International Journal of Educational Research 109 (2021) 101808

Contents lists available at ScienceDirect

International Journal of Educational Research

journal homepage: www.elsevier.com/locate/ijedures

Does Inequality Stay in the Shadow? The Cross-national Effects of

Educational Equity on Private Tutoring and Student Achievement
Alexander W. Wiseman *
Texas Tech University, College of Education, Box 41071, Lubbock, TX 79409-1071, USA

A R T I C L E I N F O A B S T R A C T

Keywords: By using multiple regression and hierarchical linear modeling, this study investigates shadow
shadow education education across and within national educational systems in order to estimate the effect of
private tutoring educational equity on the relationship between private tutoring and student achievement. Data
educational equity
are from the Programme for International Student Assessment (PISA) 2012 and supplemented by
student achievement
PISA 2012
World Bank data. Results indicate students’ participation in three types of private tutoring
(commercial company, personal, and family tutoring) are negatively related to student achieve­
ment in the cross-national sample. Evidence suggests, however, that private tutoring may
differently associate with student achievement depending on student- and nation-level educa­
tional equity factors and contexts. Finally, the validity of private tutoring as a measure of shadow
education is deconstructed given these results.

In many educational systems worldwide, the demand for private tutoring is often due to high stakes examination pressures, cre­
dentialing inflation, and unique cultural expectations (Kwok, 2004). Stevenson and Baker (1992) first defined shadow education as
educational activities happening outside of formal schooling, which mimic mainstream formal education and are primarily dedicated
to improving individual student achievement, often in preparation for examinations in high stakes educational systems. This is
consistent with what Schaub et al (2020) note regarding private investment in education; it values education as an essential component
of “the well-being of the ever more elaborated individual” (p. 2). But, shadow education, as private tutoring is often labeled, has grown
far beyond East Asian and other high stakes systems to become an established global phenomenon.
Although neither conceptually nor technically the same phenomenon, private tutoring, as one example of both shadow education
and private investment in education, is often interpreted to equate with these two much broader constructs rather than serve as an
example of each (Schaub et al, 2020). For example, Bray (1999, 2009, 2013, 2014; Liu & Bray, 2020) often uses the term shadow
education in reference to supplemental, privately-paid, academic lessons outside of school. Many researchers continue to use the
metaphor of shadow education even when speaking only of private tutoring because private tutoring, as a prime example of shadow
education, mimics the mainstream school system (e.g., Bray & Lykins, 2012; Bray & Kabakhidze, 2014; Byun, 2014; Choi, 2012; Choi &
Park, 2016; Lee & Shouse, 2011; Zhang & Bray, 2020).
Despite the fact that very few empirical studies have shown any direct effects of shadow education, broadly speaking, researchers
have been highly active in critiquing the inconsistent impact of private tutoring on student achievement ((Zhao & Wiseman, 2016)).
For example, Byun (2014) found that cram schools, private tutors, correspondence courses, and internet tutoring services had mixed
effects on student achievement with cram schools having a positive effect on student achievement, while other forms of private

* Corresponding author:
E-mail address: alexander.wiseman@ttu.edu.

https://doi.org/10.1016/j.ijer.2021.101808
Received 15 December 2020; Received in revised form 11 May 2021; Accepted 12 May 2021
Available online 4 June 2021
0883-0355/© 2021 Elsevier Ltd. All rights reserved.
A.W. Wiseman International Journal of Educational Research 109 (2021) 101808

tutoring having little effect. Others also found varying effects of shadow education depending on the type of private tutoring students
participate in along with the effects of student selection bias (Choi & Park, 2016; Dang & Rogers, 2008). Instead, variation in the effects
of what researchers have perhaps misnamed “shadow education” may instead be due to variation in the measurement of participation
in private tutoring and the confounding contextual effects of educational equity on access to or participation in private tutoring rather
than the broader institutional phenomenon of shadow education (Zhang & Liu, 2016; (Zhao & Wiseman, 2016)).

1. Shadow Education and Educational Equity

Shadow education is a form of outside-school learning activity that mimics the mainstream school educational system, and private
tutoring is its most frequent form ((Baker, Akiba, LeTendre, & Wiseman, 2001)). Bray and Kwok (2003) assert that shadow education
can be framed in terms of privateness, financial-gain, and advantaged additions to the provision of mainstream schooling, which is
another term for the formal mass education systems common around the world. This framework, however, does not fully realize the
many ways in which shadow education is a parallel institution to formal, mass education systems. Assumptions that shadow education
is the same as private tutoring are pervasive in the literature, and especially in the critical- or conflict-oriented tradition that examines
how the privatization of supplementary educational activities, which includes many types of private tutoring, reinforces social agency,
political power, and economic stratification.
Shadow education mimics those parts of the regular educational system that are the most normative and, therefore, legitimized.
This has resulted in the global institutionalization of shadow education alongside mass education systems worldwide. As a result,
shadow education also has significant symbolic value, which suggests that its effects are both actual and perceived (Grodsky, 2010).
Shadow education has been institutionalized in a schooled society (Baker, 2020; Mori & Baker, 2010). World institutions, including
education, have been highly rationalized, which creates “rationalized myths” regarding what education is and what it can do (Caruso,
2008). These myths become legitimized and taken-for-granted among populations that have passed through mass education systems
regardless of their veracity (Aurini, 2006). The shared expectations that higher performance in formal mass education systems pro­
vides opportunities for social and economic mobility among individual participants has motivated many families to privately purchase
an extensive amount of outside-of-school tutoring to help their children perform better within school.
Some evidence shows that shadow education may increase the educational inequality gaps between students from lower-income
families and those from higher-income families, and between high-achievers and low-achievers, because of the extra educational
opportunities (Bray, 2013; Bray et al, 2014; Choi & Park, 2016). With both the perceived and experienced pressure of high-stake
examinations, shadow education in the form of private tutoring has grown most rapidly. Student socioeconomic background,
school resources, and school quality demonstrate differing relationships with indicators of private tutoring.

1.1. Private Tutoring

Most of the evidence for the expansion of a shadow education system worldwide is specifically related to paid private tutoring
options (Bray, 2013; Bray & Lykins, 2012; Liu & Bray, 2020). Special attention among researchers on East Asian systems provides a
snapshot of the global spread of private tutoring. For example, in mainland China, about 65.6% and 53.5% students in lower and upper
secondary school, respectively, participated in private tutoring in 2006 (Xue & Ding, 2009). In South Korea in 2009, around 72.5% of
middle school students participated in private tutoring, while the proportion for elementary school students was significantly higher at
87.9% (Kim & Lee, 2010). In Japan in 2011, 65.2% of junior secondary students participated in private tutoring served by juku (Bray,
2013). As this evidence suggests, the shadow education system of private tutoring is a heavily institutionalized component of the
educational landscape in East Asian educational systems, as it is in much of the world.
Shadow education broadly-speaking includes several types of private tutoring. Research identifies four types of private tutoring:
commercial tutoring, personal tutoring, family tutoring, and online tutoring. Private tutoring is often provided by commercial tutoring
companies (e.g., Buxiban in Mainland China, juku in Japan, and hagwons in South Korea) (Bray & Lykens, 2012; Kim, 2016; Ozaki,
2015; Harnish, 1994). Students can choose small, medium, or large classes in tutoring companies, which focus largely on entrance or
other high-stakes examination preparation (Kwok, 2010). Personal tutoring occurs when students are tutored one-on-one by individual
tutors either at the tutors’ or students’ homes in order to meet individual students’ needs (Tansel & Bircan, 2006). Personal tutors are
often either schoolteachers supplementing their regular income or people with subject matter expertise but without teacher creden­
tials, such as college students. Private tutoring can be provided by family members (Park et al, 2011), although this type of private
tutoring has been largely ignored by researchers, possibly because there is not a direct cost involved or because family tutoring is also a
form of parental involvement (see, for example, Bray, 1999; Stevenson & Baker, 1992). A final type of private tutoring is online
tutoring, where students receive tutoring through the internet (Bray, 2006; Burch, Good, & Heinrich, 2016).

1.2. Student Achievement

The relationship between private tutoring and student achievement also varies by context. For example, in South Korea, evidence
shows that private tutoring was positively related to student achievement in middle school and, specifically, math tutoring was
beneficial for high school students in improving their math achievement on the university entrance examination (Lee, 2013). On the
contrary, private tutoring is also negatively associated with student achievement. In Singapore, students with home tutors tend to have
lower academic performance (Cheo & Quah, 2005). Also, some studies indicate private tutoring is a statistically non-significant
predictor of student achievement (Ha & Harpham, 2005; Kuan, 2011; Zhang, 2013).

2
A.W. Wiseman International Journal of Educational Research 109 (2021) 101808

Given the seemingly dichotomous relationship between private tutoring and student achievement, private tutoring may be cate­
gorized as enrichment when student achievement is above average or rising and remediation when student achievement is below
average or falling. For example, students with academic advantages may seek to capitalize on their abilities in order to be more
competitive on high-stakes tests (Lee, 2007). In contrast, private tutoring for remediation is characterized by tutoring meant to assist
students who struggle to keep up with mainstream academic expectations (Baker et al, 2001; Heinrich, Meyer, & Whitten, 2010). Of
course, these dual categories of private tutoring (for remediation and enrichment) vary between nations (Baker & LeTendre, 2005). For
example, evidence suggests that private tutoring in the United States is often focused on the remediation needs of lower achieving
students (Lee, 2007).
Overall, prior research finds inconsistencies in the relationship between private tutoring and student achievement. Research
suggests these inconsistencies are the result of many different factors, including selection bias (Dang & Rogers, 2008), students’
characteristics (Choi & Park, 2016), and data sources (Zhang & Liu, 2016), while much less evidence focuses on the types of private
tutoring as both an independent effect on student achievement as well as a product of the educational equity context. One exception is
Byun’s (2014) study, which noted different effects of private tutoring types on student achievement. Unfortunately, Byun’s (2014)
results were not generalizable (Bray, 2014).

1.3. Educational Equity

Private tutoring may disadvantage poorer households compared to richer ones (Burch, Steinberg, & Donovan, 2007; Dang &
Rogers, 2005). Some evidence, for example, suggests that a hidden marketization of education maintains social inequality (Bray,
2009). Evidence also suggests that the relationship between private tutoring and mass schooling in three East Asian educational
systems (Japan, South Korea, and Cambodia) is biased in favor of private tutoring receiving more of the available resources, and that
government intervention may be needed to promote equal opportunity of education (Dawson, 2010). Moreover, private tutoring has
been shown to widen the achievement gap between high and low achievers (Lee, 2013).
Socioeconomic status (SES) plays a significant role in private tutoring participation because its cost is typically borne by families. In
fact, students from families with higher socioeconomic standing have consistently been more likely to participate in private tutoring
(Stevenson & Baker, 1992; Bray et al, 2011). Parental education and household income were positively related to shadow education
across all different school stages in South Korea (Choi, 2012). Similarly, higher socioeconomic status households tend to spend more on
private tutoring both in Hong Kong and Macao than other cities in mainland China (Kwok, 2001), which suggests that in higher SES
communities the costs of private tutoring increase.
Evidence also suggests that the relationship between school quality and the demand for private tutoring varies by nation. For
example, a negative relationship exists between school quality and the demand for private tutoring in Armenia and Georgia, while
students in better quality schools in Japan, South Korea and Hong Kong are more likely to receive private tutoring (Bray et al, 2014). In
contrast, other research finds that students participate in private tutoring partly because of the poor quality of the schools, even in
systems where other evidence shows that students at higher quality schools have more access to private tutoring. Higher
student-teacher ratios and smaller schools also increase the number of students participating in private tutoring. Finally, private
tutoring participation differs by school type with public school students more likely to attend private tutoring than private school
students (Kim & Lee, 2010).

2. Research Questions

The following research questions seek to deconstruct the relationship between private tutoring and student achievement into its
component parts in order to estimate and distinguish the overall effects of educational equity on shadow education participation and
its effect on student learning and performance. To investigate the relationship between private tutoring and student achievement with
a generalizable sample, the following research question (RQ1) is first investigated:

RQ1: How are unique types of private tutoring related to student achievement in diverse subject areas?

This first research question goes beyond a simple overall association to ask how different types of private tutoring relate to student
achievement in different subject areas. The types of private tutoring to be investigated include personal tutoring, commercial company
tutoring, and family tutoring. The diverse subject areas include mathematics, science, and reading.
Also, private tutoring has chiefly been examined as a way to highlight the unequal opportunities that private tutoring is said to
bring where those with socio-economic advantages are benefitted and those without are not ((Zhao & Wiseman, 2016)). Hence, this
study aims to analyze private tutoring cross-nationally and within national educational system groupings in order to examine the
relationships between private tutoring and student achievement while also estimating the effect that educational equity has on this
relationship. To do this, a second research question (RQ2) asks:

RQ2: Does student socioeconomic status associate with private tutoring differently than school measures of educational equity?

In other words, can organizational level measures of educational equity, such as school quality, mediate or override the effects of
student-level socioeconomic status on private tutoring participation? This question builds upon the foundation of research suggesting
that privatization of any aspect of schooling, including forms of shadow education such as private tutoring, reduces educational equity

3
A.W. Wiseman International Journal of Educational Research 109 (2021) 101808

overall or deepens traditionally-defined social, political, and economic stratification.

Educational equity is a product of multiple, and often overlapping, factors, at the individual, organizational, and system levels. At
the individual level, the type of private tutoring and students’ background characteristics may influence their equitable access to
education, which is often reflected in students’ achievement scores (Choi, 2012). At the organizational level, the type of school and
school quality may affect students’ equitable access to education, which may also affect the relationship between students’
achievement and participation in private tutoring (Aurini, Davies, & Dierkes, 2013). And, finally, the socio-cultural region and eco­
nomic status of national educational systems may influence the relationship between students’ participation in private tutoring and
their academic achievement (Baker et al, 2001; Baker & LeTendre, 2005; Anderson, Domosh, Pile, & Thrift, 2002). This multilevel
effect model suggests that educational equity may have a significant effect on the relationship between shadow education (oper­
ationalized as private tutoring) and student learning (operationalized as student achievement scores).
After examining the effects of educational equity on both private tutoring and student achievement, the final research question
(RQ3) asks:

RQ3: How does the effect of educational equity on the relationship between private tutoring and student achievement vary by type
of private tutoring and level of status or quality?

By examining the intervening or mediating effect of educational equity indicators at the individual student and school levels, there
will be empirical evidence either supporting the assertion by some that private tutoring is essentially inequitable and test the degree to
which private tutoring is a meaningful measure of shadow education as a whole. To conduct this analysis, a multivariate and multilevel
approach is necessary.

3. Methods

3.1. Data & Sample

Data come from the Programme for International Student Assessment (PISA), which is an international assessment measuring 15-
year-old students’ mathematics, science, and reading literacy worldwide every three years since 2000. The 2012 PISA cycle repre­
sented countries and regions from around the world, including surveys of students, parents, teachers, and school principals or head
administrators (OECD, 2013). PISA 2012 data from the student and school surveys supplemented by World Bank nation-level data on
regional and economic groupings (World Bank, 2020) comprise the data included in the analyses below. Data from the PISA 2012 cycle
was specifically chosen because of the unique international survey data available on private tutoring, which was not included in the
2015 or 2018 cycles. The PISA 2012 dataset examined here includes a total representative sample of 248,308 students in 16,823
schools in 65 national educational systems, which expands to 83 separate educational administrative systems. Nation-level World Bank

Table 1
Descriptive Statistics for Student Achievement, Private Tutoring, and Equity Indicators across All Participating Educational Systems (PISA 2012).
N Mean SD Min Max

Student Achievement
Student Mathematics Achievement (MATH) 248308 473.96 89.44 47.60 962.23
Student Science Achievement (SCIE) 248308 479.22 88.71 22.42 926.56
Student Reading Achievement (READ) 248308 474.43 92.15 0.08 904.80
Private Tutoring Types
Commercial Company Tutoring (COMT) 248308 0.88 2.31 0.00 30.00
Personal Tutoring (PERT) 248308 0.90 2.07 0.00 30.00
Family Tutoring (FAMT) 248308 1.13 2.25 0.00 30.00
Student Socioeconomic Status
Home Possessions Index (HOMEPOS) 248308 -0.26 1.12 -6.88 4.15
Parent Highest Years of Education (PARED) 248308 13.16 3.28 3.00 18.00
Student Background
Female Student (FEMALE) 248308 0.51 0.50 0.00 1.00
Native Citizen (NATIVE) 248308 0.89 0.32 0.00 1.00
School Quality
Private School (PRIVATE) 16823 0.17 0.38 0.00 1.00
Teacher Shortage Index (TCSHORT) 16823 0.03 1.05 -1.09 3.60
Quality of School Educational Resources Index (SCMATEDU) 16823 -0.14 1.11 -3.59 1.98
National Economic Indicators
Economic Categories (ECONCAT) 83 2.71 0.51 1.00 3.00
Gini Coefficient (GINI2012) 83 35.52 5.98 25.60 53.50
Regional Groupings
Europe & Central Asica (ECS) 83 0.63 0.49 0.00 1.00
Latin America & Caribbean (LCN) 83 0.11 0.31 0.00 1.00
East Asia & Pacific (EAS) 83 0.16 0.37 0.00 1.00
North America 83 0.02 0.15 0.00 1.00
Middle East & North Africa (MEA) 83 0.08 0.28 0.00 1.00

4
A.W. Wiseman International Journal of Educational Research 109 (2021) 101808

data was selected specifically for the 2012 year or the nearest year with available data to measure national socio-cultural and economic
factors (World Bank, 2020).

3.2. Measures

Student achievement measures serve as the dependent variable in all analyses. Student achievement in PISA 2012 is measured
uniquely for 15-year-old students in each national educational system in three different subject areas, including mathematics, science,
and reading. Independent variables include three types of private tutoring (commercial tutoring, personal tutoring, and family
tutoring), student socioeconomic status (home possessions and parental education in years) and school quality (school type, teacher
shortage, and school educational resources). Control variables include student gender and immigrant status. Table 1 provides
descriptive information of all the key variables across all participating educational systems in PISA 2012.

3.2.1. Dependent Variables

Student achievement. PISA 2012 data provides five plausible values that were multiply-imputed for the 15-year-old students’ overall
mathematics, science, and reading achievement (OECD, 2013). In this research, all five plausible values were used with the consid­
eration of total student weights and 80 replicate weights for mathematics (MATH, X=473.96; SD=89.44), science (SCIE, X=479.22;
SD=88.71), and reading (READ, X=474.43; SD=92.15).

3.2.2. Independent Variables

Private Tutoring Types. In PISA 2012, students were asked, “Thinking about all school subjects: on average, how many hours do you spend
each week on the following?” Item 1, “Work with a personal <tutor> (whether paid or not)”. Item 2, “Attend out of school classes organized
by a commercial company, and paid for by your parents”. Item 3, “Study with a parent or other family member”. The first type of private
tutoring is “commercial company tutoring” (COMT, X=0.88; SD=2.31). The second type of private tutoring is “personal tutoring”
(PERT, X=0.90; SD=2.07). And, the third type of private tutoring is “family tutoring” (FAMT, X=1.13; SD=2.25). These three variables
are all continuous variables measured as student-reported hours per week.
Student Socioeconomic Status. PISA 2012 provides two measures of student socioeconomic status. The first is an index of home
possessions for each student (HOMEPOS, X = -0.26; SD=1.12), which is a summary of four indices of household possessions, including
family wealth possessions, cultural possessions, home educational resources, and home possessions (OECD, 2013). The second indi­
cator of student socioeconomic status is parental education background derived from “parent highest years of education” (PARED,
X=13.16; SD=3.28). It is a continuous variable with a range from 3 to 16 based on the International Standard Classification of Ed­
ucation (ISCED) (OECD, 2013).
Student Background Indicators. Student gender (FEMALE, X=0.51; SD=0.50) is a dichotomous variable, coded as “1 = Female”, and
“0 = Male”. Student immigration status (NATIVE, X=0.89; SD=0.32) is also a dichotomous variable, which includes “1= Native” and
“0 = First or Second Generation Immigrant”.
School Quality. School type is a dichotomous variable that divides schools into two types, “1 = private”, and “0 = public” (PRIVATE,
X=0.17; SD=0.38). An index of teacher shortage (TCSHORT, X=0.03; SD=1.05) was derived from four items measuring school
principals’ perceptions of potential factors hindering instruction at their school (OECD, 2013). Higher values on this index indicate
school principals’ reports of higher teacher shortage at a school. An index of the quality of school educational resources (SCMATEDU,
X = -0.14; SD=1.11) was derived from six items measuring school principals’ perceptions of potential factors hindering instruction at
their school (OECD, 2013). As all items were inverted for scaling, higher values on this index indicate better quality of educational
resources.
At the educational system (i.e., nation) level, there are variables related to a nation’s economic category, level of income inequality,
and regional grouping. The economic category and level of income inequality are both educational equity indicators at the national
educational system level, while the regional grouping is an indicator of social and cultural affinities (Anderson, Domosh, Pile, & Thrift,
2002). National economic categories (ECONCAT, X=2.71; SD=0.51) come from the World Bank (2020) country database and reflect
the following ordinal categories: low income (0), lower middle income (1), upper middle income (2), and high income (3). The
measure of income inequality is the Gini coefficient (GINI2012, X=35.52; SD=5.98), which shows economic (i.e., income) inequality
by measuring the extent to which the distribution of income among individuals or households within an economy deviates from a
perfectly equal distribution (Chiu, 2010). The World Bank data did not include Gini coefficients for all participating countries in 2012,
so the nearest year’s data was substituted.
Regional grouping data from the World Bank (2020) was also included at the national level. These categories include: Europe &
Central Asia (ECS, X=0.63; SD=0.49), Latin America & Caribbean (LCN, X=0.11; SD=0.31), East Asia & Pacific (EAS, X=0.16;
SD=0.37), and Middle East & North Africa (MEA, X=0.08; SD=0.28). North America (NAC, X=0.02; SD=0.15) was excluded from the
analysis to allow for variation. All regional category indicators are dummy coded as 1 (yes) or 0 (no).

3.3. Analytical Approach

PISA 2012 provides data on 15-year old students’ use of private tutoring using nationally representative and uniquely-diverse
samples, enabling the analysis of one type of shadow education (i.e., private tutoring) cross-nationally. To answer the research

5
A.W. Wiseman International Journal of Educational Research 109 (2021) 101808

questions, three analyses were conducted in this study.

First, correlations were run for all private tutoring and educational equity variables at the student and school levels. The purpose of
the correlation was to estimate the bivariate association between private tutoring types and estimates of educational equity. Second,
multiple linear regression analyses were conducted with student achievement as the dependent variable and considering student total
weight and the 80 replicate weights of the five plausible values for each of the three subjects (mathematics, science, and reading). The
purpose of this regression analysis was to examine the relationship between each type of private tutoring and student math
achievement, which addresses the first research question (RQ1). In this multiple regression, student gender (FEMALE), immigrant
status (NATIVE), and socioeconomic status (HOMEPOS & PARED) were all included as student background variables. School quality
variables (PRIVATE, TCSHORT, and SCMATEDU) were included as independent variables. And, personal tutoring time (PERT),
commercial company tutoring time (COMT), and family tutoring time (FAMT) were also included as independent variables. The
multivariate linear regression model equation is as follows:
Y = β1 (FEMALE) + β2 (NATIVE) + β3 (HOMEPOS) + β4 (PARED) + β5 (PRIVATE) + β6 (TCSHORT)
+β7 (SCMATEDU) + β8 (PERT) + β9 (COMT) + β10 (FAMT) + e

where Y represents the dependent variable of student achievement in mathematics (MATH), science (SCIE), or reading (READ), and e
represents the error term.
To address research question two (RQ2) and three (RQ3), a three-level hierarchical linear model (HLM) analysis examines the effect
of educational equity on the relationship between private tutoring type and student achievement. HLM is particularly relevant because
PISA is a nested dataset, with PISA 2012 student-level data nested in school-level data, which is then nested in the nation-level data
from the World Bank (2020). Restricted Maximum Likelihood (REML) is used as the estimation method because the outcome variable
is a continuous variable. REML is a traditional method which gives unbiased estimates of variance components as they estimate all
parameters simultaneously (Raudenbush & Bryk, 2002).
The first level equation (eq.1) estimates the influence of private tutoring types (π1jk, π2jk, and π3jk), student socioeconomic status
(π4jk and π5jk), and, for control of background effects, indicators of the student’s individual background (π 6jk and π7jk), as follows:
( ) ( ) ( ) ( )
Yijk =( π0jk + π1jk) COMT( ijk + π2jk PERT
) ijk (+ π3jk FAMT
) ijk + π4jk HOMEPOSijk (1)
+π5jk PAREDijk + π6jk FEMALEijk + π7jk NATIVEijk + eijk ,

where Yijk is the dependent variable (e.g., student mathematics, science, or reading achievement score) for the ith student within school
j within system k. π0jk is an estimate of the adjusted mean dependent variable for school j within system k, and eijk is a student level
residual. By assumption, E(eijk) = 0 and Var(eijk) = σ2.
For the school level equation (eq.2), the effect of private tutoring on the mean dependent variable (πxjk) is modeled as a function of
school quality (βx1k, βx2k, and βx3k) as follows,
( ) ( ) ( )
πxjk = βx0k + βx1k PRIVATEjk + βx2k TCSHORTjk + βx3k SCMATEDUjk + r0jk (2)

where r0jk is a school level residual. By assumption, E(r0jk) = 0 and Var(r0jk) = τ000. This cross-level interaction effect provides an
estimate of the impact of educational equity on the relationship between private tutoring and student achievement.
Finally, a system level equation (eq.3) was estimated where the effect of school quality on the relationship between private tutoring
type and student achievement is modeled as a function of national economic indicators and regional groupings as follows,
βx0k = γx00 + γx01 (ECONCATk ) + γx02 (GINI2012k ) + γ x03 (ECSk ) + γ x04 (LCNk )
(3)
+γx05 (EASk ) + γ x06 (MEAk ) + u00k

where βx0k is an estimate of the adjusted effect of private tutoring type on student achievement, and u00k is a national system level
residual.

Table 2
Correlation of Private Tutoring Types and Educational Equity Indicators across All Participating Educational Systems (PISA 2012).
Private Tutoring Types
Commercial Company Tutoring Personal Tutoring Family Tutoring
(COMT) (PERT) (FAMT)

Student Socioeconomic Status

Parent Highest Years of Education (PARED) -.017** -.013** -.019**
Home Possessions Index (HOMEPOS) -.042** -.021** -.059**
School Quality
Private School (PRIVATE) -.007** -1.0E-03 -.039**
Teacher Shortage Index (TCSHORT) .020** .007** .043**
Quality of School Educational Resources Index -.047** -.040** -.072**
(SCMATEDU)

** p<.01
N=269080

6
A.W. Wiseman International Journal of Educational Research 109 (2021) 101808

3.4. Limitations

The measurement of private tutoring by type was based on the suggestions of Bray and Kobakhidze (2014), yet there remain some
potential limitations. For example, the three types of private tutoring are not specific to the different academic subjects, so alignment of
tutoring to subject is not possible. In the PISA 2012 student questionnaire, there are items related to out-of-school time lessons in
mathematics, science, and reading, which were not included in these analyses, but could also refer to private tutoring participation.
Also, participation data for other types of private tutoring, such as online tutoring, are not available from the PISA 2012 data. This
analysis is also strictly cross-national, and although there is variation by school and national system, there is a specific emphasis on the
cross-national trends rather than intra-national results.

4. Research Findings

Table 2 provides the weighted correlation coefficients between private tutoring types and educational equity indicators across all
participating educational systems. Blanket statements have been made about the use of private tutoring being for advantaged students
predominantly and, consequently, reproducing inequalities rather than closing achievement gaps (Bray, 2006, 2009). If that common
assertion is true, then there would be a positive and significant global association between all types of private tutoring and educational
equity indicators. Instead, as Table 2 shows, cross-national data on private tutoring shows a negative association in all but one
instance. This suggests that private tutoring is used more often worldwide among those whose educational equity indicators are lower,
both at the individual level as well as collectively at the school level.
Each of the correlation coefficients suggests that higher student socioeconomic status or school quality, which are both measures of
educational equity, associates with lower rates of private tutoring use, and that this negative association occurs across all three types of
private tutoring (i.e., commercial company, personal, and family tutoring). Yet, we know from much prior research that higher so­
cioeconomic status and school quality indicators consistently associate with higher student achievement scores (Perry & McConney,
2010). The data also suggest that some private tutoring is a supplement for or response to lower school quality, which confirms
previous research on supplemental schooling among parents with access to additional resources (Song, Park, & Sang, 2013).
Multiple regression results in Table 3 indicate that all three types of private tutoring are negative predictors of student achievement
in all three subject areas. This linear regression model was estimated using all five plausible values of student achievement in each
subject area and the appropriate weights, based on OECD (2013) recommendations. Commercial company tutoring (COMT) has a
negative effect on student achievement in mathematics (b = -2.42, p<.001), science (b = -2.90, p<.001), and reading (b = -2.98,
p<.001). Personal tutoring (PERT) has a negative effect on student achievement in mathematics (b = -5.33, p<.001), science (b =
-5.09, p<.001), and reading (b = -4.79, p<.001). And, family tutoring also has a negative effect on student achievement in

Table 3
Linear Regression of Private Tutoring Types and Educational Equity on Student Achievement across All Participating Educational Systems (PISA
2012).
Student Achievement
Mathematics Science Reading

Private Tutoring Types

Commercial Company Tutoring (COMT) -2.42 -2.90 -2.98

(0.30) (0.29) (0.33)

Personal Tutoring (PERT) -5.33 -5.09 -4.79
(0.23) (0.25) (0.24)
Family Tutoring (FAMT) -4.59 -4.19 -3.92
(0.15) (0.14) (0.14)
Student Socioeconomic Status
Home Possessions Index (HOMEPOS) 18.48 16.70 16.76
(0.31) (0.31) (0.32)
Parent Highest Years of Education (PARED) 5.44 5.42 5.46
(0.11) (0.12) (0.12)
School Quality
Private School (PRIVATE) 11.01 9.47 13.55
(1.73) (1.78) (1.79)
Teacher Shortage Index (TCSHORT) -5.63 -5.05 -4.89
(0.56) (0.54) (0.57)
Quality of School Educational Resources Index (SCMATEDU) 2.08 2.25 2.34
(0.50) (0.49) (0.50)
Student Background
Female Student (FEMALE) -8.5 0.57 36.46
(0.51) (0.51) (0.52)
Native Citizen (NATIVE) 12.89 19.25 14.16
(2.59) (2.39) (2.62)
Standard errors in parentheses.
p <.001 for all coefficients.

7
A.W. Wiseman International Journal of Educational Research 109 (2021) 101808

mathematics (b = -4.59, p<.001), science (b = -4.19, p<.001), and reading (b = -3.92, p<.001). This negative effect of private tutoring
across all types suggests that cross-nationally private tutoring is a remediation tool rather than an enrichment mechanism (Baker et al,
2001).
Linear multiple regression of the cross-national PISA 2012 data also show that both student and school indicators of educational
equity are positively associated with student achievement in all three subject areas. In other words, as student socioeconomic status
rises, student achievement also significantly rises. And, as school quality indicators rise, student achievement also significant rises. In
other words, even though educational equity indicators are consistently negatively associated with private tutoring participation (see
Table 2), they are consistently, positively, and statistically significant predictors of student achievement in all subjects. These results
from the correlations and linear regressions concerning the relationships between private tutoring types, educational equity indicators,
and student achievement suggest that there may potentially be a significant effect of educational equity on the relationship between
private tutoring participation and student achievement.
After analyzing the relationship between three types of private tutoring participation and educational equity with student
achievement by both correlation and linear multiple regression, an estimation of the effect of educational equity on the relationship
between private tutoring participation and student achievement was conducted using HLM. First, tests of the fully unconditional
models found that a significant proportion of the variance in student achievement occurred between schools; thus, hierarchical linear
modeling is an appropriate estimation method.

Table 4
HLM Analysis of Private Tutoring Types and Educational Equity on Student Mathematics Achievement across All Participating Educational Systems
(PISA 2012).
Student Mathematics Achievement
Model 1 Model 2 Model 2 Model 2 Model 3 Model 3 Model 3
COMT PERT FAMT COMT PERT FAMT

Fixed Effects
Private Tutoring Types
Commercial Company Tutoring (COMT) -0.20 -0.16 -0.19 -0.22 -8.62* -0.19 -0.37
(0.49) (0.67) (0.49) (0.50) (3.93) (0.51) (0.51)
Personal Tutoring (PERT) -1.97*** -1.98*** -1.95*** -1.98*** -1.95*** -5.94 -2.14***
(0.50) (0.52) (0.40) (0.49) (0.50) (6.23) (0.50)
Family Tutoring (FAMT) -1.56** -1.57** -1.55** -1.83*** -1.59*** -1.58*** -1.82
(0.49) (0.50) (0.49) (0.41) (0.48) (0.48) (3.61)
Student Socioeconomic Status
Home Possessions Index (HOMEPOS) 8.95*** 8.94*** 8.95*** 8.98*** 8.91*** 8.95*** 9.11***
(2.08) (2.09) (2.07) (2.11) (2.08) (2.09) (2.13)
Parent Highest Years of Education (PARED) 1.37*** 1.37*** 1.38*** 1.37*** 1.39*** 1.38*** 1.37***
(0.33) (0.32) (0.33) (0.32) (0.32) (0.32) (0.33)
Student Background
Female Student (FEMALE) -11.28*** -11.25*** -11.30*** -11.24*** -11.30*** -11.35*** -11.07***
(1.88) (1.87) (1.88) (1.90) (1.84) (1.89) (1.93)
Native Citizen (NATIVE) 12.10* 12.17* 12.12* 12.03* 11.85* 11.84* 11.85*
(5.19) (5.22) (5.18) (5.23) (5.20) (5.21) (5.20)
School Quality
Private School (PRIVATE) -0.12 -0.62 0.90 -0.93* -1.59** -0.59
(0.63) (1.02) (0.88) (0.40) (0.59) (0.50)
Teacher Shortage Index (TCSHORT) 0.34* 0.30 0.35 0.21 0.16 0.16
(0.17) (0.24) (0.21) (0.16) (0.27) (0.19)
Quality of School Educational Resources 0.17 0.03 -0.10 0.28 0.16 0.10
Index (SCMATEDU)
(0.18) (0.20) (0.25) (0.16) (0.21) (0.16)
Economic Categories (ECONCAT) -0.06 -1.23 -2.29***
(0.66) (1.47) (0.56)
Gini Coefficient (GINI2012) . . . . 0.07 0.19 0.12
(0.08) (0.14) (0.07)
Regional Groupings
Europe & Central Asia (ECS) . . . . 5.39*** 0.41 1.31
(0.78) (1.41) (0.71)
Latin American & Caribbean (LCN) . . . . 4.73*** -5.69* -0.94
(1.42) (2.51) (0.98)
East Asia & Pacific (EAS) . . . . 7.96*** -1.09 1.21
(1.25) (3.06) (1.11)
Middle East & North Africa (MEA) . . . . 2.99** -3.20 0.25
(1.14) (1.99) (0.87)
Intercept 432.10*** 432.01*** 432.07*** 432.01*** 431.68*** 432.16*** 432.34***
(14.10) (14.08) (14.09) (14.27) (14.35) (23.18) (14.45)
Random Effects
Level 1 Variance, e 3836.32 3835.41 3834.35 3833.90 3819.58 3809.56 ######
Level 2 Variance, r0 2528.80 2529.85 2537.74 2530.13 2514.42 2525.06 ######
Level 3 Variance, u00 2175.77 2172.66 2169.41 2209.43 2237.07 2269.12 ######

8
A.W. Wiseman International Journal of Educational Research 109 (2021) 101808

Tables 4, 5, and 6 show the cross-national results of HLM analyses of private tutoring types and educational equity on student
achievement in mathematics, science, and reading, respectively. As expected, student socioeconomic status indicators (HOMEPOS and
PARED) were positively and significantly associated with student achievement in every HLM model. Similarly, student background
indicators (FEMALE and NATIVE) results were as expected, although there was some variation in the effect that student gender had on
student achievement, with there being a negative effect for mathematics (see Table 4), a statistically non-significant effect for science
(see Table 5), and a positive effect for reading (see Table 6).
Estimations of the effect of private tutoring participation varied among the different types. Commercial company tutoring (COMT)
had largely a small and statistically non-significant effect on student achievement in every subject, except when educational equity
indicators at the school and system level were regressed on the relationship between commercial company tutoring participation and
student achievement. In each case, when educational equity was regressed on the relationship between commercial company tutoring
participation and student achievement, the effect of commercial company tutoring participation was strong, negative, and significant
in mathematics (b = -8.62, p<.05), science (b = -10.21, p<.05), and reading (b = -13.05, p<.001). Otherwise, there was no effect of
commercial company tutoring on student achievement in any of the estimation models. In other words, when variation at both the
school and system level is allowed, commercial company tutoring participation has no effect on student achievement, except when

Table 5
HLM Analysis of Private Tutoring Types and Educational Equity on Student Science Achievement across All Participating Educational Systems (PISA
2012).
Student Science Achievement
Fixed Effects Model 1 Model 2 Model 2 Model 2 Model 3 Model 3 Model 3
COMT PERT FAMT COMT PERT FAMT

Private Tutoring Types

Commercial Company Tutoring (COMT) -0.78 -0.70 -0.76 -0.79 -10.21* -0.84 -0.91
(0.46) (0.63) (0.45) (0.46) (4.10) (0.48) (0.47)
Personal Tutoring (PERT) -1.83*** -1.83*** -1.81*** -1.83*** -1.82*** -7.27 -1.96***
(0.49) (0.50) (0.48) (0.49) (0.47) (5.07) (0.50)
Family Tutoring (FAMT) -1.58*** -1.59*** -1.57*** -1.68*** -1.59*** -1.62*** -2.16
(0.42) (0.42) (0.42) (0.39) (0.40) (0.41) (3.12)
Student Socioeconomic Status
Home Possessions Index (HOMEPOS) 7.78*** 7.78*** 7.78*** 7.81*** 7.73*** 7.81*** 7.94***
(1.98) (1.99) (1.97) (2.01) (1.98) (1.98) (2.05)
Parent Highest Years of Education (PARED) 1.55*** 1.55*** 1.56*** 1.55*** 1.56*** 1.55*** 1.54***
(0.34) (0.34) (0.34) (0.34) (0.33) (0.33) (0.34)
Student Background
Female Student (FEMALE) -2.47 -2.46 -2.48 -2.44 -2.55 -2.50 -2.25
(1.34) (1.32) (1.34) (1.34) (1.32) (1.35) (1.35)
Native Citizen (NATIVE) 18.00*** 18.07*** 18.05*** 17.95*** 17.83*** 18.03*** 17.78***
(4.74) (4.77) (4.74) (4.78) (4.78) (4.73) (4.78)
School Quality
Private School (PRIVATE) . -0.10 -0.75 0.23 -0.72 -1.38* -1.00*
(0.62) (0.89) (0.72) (0.50) (0.58) (0.49)
Teacher Shortage Index (TCSHORT) . 0.36 0.49* 0.45 0.20 0.53* 0.25
(0.23) (0.24) (0.26) (0.21) (0.26) (0.21)
Quality of School Educational Resources . 0.31 0.11 0.09 0.40 0.12 0.28
Index (SCMATEDU)
(0.22) (0.25) (0.25) (0.21) (0.28) (0.18)
National Economic Indicators
Economic Categories (ECONCAT) . . . . 0.04 -1.15 -1.97***
(0.60) (0.84) (0.48)
Gini Coefficient (GINI2012) . . . . 0.07 0.18 0.09
(0.08) (0.10) (0.06)
Regional Groupings

Europe & Central Asia (ECS) . . . . 8.41*** 2.43 2.61***

(1.16) (1.23) (0.67)
Latin American & Caribbean (LCN) . . . . 7.16*** -1.80 0.11
(1.51) (1.80) (0.66)
East Asia & Pacific (EAS) . . . . 10.11*** 1.55 2.00*
(1.28) (1.80) (0.88)

Middle East & North Africa (MEA) . . . . 5.66*** -0.61 1.08

(1.44) (1.89) (0.90)
Intercept 428.98*** 428.97*** 428.95*** 429.02*** 428.68*** 428.77*** 429.20***
(14.19) (14.12) (14.08) (14.18) (14.36) (13.84) (14.35)
Random Effects
Level 1 Variance, e 3910.63 3909.68 3907.49 3908.52 3891.30 3898.70 3893.61
Level 2 Variance, r0 2489.83 2487.95 2499.47 2495.07 2466.47 2490.14 2486.85
Level 3 Variance, u00 2086.58 2076.79 2078.42 2097.56 2129.23 2141.08 2229.02

9
A.W. Wiseman International Journal of Educational Research 109 (2021) 101808

Table 6
HLM Analysis of Private Tutoring Types and Educational Equity on Student Reading Achievement across All Participating Educational Systems (PISA
2012).
Student Reading Achievement
Fixed Effects Model 1 Model 2 Model 2 Model 2 Model 3 Model 3 Model 3
COMT PERT FAMT COMT PERT FAMT

Private Tutoring Types

Commercial Company Tutoring (COMT) -0.52 -0.69 -0.52 -0.53 -13.05*** -0.63 -0.68
(0.52) (0.67) (0.51) (0.53) (4.03) (0.56) (0.54)
Personal Tutoring (PERT) -1.84** -1.83** -1.87*** -1.86** -1.82** -5.07 -2.02***
(0.63) (0.64) (0.52) (0.61) (0.61) (5.19) (0.61)
Family Tutoring (FAMT) -1.79*** -1.77*** -1.79*** -2.03*** -1.79*** -1.86*** -1.38
(0.54) (0.54) (0.53) (0.52) (0.51) (0.50) (3.75)
Student Socioeconomic Status
Home Possessions Index (HOMEPOS) 7.74*** 7.76*** 7.74*** 7.76*** 7.72*** 7.78*** 7.94***
(1.82) (1.83) (1.83) (1.86) (1.82) (1.84) (1.93)
Parent Highest Years of Education (PARED) 1.44*** 1.44*** 1.44*** 1.44*** 1.46*** 1.44*** 1.43***
(0.38) (0.38) (0.39) (0.38) (0.37) (0.38) (0.38)
Student Background
Female Student (FEMALE) 29.83*** 29.83*** 29.86*** 29.89*** 29.74*** 29.87*** 30.11***
(1.24) (1.25) (1.24) (1.24) (1.27) (1.23) (1.25)
Native Citizen (NATIVE) 13.82* 13.86* 13.88* 13.78* 13.56* 13.86* 13.58*
(5.84) (5.85) (5.84) (5.87) (5.85) (5.75) (5.86)
School Quality
Private School (PRIVATE) . 0.61 0.19 1.14 -0.20 -0.90 -0.39
(0.76) (1.21) (0.85) (0.45) (0.62) (0.53)
Teacher Shortage Index (TCSHORT) . 0.35 0.30 0.49* 0.17 0.35 0.22
(0.21) (0.27) (0.24) (0.18) (0.26) (0.19)
Quality of School Educational Resources . 0.30 0.22 0.18 0.41* 0.29 0.41
Index (SCMATEDU)
(0.21) (0.32) (0.35) (0.17) (0.34) (0.26)
National Economic Indicators
Economic Categories (ECONCAT) . . . . -0.15 -1.69* -2.07***
(0.61) (0.77) (0.65)
Gini Coefficient (GINI2012) . . . . 0.11 0.13 0.05
(0.08) (0.10) (0.07)
Regional Groupings
Europe & Central Asia (ECS) . . . . 8.52*** 3.35* 2.95**
(0.98) (1.31) (0.74)
Latin American & Caribbean (LCN) . . . . 6.57*** -0.60 1.59
(1.55) (1.60) (0.95)

East Asia & Pacific (EAS) . . . . 10.46*** 2.91 3.10*

(1.39) (1.58) (1.22)
Middle East & North Africa (MEA) . . . . 5.60*** 0.31 1.78*
(1.49) (1.63) (0.82)
Intercept 413.41*** 413.49*** 413.42*** 413.43*** 413.06*** 413.13*** 413.79***
(12.27) (12.28) (12.19) (12.31) (12.69) (12.25) (12.63)
Random Effects
Level 1 Variance, e 4133.77 4132.71 4133.30 4130.90 4109.93 4118.58 4110.62
Level 2 Variance, r0 2872.46 2865.38 2868.87 2869.07 2836.21 2850.70 2856.76
Level 3 Variance, u00 1654.63 1651.82 1653.13 1675.73 1711.65 1731.99 1814.32

school and system educational equity indicators are regressed on the private tutoring and student achievement relationship. Then, and
only then, the effect of commercial company tutoring on student achievement is strong, negative, and significant.
Personal tutoring (PERT) and family tutoring (FAMT) were both negative and significant predictors of student achievement in all
subjects, except when school and system level educational equity indicators were regressed on the relationship between private
tutoring participation and student achievement. In other words, the effect of personal tutoring and family tutoring on student
achievement disappears when equity indicators at the school and system level are regressed on the relationship between private
tutoring and student achievement across all subject areas.
Indicators of educational equity at the school level mitigate the relationship between private tutoring participation and student
achievement differently for different combinations of private tutoring types, student achievement subjects, and whether or not system
level equity indicators are also estimated as part of the model. Whether or not a students’ school is a private rather than a public school
(PRIVATE) has a very small or no impact on the relationship between private tutoring participation and student achievement, except
when school-level quality indicators and system level economic and regional groupings are regressed on the relationship of student
mathematics achievement and commercial company tutoring (b = -0.93, p<.05) and personal tutoring (b = -1.59, p<.01) or when
science achievement and personal tutoring (b = -1.38, p<.05) and family tutoring (b = -1.00, p>05). There is no significant effect
when regressed on private tutoring participation and student reading achievement.

10
A.W. Wiseman International Journal of Educational Research 109 (2021) 101808

The teacher shortage index (TCSHORT) is a measure of inequity. Although it does not have a consistent or frequent statistically
significantly effect on the relationship between private tutoring participation and student achievement, when it does post a statistically
significant effect it represents an increase in a negative relationship between private tutoring participation and student achievement.
For example, in Table 4, the teacher shortage index is positive and significant (b = 0.34, p<.05) when regressed on the relationship
between commercial company tutoring (COMT) and student mathematics achievement (MATH). In Table 5, the teacher shortage index
is positive and significant when regressed on the relationship between student science achievement and personal tutoring (b = 0.49,
p<.05; b = 0.53, p<.05). And, in Table 6, the teacher shortage index is positive and significant when regressed on the relationship
between student reading achievement and family tutoring (b = 0.49, p<.05). In each of these instances a positive coefficient for the
teacher shortage index shows an increase in the negative relationship between private tutoring participation and student achievement.
Overall, the effect of the quality of the school educational resources index (SCMATEDU) on the relationship between private
tutoring participation and student achievement is statistically non-significant with only one exception. That exception is when system
level economic indicators and regional groupings are regressed on the relationship between commercial company tutoring partici­
pation and reading achievement. In that situation, the quality of school educational resources index is significant and positive (b =
0.41, p<.05), but is regressed on the effect of commercial company tutoring participation on student reading achievement (b = -13.05,
p<.001), which means that it increases the negative relationship between private tutoring and student achievement. Largely, though,
the school educational resources index is statistically non-significant.
At the system level, there are two indicator types: national economic indicators and regional groupings. National economic in­
dicators consist of World Bank-defined national economic categories (ECONCAT) and the 2012 Gini coefficient (GINI2012). The Gini
coefficient is not statistically significant in any of the estimated HLM models, although it was kept in the analyses because it has been
shown in prior research to be a significant predictor of educational equity, or, rather, inequity (Chiu, 2010). National economic
categories (ECONCAT) was a statistically significant predictor of the relationship between family tutoring and student achievement for
mathematics (b = -2.29, p<.001), science (b = -1.97, p<.001), and reading (b = -2.07, p<.001), which means that it strengthens the
negative effect that family tutoring participation has on student achievement in all subjects. It also had a similar effect of strengthening
the negative effect of personal tutoring on student reading achievement (b = -1.69, p<.05). This suggests that the economic strength of
a national system does not significantly affect the relationship between private tutoring and student achievement, except in the case of
family tutoring.

5. Conclusions and Implications for Shadow Education

Educational policy decisions at the national and international levels are often driven by estimates of the impact particular
educational programs or private investments are likely to have on student learning and performance outcomes (Schaub et al, 2020).
The research reported here shows that although a plethora of research on private tutoring indicates that both direct and indirect costs
of participation restrict access to private tutoring to some and advantages others, there is little evidence in the cross-national data to
support this critique. While local and intra-national data may tell a different story, especially in high stakes educational systems,
cross-national analyses do not suggest that inequities in students’ backgrounds compound or shift from privately held to institutionally
embedded norms or mechanisms.
There is also no evidence to support a conclusion about which type of private tutoring is more effective than the others in raising
student achievement, even though the evidence does clearly show that cross-national participation in private tutoring of all types is
more frequent among students who score lower than others. And, even though remediation both for lagging students and enhancement
for higher performing students does exist, the cross-national evidence is clear that remediation is the dominant approach to private
tutoring participation worldwide. This research also finds that educational equity indicators at the school and system levels have
largely non-significant effects on the relationship between private tutoring and student achievement. Hence, the demand for private
tutoring is not affected by educational equity, or perhaps more specifically educational inequity, as previous research suggested.
The first question (RQ1) addressed the relationship of three types of private tutoring with student math achievement. This study
indicates that commercial company tutoring time is either negatively or not statistically significantly related to student achievement in
the cross-national data, while personal tutoring and family tutoring time are negatively associated with student achievement in almost
every model. Two plausible reasons explain these two different outcomes. The first reason is that more high performing students are
likely to choose commercial company tutoring and more low performance students are likely to choose personal tutoring or receive
help from family members, even though there is not enough evidence of this effect in all systems to create a positive association
between commercial company tutoring participation and student achievement. Commercial company tutoring has tutors who spe­
cifically focus on test skills in examinations. In order to attract more students to study, these commercial tutors are trained to teach
students the test skills that could be successful in high-stake examinations. Therefore, commercial company tutoring may only post a
positive association with student achievement in high stakes systems, but not in all systems when considered collectively. Never­
theless, it is common that commercial tutoring institutions have large class sizes, and teachers cannot fulfill individual needs in a large
class, which disadvantages low performing students. Hence, lower-performing students may prefer personal tutoring or family tutoring
in order to get individualized support. These results also suggest that private tutoring may be mostly used for remediation purposes
worldwide (Baker et al, 2001; Baker & LeTendre, 2005).
The second question (RQ2) asked whether student level indicators of educational equity associate with private tutoring partici­
pation differently than school level equity indicators. Surprisingly, both student and school level indicators of educational equity
negatively associate with all three indicators of private tutoring across all participating educational systems. Despite the symbolic
value of private tutoring, lower income families are more likely to incur the fees of private tutoring or provide it at home themselves.

11
A.W. Wiseman International Journal of Educational Research 109 (2021) 101808

The evidence also suggests that students are more likely to participate in private tutoring when school quality drops, meaning that
there is some evidence of enrichment, but that the bulk of evidence still suggests remediation as the more likely practice.
Research question three (RQ3) asks about the effect of educational equity on the relationship between private tutoring type and
student achievement. Results of the HLM analyses suggest that in almost every model, the student level indicators of educational equity
predicted student achievement as expected, but that the school and system level equity indicators tended to either not affect the
relationship between private tutoring and student achievement at all, or, when they did, they did so irregularly.
As a result, the quality of mainstream schools or formal mass education systems is not a key factor influencing participation in
private tutoring, but students’ socioeconomic status (SES) has a key effect across all educational systems. This result does not support
previous research findings that private tutoring exacerbates educational inequality in society (e.g., Bray 2009) because students in
higher income families do not gain more educational resources within schools and are not necessarily supplementing their students’
education out of school at any higher rate than lower income families.
Stevenson and Baker (1992) coined the term shadow education because private tutoring mimics formal mass school (Baker, 2020).
This study shows that educational equity at the school and system levels does not vary in its impact on the relationship between private
tutoring participation and student achievement except in unique situations. This suggests that private tutoring attracts both high
performers and low performers, and that parents and students do not make choices about private tutoring based on the effectiveness or
value of private tutoring. Instead, they participate because of a shared expectation that they and others will all participate in it. The
symbolic value of private tutoring has gradually surpassed its academic value. As Mori and Baker (2010) suggest, it is highly possible
that private tutoring will emerge from the shadow and become a fully legitimized part of formal mass education. But, there is no
evidence that social or economic inequalities that may factor into a students’ access to or participation in private tutoring will move
out of the shadow and into the formal education system.
Private tutoring is the subject of many intense scholarly critiques related to educational equity (e.g., Bray, 1999; Ball & Youdell,
2008; Dawson, 2010; Schaub et al, 2020) as well as examinations of public versus private costs of education (e.g., Heyneman, 2011;
Tsang, 1997). It is worthwhile noting that student level SES has an impact on student achievement far beyond the impact of private
tutoring. This study indicates that private tutoring participation is less related to educational equity at the school and system level, and
more related to SES in the cross-national sample. While private tutoring fees are undoubtedly an extra burden for lower income
families, this research suggests that these economic inequalities related to private tutoring are not carried from the shadow into formal
mass schooling to the degree that previous research suggested.
Shadow education also has symbolic value not only because it has been normalized and legitimized in society, but also because it
has symbolic value (Baker et al, 2001; Lee & Shouse, 2011). Students and their parents may be concerned about underperforming, or
being perceived as underperforming, if they do not invest in shadow education (Schaub et al, 2020). By the same token, similar to
formal mass education, private tutoring is often accompanied by a taken-for-granted trust from parents and students (Aurini, 2006).
Even though shadow education may not exactly follow the standard educational templates and private tutoring staff may not have the
same credentials as regular or certified teachers, parents and students are willing to participate due to the primarily symbolic “logic of
confidence” (Aurini, 2006, p.90).
Because private tutoring activities tend to reflect or confirm the same rationalized myths about their functions and effects
worldwide, these activities are inevitably and rapidly becoming or have already become an institutionalized component of the global
institution of education (Baker, 2020). Therefore, with the near universal expansion of formal mass schooling worldwide, private
tutoring becomes a product of the education culture in a schooled society and is part of the “constructing force” of education (Mori &
Baker, 2010, p.40), which further “muddies the distinction between learning in the shadows and in the light of day” (Grodsky, 2010,
p.480). Private investment and government support have both facilitated the spread of shadow education activities, like private
tutoring (Schaub et al, 2020). As a result, shadow education has developed into a robust social institution, which now symbiotically
and symbolically connects individual performances in formal mass education to private tutoring access and participation in cultures
and systems worldwide (Mori & Baker, 2010).

References

Anderson, K., Domosh, M., Pile, S., & Thrift, N. (Eds.). (2002). Handbook of Cultural Geography. Thousand Oaks, CA: Sage.
Aurini, J. (2006). Crafting legitimation projects: An institutional analysis of private education businesses. Sociological Forum, 21(1), 83–111.
Aurini, J., Davies, S., & Dierkes, J. B. (Eds.). (2013). Out of the Shadows: The Global Intensification of Supplementary Education (Vol. 22, International Perspectives on
Education and Society Series). Bingley, UK: Emerald Publishing.
Baker, D. P. (2020). An inevitable phenomenon: Reflections on the origins and future of worldwide shadow education. European Journal of Education, 55(3), 311–315.
Baker, D., & LeTendre, G. K. (2005). National differences, global similarities: World culture and the future of schooling. Stanford University Press.
Baker, D. P., Akiba, M., LeTendre, G. K., & Wiseman, A. W. (2001). Worldwide shadow education: Outside-school learning, institutional quality of schooling, and
cross-national mathematics achievement. Educational Evaluation and Policy Analysis, 23(1), 1–17.
Ball, S. J., & Youdell, D. (2008). Hidden Privatisation in Public Education. Brussels, Belgium: Education International.
Bray, M. (1999). The Shadow Education System: Private Tutoring and Its Implications for Planners. Paris: UNESCO International Institute for Educational Planning.
Bray, M. (2006). Private supplementary tutoring: Comparative perspectives on patterns and implications. Compare, 36(4), 515–530.
Bray, M. (2009). Confronting the Shadow Education System: What Government Policy for What Private Tutoring. Paris: UNESCO International Institute for Educational
Planning.
Bray, M. (2013). Benefits and tensions of shadow education: Comparative perspectives on the roles and impact of private supplementary tutoring in the lives of Hong
Kong students. Journal of International and Comparative Education, 2(1), 18–30.
Bray, M. (2014). The impact of shadow education on student academic achievement: Why the research is inconclusive and what can be done about it. Asia Pacific
Education Review, 15(3), 381–389.
Bray, M., & Kobakhidze, M. N. (2014). Measurement issues in research on shadow education: Challenges and pitfalls encountered in TIMSS and PISA. Comparative
Education Review, 58(4), 590–620.

12
A.W. Wiseman International Journal of Educational Research 109 (2021) 101808

Bray, M., & Kwok, P. (2003). Demand for private supplementary tutoring: conceptual considerations, and socio-economic patterns in Hong Kong. Economics of
Education Review, 22(6), 611–620.
Bray, M., & Lykins, C. R. (2012). Shadow Education: Private Supplementary Tutoring and Its Implications for Policy Makers in Asia (Vietnamese translation). The University
of Hong Kong. Asian Development Bank & Comparative Education Research Centre, Faculty of Education.
Bray, M., Kwo, O., Lykins, C., Mang, E., Wang, D., & Zhan, S. L. (2011). Private supplementary tutoring: Scale, nature and implications for secondary schooling in Hong Kong.
GRF project (2011–2014). Hong Kong: Faculty of Education, The University of Hong Kong.
Bray, M., Zhan, S., Lykins, C., Wang, D., & Kwo, O. (2014). Differentiated demand for private supplementary tutoring: Patterns and implications in Hong Kong
secondary education. Economics of Education Review, 38, 24–37.
Burch, P., Good, A., & Heinrich, C. (2016). Improving Access to, Quality, and the Effectiveness of Digital Tutoring in K-12 Education. Educational Evaluation and Policy
Analysis, 38(1), 65–87.
Burch, P., Steinberg, M., & Donovan, J. (2007). Supplemental Educational Services and NCLB: Policy Assumptions, Market Practices, Emerging Issues. Educational
Evaluation and Policy Analysis, 29(2), 115–133.
Byun, S.-Y. (2014). Shadow education and academic success in Republic of Korea. In H. Park, & K. Kim (Eds.), Korean Education in Changing Economic and Demographic
Contexts: Education in the Asia-Pacific Region: Issues, Concerns and Prospects (pp. 39–58). Singapore: Springer.
Caruso, M. (2008). World systems, world society, world polity: theoretical insights for a global history of education. History of Education, 37(6), 825–840.
Cheo, R., & Quah, E. (2005). Mothers, maids and tutors: An empirical evaluation of their effect on children’s academic grades in Singapore. Education Economics, 13
(3), 269–285.
Chiu, M. M. (2010). Effects of inequality, family and school on mathematics achievement: Country and student differences. Social Forces, 88(4), 1645–1676.
Choi, J. (2012). Unequal access to shadow education and its impacts on academic outcomes: evidence from Korea. May. Paper presented at the annual meeting of the
American Sociological Association. Retrieved from http://www.aefpweb.org/sites/default/files/webform/Shadow_Education_Choi_0215.pdf.
Choi, Y., & Park, H. (2016). Shadow Education and Educational Inequality in South Korea: Examining Effect Heterogeneity of Shadow Education on Middle School
Seniors’ Achievement Test Scores. Research in Social Stratification and Mobility, 44, 22–32.
Dang, H. A., & Rogers, F. H. (2008). The growing phenomenon of private tutoring: Does it deepen human capital, widen inequalities, or waste resources? The World
Bank Research Observer, 23(2), 161–200.
Dawson, W. (2010). Private tutoring and mass schooling in East Asia: Reflections of inequality in Japan, South Korea, and Cambodia. Asia Pacific Education Review, 11
(1), 14–24.
Grodsky, E. (2010). Learning in the Shadows and in the Light of Day: A Commentary on Shadow Education, American Style: Test Preparation, the SAT and College
Enrollment. Social Forces, 89(2), 475–481. https://doi.org/10.1353/sof.2010.0063.
Ha, T. T., & Harpham, T. (2005). Primary Education in Vietnam: Extra Classes and Outcomes. International Education Journal, 6(5), 626–634.
Harnisch, D. L. (1994). Supplemental education in Japan: Juku schooling and its implication. Journal of Curriculum Studies, 26(3), 323–334.
Heinrich, C. J., Meyer, R. H., & Whitten, G. (2010). Supplemental Education Services Under No Child Left Behind: Who Signs Up, and What Do They Gain? Educational
Evaluation and Policy Analysis, 32(2), 273–298.
Heyneman, S. P. (2011). Private tutoring and social cohesion. Peabody Journal of Education, 86(2), 183–188.
Kim, S., & Lee, J. H. (2010). Private tutoring and demand for education in South Korea. Economic Development and Cultural Change, 58(2), 259–296.
Kim, Y. C. (2016). Shadow education, curriculum, and culture of schooling in South Korea. New York, NY: Palgrave Macmillan. https://doi.org/10.1057/978-1-137-
51324-3.
Kuan, P. Y. (2011). Effects of cram schooling on mathematics performance: Evidence from junior high students in Taiwan. Comparative Education Review, 55(3),
342–368.
Kwok, L. Y. (2001). A multi-level social analysis of demand for private supplementary tutoring at secondary level in Hong Kong. Unpublished doctoral dissertation. The
University of Hong Kong.
Kwok, P. (2004). Examination-oriented knowledge and value transformation in East Asian cram schools. Asia Pacific Education Review, 5(1), 64–75.
Kwok, P. (2010). Demand intensity, market parameters and policy responses towards demand and supply of private supplementary tutoring in China. Asia Pacific
Education Review, 11(1), 49–58.
Lee, J. (2007). Two worlds of private tutoring: The prevalence and causes of after-school mathematics tutoring in Korea and the United States. Teachers College Record,
109(5), 1207–1234.
Lee, J. (2013). Private tutoring and its impact on students’ academic achievement, formal schooling, and educational inequality in Korea. Teachers College: Columbia
University. Unpublished doctoral dissertation.
Lee, S., & Shouse, R. C. (2011). The Impact of Prestige Orientation on Shadow Education in South Korea. Sociology of Education, 84(6), 212–224.
Liu, J., & Bray, M. (2020). Accountability and (mis)trust in education systems: Private supplementary tutoring and the ineffectiveness of regulation in Myanmar.
European Journal of Education, 55(3), 361–375.
Mori, I., & Baker, D. (2010). The origin of universal shadow education: What the supplemental education phenomenon tells us about the postmodern institution of
education. Asia Pacific Education Review, 11(1), 36–48.
OECD. (2013). PISA 2012 Technical Report. Paris: Author.
Ozaki, M. (2015). A juku childhood: Children’s experiences in juku attendance and its relation to their well-being in Japan. UK: University of Bath. Unpublished doctoral
thesis.
Park, H., Byun, S. Y., & Kim, K. K. (2011). Parental involvement and students’ cognitive outcomes in Korea focusing on private tutoring. Sociology of Education, 84(1),
3–22.
Perry, L. B., & McConney, A. (2010). Does the SES of the school matter? An examination of socioeconomic status and student achievement using PISA 2003. Teachers
College Record, 112(4), 1137–1162.
Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models: Applications and data analysis methods. Thousand Oaks, CA: Sage. Vol. 1.
Schaub, M., Kim, H., Jang, D.-H., & Baker, D. P. (2020). Policy reformer’s dream or nightmare? Compare: A Journal of Comparative and International Education, 50(7),
1066–1079.
Song, K. O., Park, H. J., & Sang, K. A. (2013). A cross-national analysis of the student-and school-level factors affecting the demand for private tutoring. Asia Pacific
Education Review, 14(2), 125–139.
Stevenson, D. L., & Baker, D. P. (1992). Shadow education and allocation in formal schooling: Transition to university in Japan. American Journal of Sociology,
1639–1657.
Tansel, A., & Bircan, F. (2006). Demand for education in Turkey: A tobit analysis of private tutoring expenditures. Economics of Education Review, 25(3), 303–313.
Tsang, M. C. (1997). Cost Analysis for Improved Educational Policymaking and Evaluation. Educational Evaluation and Policy Analysis, 19(4), 318–324.
World Bank. (2020). World Bank Country and Lending Groups. April 14. World Bank: Data https://datahelpdesk.worldbank.org/knowledgebase/articles/906519-
world-bank-country-and-lending-groups.
Xue, H., & Ding, Y. (2009). A positivistic study on the private tutoring of students in urban China. Journal of Youth Studies, 12(1), 115–128.
Zhang, Y. (2013). Does private tutoring improve students’ National College Entrance Exam performance? A case study from Jinan, China. Economics of Education
Review, 32, 1–28.
Zhang, Y., & Liu, J. (2016). The effectiveness of private tutoring in China with a focus on class-size. International Journal of Educational Development, 46, 35–42.
Zhang, W., & Bray, M. (2020). Comparative research on shadow education: Achievements, challenges, and the agenda ahead. European Journal of Education, 55(3),
322–341.
Zhao, X., & Wiseman, A. W. (2016). April). The Impact of Shadow Education on Educational Equity and Achievement in East Asian Educational Systems. Washington, DC:
American Educational Research Association.

13