risk management dimension of try's per capita gross domestic product as
welfare states (Leisering 2003). Both these measavailable in the World Development Indicators
ures are obtained from the World Development for the year 2007.
Indicators for 2007. In a limited set of countries, some independent
Variation in opportunities to learn is captured variables are missing for the entire sample of
by the extent to which each school differs from schools.3 In order not to lose entire countries
the country average in terms of the index of mate from the analyses, dummy variable imputation is
rial resources, the proportion of teachers with used to account for these missing data (Allison
ISCED 5a degrees, and class size. Between-school 2002; OECD 2007b). Missing values are imputed
tracking is measured at the country level by the with the global mean, and analyses include
number of programs available for students aged a dummy variable indicating imputation. The
15 and the age of selection into these programs. most serious cases are France and Taiwan.
Both of these measures are provided by PISA. France did not administer the school question
Within-school tracking is a school-level indicator naire, and little country-level information is avail
from the PISA school questionnaire that measures able for Taiwan.
whether students in the school are tracked/ All aggregate variables were created using the
appropriate student- or school-level weights pro
streamed according to their ability in all subjects.
A measure of school selectivity—whethervided
stu by PISA, and models are weighted using
dents are admitted to the school according to aca
student weights. To control for the artificial infla
demic criteria—is also included in the models as tion of the sample size, weights are adjusted so
a measure of between-school tracking. The that the unweighted sample size equals the
breadth of the private sector is measured at the weighted sample size.
country level as the average school's percentage The analytic strategy of this study follows
of funding that comes from private tuition fees.three stages. Stage 1 describes the distribution of
All these measures are constructed or obtained the variance in achievement across countries and
from the PISA school survey. Finally, curriculumits relationship to mean achievement to rank coun
standardization is gathered from the World Data tries in terms of total achievement inequality.

168.176.5.118 on Fri, 07 May 2021 22:29:48 UTC
Figure I. Mean mathematics achievement

International Student Assessment (PISA) 20
Note: Reference lines depict average achievem
tries weighed equally. Grey bands depict the
estimated averages at the country level. Th
PISA Reports (Organization for Economic C
for the restricted sample of OECD countries
Stage 2 fits the generalized levels of total achievement

linear inequality in
models t
set of covariates measuring mathematics. variation in f
background, opportunities The relationship
to between
learn,total mathematics
and int
of schooling to identify inequality andthe
mean mathematics
individual-,achievement is sch
and country-level factors also shown in Figure
that 1. It is contribute
common to observe to
achievement inequality. the joint occurrence
Stage of high 3 and graphs
homogeneous th
tionship between opportunities achievement (as is the case to in Finland,
learn,Estonia, inte
of schooling, and achievement Macao-China, Ireland, Denmark, inequality
and Canada).
each country to better No school systems
explicate
produce high levelsthe of disper
most
tant nonsocioeconomic sources of achievement sion with low levels of achievement, but high lev
inequality in each participating country. els of dispersion in achievement are observed
among several high-achieving countries (Chinese
Taipei, Belgium, the Czech Republic, Germany,6
RESULTS Austria, and Switzerland). The correlation
between the standard deviation and mean mathe
The horizontal axis of Figure 1 shows the level ofmatics achievement at the country level is .210,
total achievement inequality in countries that par although it is not statistically significant.
ticipated in PISA.4 Indonesia,5 Thailand, Estonia, Descriptive statistics of the variables included
Finland, and Kyrgyzstan experience the lowest in the models are shown in Table 1.
amount of mathematics achievement inequality Multivariate models for the variance of mathe
as measured by the standard deviation (or, equiv matics achievement are shown in Tables 2 and 3.
Model 1 serves as a baseline model by including
alently, the variance) in math test scores. By con
trast, Belgium, Israel, the Czech Republic, all variables associated with family background
Chinese Taipei, and Germany have the highest and home environment. Model 2 adds variables

168.176.5.118 on Fri, 07 May 2021 22:29:48 UTC
Montt 57
Table I. Desc
Assessment 2006
Variable Mean Standard Deviation Minimum Maximum
Dependent variable
Math score 467.8 104.3 7.6 895.2
Math score squared deviation 7383 10618 0 207458
Family background
Parent's occupational status 42.5 15.0 16 90
Mother's educational attainment (%)

None 6.17
ISCED 1 9.42
ISCED 2 15.09
ISCED 3b, 3c 9.28
ISCED 3a, 4 27.6
ISCED 5b 13.57
ISCED 5a, 6 18.84
School's average occupational status 39.7 10.9 16.0 86.5
Gini coefficient 35.3 8.83 23 56.4
Out-of-pocket health expenditure (%) 73.9 20.6 21.9 100.0
Opportunities to learn
Resource quality index (deviation from 0.791 0.757 0.001 4.448
country mean)
Proportion of teachers with ISCED 5a 0.160 0.201 0.000 1.000
qualification (deviation from country mean)

Class size (deviation from country mean) 7.79 7.94 0.07 33.55
Proportion of schools that track all 0.208 0.461 0.000 1.000

students
Number of educational programs or tracks 2.36 1.24 1.00 5.00
available for 15-year-olds

Years before 15 that selection takes place 1.21 1.63 0.00 5.00
Proportion of schools requiring academic 0.218 0.212 0.00 0.866

record for admissions decisions
Average school's percentage of funding 14.1 14.9 0.1 55.7
from tuition fees
Curriculum standardization 1.38 0.69 0.00 2.00
Intensity of schooling
Resource quality index -0.297 0.612 -2.372 0.874
Public per-pupil expenditure at secondary 21.9 7.0 4.9 35.3
level (thousand dollars)

Proportion of teachers with ISCED 5a 0.739 0.253 0.088 1.000
qualifications
Class size 27.4 5.8 17.1 44.8
Start of compulsory education 6.16 0.66 5.00 7.00

Total hours of instruction 915.6 1 16.0 715.2 1368.0
Controls
Mean math achievement (country level) 469.0 59.8 310.6 549.4
GDP per capita (thousand dollars) 16142 13533 326 54178
Note: ISCED = International Standard Classification of Education. Individual- and school-level variables are weighted
with weights provided by the Programme for International Student Assessment. Country-level variables are weighted
such that each country has equal weight. Observations from Azerbaijan are dropped from this table.

168.176.5.118 on Fri, 07 May 2021 22:29:48 UTC
Sociology of Education 84( I)
Table 2. Generalized Linear Models Estimating the Mean Squared Deviation with Opportunities to Learn
and Intensity of Schooling Variables, Programme for International Student Assessment 2006
Model 1 Model 2
& SE £ SE
Intercept 8.343*** .132 8.417*** .139
Family background
Family occupational status deviation 0.007*** .001 0.006*** <.0011
Mother's education (none)3 0.087*** .024 0.078** .024
Mother's education (ISCED l)a -0.055*** .016 -0.064*** .016
Mother's education ISCED 2)a -0.038*** .011 -0.041*** .010

Mother's education ISCED 3b, c)a -0.029* .012 -0.039*** .012
Mother's education (ISCED 5a, 6)a 0.128*** .01 1 0.141*** .010
Mother's education (ISCED 5b)a -0.037*** .010 -0.021* .010
School mean occupational status deviation 0.043*** .002 0.041*** .002

Gini -0.004* .002 -0.001 .002
Gini2 >0.00 lc <.00lb >0.00 lc** <.0011
Out-of-pocket health expenditure 0.001 <.00 lb >0.001c <.0011
School resource quality deviation 0.018 .013
School teacher quality deviation 0.137** .047
School class size deviation <0.001b .002
School tracks all students -0.014 .021
Number of years before 15 that tracking 0.024*** .007
takes place
Number of programs available for 15-year-olds 0.030*** .007
Proportion of schools requiring academic -0.024 .046

history for admission
Average school's percentage of funding -0.001 .001
from fees
Standardized curriculum (0, 1, 2) 0.014 .013
Average school resource quality
Average school teacher quality
Per-pupil expenditure (thousands of dollars)
Average school class size
Start of compulsory education (years)
Total hours of instruction (hundred hours)
Controls
Mean mathematics achievement <0.001b <.00 lb <0.00 lb <.0011
GDP per capita (thousands of dollars) 0.003*** .001 0.004*** .001
Log likelihood -3086812 -3086456
n observations 312,327 312,327
n schools 11,742 11,742
n countries 56 56
Note: ISCED - International Standard Classification of Education. Models are weighted by individual weights provided
by the Programme for International Student Assessment.
a. Reference category is ISCED 3a, 4. Effect parameterization is used so other estimates are interpreted
assuming overall average mother's education level.
b. Value of the estimate is less than 0.001 but greater than 0.
c. Value of the estimate is greater than —0.001 but less than 0.
*p < .05. **p < .01. ***p < .001.

168.176.5.118 on Fri, 07 May 2021 22:29:48 UTC
Montt 59
Table 3. Ge
Conditions
Model 3 Model 4
£ S£ £ SE
Intercept 9.123*** .233 9.136*** .243

Family background
Family occupational status deviation 0.006*** <.00 lb 0.007*** .000
Mother's education (none)3 0.088*** .025 0.081 *** .024
Mother's education (ISCED l)a -0.041* .016 -0.049** .016

Mother's education (ISCED 2)a -0.039*** .011 -0.041 *** .010
Mother's education ISCED 3b, c)a -0.050*** .012 -0.058*** .012
Mother's education (ISCED 5a, 6)a q |29*** .011 0.138*** .010
Mother's education (ISCED 5b)a -0.034** .010 -0.022* .010
School mean occupational status deviation 0.043*** .002 0.041*** .002
Gini -0.001 .002 -0.001 .002
Gini2 >0.00 lc <.001b >0.00 lc** <.00lb
Out-of-pocket health expenditure >0.00 lc .001 -0.001 .001
School resource quality deviation 0.020 .013
School teacher quality deviation 0.119* .049

School class size deviation 0.001 .002
School tracks all students -0.011 .021
Number of years before 15 that tracking 0.014* .008
takes place
Number of programs available for 15-year-olds 0.026*** .008
Proportion of schools requiring academic 0.094 .049
history for admission

Average school's percentage of funding <0.0016 .001
from fees
Standardized curriculum (0, 1, 2) 0.006 .015
Average school resource quality 0.107*** .031 0.098** .032
Average school teacher quality -0.166*** .031 -0.127*** .034
Per-pupil expenditure (thousands of dollars) 0.003 .002 0.004* .002

Average school class size -0.006** .003 -0.005 .002
Start of compulsory education (years) 0.005 .012 -0.001 .015

-0.012
Total hours of instruction (hundred hours) .008 -0.013 .009
Controls
Mean mathematics achievement 0.001 <.00lb -0.001 <.001b
GDP per capita (thousands of dollars) -0.001 .001 <0.001b .001
Log likelihood -3086398 -3086165

n observations 312,327 312,327
n schools 11,742 11,742
n countries 56 56
Note: ISCED = International Standard Classification of Education. Models are weighted by individual weights provided
by the Programme for International Student Assessment.
a. Reference category is ISCED 3a, 4. Effect parameterization is used so other estimates are interpreted
assuming overall average mother's education level.
b. Value of the estimate is less than 0.001 but greater than 0.
c. Value of the estimate is greater than —0.001 but less than 0.
*p < .05. **p < .01. ***p < .001.

168.176.5.118 on Fri, 07 May 2021 22:29:48 UTC
associated with opportunities to learn, whereas all schools have the same proportion of ISCED
Model 3 provides estimates for intensity of 5 a teachers) reduces the variance in achievement
schooling. Finally, Model 4 includes all variables by 7.1 percent.8
jointly. Between-school tracking is also associated
Model 1 in Table 2 shows how variations in
with greater total achievement inequality. Each
family background are related to the varianceadditional
in educational program available for 15
achievement. Within a country, family back
year-old students is related to a 3.0 percent
ground varies both within and between schools. increase in achievement inequality. Shifting selec
As expected, increased variation in school-mean tion into these programs earlier in the course of
a student's schooling increases the variance in
parents' occupational status is strongly associated
with total achievement inequality: A 1 standard achievement by an additional 2.4 percent per
deviation increase in the occupational status devi
year. Surprisingly, however, within-school track
ing shows no association with achievement
ation of the school is associated with a 28.2 per
cent increase in achievement inequality. Netinequality.
of This finding, which is contrary to the
school composition, greater polarization in the large body of research that shows how ability
occupational status and educational attainmentgrouping
of within schools increases achievement
inequality, may be a result of an unreliable mea
parents is related to greater achievement inequal
ity. In particular, students with parents who sure.
are The measure for within-school tracking pro
one standard deviation above or below the coun vided by PISA asks school administrators whether
try's average occupational status contribute 6.0the school groups students by ability "in no sub
percent more to the variance of achievement ject," "in some subjects," or "in all subjects."
inequality than students living with parents Consistent
of with the PISA reports, the fitted mod
average occupational status.7 Similarly, greaterels use a dichotomized measure in which only
polarization in the educational level of parentsthe "in all subjects" category is indicative of
contributes more to a country's achievementwithin-school tracking (OECD 2007a). Even
inequality than a more concentrated distributionthough there is considerable variation in this
of parents' educational attainment, which is indi
dichotomized measure of tracking, both within
cated by the positive and significant estimates and
of between countries, tracking is often complex
high (ISCED 5a or 6) or low (less than ISCED(Kelly 2007, 2008), and this simple measure
1) parents' educational attainment. These back might not capture it reliably. For example, only
ground and contextual effects remain stable and 5 percent of the schools in the United States qual
statistically significant across models once oppor
ify as tracked schools according to this measure,
yet tracking is far more common in the United
tunities to learn or the intensity of schooling is
included. Net of the socioeconomic status of States (Kelly 2004; Lucas and Berends 2002;
Turner 1960).
schools and students, nations with a more unequal
income distribution have lower achievement Increased variability in resource quality and
class size across schools is not related to increases
inequality, but this estimate is substantively small
and no longer significant when other variables inare
the dispersion of achievement, supporting the
weak or null findings relating school resources
added to the models. Also, out-of-pocket health
andin
expenditures are not related to the variance class size to achievement. The breadth of
achievement. the private sector, the extent to which schools
Beyond the expected effects of socioeconomic require academic histories for selection into
background on inequality, variations in opportuni schools, and curriculum standardization are also
ties to learn contribute to achievement dispersion unrelated to the variance in math achievement.
in a school system. These relationships are as Model 3 considers the relationship between
sessed in Model 2. Uneven distributions of teach measures of the intensity of schooling and
ers' formal training among schools and tracking achievement inequality, adjusting for socioeco
are related to greater dispersion in math achieve nomic factors. Average levels of teacher and
ment. This is consistent with research showing resource quality, as well as average class size,
the effects of teacher quality and tracking on are associated with variation in math achievement.
achievement. Estimates indicate that a shift in An increase in the proportion of teachers with
the distribution of teachers from maximum hetero ISCED 5a qualifications in the school system
geneity to one of complete homogeneity (where from one quarter to three quarters is associated

168.176.5.118 on Fri, 07 May 2021 22:29:48 UTC
tAontt
with a reduction in total achievement inequality of corresponding set of values in the model equations
8.4 percent. An increase in the average quality of using Model 4's estimated coefficients while fixing
resources of the schools, however, is associated the values for socioeconomic background and
with an even greater increase in achievement dis intensity of schooling at their sample means. This
persion: a one-unit increase in the school resource assumes that all countries share similar distributions
index at the country level results in an increase in in both socioeconomic background and intensity of
achievement inequality of 10.1 percent (at the schooling. Conversely, the contribution of intensity
country level, the index ranges from —2.4 to of schooling is calculated by fixing the values for
0.8). Larger class sizes are negatively, and signif socioeconomic background and opportunities to
icantly, associated with increases in achievement learn at their means. Observations per country are
inequality, although the magnitude of the effect then averaged to calculate the coordinates for
is not substantively relevant and losses signifi each country.
cance in the final model. Other measures of inten The net contribution of intensity of schooling (x
sity of schooling such as the onset of compulsory axis) is always negative. As predicted, intensity of
education and total hours of instruction are not schooling serves to reduce total achievement
related to total achievement inequality. inequality. The countries that have the greatest
The results discussed for Models 1 through 3reduction in achievement inequality due to intensity
do not change substantially once all variables of schooling are Indonesia, Kyrgyzstan, Jordan,
are entered simultaneously (Model 4). As dis Thailand, Brazil, and Colombia. The greatest con
cussed above, Model 4 suggests that there istributions to total achievement inequality from var
a strong relationship between variability in familyiations in opportunities to learn (y-axis) are in
background and total achievement inequality.Liechtenstein, Germany, the Czech Republic,
Similarly, greater variation in opportunities to Slovakia, Switzerland, and the Netherlands. By
learn is related to greater achievement inequality,contrast, Canada, the United States, Australia,
New Zealand, and all five Nordic countries are
particularly in the form of inequality in the distri
bution of teachers across schools and between among those that produce the least achievement
school tracking. In terms of the intensity of
inequality from variations in opportunities to learn.
schooling, increases in overall teacher quality (as
It is noteworthy to compare the position of the
measured by teachers' educational attainment) United States in Figure 2 and in Figure 1. From
are related to reductions in total achievement Figure 2 we conclude that variations in opportuni
inequality. Increased overall resource quality tiesof
to learn and intensity of schooling contribute
the schools in a country, however, is related equally
to to total inequality in the United States
higher levels of achievement inequality. and in the Nordic countries. Yet from Figure 1
we learn that total achievement inequality in the
Collectively, the measures in Model 4 predict
United States is much greater than in the Nordic
total achievement inequality well. At the country
level, the correlation between the observed countries.
and This is the result of both the greater dis
predicted (by Model 4) variance in achievement persion in distribution of socioeconomic back
is .65. If Macao China and Hong Kong China ground in the United States and the different
are not considered in the correlation, its value in
effects socioeconomic background has on achieve
creases to .78. This indicates that with information ment in the United States when compared to the
about the distribution of socioeconomic back Nordic countries. The greater achievement inequal
ground, variations in opportunities to learn, ity
andin the United States, therefore, stems from the
greater variability in students' background and
intensity of schooling, the model predicts more
than 60 percent of the country-level variance the
instronger effects student background, rather
achievement inequality. than the institutional arrangements of the educa
tional system, has on achievement.
Figure 2 shows a scatterplot with the contribu
tion of each set of variables (opportunities to learn
and intensity of schooling) to the logged variance of
achievement. It portrays graphically how intensity
DISCUSSION
of schooling and variations in opportunities to learn
In the sociology of education, research on educa
contribute to total achievement inequality in each
tional inequality has often focused on equality of
country. The contribution of opportunities to learn
opportunity, the strength of the relationship
is calculated by substituting each observation's

168.176.5.118 on Fri, 07 May 2021 22:29:48 UTC
DEU
■
^^•SVK
NLD
EX. MAC
HRV
•^S HUN
n
MN^
g»PRT
SVN
KCZ
im % GBR^
Bfi* CRC LTU
CAN^NZL^
-0.5 -0.4 —0.3 -0.2
Contrtoutton of Intensity of Schooling
Figure 2. Predicted contributio

logged variance of achievemen
between background charac

Importantly though, achievement inequality is
tional also a function of
outcomes. characteristics ofapproac
This educational
the total inequality
systems themselves such that
as variations inscho
opportu
which is important
nities to learn, particularly inbecause
the form of inequi
are expected ties to in the endure or
distribution of teachers among eve
schools
cohorts advance in their educational careers and and the extent of between-school tracking.
enter the labor market. Inequalities in achieve Total inequality is also a function of the intensity
ment may then become inequalities in attain of schooling, such that school systems with better
ment, occupational status, income, or othertrained teachers have less dispersion in achieve
adult outcomes. ment. Overall resource quality is an exception
Consistent with the large body of literaturehowever; countries with higher overall resource
linking socioeconomic background and achieve quality actually have greater variation in achieve
ment, this article shows that achievement inequalment. This result is at odds with the hypotheses
ity is partly a function of the distribution of
of this article and could indicate differential
socioeconomic status. I find greater achievement capacity of schools to use resources effectively
inequality in countries with higher segregation inor an imprecise measurement of resource quality
school socioeconomic status composition andused in the analyses.9
higher individual socioeconomic status differen These results reinforce the positive relation
ship between teacher quality and achievement
ces within the school. In other words, by reducing
the strength of the relationship between socioeco(Greenwald et al. 1996; Rivkin et al 2005). As
nomic status and achievement in a school system,
this article suggests, a better trained teacher work
the overall distribution of achievement is also force is also related to lower total achievement
equalized. Hence, improving the equality ofinequality. Stronger investments in the teacher
opportunity of a school system not only hasworkforce increase the performance of educa
important meritocratic consequences for individu
tional systems and make the distribution of out
comes more equitable. This investment may
als and their life chances but also has important
consequences in shaping the achievement distri have the threefold benefit of also reducing in
bution of a school system. equalities of opportunity inasmuch as certain

168.176.5.118 on Fri, 07 May 2021 22:29:48 UTC
Montt
teacher characteristics have been linked to higher achievement through educational practices such
achievement among disadvantaged students as tracking (OECD 2007a), shadow education
(Bryk, Lee, and Holland 1993; Gamoran 1993; where it is private (Baker et al. 2001; Stevenson
Nye et al. 2004). Similarly, prior research shows and Baker 1992), and inequalities in the allocation
that reducing or eliminating tracking leads to in of resources such as those that stem from educa
creases in equality of opportunity (Gamoran and tional funding mechanisms in the United States
Mare 1989; OECD 2007a). This article extends (Moser and Rubenstein 2002).
this finding by showing that comprehensive Cross-national studies of education have an
school systems produce a more even distribution important advantage over studies that concentrate
of achievement across students (Hanushek and on one educational system: They enable research
Wo(3mann 2005). ers to evaluate the impact of broad educational
The models in this article also show that cur policies that remain constant within school sys
riculum standardization, starting age of compultems (in this case, overall teacher quality, overall
sory education, and total hours of instruction areresource quality, total hours of instruction, and cur
not related to achievement inequality. Although riculum
I standardization) on total achievement
expected that these variables would be related toinequality. However, most of these studies are
achievement inequality, the lack of an observedcross-sectional and are not able to assess causality
relationship could be a result of the low levelsin the observed relationships (Loveless 2009).
of variance in these measures of intensity of These relationships can become the basis for
schooling. As neo-institutional research showsfuture cross-national studies on achievement.
(Baker and LeTendre 2005; Meyer and Rowan They should adopt a longitudinal perspective
1977), even though countries may vary in curricand include a more comprehensive set of school
ulum standardization, it is likely that implemented
and institutional practices to address the particular
curriculums do not vary as much across schools,mechanisms that explain how intensity of school
districts, or states. Similarly, there does not seeming or variations in opportunities to learn affect
to be sufficient variation in the total hours of achievement inequality.
instruction or the starting age of compulsory edu A longitudinal account of achievement inequal
cation to produce meaningful differences in total
ity could incorporate the recently observed trend
achievement inequality. toward greater schooling equality (Baker and
The comparison of the effects of schooling LeTendre 2005) such as that which results from
variables on achievement inequality and the increased enrollment in secondary and higher edu
observed distribution of achievement inequalitycation (Meyer, Ramirez, and Soysal 1992; Schofer
and Meyer 2005), investment of resources (Baker
illustrates the importance of socioeconomic diver
sity in shaping total achievement inequality. et al. 2002), and the changes some school systems
School systems such as those of Finland and thehave adopted with respect to their tracking policies
United States produce similar levels of achieve(a change toward general education in Poland, de
tracking in the United States, and a wider preva
ment inequality as a result of intensity of school
lence of comprehensive high schools in Germany
ing and variations in opportunities to learn. Their
observed achievement inequality, however, differsand France). A longitudinal perspective on achieve
greatly as a result of different distributions ment
of inequality could also relate the decreasing
socioeconomic status both between and within inequality of opportunity observed in Sweden, the
schools and as a result of the different impact Netherlands, and other countries (Breen and
socioeconomic background has on achievement Jonsson 2005; Shavit and Blossfeld 1993) to total
in these two countries. achievement inequality. These longitudinal per
This finding supports the assertion that reduc spectives on total achievement inequality would
tions in equality of educational outcomes can be shed light on the trend toward greater schooling
efficiently overcome by equalizing life conditions equality.
(i.e., reducing the spread of the distribution; The immediate costs and benefits of more
Jencks et al. 1972; Shavit and Blossfeld 1993) equal distributions of schooling also offer fruitful
through, for example, comprehensive welfare pro possibilities for research. Two possible costs of an
grams (Pong, Dronkers and Hampden-Thompson egalitarian education are a loss in academic excel
2003). Educational arrangements can amplify lence and higher financial costs. Hanushek and
the effect of socioeconomic background on Wo(3mann (2005) conclude that tracking increases

168.176.5.118 on Fri, 07 May 2021 22:29:48 UTC
the magnitude of this bias

total achievement inequality but is impossible
does to determine.
not re
The inabilityalso
mean achievement. They to account note
for selectionthat
processes is off
a limitation of this study, and inferences should
a comprehensive education that leads to less
thus be limited to total achievement inequality
achievement inequality is no more expensive
among students who are enrolled in school.
offering a stratified educational system that
Attempts were made, however, to account for the
ces more total achievement inequality.
possibility of selection bias. Adding a control for the
The long-term costs gross
and benefits
enrollment or
ratio in secondary more
schools does not eq
distributions of schooling also
alter the results offer
presented in this study. new
Limiting thegro
for research. Totalachievement inequality
sample to the 30 countries that were members of the
be related to formsOrganization
of inequality
for Economic Co-Operationthat and ha
increased, mainly Development inequality
income (OECD) in 2006—under thein
presump
indu
ized countries tion that selection
(Smeeding processes Gottschalk
and in developed countries
and the stagnation in differ
the as compared to developing countries—only
reduction of povert
alters the results concerning the distribution of teach
some parts of the world (Collier and Dol
ers and the overall quality of teachers: Both estimates
2001). Furthermore, as a student cohort prog
are not statistically significant in the sample for
into higher education
and enters the labor ma
OECD-only countries. Furthermore, results are not
total achievement inequality may
altered substantively after become
restricting the sample of sk
equalities that affect countries
economic growth and
to those school systems with gross enroll in
inequality (Castello-Climent and
ment rates greater than Domenech
95 (36 countries).
Castello and Domenech 2002;
2. In particular, De
the following Gregorio
MODEL options are
Lee specified in PROC
2002), although these are GENMOD to estimate models
strongly med
by the characteristics using athe
of gamma distribution
labor with 1 degree of free
market and o
institutions such as dom: dist=gamma link=log
collective bargainingscoring=100 noscale
and u
scale=.5. The nested design is incorporated in the
membership (Blau and Kahn 2005; Carbo
analyses by including the school_id * country inter
2005, 2006; Devroye action
and Freeman 2001). The
in the REPEATED statement in SAS.
sequences of total achievement inequality
Individuals within each school are assumed to be
spread beyond economic grounds
independent; thus, the type=ind optionas educa
is specified.
has been linked to other adult life outcomes
This specification of the REPEATED statement pro
(Pallas 2000), and equality in educational outcomes
duces empirical (GEE) standard errors for each
fosters political equality (Labaree 1997). These parameter estimate by specifying the adequate
analyses may lead to discussion about the amount covariance structure (see SAS 2007 for details).
SAS 9.1.3 is used for estimation. Because of the
of total achievement inequality that is desirable
for each country to produce. nested structure of the data, hierarchical linear mod
els appear as an alternative technique to analyze this
data. Unfortunately, current versions of the hierarchi
ACKNOWLEDGMENT cal linear models software allow for the analysis of
variables only with a Normal, Poisson, or Bernoulli
The author is grateful for the guidance and comments
distribution, but not—as is the case in modeling the
offered by Sean Kelly as well as the commentsvariance
offered through the mean squared deviation—errors
by William Carbonaro and three anonymous reviewers.
with a Gamma distribution.
The content and particularly the errors of this manuscript
3. Several countries have countrywide missing values
are of the sole responsibility of the author. on different variables. These countries are Austria,
Canada, Chinese Taipei, France, Ireland, Israel,
Hong Kong China, Japan, Jordan, Liechtenstein,
NOTES
Macao-China, Montenegro, Qatar, the Russian
Federation, Serbia, Spain, and Turkey.
4. Azerbaijan presents abnormally low levels of
1. Because schools are the primary sampling unit, all
15-year-olds not in school the day of the test are achievement inequality: Its standard deviation in
excluded from the sampling frame. Therefore, results math scores is almost half that of other countries
are only generalizable to the enrolled population of with low levels of total achievement inequality.
students. To generalize to the entire 15-year-old pop Due to this irregularity as well as the uncommon dis
ulation, it is important to keep in mind the possibility tribution of other variables used in this analysis, all
of selection bias in countries where enrollment ratios cases belonging to Azerbaijan are dropped from the
are lower. Achievement inequality in countries with analyses. Its elimination from the models does not
lower enrollment ratios may be underestimated, but alter the main conclusions of this study.

168.176.5.118 on Fri, 07 May 2021 22:29:48 UTC
Montt 65
5. As discussed in the Data and Method section and possible that these measures address the principals'
Note 1, these estimates have to be interpreted taking
involvement or expectations of what constitutes ade
into account the potential sample selection bias for
quate instruction and not the quality or even the use
each country. Indonesia has the lowest gross enrollof resources for instructions.
ment ratios for all secondary school systems ana
lyzed, so low achievement inequality in this case
could be a result of selection into the secondary REFERENCES
school system; therefore, inequality for the adoles
Allison, Paul. 1978. "Measures of Inequality." American
cent population is highly underestimated. Inequality
for the student population, however, is not biased.Sociological Review 43:865-80.
Allison, Paul. 2002. Missing Data. Thousand Oaks, CA:
Models that control for gross enrollment ratios that
restrict the sample to the 30 OECD countries of Sage.
2006 and that restrict the sample to 36 countriesAmadio, Massimo, ed. 2007. World Data on Education.
with gross enrollment rates greater than 95 do not Paris, France: UNESCO.
Arnett, Stephanie. 2007. "National Variation in the
alter substantially the results presented here (see
Effects of Socioeconomic Status on Student
Note 8).
6. In OECD reports, Germany is not statistically differ Learning: Inequality and Stratification in
ent than the OECD average and is thus not consid Comparative Perspective." Department of Sociology,
ered a high-achieving country. As noted in the University of Notre Dame, IN. Unpublished
figure, the country-level average used as a reference manuscript.
in this article is calculated for the entire sample of Baker, David, Motoko Akiba, Gerald LeTendre, and
countries and not restricted to OECD countries. Alexander Wiseman. 2001. "Worldwide Shadow
7. Discrete percentage change (z) calculations are Education: Outside-school Learning, Institutional
computed as follows: z = e(PkX<rk) X 100, where Quality of Schooling and Cross-national Mathematics
Pkis the estimates from the multivariate models for Achievement." Educational Evaluation and Policy
variable k, and cr^ is the observed standard deviation Analysis 23(1):1-17.
for variable k. For the following examples, the stan Baker, David, Brian Goesling, and Gerald Letendre. 2002.
dard deviations used in the computations are 8.32 for "Socioeconomic Status, School Quality, and National
absolute deviation in occupational status and 5.78 in Economic Development: A Cross-national Analysis of
school deviation from mean school occupational sta the 'Heyneman-Loxley Effect' on Mathematics and
tus. For categorical variables, and variables described Science Achievement." Comparative Education
Review 43:291-312.
as unit increases, discrete percentage changes (z) are
computed as z = e^ X100. Baker, David and Gerald LeTendre. 2005. National
8. These results are not statistically significant when re Differences, Global Similarities: World Culture
stricting the sample to the 30 OECD countries of and the Future of Schooling. Stanford, CA:
Stanford Social Sciences.
2006. This might be due to the fact that teacher cer
tification is less relevant in explaining achievement Blau, Francine and Lawrence Kahn. 2005. "Do
in OECD countries. This change of effect of teacher Cognitive Test Scores Explain Higher US Wage
certification does not seem to be a result of different Inequality?" Review of Economics and Statistics
87:184-93.
levels of variability among OECD and non-OECD
countries: Descriptive statistics for these variables Brady, David. 2005. "The Welfare State and Relative
signal no substantial difference in the level and var Poverty in Rich Western Democracies, 1967
1997." Social Forces 83:1329-64.
iability of this variable across the two sets of coun
tries. The other results do not change substantially Breen, Richard and Jan Jonsson. 2005. "Inequality of
by limiting the sample. A hypothesis behind the Opportunity in Comparative Perspective: Recent
null result of teacher certification in OECD countries Research on Educational Attainment and Social
is that the selection of teachers (through professional Mobility." Annual Review of Sociology 31:223-43.
tests or long selection processes) ensures a homoge Broaded, C. Montgomery. 1997. "The Limits and
neous and qualified teaching workforce, beyond that Possibilities of Tracking: Some Evidence from
provided by the professional credential. Taiwan." Sociology of Education 70:36-53.
9. The measure for resource quality used by the Bryk, Anthony, Valerie Lee, and Peter Holland.
1993. Catholic Schools and the Common Good.
Programme for International Student Assessment in
volves the school principals' assessing the way cer Cambridge, MA: Harvard University Press.
tain resources affect instruction. These resources Buchmarm, Claudia and Emily Hannum. 2001.
include Internet connectivity, library materials, "Education and Stratification in Developing
Countries: A Review of Theories and Research."
audio-visual resources, computers, instructional ma
terials, and science laboratory equipment. It is Annual Review of Sociology 27:77-102.

168.176.5.118 on Fri, 07 May 2021 22:29:48 UTC
Carbonaro, William. 2005. "Explaining Variable Achievement in American Society. New York, NY:
Returns to Cognitive Skill across Occupations." Academic Press.
Social Science Research 34:165-88. Fuchs, Thomas and Ludger Wopmann. 2007. "What
Carbonaro, William. 2006. "Cross-national DifferencesAccounts for International Differences in Student
in the Skills-earnings Relationship: The Role of Performance? A Re-examination Using PISA
Labor Market Institutions." Social Forces 84:1819 Data." Empirical Economics 32:433-64.
42. Fuller, Bruce. 1987. "What School Factors Raise
Carbonaro, William and Elizabeth Covay. 2010. Achievement in the Third World?" Review of
"Sector Differences in Student Experiences and Educational Research 57:255-92.
Achievement: An Update." Sociology of Education Gamoran, Adam. 1993. "Alternative Uses of Ability
83(2): 160-82. Grouping in Secondary Schools: Can We Bring
Castello, Amparo and Rafael Domenech. 2002. "Human High-quality Instruction to Low-ability Classes?"
Capital Inequality and Economic Growth: Some American Journal of Education 102:1-22.
New Evidence." Economic Journal 112:C187-C200. Gamoran, Adam. 2009. "Tracking and Inequality: New
Castello-CIiment, Amparo and Rafael Domenech. 2008. Directions for Research and Practice." WCER
"Human Capital Inequality, Life Expectancy and Working Paper No. 2009-6, Wisconsin Center for
Economic Growth." Economic Journal 118:653-77. Education Research, Madison, WI.
Chubb, John and Terry Moe. 1990. Politics, Markets,Gamoran, Adam and Robert Mare. 1989. "Secondary
and America's Schools. Washington, DC: School Tracking and Educational Inequality:
Brookings Institution. Compensation, Reinforcement, or Neutrality?"
Coleman, James Samuel. 1990. Equality and American Journal of Sociology 95(5): 1146-83.
Gamoran, Adam, Walter Secada, and Cora Marrett.
Achievement in Education. Boulder, CO: Westview.
Coleman, James Samuel and Thomas Hoffer. 1987. 2000. "The Organizational Context of Teaching
Public and Private High Schools: The Impact of and Learning." Pp. 37-64 in Handbook of the
Communities. New York, NY: Basic Books. Sociology of Education, edited by M. Hallinan.
Collier, Paul and David Dollar. 2001. "Can the World New York, NY: Kluwer Academic/Plenum.
Cut Poverty in Half? How Policy Reform andGreenwald, Rob, Larry Hedges, and Richard Laine.
Effective Aid Can Meet International Development 1996. "The Effect of School Resources on Student
Goals." World Development 29:1787-1802. Achievement." Review of Educational Research
Condron, Dennis and Vincent Roscigno. 2003. 66:361-96.
"Disparities Within: Unequal Spending and Urban Hanushek, Eric and Ludger W6(3mann. 2005. "Does
School Achievement in a District." Sociology of Educational Tracking Affect Performance and
Education 76:18-36. Inequality? Difference-in-Differences Evidence Across
De Gregorio, Jose and Jong-Wha Lee. 2002. "Education
Countries." Ifo Working Paper No. 1, Institute for
and Income Inequality: New Evidence from Cross Economic Research, University of Munich,
country Data." Review of Income and Wealth Germany.
48:395-416. Heyneman, Stephen and William Loxley. 1983. "The
Devroye, Dan and Richard Freeman. 2001. Does Effect of Primary-school Quality on Academic
Inequality in Skills Explain Inequality of Earnings Achievement Across Twenty-nine High- and Low
Across Advanced Countries? Cambridge, MA: income Countries." American Journal of Sociology
National Bureau of Economic Research. 88:1162-94.
Downey, Doug, Paul von Hippel, and Beckett Broh.
Huang, Min-Hsiung. 2009. "Classroom Homogeneity
2004. "Are Schools the Great Equalizer? Cognitive and the Distribution of Student Math Performance:
Inequality During the Summer Months and the A Country-level Fixed-effects Analysis." Social
School Year." American Sociological Review Science Research 38:781-91.
69:613-35.
Jencks, Christopher, Marshall Smith, Henry Ackland,
Ehrenberg, Ronald, Dominic Brewer, Adam Gamoran,Mary Jo Bane, David Cohen, Herbert Gintis,
and Douglas Willms. 2001. "Class Size and Barbara Heyns, and Stephen Michaelson. 1972.
Student Achievement." Psychological Science in Inequality: A Reassessment of the Effect of Family
the Public Interest 2:1-30. and Schooling in America. New York, NY: Basic
Entwisle, Doris, Karl Alexander, and Linda Olson. 2000. Books.
"Summer Learning and Home Environment." Pp. 9 Kelly, Sean. 2004. "Are Teachers Tracked? On What
30 in A Notion at Risk: Preserving Public Education Basis and with What Consequences." Social
as an Engine for Social Mobility, edited by R. D. Psychology of Education 7:55-72.
Kahlenberg. New York, NY: Century Foundation. Kelly, Sean. 2007. "Social Class and Tracking within
Featherman, David L., Robert Mason Hauser, and Schools." Pp. 210-24 in The Way Class Works, edi
William Hamilton Sewell. 1976. Schooling and ted by L. Weis. New York, NY: Routledge.

168.176.5.118 on Fri, 07 May 2021 22:29:48 UTC
Montt 67
Kelly,Students Benefit
Sean. More from Small Classes?
Evidence from the Tennessee Class Size
Carolina." H
Kerckhoff
Experiment." Educational Evaluation and Pol
Analysis 24:201-17.
Cambridge,
Konstantop Nye, Barbara, Spyros Konstantopoulos, and L
Effects on Student Achievement: Evidence from Hedges. 2004. "How Large Are Teacher Effec
NLS : 72, HSB : 82, and NELS : 92." Teachers Educational Evaluation and Policy Analysis
College Record 108:2550-81. 26:237-57.
Korpi, Walter and Joakim Palme. 1998. "The Paradox ofOrganization for Economic Co-Operation and
Redistribution and Strategies of Equality: Welfare Development. 2007a. PISA 2006. Science
State Institutions, Inequality, and Poverty in the Competencies for Tomorrow's World. Paris,
Western Countries." American Sociological Review France: Organization for Economic Co-Operation
63:661-87. and Development.
Organization for Economic Co-Operation and
Labaree, David. 1997. "Public Goods, Private Goods:
The American Struggle Over Educational Goals." Development. 2007b. PISA 2006 Annex A8:
American Educational Research Journal 34:39-81. Technical Notes on Multilevel Regression Analysis.
Leisering, Lutz. 2003. "Government and the Life Paris, France: Organization for Economic Co
Course." Pp. 205-28 in Handbook of the Life Operation and Development.
Course, edited by J. Mortimer and M. Shanahan.
New York, NY: Kluwer Academic/Plenum. Development. 2007c. PISA 2006 Volume 2: Data.
LeTendre, Gerald, Barbara Hofer, and Hidetada Paris, France: Organization for Economic Co
Shimizu. 2003. "What Is Tracking? Cultural Operation and Development.
Expectations in the United States, Germany, and
Japan." American Educational Research Journal Development. 2009a. Creating Effective Teaching
40:43-89. and Learning Environments: First Results from
Loveless, Tom. 2009. How Well Are American Students TALIS. Paris, France: Organization for Economic
Learning? The 2008 Brown Center Report on Co-Operation and Development.
American Education: With Sections on Organization for Economic Co-Operation and
International Assessments, the Misplaced Math 2009b. Education at a Glance 2009:
Development.
Student, and Urban Schools. Washington, DC:
OECD Indicators. Paris, France: Organization for
Brookings Institution. Economic Co-Operation and Development.
Lucas, Samuel and Mark Berends. 2002. Pallas, Aaron. 2000. "The Effects of Schooling on
"Sociodemographic Diversity, Correlated Individual Lives." Pp. 499-528 in Handbook of the
Achievement, and De Facto Tracking." Sociology Sociology of Education, edited by M. Hallinan.
of Education 75:328-48. New York, NY: Kluwer Academic/Plenum.
Ma, Xin and Don Klinger. 2000. "Hierarchical Linear Pong, Suet-Ling, Japp Dronkers, and Gillian Hampden
Modelling of Student and School Effects on Thompson. 2003. "Family Policies and Children's
Academic Achievement." Canadian Journal of School Achievement in Single- Versus Two-parent
Education 25:41-55. Families." Journal of Marriage and Family
Meyer, John, Francisco Ramirez, and Yasemin Soysal. 65(3):681-99.
1992. "World Expansion of Mass Education, 1870 Quadagno, Jill. 1987. "Theories of the Welfare-State."
1980." Sociology of Education 65:128-49. Annual Review of Sociology 13:109-28.
Ramirez, Francisco. 2006. "Beyond Achievement and
Meyer, John and Brian Rowan. 1977. "Institutionalized
Organizations—Formal-Structure as Myth and Attainment Studies—Revitalizing a Comparative
Ceremony." American Journal of Sociology Sociology of Education." Comparative Education
83:340-63. 42:431-49.
Rivkin, Steven, Eric Hanushek, and John Kain. 2005.
Meyers, Marcia, Dan Rosenbaum, Christopher Ruhm,
and Jane Waldfogel. 2004. "Early Childhood "Teachers, Schools, and Academic Achievement."
Education and Care." Pp. 223-270 in Social Econometrica 73:417-58.
Rowan, Brian, Richard Correnti, and Robert Miller.
Inequality, edited by K. Neckerman. New York,
NY: Russell Sage. 2002. "What Large-scale, Survey Research Tells
Moser, Michele and Ross Rubenstein. 2002. "The Us About Teacher Effects on Student
Equality of Public School District Funding in the Achievement: Insights from the Prospects Study o
United States: A National Status Report." Public Elementary Schools." Teachers College Record
Administration Review 62(l):63-72. 104:1525-67.
Nye, Barbara, Larry Hedges, and Spyros SAS Institute Inc. 2007. SAS/STAT 9.1.3 User's Guide.
Konstantopoulos. 2002. "Do Low-achieving Vol. 2. Cary, NC: SAS Institute.

168.176.5.118 on Fri, 07 May 2021 22:29:48 UTC
Scheerens, Jaap and Roel Bosker. 1997. The across Countries: A Comparison of Methods."
Foundations of Educational Effectiveness. Oxford, Labour Economics 11:487-506.
UK: Pergamon. Van de Werfhorst, Herman and Johnathan Mijs. 2010.
Schofer, Evan and John Meyer. 2005. "The Worldwide "Achievement Inequality and the Institutional
Expansion of Higher Education in the Twentieth Structure of Educational Systems: A Comparative
Century." American Sociological Review 70:898-920. Perspective." Annual Review of Sociology 36:407-28.
Shavit, Yossi and Hans-Peter Blossfeld. 1993. PersistentWestern, Bruce and Deirdre Bloome. 2009. "Variance
Inequality: Changing Educational Attainment in Function Regressions for Studying Inequality."
Thirteen Countries. Boulder, CO: Westview. Sociological Methodology 39:293-326.
Smeeding, Timothy and Peter Gottschalk. 1999. "CrossWestern, Bruce, Deirdre Bloome, and Christine
National Income Inequality: How Great Is It and Percheski. 2008. "Inequality among American
What Can We Learn from It?" International Families with Children, 1975 to 2005." American
Journal of Health Services 29:733-41. Sociological Review 73:903-20.
Sorensen, Aage and Maureen Hallinan. 1977. World Bank. 2008. World Development Indicators.
"Reconceptualization of School Effects." Sociology Washington, DC: World Bank.
of Education 50:273-89.
Stevenson, David and David Baker. 1991. "State Control
of the Curriculum and Classroom Instruction."
Sociology of Education 64:1-10.
BIO
Stevenson, David and David Baker. 1992. "Shadow
Education and Allocation in Formal Schooling: Guillermo Montt is a graduate student at the Center for
Transition to University in Japan." American Research on Educational Opportunity in the Department
Journal of Sociology 97(6): 1639-57. of Sociology, University of Notre Dame. He has worked
Turner, Ralph. 1960. "Sponsored and Contest Mobilityas a consultant for the Education Directorate of the
and the School-system." American SociologicalOrganization for Economic Co-Operation and
Review 25:855-67. Development, and his interests include cross-national
Valenzuela, J. Samuel, Eugenio Tironi Barrios, and
studies on education, social stratification, and quantita
Timothy Scully. 2006. El Eslabon Perdido: tive research methods. He is currently working on poten
Familia, Modernizacion y Bienestar en Chile. tial trade-offs between educational inequality and
Santiago, Chile: Taurus. achievement excellence and the effects of educational
Vandenberghe, Vincent and Stephane Robin. 2004. inequality on economic growth, civic engagement, and
"Evaluating the Effectiveness of Private Education income inequality.

168.176.5.118 on Fri, 07 May 2021 22:29:48 UTC

Cross-National Differences in Educational Achievement Inequality

Uploaded by

Copyright:

Available Formats

You might also like

Cross-National Differences in Educational Achievement Inequality

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cross-National Differences in Educational Achievement Inequality

Uploaded by

Copyright:

Available Formats

Cross-national Differences in Educational Achievement Inequality

Author(s): Guillermo Montt

Stable URL: https://www.jstor.org/stable/23057035

This content downloaded from

Educational Achievement http://soe.sagepub.com

This content downloaded from

secondary schooling are important

This content downloaded from

Although informative, studies of educational achievement and thus in generating inequality.

This content downloaded from

This content downloaded from

resource (OECD 2009a). Per-pupil expenditures on queduca

This content downloaded from

a random sample of ratios—are largely functions of the

This content downloaded from

This content downloaded from

Figure I. Mean mathematics achievement

Stage 2 fits the generalized levels of total achievement

This content downloaded from

Variable Mean Standard Deviation Minimum Maximum

Mother's educational attainment (%)

qualification (deviation from country mean)

Proportion of schools that track all 0.208 0.461 0.000 1.000

Number of educational programs or tracks 2.36 1.24 1.00 5.00

available for 15-year-olds

Proportion of schools requiring academic 0.218 0.212 0.00 0.866

Public per-pupil expenditure at secondary 21.9 7.0 4.9 35.3

level (thousand dollars)

Start of compulsory education 6.16 0.66 5.00 7.00

GDP per capita (thousand dollars) 16142 13533 326 54178

This content downloaded from

Intercept 8.343*** .132 8.417*** .139

Mother's education (ISCED l)a -0.055*** .016 -0.064*** .016

Mother's education ISCED 2)a -0.038*** .011 -0.041*** .010

Mother's education (ISCED 5a, 6)a 0.128*** .01 1 0.141*** .010

Mother's education (ISCED 5b)a -0.037*** .010 -0.021* .010

School mean occupational status deviation 0.043*** .002 0.041*** .002

Proportion of schools requiring academic -0.024 .046

This content downloaded from

Intercept 9.123*** .233 9.136*** .243

Mother's education (none)3 0.088*** .025 0.081 *** .024

Mother's education (ISCED l)a -0.041* .016 -0.049** .016

School teacher quality deviation 0.119* .049

Proportion of schools requiring academic 0.094 .049

history for admission

Average school teacher quality -0.166*** .031 -0.127*** .034

Per-pupil expenditure (thousands of dollars) 0.003 .002 0.004* .002

Start of compulsory education (years) 0.005 .012 -0.001 .015

Log likelihood -3086398 -3086165

This content downloaded from

This content downloaded from

This content downloaded from

-0.5 -0.4 —0.3 -0.2

Contrtoutton of Intensity of Schooling

Figure 2. Predicted contributio

between background charac

This content downloaded from

Intercept 8.343* .132 8.417* .139

Mother's education (ISCED l)a -0.055* .016 -0.064* .016

Mother's education ISCED 2)a -0.038* .011 -0.041* .010

Mother's education (ISCED 5a, 6)a 0.128* .01 1 0.141* .010

School mean occupational status deviation 0.043* .002 0.041* .002

Intercept 9.123* .233 9.136* .243

Mother's education (none)3 0.088* .025 0.081 * .024

Average school teacher quality -0.166* .031 -0.127* .034