University of Southern California Dissertations and Theses

Essays on economics of education and private tutoring

(USC Thesis Other)

Essays on economics of education and private tutoring

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content ESSAYS ON ECONOMICS OF EDUCATION AND PRIVATE
TUTORING
by
Ahram Moon
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulllment of the
Requirements of the Degree
DOCTOR OF PHILOSOPHY
(Economics)
August 2016
Copyright 2016 Ahram Moon
to my parents,
and my sister and brother
for their unconditional love and support
ii
Acknowledgements
Foremost, I would like to express my special appreciation to my advisor, Professor John
Strauss, for his continuous support I received during my graduate training. I am very
grateful for his guidance and patience for my research. Without his consistent encourage-
ment, tremendous knowledge, and insightful discussion, this Ph.D. would not have been
achievable.
Besides my advisor, I would like to thank Professor Jerey Nugent, Professor Hyungsik
Roger Moon, and Professor Gary Painter for their professional advisement and valuable
comments which augmented my dissertation with various perspectives. My sincere thank
also goes to Professor Hak Kil Pyo who made it possible for me to pursue the doctoral
degree.
I thank my colleagues at USC for stimulating discussions, and for all the fun we had had.
I extend my sincere gratitude to my friends Mihye Lee, Kyoung-Eun Kim, and Kunhwa
Kim who all helped me in numerous ways during every stages of my doctoral degree. I
would like to give a very special thanks to Jung Hyun Choi and Ahyoung Song for their
personal and scholarly interactions.
Last but not least, I am very much indebted to my family. I deeply thank my father and
mother who sustained me during last six years. I am also thankful for having my sister
and brother who made me more pleasant throughout my life in general.
iii
Table of Contents
Dedication ii
Acknowledgements iii
List of Tables v
Abstract ix
Chapter 1 Introduction 1
Chapter 2 Demand Systems of Private Tutoring in South Korea 5
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Educational Policy and Private Tutoring in Korea . . . . . . . . . . . . . 14
2.4 Econometric Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4.1 Modeling quantity, quality, and unit values . . . . . . . . . . . . . 19
2.4.2 Specication and Estimation . . . . . . . . . . . . . . . . . . . . . 24
2.5 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.5.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.6 Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.7 Conclusions and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Chapter 3 Eect of Private Tutoring on Academic Achievement: Evidence from
South Korea 70
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.2 Previous Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.3 Specication and Estimation . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.5.1 Eect of private tutoring on test score . . . . . . . . . . . . . . . . 86
3.5.2 Heterogeneous eect of private tutoring . . . . . . . . . . . . . . . 91
3.5.3 Comparison with self-study hours . . . . . . . . . . . . . . . . . . . 94
3.6 Conclusions and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 96
Chapter 4 Conclusion 126
References 129
Appendix 135
iv
List of Tables
Table 2.1 Participation rate and monthly expenditure in private tutoring from
2007 to 2014 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Table 2.2 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
Table 2.3 Summary statistics for private tutoring . . . . . . . . . . . . . . . . 59
Table 2.4 Quality elasticity (
1
): the rst stage estimation of unit values . . . 60
Table 2.5 The coecient
0
on the total educational expenditure per capita:
the rst stage estimation of budget share . . . . . . . . . . . . . . . 61
Table 2.6 Average budget share ( w) . . . . . . . . . . . . . . . . . . . . . . . . 62
Table 2.7 Total educational expenditure elasticities (e = 1
1
+
0
w
1
) . . . 62
Table 2.8 Own- and cross-price elasticities: All . . . . . . . . . . . . . . . . . 63
Table 2.9 Own- and cross-price elasticities: Boys and girls . . . . . . . . . . . 64
Table 2.10 Own- and cross-price elasticities: Rural and urban sections . . . . . 65
Table 2.11 Own- and cross-price elasticities: Father's education . . . . . . . . . 66
Table 2.12 Own- and cross-price elasticities: Mother's education . . . . . . . . 67
Table 2.13 Own- and cross-price elasticities: Mother's employment . . . . . . . 68
Table 2.14 Own- and cross-price elasticities: Baseline test score . . . . . . . . . 69
Table 3.1 Summary statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
Table 3.2 Summary statistics for private tutoring and academic achievement . 101
Table 3.3 Eect of private tutoring on English Z-score . . . . . . . . . . . . . 102
Table 3.4 Eect of private tutoring on mathematics Z-score . . . . . . . . . . 103
Table 3.5 Eect of private tutoring on English Z-score: Bottom 50% of per-
centile baseline score . . . . . . . . . . . . . . . . . . . . . . . . . . 104
Table 3.6 Eect of private tutoring on mathematics Z-score: Bottom 50% of
percentile baseline score . . . . . . . . . . . . . . . . . . . . . . . . . 105
Table 3.7 Eect of private tutoring on English Z-score: Top 50% of percentile
baseline score . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
Table 3.8 Eect of private tutoring on mathematics Z-score: Top 50% of per-
centile baseline score . . . . . . . . . . . . . . . . . . . . . . . . . . 107
v
Table 3.9 Eect of private tutoring on English Z-score: Boys . . . . . . . . . . 108
Table 3.10 Eect of private tutoring on mathematics Z-score: Boys . . . . . . . 109
Table 3.11 Eect of private tutoring on English Z-score: Girls . . . . . . . . . . 110
Table 3.12 Eect of private tutoring on mathematics Z-score: Girls . . . . . . . 111
Table 3.13 Eect of private tutoring on English Z-score: Urban sector . . . . . 112
Table 3.14 Eect of private tutoring on mathematics Z-score: Urban sector . . 113
Table 3.15 Eect of private tutoring on English Z-score: Rural sector . . . . . 114
Table 3.16 Eect of private tutoring on mathematics Z-score: Rural sector . . 115
Table 3.17 Eect of private tutoring on English Z-score: Lower educated father
group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
Table 3.18 Eect of private tutoring on mathematics Z-score: Lower educated
father group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
Table 3.19 Eect of private tutoring on English Z-score: Higher educated father
group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
Table 3.20 Eect of private tutoring on mathematics Z-score: Higher educated
father group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
Table 3.21 Eect of private tutoring on English Z-score: Lower educated mother
group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
Table 3.22 Eect of private tutoring on mathematics Z-score: Lower educated
mother group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
Table 3.23 Eect of private tutoring on English Z-score: Higher educated mother
group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
Table 3.24 Eect of private tutoring on mathematics Z-score: Higher educated
mother group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
Table 3.25 Comparison between eects of private tutoring hours and self-study
hours on average test score . . . . . . . . . . . . . . . . . . . . . . . 124
Table 3.26 Comparison between eects of private tutoring hours and self-study
hours . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Table A1.1 Cause of deletion in re-sampling process: the number of missing values135
Table A1.2 Attrition of the KELS . . . . . . . . . . . . . . . . . . . . . . . . . . 136
Table A1.3 Quality elasticity with respect to total educational expenditure: the
rst stage results of unit values . . . . . . . . . . . . . . . . . . . . . 137
vi
Table A1.4 Quantity elasticity with respect to total educational expenditure: the
rst stage results of budget shares . . . . . . . . . . . . . . . . . . . 138
Table A2.1 Comparison between eects of private tutoring hours and self-study
hours: Boys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
Table A2.2 Comparison between eects of private tutoring hours and self-study
hours: Girls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
Table A2.3 Comparison between eects of private tutoring hours and self-study
hours: Urban . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
Table A2.4 Comparison between eects of private tutoring hours and self-study
hours: Rural . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
Table A2.5 Comparison between eects of private tutoring hours and self-study
hours: Lower educated father . . . . . . . . . . . . . . . . . . . . . . 143
Table A2.6 Comparison between eects of private tutoring hours and self-study
hours: Higher educated father . . . . . . . . . . . . . . . . . . . . . 144
Table A2.7 Comparison between eects of private tutoring hours and self-study
hours: Lower educated mother . . . . . . . . . . . . . . . . . . . . . 145
Table A2.8 Comparison between eects of private tutoring hours and self-study
hours: Higher educated mother . . . . . . . . . . . . . . . . . . . . . 146
Table A2.9 Comparison between eects of private tutoring hours and self-study
hours: Bottom 50 % of percentile baseline score . . . . . . . . . . . 147
Table A2.10Comparison between eects of private tutoring hours and self-study
hours: Top 50 % of percentile baseline score . . . . . . . . . . . . . 148
vii
Abstract
This dissertation asks two questions about private supplementary tutoring: What is the
price responsiveness of private tutoring, and does private tutoring improve children's
academic achievement?
Chapter 2 examines the expenditure and own- and cross-price elasticities for private
tutoring. Due to the school equalization policy, Korean students cannot choose schools
they prefer, and almost every school follows a uniform curriculum. Extra-study opportu-
nities such as private tutoring have arisen to satisfy individual demand. Besides, getting
into a prestigious university is very competitive and is based to a large extent on the
National College Scholastic Ability Test accepted for admission. Hence, students are re-
quired to prepare rigorously for this high-stakes test, especially by getting support from
private tutoring. In recent years, household expenditure on private tutoring has been
constituting an estimated 2% of GDP in Korea. However, despite this signicant amount
of expenditure, comparatively little attention has been paid to the demand for tutoring.
In order to analyze the demand system of tutoring activities, unit values are used as a
proxy for price, and Angus Deaton's approach is considered. This paper discusses two
kinds of private tutoring: one-to-one or small group tutoring and cram school tutoring.
This essay nds that educational expenditure elasticities for most types of tutoring ser-
vices are larger than unity. Own-price elasticities for tutoring are negative and smaller
than unity. Small-group private coaching is almost three times as own-price elastic as
cram school tutoring. Cross-price elasticities are very small in magnitude.
In Chapter 3, the eectiveness of private tutoring on test scores is investigated. Since
the two primary purposes of private tutoring are to help less able students with their
viii
insucient background and to enrich the educational excellence of the more able students,
heterogeneity in the eect across dierent types of students is also studied. In order to
deal with selectivity issues on school district choice, this study makes use of the dataset,
which is the KELS linked to the school lists extracted from each city and provincial
Oce of Education in Korea using school characteristics. This unique data makes it
possible to identify school district information. Findings indicate the positive association
between private tutoring and test score of English and mathematics. The Arellano-Bond
estimation, one type of dynamic panel data approach, suggests that an increase in English
tutoring may raise the student academic achievement. Since the eect of hours spent on
studying by oneself is not signicantly dierent from hours dedicated to tutoring, private
tutoring may not be considered a very eective way to improve students' performance.
From heterogeneous eects based on students' baseline test scores, we nd that private
tutoring may better serve the enrichment role for more able students.
ix
Chapter 1
Introduction
Private supplementary tutoring is widely performed all over the world including both
developing and developed countries. In developing countries, the lack of scal resources
for public school systems creates very low school teachers' salaries and poor quality of
teaching and facilities. Although parents choose supplementary tutoring for their chil-
dren's education, school teachers oer this service for extra money, for example in Kenya,
Bangladesh, and Cambodia. On the other hand, in developed countries where public ed-
ucation systems generate high quality of learning, private tutoring supplements a paucity
of knowledge. In some countries, such as Japan, Taiwan, or Korea, private tutoring be-
comes an attractive tool to enter elite universities because of high perceived nancial or
non-nancial benet of the top universities.
1
This dissertation consists of two essays on private tutoring issues in South Korea.
Korea has the largest private tutoring industries in the world. In Korea, household ex-
penditure for private tutoring in 2006 was estimated to be US$24 billion or 2.8 percent of
GDP, which was similar in size to the Korean automobile industry. Given the magnitude
of this spending, its analysis and research for appropriate public policy have attracted the
authorities' attention.
The rst essay investigates the demand for private tutoring, which has not been well-
developed, using the Korean Educational Longitudinal Survey (KELS). One of the chal-
lenges in tutoring studies that examine price responsiveness is the lack of high-quality
data on the price of tutoring. I use unit values as a proxy for prices by dividing the
expenditure on particular types of private tutoring by hours spent. When the quality
of goods and services such as tutoring varies, the usage of unit values can then lead to
biased estimates of price eects. In order to correct this bias, the rst essay considers an
approach developed by Angus Deaton.
Findings indicate positive quality elasticities with respect to total educational expen-
diture per capita; better-o households pay more per hour. Educational expenditure
elasticities for small group of private tutoring are larger than unity so that this type of
tutoring activities is luxury goods. Own-price elasticities for all kinds are negative and
inelastic, and small-group tutoring is almost three times as elastic as cram school type of
training. Regarding cross-price elasticities, the eects are very small. According to the
inverse-elasticity rule, cram schools with smaller price elasticities should be taxed more
2
heavily than small-group types. However, the equity-eciency trade-o can more severely
impact the welfare of households of a lower economic status.
The second essay is an empirical study that discusses whether private tutoring in
u-
ences academic achievement. The eectiveness of tutoring activities is investigated based
on educational production function. Even though most previous studies can detect the
average eect of private tutoring, this paper additionally deals with the impact of private
instruction estimated from the dierence in the standardized test scores of students who
increase/decrease private tutoring spending or hours, taking into account student xed
eects through the dynamic panel data model.
Findings indicate a positive relation between tutoring activities and the test scores
of students in English and mathematics. Even though the xed eects model results in
the positive association, the Arellano-Bond estimation, one type of GMM estimation of
dynamic panel data models, suggests that only English tutoring has an actual eect on
academic achievement. The impact of hours spent on studying has no signicant dierence
from that of hours devoted to tutoring for English. In addition, math tutoring does not
have a signicant eect, but self-study hours for mathematics do. Hence, private tutoring
may not be the best way to improve students' performance. From the heterogeneous
eects, students who live in the urban sector have a signicant eect on the mathematics
test scores. The results between students with lower and higher educated parents have
similar magnitudes of tutoring estimates for English, while only the higher educated
parents group has a signicant eect on math tutoring. According to the results of the
3
Arellano-Bond approach, students in top 50% of the percentile baseline score have eects
on math private tutoring.
The dissertation is organized as follows. Chapter 2, \Demand Systems of Private Tu-
toring in South Korea", has seven sections. Section 2.1 and 2.2 are an introduction and
literature reviews, respectively. Section 2.3 describes background on Korean private tu-
toring. Section 2.4 presents Deaton's approach and section 2.5 explains data. Section 2.6
provides the results, and section 2.7 is the conclusion. Chapter 3, \Eect of Private Tu-
toring on Academic Achievement: Evidence from South Korea", has six sections. Section
3.1 and 3.2 give an introduction and overview previous literature. Section 3.3 presents the
estimation strategy and section 3.4 explains data. Section 3.5 gives the empirical results.
Section 3.5 is the conclusion and discussion.
4
Chapter 2
Demand Systems of Private
Tutoring in South Korea
2.1 Introduction
With rapidly increasing household incomes and competition to enter into the elite uni-
versities, South Korea has long had one of the largest private tutoring industries in the
world (Baker and LeTendre, 2005; Bray, 2009). In 2006 and 2011, expenditures on pri-
vate tutoring by Korean households were estimated to be at US$24 billion (2.8 percent of
GDP) and at US$18 billion (1.7 percent of GDP), respectively. In 2014, 56.4% of Korean
students either attended large-scale classes or hired a private tutor for academic subjects
as supplementary tutoring (73.1% those in middle school and 55.7% those in general high
5
school students). Households spent US$174 per child on average monthly (about US$225
of middle school students and US$209 of general high school students).
Policy makers have focused on the appropriate role of the government in the educa-
tional sector, especially with respect to private tutoring. In general, private supplemen-
tary tutoring has three objectives: tutoring in academic subjects, nancial gain, and as
a supplement to mainstream schooling (Bray, 1999). In addition, the high-stakes exams
and test-oriented curricula in both public and private schools in Korea have made private
tutoring an attractive means to enhance students' competitiveness. However, the costs
and eorts associated with private tutoring may have the eect of widening educational
achievement gaps between households of socio-economic backgrounds.
1
Even though the features of private tutoring can be explained from both educational
and economic perspectives, most policies that regulate private tutoring have been based
on the educational component of private tutoring, such as by reducing the burden on
the preparation for academic tests or by improving the quality of public schools. Specif-
ically, policies to reduce the coverage that exams require or increasing in the number
of dierent channels to college entrance so as reduce the burdens associated with the
system of standardized tests have been proposed as well as after-school classes through
public schools and/or improving the quality of teachers. However, policies related to the
economic aspects of tutoring driven have drawn less attention in Korea.
1
The source is the Survey of Private Education Expenditure conducted by Statistics Korea. The values
of expenditure are calculated at the exchange rate of 1,100 Korean won to the U.S. dollar.
6
In addition to East Asian countries, such as Japan and Taiwan, other countries have
also experienced recent growth in the private tutoring market (Dang and Rogers, 2008).
In France, the private tutoring sector in 2006 was estimated to be worthe2.21 billion and
is growing annually at 10%. In Turkey, the estimates of tutoring cost in 2004 was $2.9
billion, which accounted for 0,96% of GNP (Bray, 2009, 2011). Expenditure on tutoring
in the U.S is also growing at about 5% per year (Sullivan, 2010). Studies of tutoring have
investigated various aspects of the industry, such as the size of the market (Bray and
Kwok, 2003), the participants involved (Dang, 2007; Kim and Lee, 2010), and the impact
of private tutoring on test scores and academic achievement (Briggs, 2001; Gurun and
Millimet, 2008), among other aspects. However, compared to the number of studies on
the determinants of demands for private tutoring, few papers analyze demand for private
tutoring.
Because of the paucity of economic analyses of the demand for tutoring, this paper
attempts to use an economic approach to explain private tutoring as one of consumption
goods and to analyze a demand system for private tutoring in order to estimate income
and price elasticities. Unlike previous research, the estimation of demand systems in
this study provides better estimates of the income, own- and cross-price elasticities for
dierent tutoring types in order to better understand consumer's behavior on private
tutoring. Also, the estimates of elasticities may suggest a glimpse of tax policy on tutoring
industry according to the inverse-elasticity rule given from Ramsey (1927).: The optimal
tax rates and price elasticities of demand are inversely correlated.
7
One of the challenges is the lack of price data on private tutoring. Although prices
play an important role in households' decision making, there is little data on the prices
that each consumer actually pays. In spite of the lack of high quality data on prices, some
surveys collect information on both the expenditure and quantities purchased for specic
set of goods. In particular, the Korea Educational Longitudinal Survey (KELS) panel data
used in this paper produces data on monthly average expenditures and weekly average
hours for each type of private tutoring. Hence, this information makes it possible to
calculate unit values for individuals by dividing expenditure by hours spent. Unit values
are attractive proxies for price because aggregate prices are just averages and cannot
indicate how much households actually pay. Consequently, variations in unit values over
households can explain preferences in household choice with respect to the characteristics
(or quality) of the goods purchased.
Deaton (1987, 1988, 1990), however, identies a possible drawback in the use of unit
values as a proxy for price. If one commodity group (e.g., private tutoring) includes several
specic commodities of dierent qualities (e.g., in tutor quality), an increase in the unit
value could indicate an increase in price or a shift to relatively expensive commodities
within consumption bundle of tutoring. For example, if households respond to an in-
crease in the price by choosing cheaper types of private tutoring and consuming the same
quantity (hours for private tutoring), the change in unit value may understate the true
price change. The details of Deaton's approach are presented in the section on empirical
strategy below. In order to adopt his method, there is an important assumption related
8
to a spatial variation in prices, and the original KELS data does not provide regional
location information on such observations. Hence, I linked the KELS panel to the school
lists provided from each city and provincial Oces of Education in Korea. The use of
school characteristics enable to identify the school district of each student.
While previous tutoring study to estimate price elasticity uses unit values, the quality
issues are not treated and also it can only identify an own-price elasticity. In order to
address the quality eect of unit values as price proxies, we use Deaton's method to
model the determination of unit values. Compared to previous studies, this paper is
able to examine demands for dierent types of tutoring and estimate income and both
own- and cross-price elasticities. To be specic, we consider two kinds of tutoring, large-
scale classes (called cram school or hakwon) and one-to-one or small group tutoring by
an individual tutor. Large-scale classes of tutoring provide sessions to students in a
classroom of learning institutions which have classes and tutors. Students attend these
cram school tutoring out of school hours. Unlike cram schools, one-to-one or small group
tutoring is provided at tutee's house. Tutors can be professional or university students.
Because of oering more intensive teaching services, most individual tutoring types are
more expensive.
This paper nds the quality eect on unit values because of positive estimates of
quality elasticities for all dierent types of tutoring with respect to total educational
expenditure per capita. As a result of total educational expenditure elasticities, small-
group and large-scale class types of tutoring are normal goods and small-group types are
9
strongly demonstrated to be luxury goods. But, households which have higher educated
parents are more likely to consume cram-school type as a necessity good. The own-price
elasticities for all types are negative and less than unity, and also individual tutoring
is more elastic than cram school type. In terms of cross-price elasticities, very small
substitutability between dierent types are estimated.
Thus, the inverse-elasticity rule proposes that all types of private tutoring as inelastic
goods can be taxed, and specically, cram schools with less inelastic demand should be
taxed more to avoid the distortion of taxation. While it could satisfy eciency, it would be
detrimental to equity because this taxation can reduce the welfare of worse-o households,
which have relatively elastic demand, such as less-educated parents (the equity-eciency
trade-o).
This paper is organized as follows.: The next section provides a review of previous
literature on private tutoring. Section 3 describes backgrounds of private tutoring in
Korea and Korean educational policy, and Section 4 introduces the analytical approach
used to evaluate the price responsiveness of private tutoring. The data and sampling are
described in Section 5. Section 6 is devoted to the estimation results of private tutoring.
Finally, Section 7 concludes and discusses.
10
2.2 Literature Review
After Bray's research (Bray, 1999; Bray and Kwok, 2003; Bray, 2011) who is largely
contributed to making the perspectives of private tutoring, impacts of tutoring on educa-
tional outcome have been studied very much. Regarding the two main purposes of private
tutoring, remediation for less able students and enrichment for more able students, the
academic achievement of student, such as test score, has been examined with respect to
the decision to send children to private tutoring. However, there is no clear conclusion
about this correlation and causation between educational outcomes and tutoring. While
Briggs (2001) nds heterogeneous eects, Tansel and Bircan (2006), Dang (2007), and
Ono (2007) identify positive eects on academic achievement. Suryadarma et al. (2006)
and Zhang (2013) show the eect to be insignicant except in certain cases. Even though
Korea has relatively abundant data sets on private tutoring, the results obtained from
Korean studies are also inconsistent. While Lee et al. (2004) nds no signicant eect,
Byun (2014), Ryu and Kang (2013), Choi (2012) and Moon (2015) nd positive eects.
But, these papers also nd that the eectiveness of private tutoring on academic outcome
can be modest and heterogenous.
2
2
Choi (2012) uses the Seoul Education Longitudinal Study and nds a positive eect on English and
math test scores. Moon (2015) also nds modest positive eect on academic achievement for middle school
students, but hours spent on tutoring of English and math has no signicant dierent eect of hours spent
on study excluding tutoring and preparation for tutoring. Lee et al. (2004) nds no signicant eect of
pre-class tutoring on the academic achievement by using simple mean dierences in test rankings between
2006 and 2007.
11
On the other hand, in studies of the determinants of demand for private tutoring,
private tutoring can be analyzed at several levels, such as the society, cultural, and in-
dividual levels. As society or cultural factors, in East Asian countries, such as Taiwan,
Korea, Hong Kong or Japan, Confucian ideas place enormous emphasis on education,
leading parents to adopt private tutoring programs on a wide scale (Kwok, 2010; Bray,
1999; Liu, 2012). In some developing countries which suer from the lack of scal re-
sources, school teachers may be poorly paid giving them incentives to provide tutoring
in order to supplement teacher salaries, such as in Bangladesh and Kenya (Buchmann,
2002; Nath, 2008). As far as individual level factors are concerned as educational inputs,
households' socio-economic characteristics are also related to the demand for private tu-
toring: households' wealth (Bray and Kwok, 2003; Kenayathulla, 2013), the size of family
(de Castro and de Guzman, 2014; Liu, 2012), students' gender (Nath, 2008; Dang, 2007;
Kim and Lee, 2010), and so on.
Compared to the number of studies on the determinants of private tutoring, few papers
estimate a demand function for private tutoring and most of these papers nd only the
income elasticity. Dang (2007) nds that private tutoring is a necessity good for primary
and lower secondary students in Vietnam. In Greece, the income elasticity is 0.2-0.3 for
all grades of students so that tutoring is more likely to be a necessity (Psacharopoulos
and Papakonstantinou, 2005). Tansel and Bircan (2006) nds an almost unit elasticity in
Turkey. In Kim and Lee (2010), the income elasticity for all Korean students is found to
12
be about 0.5. Similarly, Kim and Park (2010) nds that the income elasticity is less than
one for the 12th grade students in Korea.
There is a paper to develop a theoretical model for demands for private tutoring. Kim
and Lee (2010) models how an individual decides the demand for private tutoring condi-
tioning on the xed public educational expenditure. Although their theoretical framework
includes the price of private tutoring, their empirical estimation is not attracted to price
eect on the demand. Instead, they examine the impact of other factors.: Observations
who are likely to have highly educated parents, higher income and/or higher ability spend
more money on tutoring services. We empirically extend their study to be able to iden-
tify price elasticities. In addition to that, we analyze disaggregated demands for private
tutoring.
Kim (2007), as in the earlier version of Kim and Park (2010), shows the price elasticity
of tutoring by including the unit value and nds a negative own-price elasticity (around
-0.5). However, the quality issue associated with unit values is not treated in their study
and it is able only to identify an own-price elasticity of aggregate demand for tutoring.
Compared to previous studies, this paper examines demand for dierent types of tutoring,
making it possible to estimate quality, income, and both own- and cross-price elasticities
with Deaton's approach in order to deal with the possible bias due to quality dierence
eect when unit value as our price measure.
13
2.3 Educational Policy and Private Tutoring in Korea
Two education policies have signicantly in
uenced the growth of private tutoring in
Korea. In 1968, the secondary school equalization policy was introduced and students
were randomly assigned to middle and high schools located within the school district of a
student. Since students cannot choose schools they prefer and almost every school follows
a uniform curriculum, extra-study opportunities such as private tutoring have arisen to
satisfy individual demand. In addition, there is only one national College Scholastic
Ability test accepted for admission to the prestigious universities. Hence, students are
required to prepare rigorously for this high-stakes test, particulary with support from
private tutoring.
Recently, however, the educational inequality caused by private tutoring has became a
growing concern among policymakers. Households with lower socioeconomic backgrounds
can be disadvantaged from additional educational opportunity because of the high cost
of private tutoring. The rst graph in Figure 2.2 shows the trends in private tutoring
costs by deciles of household income levels in Korea. The two bottom graphs suggest time
trends in private tutoring expenditure proportional to household income and expenditure
by household income level. Figure 2.2 shows both that households in higher income groups
spend more on private tutoring, and that the share of expenditures on private tutoring
is increasing over time. To the extent that private tutoring helps to raise academic
achievement, unequal access to private tutoring may aggravate educational inequality
14
among students because of their dierent socioeconomic status. Tutoring may therefore
primarily help more the children of parents with more education.
Because of the extraordinary expansion of private tutoring and the possible educa-
tional inequality induced by private tutoring, the Korean government has launched sev-
eral policies. In the 1980's a new government led by General Chun Doo-Hwan banned
all types of private tutoring. Despite this draconian measure, the ban was ineective.
Even after a new government came into power in 1996, the strict regulations on private
tutoring were enforced but it was hard to control private tutoring eectively. In addition,
the Constitutional Court ruled in 2000 that the government's prohibition of private tu-
toring was unconstitutional. After that, the private tutoring industry was substantially
enlarged. The number of cram schools (called hakwon) for general academic subjects
increased substantially from 381 in 1980 to 38,665 in 2014. Also, the number of attendees
and instructors in hakwon also increased from 117,618 and 3,051, respectively, in 1980
to 4,424,033 and 173,277 in 2014 (Statistical yearbook of Education published by Korea
Ministry of Education and Human Resources Development).
Besides the ban on private tutoring, the government has pursued policies to relieve
the burden of preparation on the competitive college entrance exam and reduce the de-
pendence of students on private tutoring, for example, by lowering the level of diculty
in the college entrance exams and narrowing the range of subjects that the exams cover.
Also, institutions of higher education have been allowed to provide alternative channels
for entering college besides the college entrance test. The government has also tried
15
to improve public education by decreasing the number of pupils per teacher, evaluat-
ing teachers' performance, and providing dierentiated instructions by student' academic
achievement levels. At the same time, as a substitute to private tutoring, the public ed-
ucation system provides after-school programs and educational broadcasting services on
TV and the Internet. In addition, the government has attempted to regulate the private
tutoring market, for example, by forcing individual tutors to register with the provincial
Educational oce, by posting cram school information including tuition on the Internet,
and by mandating closing of hakwon before 10pm.
In spite of these eorts, the policies have not been able to curtail the demand for
private tutoring and the expansion of the industry. For instance, alternative options for
admission to universities have created new demands for private tutoring to prepare for
them. Educational broadcasting services on TV and the Internet generate an additional
burden on students so that they now seek out new services of private tutoring. There-
fore, before new policies to control the demand for private tutoring are pursued, detailed
evaluations are needed on the eectiveness and outcomes of such policies.
Taxation on the private tutoring sector can be considered as a new policy. In fact, most
cram schools, which require government's approval to operate the business, are exempted
from value-added tax since educational services are considered as goods subject to value-
added tax exemption in Korea. Following the general inverse elasticity rule of taxation,
the tax rates on goods are inversely correlated with elasticities of demand. The study
16
of income and price elasticities for private tutoring is thus necessary to approximate the
form of taxation.
This economic policy can con
ict with equity and eciency. Based on the inverse
elasticity rule, it is ecient to tax goods with low elasticities of demand. However,
goods with lower elasticities are likely to be necessity goods that worse-o households
disproportionately consume, and then taxing heavily on these goods may induce relatively
larger losses in the welfare of less well-o households. Tax policy to satisfy economic
eciency can be inequitable (Myles, 1995). In particular, this trade-o between equity
and eciency of taxation on private tutoring may cause an academic achievement gap
between dierent demographic groups through unequal accessibility to supplementary
education. Hence, tax programs need to be proposed carefully.
Furthermore, even though suppliers for both small-group and cram school types should
report their business status including total income, fees, subjects that they teach to tutees,
the number of registered students, and so on, to the authorities, tutors who oer one-
to-one or small-group types of instruction are highly likely to avoid registering with the
government. For example, the gross sales of private tutoring industry reported by the
Korean National Tax Service are about half of estimated total expenditure in private
tutoring reported by the Korean National Statistical Oce from 2010 to 2014.: The gross
sales for ve years were about US$56 billion, while the total expenditure in tutoring was
US$88 billion. The evidence implies that there is a large scale underground economy
17
based on private tutoring. Therefore, taxation on the overall tutoring sector can more
heavily aect cram school types.
2.4 Econometric Framework
Due to lack of price data for each type of private tutoring, the (calculated) unit value
can be an attracted measure (Kim, 2007). However, households' choices of quality or
the actual price of goods are related to the unit value and thus it cannot the same as
price. While price is given, unit values are choice variables and more informative means
to indicate households' preferences of demand. For example, since households that have
higher SES prole tend to hire more qualied tutor or choose better quality of cram schools
with higher unit values, levels of unit values can be determined with both households'
incomes and quantities demanded at the same time. Hence, using unit values to investigate
demand patterns can induce the simultaneous bias. In particular, private tutoring, as a
kind of educational services, can be more heterogenous than foods that demand system is
analyzed by Deaton using unit values. Hence, unit values of private tutoring would cause
greater quality eect than those of foods.
In addition to the quality eect, using unit values as a proxy of price can exaggerate
quantity elasticity with respect to price. To be specic, corresponding to an increase in
prices, unit values are also increased. At the same time, consumers respond to buy less
and/or choose less-quality items which imply lower unit values. Thus, changes in unit
values can not fully account for changes in prices. Since price elasticities are constructed
18
by the percentage change in quantity divided by the percentage change in unit values, the
price elasticities can be overstated because the size of an increase in unit values would
be smaller than the size of actual price changes. This kind of quality issues is called
the quality shading by Deaton. The relation between unit values and quality should be
explained in the model in order to avoid the biases.
This section starts by introducing Deaton's approach to modelling the determination
of quality and quantity associated with unit values. Assumptions and notations in this
paper follow his works on the subject of demand analysis using the unit value specication
(Deaton, 1987, 1990, 1997). Two assumptions related to spatial variation of prices make
it possible to estimate own- or cross-price elasticities. The rst assumption is that there
exists no variation in market prices within a cluster, so that within-cluster estimators of
budget share and unit value equations can identify the income and other eects on quality
or quantity without contamination from variation in prices. Secondly, we assume that
prices vary between clusters. Thus variation in prices between clusters will be used to
identify price eect.
2.4.1 Modeling quantity, quality, and unit values
Following Deaton's work, quality is dened as a property of commodity aggregate. That
is, higher-quality consumption bundle means higher proportion of relatively expensive
items. Some reference price vector can dene relatively expensive or cheap. Commodities
are divided intom groups and a group of goods, denoted by the letterG, has the quantity
vector q
G
and the unobserved price vector p
G
. Specically, this paper has six groups of
19
goods, which are composed of two types (small-group type and cram school) and three
academic subjects (English, mathematics and Korean) private tutoring: small-group for
English, cram school for English, small-group for math, cram school for math, small-group
for Korean, cram school for Korean. An observed group quantity indexQ
G
is determined
byQ
G
1
0
G
q
G
, where1
0
G
is a vector of ones. The unit of quantity in this paper is hours
spent on tutoring. Expenditure on the commodity group G is denoted byE
G
, which is is
written as E
G
=p
G
q
G
and a unit value is calculated as V
G
E
G
=Q
G
.
Based on the supposition that the cluster of household's residence is distinguishable,
two important assumptions are used regarding the spatial variation in the prices. The rst
assumption is that relative prices within each group of goods are constant over clusters.
The vector of price p
G
can be decomposed as
p
G
=
G
p
0
G
; (2.1)
where p
0
G
is a vector capturing the relative price of commodities within goods G and
G
is a scalar measure of group price levels. This equation implies that changing the scalar
measure
G
can simply change the level of price while maintaining the xed structure of
relative prices within the group. Since the reference price vector p
0
G
within a commodity
group is unknown but xed, the assumption means that variation inp
G
between clusters is
dominated by variation in
G
between clusters than variation inp
0
G
. Secondly, we assume
that the price vector
G
is xed within a regional cluster while it is varied between clusters.
Hence, the price vector can be denoted as
Gc
for clusters c = 1;:::;C.
20
The unit value and expenditure on a group of goods G are given by
V
G

E
G
Q
G
=
G
(
p
0
G
q
G
Q
G
) (2.2)
E
G
=p
0
G
q
G
=
G
p
0
G
q
G
=
G
(
p
0
G
q
G
Q
G
)Q
G
: (2.3)
The term in brackets of equations (2.2) and (2.3), which is denoted as
G
, is a measure
of quality of goods G consumption,
G
p
0
G
q
G
=Q
G
. At the xed relative price within
group p
0
G
, the quality measure is a function of the consumption bundle q
G
. Choosing
higher quality of consumption bundle can be interpreted as change in the consumption
bundle towards more expensive goods and away from cheaper goods.
Expenditure on a group of goods G can be expressed with the price
G
, quality
G
,
and quantity Q
G
, that is E
G
=
G

G
Q
G
, and it can be expressed as
lnE
G
= ln
G
+ ln
G
+ lnQ
G
: (2.4)
Besides, the unit value also can be written as price and quality measure V
G
=
G

G
and
after taking logarithm it:
lnV
G
= ln
G
+ ln
G
: (2.5)
To derive the link between the income and price elasticities of quality, the next step
starts with a weakly separability assumption of preferences.
3
Assume that the total
21
utility function depends on the consumption of goods within the group through a sub-
utility function of a group of goods. Thus, the maximization of the total utilityu requires
the maximization of subutility
G
(q
G
)
u =f
1
(q
1
);
2
(q
2
);:::;
G
(q
G
)::::;
m
(q
m
)g: (2.6)
Therefore, the vector of subgroup demands q
G
depends only on expenditure and price
vector of goods G,
q
G
=g
G
(E
G
;p
G
) =g
G

E
G

G
;p
0
G

; (2.7)
where the second equality is derived by assuming zero degree homogeneity of demand
functions in group expenditure and prices.
By the denition, the quality measure
G
can be expressed with p
0
G
and q
G
. Due to
the constant p
0
G
, the eect of price of other goods H (
H
) on quality (
G
) is appeared
only through q
G
, which depends only on E
G
=
G
in the subgroup demand (2.7). Thus,
@ ln
G
@ ln
G
=
@ ln
G
@ lnE
G

GH

@ lnE
G
@ ln
G

; (2.8)
3
The denitions we settled so far can help to address what happen the quality measureG when there is
a change in priceG and tie the price and quantity elasticities to the price elasticity of quality. But, there
exists no link between income(or total expenditure) and price elasticities of quality. Weakly separability
assumption supports to construct the links.
22
where
GH
is the Kronecker delta. By (2.4), the last term of the derivative of
G
(2.8) is
@ lnE
G
@ ln
H
=
GH
+
@ ln
G
@ ln
H
+
@ lnQ
G
@ ln
H
=
GH
+
@ ln
G
@ ln
H
+"
GH
; (2.9)
where "
GH
is the own- or cross-price derivative of G with respect to H.
Also, a derivative of quality with respect to total expenditure (or income) x is the
Prais-Houthakker quality elasticity
G
and the derivative @ ln
G
=@ lnx works through
only the group expenditure E
G
because of the separability:
@ ln
G
@ lnx
=
@ ln
G
@ lnE
G

@ lnE
G
@ lnx
=
@ ln
G
@ lnE
G
("
G
+
G
); (2.10)
where "
G
and
G
are the total expenditure (or income) elasticities of quantity and qual-
ity. The equation (2.10) can be transformed as @ ln
G
=@ lnE
G
=
G
=("
G
+
G
). After
substituting (2.9) into (2.8),
@ ln
G
@ ln
H
=
"
GH
@ ln
G
=@ lnE
G
1@ ln
G
=@ lnE
G
=
"
GH

G
"
G
: (2.11)
In the case of unit values, the derivative of logarithm lnV
G
with respect to the price
H
becomes
@ lnV
G
@ ln
H
=
GH
+
"
GH

G
"
G
: (2.12)
23
Hence, if the quality elasticity
G
is zero, the unit value of G responds one for one with
the groupG's price, and then is independent of prices of other groupH. However, if
G
is
non-zero, the unit value moves less one for one with the price ofG as the quality tends to
be downward through the income eect in response to increase in the price. In addition,
if the quality elasticity is non-zero, the eects of change in the price of one goods H on
quality of another G can be controlled by the cross-price elasticity, which indicates how
much an increase in the price aects quantity. Therefore, the equation (2.12) plays a role
as the linkage between the eect of prices on quality and the eects of total expenditure
on quality in the estimation stage.
2.4.2 Specication and Estimation
For empirical work, Deaton's approach begins with two equations for the budget shares
(w
G
) and the logarithms of unit value (ln
G
) as a function of demographic variables,
cluster xed eects and errors. For good G, household i in cluster c,
w
Gic
=
0
G
+
0
G
lnx
ic
+
0
G
z
ic
+
M
X
H=1

GH
ln
Hc
+ (f
Gc
+u
0
Gic
); (2.13)
ln
Gic
=
1
G
+
1
G
lnx
ic
+
1
G
z
ic
+
M
X
H=1

GH
ln
Hc
+ (f
Gc
+u
1
Gic
); (2.14)
where the budget shares are constructed by w
G
=E
G
=x. z
ic
is a vector of observation's
characteristics (student's gender, the number of siblings, Z-scores of English, mathematics
and Korean, father and mother's education attainment, father's age and age squared,
24
employment status of mother, home owned). The unobserved prices eect on the share
and unit values can appear through the matrix and that include the elements
GH
and
GH
, respectively. Particularly, if is the identity matrix, unit values can be the
same as prices. If the quality shading is existed as stated at the beginning of this section,
we would expect that
GG
< 1 and some non-zero o-diagonal elements.
In order to estimate the parameter, this approach includes a two-stage process. From
the assumption that households in the same clusters face the same price, the within-
cluster estimation results in the quality elasticity and total expenditure(or income) elas-
ticity. And, the price elasticity will be constructed by between-cluster estimation from
the assumption of price variation between clusters.
The rst stage uses within-cluster estimation with a two stage least square (2SLS)
framework. This stage performs to demean all variables in the budget share equations
(2.13) and unit values (2.14) by their cluster means and run two regressions using the
demeaned variables. Then, it can sweep away the unobservable price from the share
and unit value equations, as well as cluster-xed eect from the share equation.
Endogeneity issue of total expenditurex is addressed by using instrumental approach
at the rst stage suggested by Crawford et al. (2003). For example, if an explanatory
variable x is total expenditure, the x may be endogenous because of the mismeasured
data. This errors-in-variable can arise from under/over-reports of the actual household
expenditure. Also, since the budget share is calculated from expenditure on one good di-
viding by total expenditure, the measurement error may be appeared in both dependent
25
and explanatory parts. The correlation of errors between one good's and total expendi-
tures can cause the spurious correlation between the budget share and total expenditure.
Therefore, endogenous total expenditure in the demand system can be controlled with
instruments, such as income.
In addition to an appropriate instrumental variable, cluster averages excluding the
current observation can be considered as instruments to correct for random measurement
errors, not for omitted variables.
4
Since total educational expenditure is considered as the
variable x in this paper, we treat as endogenous the log of total educational expenditure
per child, lnx. Instruments include the log of household income (which is correlated with
ln x) and cluster means of total educational expenditure per child excluding the current
observation.
In order to calculate the parts of average cluster share and unit values that cannot
account for by the rst stage variables, the second stage uses between-cluster information
with the rst stage estimates. Because of the assumption that prices are constant within
4
Crawford et al. (2003) supports this type of instrument as follows. Let

Xj = 1=(nc 1)
P
j2c;j6=i
Xj
be the cluster means excluding the current observation. In the within-cluster estimation yi y
c
= (Xi

X
c
) + (ui u
c
), the asymptotic covariance between

Xj and ui u
c
goes to zero when the number of
households per cluster nc goes to innity because
E(

Xj (ui u
c
)) =E(

Xj u
c
) =
1
nc
1
nc 1
X
j2c;j6=i
E(Xjuj ) =
1
nc
X
E(Xu):
26
clusters, demeaning variables can pull out all price variables in these equations and re-
sult in consistent estimates,
~
's and ~
's. The rst-stage estimates are used to compute
the cluster averages of shares and unit values purged of expenditure (or income) x and
demographic variables z
~ y
0
Gc
=n
1
c
X
i2c
(w
Gic

~

0
G
lnx
ic
~
0
G
z
ic
) (2.15)
~ y
1
Gc
= (n
+
Gc
)
1
X
i2c
(ln
Gic

~

1
G
lnx
ic
~
1
G
z
ic
); (2.16)
wheren
c
is the number of households in clusterc andn
+
Gc
is the number of households in
the cluster c who purchased goods G.
As the number of observations increases at the rst stage, cluster means of the purged
shares and unit values (2.15) and (2.16) become the true cluster means. From (2.13) and
(2.14), the population counterparts to (2.15) and (2.16) can be expressed as
y
0
Gc
=
0
G
+
M
X
H=1

GH
ln
Hc
+f
Gc
+u
0
Gc
(2.17)
y
1
Gc
=
1
G
+
M
X
H=1

GH
ln
Hc
+f
Gc
+u
1
Gc
; (2.18)
, where u
0
Gc
and u
1
Gc
are cluster means of error terms. The cluster means contain infor-
mation on prices which are excluded in the within-cluster estimation.
27
The between-cluster variance-covariance matrix of they
0
can be denoted asQ, that of
they
1
isS and their covariance matrix isR. Elements of these matrices can be estimated
from (2.17) and (2.18),
~ q
GH
=cov(^ y
0
Gc
; ^ y
0
Hc
) , ~ s
GH
=cov(^ y
1
Gc
; ^ y
1
Hc
); and ~ r
GH
=cov(^ y
1
Gc
; ^ y
0
Hc
): (2.19)
In the setting, the result of the between-cluster regression can be written as
B
OLS
=
~
S
1
~
T; (2.20)
where the Gth column of B
OLS
is the coecient vector of ^ y
0
regressed on all ^ y
1
G
s. How-
ever, the OLS estimator cannot capture the spurious correlation between budget share
and price caused by measurement errors. The variance-covariance matrix S includes
the measurement error u
1
Gc
of unit values (2.18) and the matrix S can overestimate the
variance-covariance matrix of the true price. Also, the matrixR, which is associated with
(2.18), is contaminated from the measurement error between two equations (2.17) and
(2.18). In order to correct the measurement errors, the variance-covariance matrices of
errors estimated from the rst stage are used.
Let be the variance-covariance matrix ofu
0
Ghc
with element
GH
from the rst stage
estimation. Also, suppose that
is the variance-covariance matrix of u
1
Ghc
with element
28
!
GH
and that is the covariance matrix of u
1
Ghc
and u
0
Ghc
with element
GH
. Elements
of these matrices can be estimated by
~
GH
= (nCk)
1
X
c
X
i2c
e
0
Ghc
e
0
Hhc
;
~ !
GH
= (n
+
G
Ck)
1
X
c
X
i2c
e
1
Ghc
e
1
Hhc
; (2.21)
~
GH
= (n
+
G
Ck)
1
X
c
X
i2c
e
1
Ghc
e
0
Hhc
;
where e
1
and e
0
are the residual from the within-cluster regression of the rst stage.
Subscript tildes shows estimates from the rst stage. Thus, from (2.17) and (2.18)
S = M
0
+
N
1
, R = M
0
+ N
1
+
; (2.22)
where the matrixM is the unobserved variance-covariance matrix of the true price vector.
The matrix
~
N
1
+
= C
1
P
c
D(n
+
c
)
1
, D(n
+
c
) is a diagonal matrix from the elements of
n
+
Gc
and the matrix
~
N
1
is the corresponding quantity from the n
c
's.
Thus the OLS estimator using the corrected variance-covariance matrices is calculated
by
~
B = (
~
S
~

~
N
1
+
)
1
(
~
T
~

~
N
1
); (2.23)
plim
~
B =B = (
0
)
1

0
: (2.24)
29
The equation (2.24) represents that its probability limit is the ratio of to when the
quality eects exist.
At the nal stage of estimation, the corrected estimators from the second stage are
used to extract the eect of price on the budget share by adopting the model linking
quality and quantity elasticities. The equation (2.12) is expressed as the matrix notation:
=I +D(
1
)D(e)
1
E; (2.25)
where D() is diagonalization operator that its vector arguments into a diagonal matrix
andE is the matrix of price elasticities. The matrix of price elasticities E and the vector
of total expenditure elasticities e can be linked to the model parameters by
E = +D( w)
1
; (2.26)
e =
1
+
0
D( w)
1
: (2.27)
In order to regain and E, equations (2.25) and (2.26) incorporate the equation (2.24),
then
=B
0
=B
0
[ID()B
0
+D()D( w)]
1
; (2.28)
E = [D( w)
1
I] = [D( w)
1
I][ID()B
0
+D()D( w)]
1
; (2.29)
30
where the elements of a vector are dened by

G
= [(1
1
G
)D( w) +
0
G
]
1

1
G
: (2.30)
Finally, the rst-stage's estimates and residuals are used to construct the variance-
covariance matrices of the error terms e
0
ande
1
from (2.19) to (2.21) and these matrices
also apply to constructing the matrix
~
B by (2.23). Consequently, the own- and cross-price
elasticity matrix E is corrected by using (2.28).
In this paper, variance-covariance matrices for the estimated parameters and elastici-
ties are obtained by bootstrapping. The details on the derivation of estimates of variances
and covariances of elasticities can be found in the appendix part of Deaton (1987, 1990). In
practice, Deaton (1997) bootstraps only the second stage of estimation. This is defended
by the asymptotic variance formulas, which show that contribution from the second stage
is dominant, and also practical experience, which shows that contribution of the rst stage
to the variance of the results can be ignored.
Completing the system and imposing the Slutsky symmetry
In order to complete the demand system, some restrictions need to be imposed. Because
the budget shares add to unity, the sum of the constant terms
0
in (2.13) must be one.
The vector
0
and each column of the matrix need to sum to zero, which make the
31
system homogeneous of degree zero in total expenditure and prices. Also, linear homo-
geneity of the unit value equation is also required. Thus both homogeneity restrictions
will occur if and only if that, for all G
X
H

GH
+
0
G
= 0 and
X
H

GH
+
1
G
= 1: (2.31)
This adding-up and homogeneity restrictions complete the demand system by adding a
single commodity, such as "educational goods except private tutoring"
Technically, after calculating the mm matrix by (2.28), the corresponding (m +
1) (m + 1) matrix
x
is constructed with an additional row and column using the
homogeneity restriction (2.31) and adding-up restriction of (for the nal row of
x
),

x
GM+1
=
0
G

M
X
H=1

GH
and
x
M+1G
=
M
X
H=1

x
GH
: (2.32)
Also, the adding-up restriction extends the vectors
0
,
0
, and w to
0x
,
0x
, and w
x
by
adding one more element. From the extended vectors by the adding-up restriction, the
extended
x
is computed by (2.4.2). Consequently, (2.28) completes the system, then

x
=B
x0

x
; (2.33)

x
=ID(
x
)B
x0
+D(
x
)D( w
x
): (2.34)
32
By eliminating B
x
,

x
= [I +D(
x
)D( w
x
)]
1
[I +D(
x
)
x
]: (2.35)
Hence, the price and total expenditure elasticity in the full demand system can be derived
by (2.26).
In addition, Deaton imposes the Slutsky symmetry restriction.:

GH
+
0
G
w
H
=
HG
+
0
H
w
G
: (2.36)
In practice, the symmetry condition is that the matrix
B + w
0
(2.37)
is symmetric. This restriction can be transformed with notations of calculation as
L(IK)(vecB +
0

w) = 0; (2.38)
where the vec operator makes a matrix to a vector by stacking its column vertically.
The matrix L is a selection matrix which picks out from the vec of a square matrix the
element that lie below the diagonal in the original matrix. The commutation matrix K
rearranges the vec of a matrix B so that it becomes the vec of the transpose of B, that
is KvecB =vecB
0
.
33
The symmetry restriction (2.38) is written as a linear form Rb =r. Letb meanvecB.
Since the unrestricted estimate of B is derived from (2.23), the restricted estimate of B,
~
B
R
, is dened as
vec
~
B
R
=vec
~
B + (I
A)
1
R
0
[R(I
A)
1
R
0
]
1
(rRvec
~
B); (2.39)
where A =
~
S
~

~
N
1
+
.
Test for Slutsky Symmetry Restriction
Deaton also performs test for Slutsky symmetry restriction. Since the restriction equation
(2.38) can be expressed as the linear restriction form,
Rvec(
~
B) =r; (2.40)
an Wald test can be available from
W = (rRvec(
~
B))
0
(RVR
0
)
1
(rRvec(
~
B))
2
J
; (2.41)
where a matrix V is variance-covariance matrix of vec
~
B, denoted as V [vec(
~
B)]. J is the
degree of freedom. Here, the degree of freedom is the number of elements in upper/lower
diagonal of the matrix B.
34
Due to the complexity of variance-covariance matrix ofvec
~
B derived in the appendix of
Deaton (1990), Deaton et al. (1992) and Deaton and Parikh (1994) simplies its formulae
as
V [vec(
~
B)]'C
1
(P
0
HP
A
1
JHJ
0
A
1
) +C
1
(P
0
HJ
0
A
1

A
1
JJ
0
A
1
)K; (2.42)
where P
0
= (I
M
jB
0
), J = (0
M
jI
M
). The matrices H and are constructed from
variance-covariance matrices of (2.19) and (2.21):
H =
0
B
B
@
Q R
0
R S
1
C
C
A
; =
0
B
B
@

0

1
C
C
A
: (2.43)
2.5 Data
The data used in this paper are obtained from the 2006 and 2007 samples of the Ko-
rean Educational Longitudinal Study (KELS). This survey was launched by the Korean
Educational Development Institute (KEDI) in 2005. Students, parents, teachers, school
principals and school administrators answered questionnaires about student academic per-
formance, family backgrounds, and teacher and school characteristics. The sample in the
KELS was drawn in two stages. In the rst stage, the KEDI stratied all Korean middle
schools into four types of regions by city size and randomly chose a sample of 150 schools
of 2,929 middle schools in Korea. Then, at the second stage, 50 students in the 7th grade
of each sampled middle school were selected at random.
35
After the two-stage stratied cluster sampling, the survey started with 6,908 7th grade
students in 150 schools, and 6,502 (94%) actually answered all questionnaires. Although
this survey tracked transferred students, attrition occurred by emigration or dropping-out
from the survey. Then, 5,943 samples in the 2nd wave of 2006 and 5,594 samples in the
3rd wave of 2007 answered all questionnaires. All questionnaires contain academic test
scores of students, as well as the questionnaires completed by students, their parents,
teachers and schools.
Table A1.2 in appendix shows the mean value of 2005 characteristics of observations
by their attrition as of 2006 and 2007, respectively. The rst four columns show these
characteristics for nonattritions and attritors in 2006 and the set of middle columns com-
pares mean values of attriting samples of 2006 in 2007. The last three columns suggest a
dierence between the non-attriting and attriting households of 2005 in 2007. The overall
sub-sample of attritors are more likely to have higher educated parents (4-year university
graduate). But, the non-attriting students have a relatively higher test scores. In terms
of tutoring measures, there is no dierence in attendence but non-attritors tend to spend
less money on private tutoring.
This paper limits the sample of students who consistently responded to this survey
during two years (2006 and 2007) because the baseline (2005) data does not contain
information on the expenditure of each type of private tutoring. The two years cover
only the second and third grades of junior high school. Since measures of students'
academic achievement and private tutoring play important roles in the estimation of this
36
study, this paper eliminates observations with missing values on these variables and also
limits attention to those students living with both parents. Table A1.1 of the appendix
represents the number of observations dropped due to missing values. The nal number
of observations used in this study contains 2,910 students who stayed in the survey until
2007 and the balanced panel data has a total of 5,820 person-year observations on these
students.
However, in order to adopt Deaton's method, there is an important assumption related
to a spatial variation in prices, and the original KELS data does not provide specic
regional location information on such observations. Hence, I linked the KELS panel to
the school lists provided from each city and provincial Oces of Education in Korea.
The use of school characteristics makes it possible to identify the school district of each
student. Finally, this paper investigates demand systems of private tutoring in Korea
using the 2006 and 2007 samples of the unique data set.
PrivateTutoring As a very rich panel data set on Korean education, the KELS includes
the detailed information on private tutoring. Since there are dierent types of private
tutoring, the KELS provides information on the dierent types of private tutoring that
students were involved in. This survey categorizes 5 types of private tutoring for Korean,
English and mathematics, such as private tutoring in a large-scale class (cram schools or
hakwon), one-to-one or small-group tutoring by individual tutors, a paper correspondence
type, tutoring via the Internet, and broadcasting media. This study focuses on the rst
two types of private tutoring (the cram school and small-group tutoring) because these
37
are the dominant forms of private tutoring in Korea and account for a higher demand for
private tutoring. The Survey of Private Education Expenditure conducted by Statistics
Korea and the Ministry of Education reports that each 77.2 and 73.1 percent of middle
school students participated in small-group or large-scale classes of private tutoring in
2007 and 2014. Even on the side of private tutoring expenditure, 94.1 and 97.7 percent of
monthly expenditure on private tutoring for academic subjects focus on hiring individual
tutors or attending cram school types (Table 2.1).
Individual parents also answered the relevant questions for each private tutoring type
separately for Korean, mathematics and English: "Did the student attend this type of
private tutoring during this year?", "On average, how many hours did the student spend
on this type of private tutoring per week during this year?" and "On average, how much
money did you spend on this type of private tutoring for the student per month during
this year?" In addition to these three private tutoring variables, the unit value of private
tutoring is also calculated. The unit value implies the expenditure of private tutoring per
hour. This was not asked in the survey, but is constructed from the monthly expenditure
on each type of private tutoring divided by the hours of private tutoring type.
Academic Achievement At the end of each year (December), the KEDI conducts
achievement tests in Korean, English, and mathematics. The raw test scores of the KEDI
achievement tests for English, Korean and mathematics are scaled from 0 to 100. Instead
of the raw test score, this paper considers the standardized test score as an academic
achievement variable. The standardized test score, called a Z-score, is calculated from
38
the normalization of the raw test scores for the subject to have a mean of 0 and a standard
error of 1.
OtherCharacteristics The survey also delivers socioeconomic proles of students. Par-
ents reported their ages, the highest level of education attainment, and the number of
children at the baseline (2005). The KELS originally categorized parents' education at-
tainment levels into seven types: primary, middle, high school graduates, 2-year junior
college, 4-year university, master degree, and doctorate degree. In this paper, both par-
ents' education attainment variables are regrouped into four categories: middle school (or
less) graduates, high-school graduates, 2-year junior college graduates, 4-year (or more)
university graduates. Parents also provided their average monthly income, their employ-
ment status, and the monthly amount spent on education for each student. The student's
gender and the number of siblings are surveyed. The gender of siblings cannot be identied
but the KELS data asks how many younger/older siblings the student has.
The location variable in the original data admits only the four types based on the
size of the region: Seoul (the capital and the largest city in Korea), six other large cities
(the metropolitan areas), medium-sized cities, and small towns (called the up or myeon
and rural area). After matching the KELS data with the school lists to identify school
district of each sample, the spatial variables are able to specify the detailed location of
each observation in the sample including information on school districts. There exist 131
school districts in the nal sample. Therefore, this paper treats clusters in dierent years
and school districts as dierent clusters, so that there are 262 clusters.
39
2.5.1 Summary Statistics
Table 2.2 describes summary statistics of observed socio-economic variables. While only
20% of students had a mother with a 4-year university degree, more than one-third of the
total sample had a father who had graduated from a 4-year university. 50% of all samples
are male students and 46% live in large city area from statistics of location variables. The
average number of siblings per student is 1.27 and 24% has an older sibling.
In terms of time-varying variables, self-study hours for both English and mathematics
were increasing slightly over time and the variation of self-study hours among individuals is
similar to individual's variation over time. The average monthly household income in the
pooled data over 2006 and 2007 is equivalent to about 3,980 US dollars (4,384,200 KRW)
at the currency rate of 1,100 KRW/$ in the data. Although the father was employed in
almost 98% of the sample, in about 55% of the households, the mother was employed.
Table 2.3 represents summary statistics of private tutoring measured by the partici-
pation rate, expenditure and hours spent on tutoring and the calculated unit value. Since
there are three subjects and two types of private tutoring, all six dierent kinds of private
tutoring exist. One-to-one or a small-group type of English private tutoring is denoted as
"Eng. Small" in the table and a large-scale classes or cram school type of English private
tutoring is denoted as "Eng. Cram". Similarly, each type of private tutoring in other
academic subjects, mathematics and Korean, can be written as "Math Small", "Math
Cram", "Kor. Small", and "Kor. Cram" in the table. Panel A and B report the results
of the 2006 and 2007 samples and the panel C shows the statistics of total sample.
40
The participation rate in each type of private tutoring is shown in the rst column.
For both English and mathematics cases, participation rates of most types are slightly
higher at higher grades. Overall, about fteen percent of students choose one-to-one or
small-group private tutoring for English or mathematics, and 55% of students participate
in a cram school type of private tutoring in English or math. n
+
shows the number of
households that purchased a certain tutoring services.
In the third column, expenditure share is computed by dividing the total educational
expenditure by expenditure of each type of private tutoring. The fourth column indicates
average expenditure on private tutoring conditional on students purchasing it. While
total educational expenditure is 411,000 KRW on average, the average expenditure of
the small-group type of private tutoring is 182,000 to 190,000 KRW (about 44 percent of
total educational expenditure on average). And about 130,000 KRW (32 percent of the
total) is spent on the cram school type. In the fth column, we see that students are
likely to spend fewer hours per week on small-group typez of private tutoring than on
the large-class type. Consequently, once students decide to attend private tutoring, the
unit value of the small-group case of private tutoring tends to be greater than that for
the cram school type.
2.6 Empirical Results
Table 2.4 presents the quality elasticities
1
, the coecient of log total educational ex-
penditure per capita on log unit values in (2.14) as the result of the rst-stage estimation
41
along with the standard errors. These are estimates of the within-cluster 2SLS regression
for six dierent types of private tutoring: two kinds of tutoring (small-group tutoring and
cram schools) and three academic subjects (English, math, and Korean). Although the
regressions include a full set of observation characteristic variables (z), only the coe-
cients on the total educational expenditure are shown. The coecients on other variables
from the rst-stage estimation using all samples are shown in Table A1.3 of the appendix.
The rst column of Table 2.4 shows the result using the total sample. The other columns
show results using dierent sub-samples stratied by the household's characteristics.
In the modelling part, the quality eect of unit values implies that better-o house-
holds pay more per hour. Thus, the positive coecients on total educational spending
are related to the quality eect. In the rst column, all quality elasticities with respect
to total educational expenditure per capita are signicantly positive and the size of the
eects is modest ranging from 10% to 21%. Since small-group types have higher quality
elasticities than do cram schools, small-group types are more likely to be heterogenous
goods.
The quality elasticities can be varied by dierent demographic group. Because the
urban section can oer a greater variety of choices for tutoring than the rural areas can,
the quality eect in rural areas is much smaller than that of urban areas. The rural
and urban areas can be stratied by using the KELS location variable which categorizes
four regions: Seoul, six other metropolitan cities, medium- or small-sized cities, and rural
areas. The rst two locations, Seoul and six other metropolitan cities, are considered
42
as urban and the rest is treated as the rural sector in this paper. The urban quality
elasticity is signicantly positive, ranging from 15% to 24% in the second column. Also,
boys have larger elasticities than girls, and this is particularly the case in cram schools.
After separating baseline test scores into two groups, we nd that those in the top 50%
have signicant larger elasticities than the less able group. This result suggests that more
able students are likely to choose better qualities of tutoring in response to a change in
total educational expenditure.
Also, parents who have higher educational attainment are more quality elastic for
small-groups while quality elasticities for large-scale class types are similar between par-
ents who had higher and lower education. 2-year or 4-year university graduates and
high-school (or less) graduates are categorized into higher educated parents and lower ed-
ucated parents, respectively. In particular, lower educated parents are insignicant eect
for the small-group type. Since hiring individual tutors tends to require greater prices,
it is dicult for less educated parents to choose this as an option, but better educated
parents are more readily able to do so.
In addition to parents' education level, the mother's employment status causes dif-
ferences in quality eects. Working mothers have higher quality eects for cram schools,
whereas housewives have higher elasticities for the small-group type. Since most individ-
ual tutors for small-group tutoring visit students' houses, this type requires the mothers
to stay at home in addition to the nancial support (Kim, 2008). Also, employed mothers
can prefer large-scale classes tutoring, which does not require time investment of mother.
43
Moreover, sending children to cram schools may be regarded as a kind of child-care service
for working mothers.
The quality elasticities that this paper nds are higher than what Deaton estimates.
For example, analyzing the food demand system in Indonesia, Deaton (1990) nds small
responses of unit value to total food expenditure with estimates ranging between 2.9%
and 10% except for fresh sh. Deaton et al. (1992) compares the quality eects of foods
in urban with rural areas in Pakistan and the estimates range between 0.3% to 15% ex-
cept for meats. In their study, relatively heterogeneous meat commodity has the highest
quality elasticity in both areas and is higher in urban areas. Compared to his estimated
elasticities, the quality eects for tutoring are larger than those for foods since private
tutoring as a type of educational services has more heterogeneity than does food com-
modity. Therefore, modeling the quality related to unit values suggested by Deaton can
also be useful in this paper.
In addition to quality elasticities, Table A1.3 reports other coecients of the unit
value equation when using all samples. The father's education level has a signicant
positive eect on the unit values of cram schools. Households with more highly educated
fathers are likely to choose more higher unit values for cram school types. The group
of fathers who graduated from two-year colleges has a signicantly negative estimate on
small-group tutoring in mathematics, while the coecients on fathers with other education
attainment are insignicant. Even though households which have fathers with a two-year
college degree choose small-group types for math, it is hard for them to choose more
44
expensive small-group tutoring because of the higher costs. Unlike fathers, mothers who
are two or four-year university graduates are more likely to consume the small-group type
with higher unit values. This shows that higher educated mothers tend to send their
children to the small-group type with higher quality. In terms of gender, boys tend to
choose private tutoring, which has a lower unit value. This can be interpreted with the
results of Table 2.4 to understand gender dierence in the quality elasticities. Since boys
already tend to choose lower unit values of private tutoring as a result of Table A1.3, it
would be easier to increase the level of unit value in response to the change in expenditure
for the boy group. Hence, boys can have larger quality elasticities.
As a result of the within-cluster 2SLS regression, Table 2.5 represents the coecients

0
, the eect of the logarithm of the educational expenditure per child on the budget
shares, estimated from dierent samples. Since all estimates are signicantly positive in
the rst column of the table, total educational expenditure elasticities are larger than
unity. Thus all types of tutoring can be considered as luxury goods. Note that the
expenditure elasticity can be constructed by (2.27) using the
0
,
1
coecients (Table
2.4 and 2.5) and the average budget share w (Table 2.6). Hence, the total educational
expenditure elasticities are calculated in Table 2.7. Most types of private tutoring have
greater-than-unity elasticities with respect to total educational expenditure except cram
schools for the group with parents with higher education and some Korean tutoring. The
lower educational expenditure elasticity for cram schools indicates that cram schools are
less of luxury goods than are small-groups.
45
Table 2.8 - Table 2.14 present own- and cross-price elasticities for heterogenous sam-
ples divided by observation's characteristics, as well as total sample, with bootstrapped
standard errors.
5
The numbers in boldface are the estimates which are more than twice as
large as their bootstrapped standard errors. Tables of price elasticities can be expressed
as matrices so that each element in rowi and columnj of these tables implies the response
of consumption of one good i to the price of another good j. Table 2.8 - Table 2.14 re-
sults are obtained by completing the system with and without imposing three restrictions:
adding-up, homogeneity and Slutsky symmetry constraints.
The top panel of these tables shows price elasticities that impose no such constraint.
The bottom panel results from the estimation with adding-up, homogeneity, and Slutsky
symmetry constraints. Although adding-up and homogeneity conditions to complete the
system cause the changes mostly in own-price elasticities, imposing symmetry condition
makes for changes in cross-price estimates. A Wald test for the symmetry restriction is
shown in the last row of each table. The Wald test for all samples has a value of 12.0
less than the 95% critical values of
2
with 15 degree of freedom. Since the symmetry
restriction needs only to be applied to the lower(upper) triangle in the matrix (2.36), the
degree of freedom is equal to 15, which is the number of elements in the lower triangle
5
While the standard error of the rst-stage estimates are calculated by STATA, Deaton obtained
variance-covariance matrices for the elasticities by bootstrapping. The cluster-level data set calculated
estimates of the cluster averages of the corrected unit values and budget shares are considered as the
base data from which bootstrap samples are drawn. The bootstrapped standard errors in this paper are
calculated by making 1,000 draws from the second-stage data.
46
of the 6 6 matrix. As the 95% critical value of
2
15
is 25, we can accept the symmetry
restriction.
Diagonal elements in each table describe own-price elasticities. After imposing all three
restrictions, own-price elasticities for all types of tutoring are negative and less than unity
in absolute value. Small-group types tend to be almost three times more price-elastic
than are cram schools. Students who cannot aord to hire individual tutors consume
cram school types and they are less likely to change their consumption in response to an
increase in tuition for large-class type.
Cross-price elasticities are o-diagonal elements in the same tables that the column
of the table is the tutoring type whose price is changing and the row is the type af-
fected. With symmetry restriction, Table 2.8 for the total sample shows positive but
small cross-price elasticities. Small-group and large-scale class types of tutoring are very
weak substitutes less than 2%. As long as households decide to consume a certain type
of tutoring, they tend to stick with their choice. Hence, only very weak substitutability
is found.
Gender Under the belief that gender preference in Korea has existed, boys group may
have relatively smaller elasticities. From Table 2.9, boys and girls have similar own-price
elasticities for small-group types of private tutoring, but only girls' own-price elasticities
for cram schools are signicant. This can be interpreted as the existence of gender dis-
crimination in that parents prefer investments in boys rather than girls. Or, a signicant
dierence in test scores between boys and girls can explain dierences in price elasticities.
47
Tests for dierences in Z-scores between boy and girl groups show that girl groups have
signicantly higher average Z-scores than boys do,: T-statistics for testing dierences in
current average scores is 10.1 and t-statistics for testing dierences in baseline score is
8.4. Because of private tutoring's remedial purpose for less able students, parents who
have a boy would send their child to cram school despite a price increase in fees in order
to enhance the boy's academic achievement. As a result, boys' price elasticities for cram
schools can be insignicant.
Location I expect that accessibility to tutoring service aects students' demand. Since
the supply for cram schools or individual tutors varies over locations, results from each
urban and rural area can test the hypothesis. Each left and right-hand sides of Table
2.10 signify estimates using observations that reside in the rural and urban sector. The
price responsiveness for tutoring in large-scale classes in urban areas is insignicant (or
positive), but the rural area has signicantly negative price elasticities for all types. Since
an urban sector can provide more access to a variety of cram schools, whereas rural areas
cannot, it is easier to nd other cram schools oering the same tuition in cities and then
the own-price elasticity can be insignicant. While the estimates obtained by imposing
only adding-up and homogeneity conditions are not shown in the paper, they have similar
patterns of signicance to the unrestricted estimates and the sizes of own-price elasticities
become smaller. In addition, relatively higher pressure from peers in urban sector can
aect lower price responsiveness.
48
Parentaleducation I speculate that higher educated parents may be more likely to con-
centrate on children's education because they experienced nancial/non-nancial benet
from higher education and tended to become rich. In terms of the hypothesis based on
dierent parental characteristics, own-price elasticities for highly educated parents are
relatively inelastic (Table 2.11 and 2.12). For the small-group type, own-price elasticities
for households that have higher educated fathers are less than unity while households
with less educated fathers have the elasticities closed to one. Mother's educational at-
tainments also show a similar pattern to father's. Hence, parental education can play an
important role as determinants of demand for tutoring activities. Additionally, own-price
elasticities of cram schools for higher educated parents are not signicant. Besides, as
higher educated parents may indicate better-o households, they are better able to keep
the consistency of the level of consumption for cram school in response to price.
Mother's employment Mother's employment status may aect demand for tutoring
service through two sides.: One is income eect and the other is time-constraint eect.
Although mother's employment makes spending more on children's education because of
increases in household income, it can reduce the maternal contribution of time in sup-
plementary tutoring. Since Jung and Lee (2010) shows a negative correlation between
mother's employment status and tutoring expenditure, I suppose that students who have
49
non-working mother have less elastic demand. In terms of mothers' employment as de-
picted in Table 2.13, students who have working mothers are slightly inelastic than stu-
dents whose mothers do not work but the dierence in the size of elasticities is very small.
Hence, the eect of mother's employment on tutoring demand can be mixed.
Baseline test score Because the primary reason of choosing tutoring is an increase in
test score, price elasticities for less able students may be relatively smaller than that for
more able students. Table 2.14 presents the results from two dierent groups divided
into the top and bottom 50 percent in terms of the baseline test score (the average Z-
score in 2005). Similar to the gures for lower-educated fathers (high school or fewer
graduates), the bottom 50% of baseline test scores tend to be more aected by the change
in own-prices than are more able students. Particularly, both lower educated fathers
and less able student groups have larger own-price elasticities for small-group types than
unity. Although supplementary education relatively captivates students with the lower
test score, the price can easily in
uence demand for less able students. It is possible that
lower able students came from lower SES group, which mainly consists of lower educated
parents. Hence, the students group can be heavily aected by price.
The estimated own- and cross- price elasticities provide insight into taxation on private
tutoring. Following Ramsey's inverse-elasticity rule to minimize distortion, taxable goods
tend to have fewer substitutes and inelastic demand. Since all types of tutoring have own-
price elasticities less than unity and very small cross-price elasticities, Ramsey's framework
can be used to understand tax policy. As price elasticity for small-group tutoring is more
50
than three times of that for the cram schools, the inverse-elasticity rule indicates heavier
taxes on cram schools to achieve the eciency.
However, when people with dierent demographic characteristics face consistent tax
rates, this eciency policy can be inequitable. Because of the relatively elastic demand
for households with lower SES proles, such as less-educated parents, their welfare can
be negatively impacted by tax policy. Also, if cram schools with low elasticity are taxed
more, households that choose cram school type, especially those who have more elastic
demand for cram schools, are more likely to be aected. From the estimated results,
taxing these goods highly may more severely aect the welfare of households that have
less-educated parents with higher elastic demand. Therefore, the optimal taxation on the
private tutoring industry may need to consider the resolution of the equity and eciency
trade-o (Myles, 1995).
2.7 Conclusions and Discussion
We nd positive quality elasticities for small-group and cram school types of tutoring with
respect to total educational expenditure that better-o households would pay more per
hour. Thus in order to use unit values as a proxy of price, the quality eect needs to be
addressed. Deaton's demand analysis models the determination of quantity and quality
related to unit values and develops an applicable estimation strategy. Since the quality
elasticities are mostly larger than Deaton's results of quality eects, this paper using unit
values adopts his approach for the analysis of the demands for private tutoring
51
As a result of the rst stage estimation, both types of private tutoring are normal
goods with positive estimated elasticities of total educational expenditure. Especially,
one-to-one or small-group types are strongly demonstrated to be luxury goods. The
own-price elasticities for all kinds are negative and small-group tutoring is almost three
times as elastic as large-scale class tutoring. Since the price of small-group tutoring is
higher than that of the cram schools, students who choose cram school types because
of relatively expensive small-group types tend to prefer to consume continuously even if
prices are increased. Also because of the higher price for small-group tutoring, consumers
who choose small-group tutoring are more sensitive in response to the same percentage
increase in prices of both tutoring. By highlighting the substitutability between dierent
types, a very weak substitute relationship between small-group and cram school tutoring
is determined because most cross-price elasticities between both goods are less than two
percent.
Own-price elasticities for cram schools in the urban areas are insignicant because
students who live in urban sectors can more easily nd other cram schools oering the same
tuition. More educated parents are likely to have inelastic demand because households
of parents with a higher level of educational attainment are by implication better-o and
can more easily aord to maintain the level of consumption. Also, students who perform
better academic achievement in the baseline year are less likely to change in their demand
for tutoring by increases in prices.
52
Deaton's procedure has been applied in several other studies (Deaton et al., 2004;
Friedman and Levinsohn, 2002) using unit values as price proxies. However, it cannot
address sample selection eects because unit values are not reported for households that
do not purchase tutoring. Because of the weakness of the unit value method, even the
estimation methodology developed by Deaton (1990) to correct quality biases can produce
inaccurate and imprecise results. Gibson and Rozelle (2005) and McKelvey (2011) suggest
potential solution: One possibility is that a demand structure is estimated for narrowly
dened goods. Another is to use quantity as the dependent variable instead of the budget
share to eliminate bias due to quality substitution. The last suggestion is to collect both
price and unit value variables at the survey stage. Particularly, price data reported by
consumer is better than collecting the market price in surveys (Gibson and Rozelle, 2005).
The price elasticities we estimate can illustrate the rst approximation for taxation
on the private tutoring industry. According to the inverse-elasticity rule, cram schools
with inelastic demand are taxed more heavily than are small-groups. But the equity-
eciency trade-o can more severely impact the welfare of households of a lower economic
status. Moreover, since private tutoring is regarded as a tool to increase competitiveness
in college entrance exams, tax programs might more lead lower SES households to reduce
their spending on private tutoring, which in turn could lead to educational inequality.
Therefore, an imposed tax policy needs to correct the trade-o between two aims of
eciency and equity while considering distributional issues. Although this paper does
not conclude what the optimal tax rate is, future research to resolve the con
ict between
53
eciency and equity using the Diamond-Mirrlees tax rule will be helpful to expand the
policy implication (Myles, 1995).
As we pointed out, taxing private tutoring should require a necessary condition that
tutors who oer only small-group types of instruction should be forced to register or report
their business to the provincial oces of education. Unlike cram schools, one-to-one or
small-group tutoring is likely to be conduct under-the-table and thus avoid taxation.
Hence, policy design would include a means to prevent private tutoring from becoming
an illicit activity.
54
Figure 2.1: Time Trend of Public and Private Education Expenditure
Source: Korean Household Consumption survey conducted by the Korean National Statistics Oce and
Kim and Yang (2011)
55
Figure 2.2: Time Trend of Private Tutoring Expenditure
Source: Korean Household Consumption survey conducted by the Korean National Statistics Oce and
Kim and Yang (2011); Note: Each line shows private tutoring expenditure by deciles of household income
levels. The rst decile group is the lowest 10% and the tenth decile group means the highest 10%
56
Table 2.1: Participation rate and monthly expenditure in private tutoring from 2007 to
2014
Year Total
Small- Cram Paper
Internet
group school corresp.
Participation rate (%)
2007 71 20.8 56.4 14.5 3.2
2008 68.8 19.8 55.9 11.7 3.1
2009 70.1 23.7 55.7 11.6 3.6
2010 68.5 23.1 53.8 10.5 4.2
2011 66.6 22.7 52 8.9 3.3
2012 65.9 24.9 51.5 7.4 3.4
2013 64.1 22.2 50.8 6.5 2.7
2014 63.9 22.8 50.3 5.8 2.6
Monthly expenditure (10,000KRW)
2007 21.9 5.4 15.2 1 0.3
2008 22.6 5.1 16.3 0.8 0.3
2009 24.3 6.4 16.7 0.8 0.3
2010 23.8 6.3 16.4 0.7 0.4
2011 24.3 6.6 16.8 0.7 0.3
2012 25.5 7.2 17.4 0.6 0.3
2013 24.5 6.4 17.4 0.5 0.2
2014 24.8 6.7 17.4 0.5 0.2
Source: The Survey of Private Education Expenditure conducted
by Statistics Korea
Note: Each panel is participation rate and total average expen-
diture for all types and each type of private tutoring. Average
expenditure for private tutoring is calculated only for students
who attend private tutoring. `Small-group' and `cram school'
types indicate hiring private tutor and large-class type, respec-
tively. `Paper correspondence' type means private tutoring that
materials to study deliver to students via mail or tutor's visit.
`Internet' implies private tutoring through broadcasting services
or (paid) internet.
57
Table 2.2: Summary Statistics
Variables Mean S.D Variables Mean S.D
Gender (boy=1, girl=0) 0.50 0.50 Number of Siblings 1.27 0.56
ratio of upper siblings 0.24 0.25
Father's Education ratio of lower siblings 0.30 0.27
Middle school grad or less 0.07 0.25
High school grad 0.47 0.50 ln (Household income)
2-year college 0.13 0.33 in 2006 5.76 0.72
4-year university or more 0.34 0.47 in 2007 5.87 0.72
ln (Edu. Expenditure per child)
Mother's Education in 2006 3.47 0.71
Middle school grad or less 0.07 0.26 in 2007 3.50 0.82
High school grad 0.64 0.48
2 years university 0.10 0.30 Mother's Employment
4 years university or more 0.19 0.39 in 2006 0.50 0.50
in 2007 0.62 0.49
Location
Seoul 0.18 0.39 Self Study Hours of English
6 Metropolitan cities 0.28 0.45 in 2006 2.03 2.04
Medium and small size cities 0.46 0.50 in 2007 2.14 2.15
Rural area 0.08 0.27 Self Study Hours of Math
in 2006 2.12 2.04
House owned 0.69 0.46 in 2007 2.27 2.20
Observations 5820
58
Table 2.3: Summary statistics for private tutoring
Participate
n
+
Budget share Expenditure Hours ln(unit value)
rate purchased purchased purchased purchased
(%) Mean S.D Mean S.D Mean S.D Mean S.D
Panel A: year 2006 (n=2910)
Eng. Small 12% 351 0.31 0.18 15.30 11.80 3.41 2.32 0.79 0.40
Eng. Cram 55% 1561 0.29 0.17 11.30 7.49 4.48 2.19 0.52 0.29
Math. Small 13% 381 0.32 0.18 16.00 11.50 3.74 1.85 0.74 0.36
Math. Cram 55% 1555 0.29 0.17 11.10 7.64 4.46 2.18 0.51 0.29
Kor. Small 3% 98 0.17 0.10 8.36 6.33 3.11 1.98 0.58 0.35
Kor. Cram 37% 1077 0.22 0.14 7.94 4.63 3.75 2.17 0.48 0.27
Panel B: year 2007 (n=2910)
Eng. Small 16% 428 0.31 0.18 20.50 12.10 3.51 1.99 0.93 0.35
Eng. Cram 54% 1559 0.34 0.20 14.90 9.93 4.80 3.15 0.62 0.33
Math. Small 16% 455 0.43 0.22 21.50 12.40 3.64 1.89 0.93 0.36
Math. Cram 55% 1578 0.34 0.20 14.60 9.26 4.76 2.86 0.61 0.33
Kor. Small 5% 125 0.30 0.19 14.60 10.40 3.25 2.32 0.81 0.39
Kor. Cram 36% 1043 0.25 0.16 10.30 6.28 4.01 2.58 0.55 0.30
Panel C: Total Sample (n=5820)
Eng. Small 14% 779 0.36 0.20 18.20 12.20 3.46 2.14 0.87 0.38
Eng. Cram 55% 3120 0.32 0.19 13.10 8.97 4.64 2.72 0.57 0.31
Math. Small 15% 836 0.38 0.21 19.00 12.30 3.68 1.87 0.85 0.37
Math. Cram 55% 3133 0.31 0.19 12.90 8.67 4.61 2.55 0.56 0.31
Kor. Small 4% 223 0.24 0.17 11.80 9.36 3.19 2.17 0.71 0.39
Kor. Cram 37% 2120 0.23 0.15 9.10 5.63 3.88 2.38 0.52 0.29
Note: The second column n
+
means the number of observations who purchased a certain type of private tutoring.
Also, other columns are the average values over observations who purchased each type of private tutoring.
59
Table 2.4: Quality elasticity (
1
): the rst stage estimation of unit values
All
Urban Gender Father's education Mother's education Mother's labor Baseline score
Rural Urban Boys Girls Lower Higher Lower Higher Unemp. Emp. < 50% 50%
Eng. Small 0.213*** 0.167* 0.235*** 0.169** 0.145* 0.057 0.242** 0.083 0.238* 0.227*** 0.139* 0.052 0.292***
(S.E.) (0.06) (0.08) (0.07) (0.06) (0.07) (0.05) (0.08) (0.07) (0.10) (0.06) (0.06) (0.05) (0.07)
Eng. Cram 0.121*** 0.080* 0.159*** 0.123*** 0.098** 0.123*** 0.108** 0.115*** 0.118* 0.072* 0.144*** 0.093** 0.143***
(S.E.) (0.03) (0.03) (0.04) (0.03) (0.03) (0.03) (0.03) (0.03) (0.05) (0.03) (0.03) (0.03) (0.03)
Math Small 0.164** 0.051 0.256*** 0.217** 0.112 0.040 0.239** -0.035 0.342*** 0.133 0.114 0.114 0.275***
(S.E.) (0.06) (0.07) (0.08) (0.08) (0.06) (0.05) (0.08) (0.07) (0.10) (0.07) (0.06) (0.07) (0.08)
Math Cram 0.125*** 0.093** 0.153*** 0.156*** 0.074** 0.136*** 0.115*** 0.113*** 0.146** 0.048 0.180*** 0.099** 0.144***
(S.E.) (0.02) (0.03) (0.04) (0.03) (0.03) (0.03) (0.03) (0.03) (0.05) (0.03) (0.03) (0.03) (0.03)
Kor. Small 0.163** 0.143 0.201** 0.153** 0.105 -0.011 0.229*** 0.013 0.311 0.056 0.028 0.099 0.176
(S.E.) (0.05) (0.09) (0.07) (0.06) (0.07) (0.06) (0.07) (0.05) (0.16) (0.04) (0.04) (0.06) (0.12)
Kor. Cram 0.106*** 0.052 0.167*** 0.114** 0.072* 0.073* 0.109*** 0.088** 0.147*** 0.062 0.120*** 0.130*** 0.100***
(S.E.) (0.03) (0.03) (0.05) (0.04) (0.03) (0.03) (0.03) (0.03) (0.04) (0.04) (0.03) (0.04) (0.03)
Note: Standard errors are shown in brackets. * p<0.05, ** p<0.01, *** p<0.001
60
Table 2.5: The coecient
0
on the total educational expenditure per capita: the rst stage estimation of budget share
All
Urban Gender Father's edu Mother's edu Mother's labor Ability
Rural Urban Boys Girls Lower Higher Lower Higher Unemp. Emp. < 50% 50%
Eng. Small 0.041*** 0.037*** 0.046*** 0.052*** 0.031** 0.034*** 0.036*** 0.034*** 0.047*** 0.044*** 0.039*** 0.036*** 0.032**
(S.E.) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01)
Eng. Cram 0.027** 0.031** 0.024 0.024 0.027* 0.039*** 0.005 0.034*** 0.014 0.028** 0.024 0.019 0.019
(S.E.) (0.01) (0.01) (0.02) (0.01) (0.01) (0.01) (0.02) (0.01) (0.02) (0.01) (0.01) (0.01) (0.01)
Math Small 0.048*** 0.046*** 0.049*** 0.060*** 0.033*** 0.039*** 0.042*** 0.039*** 0.063*** 0.045*** 0.044*** 0.036*** 0.043***
(S.E.) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01)
Math Cram 0.035*** 0.037** 0.036* 0.024 0.045*** 0.043*** 0.018 0.037*** 0.028 0.035** 0.032* 0.023* 0.032*
(S.E.) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.02) (0.01) (0.01) (0.01) (0.01)
Kor. Small 0.010*** 0.008** 0.011** 0.016*** 0.004 0.006* 0.013*** 0.007** 0.012* 0.012*** 0.008* 0.010*** 0.006
(S.E.) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
Kor. Cram 0.005 0.004 0.005 -0.008 0.017* 0.013 -0.007 0.009 -0.003 0.007 0.009 0.011 -0.006
(S.E.) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01)
Note: Standard errors are shown in brackets. * p<0.05, ** p<0.01, *** p<0.001
61
Table 2.6: Average budget share ( w)
All
Urban Gender Father's edu Mother's edu Mother's job Ability
Rural Urban Boys Girls Lower Higher Lower Higher Unemp. Emp. < 50% 50%
Eng. Small 0.05 0.05 0.05 0.05 0.05 0.04 0.06 0.04 0.07 0.05 0.05 0.04 0.05
Eng. Cram 0.17 0.16 0.19 0.18 0.17 0.15 0.20 0.15 0.21 0.18 0.17 0.13 0.20
Math Small 0.05 0.05 0.06 0.05 0.06 0.04 0.07 0.04 0.08 0.06 0.05 0.04 0.06
Math Cram 0.17 0.15 0.18 0.18 0.16 0.14 0.20 0.15 0.21 0.17 0.17 0.13 0.20
Kor. Small 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01
Kor. Cram 0.09 0.08 0.09 0.10 0.07 0.09 0.09 0.09 0.08 0.08 0.09 0.08 0.09
Table 2.7: Total educational expenditure elasticities (e = 1
1
+
0
w
1
)
All
Urban Gender Father's education Mother's education Mother's job Ability
Rural Urban Boys Girls Lower Higher Lower Higher Unemp. Emp. < 50% 50%
Eng. Small 1.630 1.576 1.742 1.905 1.483 1.878 1.343 1.780 1.414 1.695 1.651 1.846 1.296
(S.E.) (0.060) (0.080) (0.110) (0.070) (0.070) (0.070) (0.060) (0.060) (0.080) (0.090) (0.070) (0.090) (0.060)
Eng. Cram 1.040 1.119 0.971 1.016 1.064 1.141 0.920 1.107 0.948 1.088 1.001 1.057 0.951
(S.E.) (0.010) (0.010) (0.030) (0.010) (0.020) (0.010) (0.000) (0.000) (0.010) (0.010) (0.010) (0.020) (0.010)
Math Small 1.716 1.836 1.596 1.946 1.467 1.948 1.357 1.940 1.425 1.687 1.698 1.725 1.414
(S.E.) (0.040) (0.060) (0.090) (0.050) (0.060) (0.050) (0.040) (0.040) (0.050) (0.070) (0.050) (0.070) (0.040)
Math Cram 1.082 1.149 1.040 0.978 1.200 1.167 0.977 1.129 0.983 1.153 1.014 1.077 1.017
(S.E.) (0.010) (0.020) (0.030) (0.010) (0.020) (0.010) (0.010) (0.010) (0.010) (0.010) (0.000) (0.010) (0.010)
Kor. Small 1.897 1.747 2.052 2.361 1.345 1.675 2.047 1.788 1.751 2.454 1.745 1.745 1.673
(S.E.) (0.030) (0.090) (0.160) (0.050) (0.080) (0.060) (0.060) (0.050) (0.050) (0.070) (0.040) (0.070) (0.040)
Kor. Cram 0.952 1.000 0.895 0.799 1.154 1.076 0.811 1.012 0.822 1.026 0.986 1.002 0.831
(S.E.) (0.010) (0.010) (0.040) (0.010) (0.020) (0.010) (0.010) (0.010) (0.020) (0.010) (0.000) (0.020) (0.010)
Note: Standard errors are shown in brackets. Estimates in the boldface are twice greater than its standard error.
62
Table 2.8: Own- and cross-price elasticities: All
English Math Korean
Small Cram Small Cram Small Cram
Unconstrained estimates
Eng. Small -0.702 0.020 0.018 0.015 -0.005 0.009
(S.E.) (0.014) (0.004) (0.003) (0.005) (0.001) (0.003)
Eng. Cram 0.011 -0.209 0.007 0.060 -0.002 0.042
(S.E.) (0.007) (0.049) (0.005) (0.011) (0.001) (0.009)
Math Small 0.016 0.012 -0.692 0.011 -0.004 0.009
(S.E.) (0.003) (0.003) (0.015) (0.004) (0.001) (0.002)
Math Cram 0.009 0.063 0.007 -0.214 0.000 0.044
(S.E.) (0.008) (0.010) (0.006) (0.049) (0.002) (0.008)
Kor. Small -0.004 -0.002 -0.003 0.000 -1.069 -0.002
(S.E.) (0.002) (0.001) (0.001) (0.002) (0.003) (0.001)
Kor. Cram 0.006 0.048 0.006 0.048 -0.002 -0.241
(S.E.) (0.004) (0.008) (0.003) (0.008) (0.001) (0.045)
Symmetry-constrained estimates
Eng. Small -0.696 0.020 0.018 0.015 0.004 0.009
(S.E.) (0.014) (0.003) (0.003) (0.004) (0.001) (0.003)
Eng. Cram 0.009 -0.171 0.006 0.049 0.007 0.035
(S.E.) (0.007) (0.040) (0.005) (0.010) (0.001) (0.009)
Math Small 0.016 0.012 -0.677 0.011 0.003 0.009
(S.E.) (0.003) (0.003) (0.015) (0.003) (0.001) (0.002)
Math Cram 0.007 0.052 0.006 -0.175 0.007 0.036
(S.E.) (0.008) (0.009) (0.006) (0.039) (0.002) (0.008)
Kor. Small 0.011 0.010 0.007 0.010 -1.071 0.005
(S.E.) (0.002) (0.001) (0.001) (0.001) (0.004) (0.001)
Kor. Cram 0.006 0.045 0.006 0.045 0.009 -0.228
(S.E.) (0.004) (0.007) (0.003) (0.006) (0.001) (0.043)
Wald test. 18.69 (0.811)
Note: Bootstrapped standard errors are shown in brackets. Estimates in the bold-
face are twice greater than its standard error. For the Wald test statistics in the
last row, the number in parenthesis is p-value at the 95% critical value of
2
(25).
63
Table 2.9: Own- and cross-price elasticities: Boys and girls
Boys Girls
English Math Korean English Math Korean
Small Cram Small Cram Small Cram Small Cram Small Cram Small Cram
Unconstrained estimates Unconstrained estimates
Eng. Small -0.781 0.010 0.010 0.010 -0.016 0.004 -0.779 0.005 0.005 0.004 -0.007 0.003
(S.E.) (0.009) (0.003) (0.002) (0.003) (0.002) (0.002) (0.011) (0.002) (0.001) (0.001) (0.002) (0.001)
Eng. Cram 0.006 0.080 0.010 0.075 -0.004 0.044 0.003 -0.313 0.002 0.030 -0.001 0.017
(S.E.) (0.005) (0.046) (0.007) (0.016) (0.005) (0.008) (0.003) (0.036) (0.002) (0.005) (0.007) (0.003)
Math Small 0.011 0.019 -0.724 0.018 -0.017 0.012 0.005 0.003 -0.759 0.004 -0.008 0.004
(S.E.) (0.002) (0.004) (0.013) (0.004) (0.002) (0.002) (0.001) (0.001) (0.012) (0.001) (0.003) (0.001)
Math Cram 0.006 0.082 0.010 0.093 -0.001 0.048 0.003 0.031 0.003 -0.306 0.003 0.016
(S.E.) (0.005) (0.015) (0.008) (0.048) (0.006) (0.008) (0.002) (0.005) (0.002) (0.037) (0.006) (0.003)
Kor. Small -0.008 -0.004 -0.008 -0.001 -1.285 -0.008 -0.003 0.000 -0.004 0.002 -1.290 -0.003
(S.E.) (0.005) (0.006) (0.005) (0.007) (0.015) (0.005) (0.005) (0.011) (0.006) (0.009) (0.027) (0.007)
Kor. Cram 0.002 0.046 0.006 0.047 -0.009 -0.169 0.002 0.019 0.003 0.017 -0.005 -0.415
(S.E.) (0.003) (0.007) (0.005) (0.008) (0.005) (0.043) (0.001) (0.003) (0.001) (0.003) (0.004) (0.031)
Symmetry-constrained estimates Symmetry-constrained estimates
Eng. Small -0.767 0.010 0.010 0.009 -0.015 0.004 -0.766 0.005 0.005 0.004 -0.007 0.003
(S.E.) (0.009) (0.002) (0.002) (0.003) (0.014) (0.001) (0.011) (0.001) (0.001) (0.001) (0.029) (0.006)
Eng. Cram 0.006 0.065 0.009 0.061 0.017 0.036 0.003 -0.244 0.002 0.023 0.006 0.013
(S.E.) (0.004) (0.039) (0.007) (0.014) (0.006) (0.007) (0.003) (0.028) (0.002) (0.003) (0.201) (0.012)
Math Small 0.011 0.019 -0.715 0.018 -0.017 0.011 0.004 0.003 -0.731 0.004 -0.008 0.004
(S.E.) (0.002) (0.003) (0.012) (0.004) (0.019) (0.002) (0.001) (0.001) (0.011) (0.001) (0.043) (0.005)
Math Cram 0.006 0.071 0.010 0.080 0.018 0.041 0.002 0.022 0.002 -0.217 -0.002 0.011
(S.E.) (0.005) (0.012) (0.008) (0.043) (0.007) (0.008) (0.002) (0.004) (0.002) (0.027) (0.164) (0.010)
Kor. Small -0.019 0.105 -0.028 0.098 -1.289 0.125 -0.009 0.112 -0.009 -0.093 -1.293 0.318
(S.E.) (0.003) (0.005) (0.004) (0.006) (0.015) (0.005) (0.004) (0.009) (0.006) (0.006) (0.460) (0.602)
Kor. Cram 0.005 0.044 0.008 0.044 0.029 -0.160 0.003 0.017 0.004 0.015 0.033 -0.383
(S.E.) (0.003) (0.006) (0.004) (0.007) (0.006) (0.042) (0.001) (0.002) (0.001) (0.002) (6.066) (0.085)
Wald test 12.37 (0.983) 2.75 (1.000)
Note: Bootstrapped standard errors are shown in brackets. Estimates in the boldface are twice greater than its standard error. For the Wald test
statistics in the last row, the number in parenthesis is p-value at the 95% critical value of
2
(25).
64
Table 2.10: Own- and cross-price elasticities: Rural and urban sections
Rural Urban
English Math Korean English Math Korean
Small Cram Small Cram Small Cram Small Cram Small Cram Small Cram
Unconstrained estimates Unconstrained estimates
Eng. Small -0.756 0.006 0.004 0.005 -0.003 0.005 -0.668 0.071 0.050 0.066 -0.007 0.038
(S.E.) (0.015) (0.002) (0.001) (0.003) (0.002) (0.002) (0.028) (0.032) (0.025) (0.030) (0.003) (0.018)
Eng. Cram 0.004 -0.411 0.000 0.026 -0.001 0.018 0.041 0.302 0.056 0.203 -0.007 0.131
(S.E.) (0.003) (0.044) (0.001) (0.006) (0.002) (0.005) (0.056) (0.129) (0.075) (0.091) (0.006) (0.054)
Math Small 0.003 0.000 -0.830 -0.001 -0.002 0.002 0.052 0.100 -0.553 0.091 -0.009 0.062
(S.E.) (0.002) (0.001) (0.012) (0.002) (0.002) (0.001) (0.024) (0.042) (0.041) (0.038) (0.003) (0.021)
Math Cram 0.003 0.029 -0.001 -0.391 0.003 0.017 0.037 0.200 0.050 0.220 -0.007 0.121
(S.E.) (0.004) (0.005) (0.002) (0.047) (0.003) (0.004) (0.053) (0.092) (0.070) (0.125) (0.006) (0.051)
Kor. Small -0.003 -0.001 -0.002 0.004 -1.125 -0.002 -0.006 -0.011 -0.008 -0.011 -1.072 -0.007
(S.E.) (0.003) (0.001) (0.001) (0.003) (0.009) (0.002) (0.003) (0.004) (0.004) (0.004) (0.005) (0.004)
Kor. Cram 0.003 0.021 0.002 0.018 -0.002 -0.384 0.024 0.144 0.038 0.135 -0.005 0.062
(S.E.) (0.003) (0.004) (0.001) (0.004) (0.002) (0.049) (0.028) (0.049) (0.035) (0.046) (0.005) (0.093)
Symmetry-constrained estimates Symmetry-constrained estimates
Eng. Small -0.744 0.006 0.004 0.004 0.008 0.005 -0.664 0.071 0.050 0.065 0.002 0.038
(S.E.) (0.014) (0.001) (0.001) (0.002) (0.002) (0.002) (0.026) (0.025) (0.022) (0.022) (0.003) (0.016)
Eng. Cram 0.003 -0.310 0.000 0.020 0.006 0.013 0.035 0.257 0.048 0.173 0.009 0.112
(S.E.) (0.003) (0.033) (0.001) (0.005) (0.002) (0.004) (0.050) (0.100) (0.067) (0.067) (0.006) (0.046)
Math Small 0.003 0.000 -0.760 -0.001 0.002 0.001 0.051 0.100 -0.549 0.091 0.004 0.061
(S.E.) (0.001) (0.001) (0.010) (0.001) (0.002) (0.001) (0.022) (0.032) (0.038) (0.029) (0.003) (0.019)
Math Cram 0.002 0.023 -0.001 -0.306 0.005 0.014 0.031 0.167 0.042 0.183 0.008 0.101
(S.E.) (0.004) (0.004) (0.001) (0.037) (0.003) (0.003) (0.047) (0.069) (0.062) (0.097) (0.006) (0.044)
Kor. Small 0.036 0.021 0.010 0.013 -1.126 0.015 -0.001 -0.001 0.002 -0.001 -1.078 -0.003
(S.E.) (0.002) (0.001) (0.001) (0.002) (0.008) (0.002) (0.003) (0.003) (0.004) (0.003) (0.005) (0.003)
Kor. Cram 0.003 0.018 0.002 0.016 0.008 -0.339 0.023 0.141 0.037 0.132 0.010 0.060
(S.E.) (0.002) (0.003) (0.001) (0.003) (0.002) (0.044) (0.026) (0.036) (0.032) (0.033) (0.005) (0.090)
Wald test 10.14 (0.996) 6.78 (0.999)
Note: Bootstrapped standard errors are shown in brackets. Estimates in the boldface are twice greater than its standard error. For the Wald test
statistics in the last row, the number in parenthesis is p-value at the 95% critical value of
2
(25).
65
Table 2.11: Own- and cross-price elasticities: Father's education
Lower educated father Higher educated father
English Math Korean English Math Korean
Small Cram Small Cram Small Cram Small Cram Small Cram Small Cram
Unconstrained estimates Unconstrained estimates
Eng. Small -0.959 0.001 0.000 0.000 0.000 0.000 -0.660 0.005 0.027 0.006 -0.055 0.003
(S.E.) (0.002) (0.000) (0.000) (0.000) (0.000) (0.000) (0.015) (0.002) (0.005) (0.002) (0.007) (0.002)
Eng. Cram 0.001 -0.246 0.000 0.041 0.010 0.015 0.002 -0.040 0.004 0.030 -0.019 0.024
(S.E.) (0.000) (0.036) (0.000) (0.006) (0.001) (0.002) (0.004) (0.044) (0.005) (0.006) (0.007) (0.005)
Math Small 0.000 0.000 -0.930 0.000 -0.002 0.000 0.024 0.008 -0.597 0.008 -0.060 0.008
(S.E.) (0.000) (0.000) (0.003) (0.001) (0.000) (0.000) (0.005) (0.003) (0.020) (0.003) (0.010) (0.003)
Math Cram 0.001 0.044 0.000 -0.202 0.021 0.018 0.003 0.034 0.005 0.036 -0.015 0.029
(S.E.) (0.000) (0.006) (0.001) (0.041) (0.002) (0.003) (0.004) (0.005) (0.005) (0.047) (0.008) (0.005)
Kor. Small 0.000 0.001 0.000 0.003 -1.155 0.000 -0.022 -0.016 -0.027 -0.011 -1.377 -0.015
(S.E.) (0.001) (0.009) (0.001) (0.015) (0.007) (0.007) (0.018) (0.009) (0.021) (0.010) (0.033) (0.010)
Kor. Cram 0.000 0.015 0.000 0.017 0.002 -0.419 0.002 0.026 0.004 0.027 -0.018 -0.183
(S.E.) (0.000) (0.002) (0.000) (0.003) (0.001) (0.028) (0.004) (0.004) (0.005) (0.005) (0.008) (0.036)
Symmetry-constrained estimates Symmetry-constrained estimates
Eng. Small -0.922 0.001 0.000 0.000 0.000 0.000 -0.656 0.005 0.027 0.005 -0.055 0.003
(S.E.) (0.001) (0.000) (0.000) (0.000) (0.002) (0.000) (0.016) (0.002) (0.008) (0.002) (0.088) (0.002)
Eng. Cram 0.001 -0.208 0.000 0.035 0.008 0.013 0.009 -0.030 0.009 0.023 0.030 0.020
(S.E.) (0.000) (0.029) (0.000) (0.006) (0.670) (0.002) (0.005) (0.035) (0.006) (0.004) (0.020) (0.004)
Math Small 0.000 0.000 -0.865 0.000 -0.001 0.000 0.024 0.007 -0.590 0.008 -0.059 0.007
(S.E.) (0.000) (0.000) (0.002) (0.001) (0.001) (0.000) (0.009) (0.002) (0.021) (0.002) (0.180) (0.003)
Math Cram 0.001 0.038 0.000 -0.175 0.018 0.016 0.008 0.027 0.009 0.028 0.028 0.023
(S.E.) (0.000) (0.005) (0.001) (0.034) (0.853) (0.003) (0.004) (0.004) (0.006) (0.038) (0.018) (0.004)
Kor. Small 0.000 0.002 0.000 0.003 -1.168 0.000 -0.045 0.498 -0.057 0.462 -1.422 0.276
(S.E.) (0.000) (0.033) (0.000) (0.044) (0.058) (0.014) (0.015) (0.007) (0.026) (0.008) (0.059) (0.010)
Kor. Cram 0.000 0.014 0.000 0.015 0.002 -0.380 0.011 0.026 0.013 0.027 0.045 -0.174
(S.E.) (0.000) (0.002) (0.000) (0.003) (0.149) (0.025) (0.005) (0.003) (0.006) (0.004) (0.013) (0.037)
Wald test 4.35 (0.999) 16.55 (0.897)
Note: Bootstrapped standard errors are shown in brackets. Estimates in the boldface are twice greater than its standard error. For the Wald test
statistics in the last row, the number in parenthesis is p-value at the 95% critical value of
2
(25).
66
Table 2.12: Own- and cross-price elasticities: Mother's education
Lower educated mother Higher educated mother
English Math Korean English Math Korean
Small Cram Small Cram Small Cram Small Cram Small Cram Small Cram
Unconstrained estimates Unconstrained estimates
Eng. Small -0.856 0.005 0.000 0.004 -0.002 0.002 -0.801 0.003 0.005 0.002 -0.044 0.004
(S.E.) (0.006) (0.001) (0.000) (0.001) (0.001) (0.001) (0.010) (0.001) (0.002) (0.001) (0.007) (0.001)
Eng. Cram 0.005 -0.206 -0.001 0.057 0.000 0.029 0.002 0.074 0.002 0.033 -0.049 0.030
(S.E.) (0.002) (0.038) (0.001) (0.012) (0.003) (0.005) (0.002) (0.050) (0.005) (0.008) (0.028) (0.009)
Math Small 0.000 -0.001 -0.870 -0.001 -0.002 0.000 0.006 0.005 -0.682 0.001 -0.077 0.007
(S.E.) (0.000) (0.001) (0.005) (0.001) (0.000) (0.001) (0.001) (0.002) (0.015) (0.002) (0.012) (0.003)
Math Cram 0.004 0.058 -0.001 -0.195 0.004 0.030 0.001 0.041 0.001 0.071 -0.045 0.032
(S.E.) (0.002) (0.011) (0.001) (0.040) (0.004) (0.005) (0.002) (0.007) (0.004) (0.054) (0.023) (0.009)
Kor. Small -0.001 0.000 -0.001 0.003 -1.183 -0.001 -0.033 -0.075 -0.052 -0.055 -1.835 -0.069
(S.E.) (0.001) (0.004) (0.001) (0.006) (0.008) (0.004) (0.009) (0.019) (0.017) (0.019) (0.115) (0.022)
Kor. Cram 0.002 0.029 0.000 0.030 -0.002 -0.293 0.003 0.040 0.004 0.035 -0.062 -0.152
(S.E.) (0.001) (0.005) (0.001) (0.005) (0.003) (0.034) (0.002) (0.007) (0.005) (0.008) (0.025) (0.044)
Symmetry-constrained estimates Symmetry-constrained estimates
Eng. Small -0.831 0.005 0.000 0.004 -0.002 0.002 -0.789 0.003 0.005 0.002 -0.043 0.003
(S.E.) (0.006) (0.001) (0.000) (0.001) (0.003) (0.001) (0.010) (0.001) (0.002) (0.003) (0.162) (0.068)
Eng. Cram 0.004 -0.170 0.000 0.047 0.006 0.024 0.018 0.059 0.021 0.028 0.039 0.028
(S.E.) (0.001) (0.032) (0.001) (0.010) (0.003) (0.005) (0.004) (0.037) (0.007) (0.011) (0.939) (0.197)
Math Small 0.000 -0.001 -0.971 -0.001 -0.002 0.000 0.006 0.005 -0.682 0.001 -0.077 0.006
(S.E.) (0.000) (0.001) (0.006) (0.001) (0.002) (0.001) (0.002) (0.002) (0.017) (0.004) (0.263) (0.103)
Math Cram 0.003 0.048 -0.001 -0.161 0.004 0.025 0.025 0.037 0.027 0.061 0.069 0.033
(S.E.) (0.002) (0.009) (0.001) (0.036) (0.004) (0.005) (0.003) (0.005) (0.006) (0.094) (4.511) (0.174)
Kor. Small -0.003 0.043 -0.001 0.001 -1.172 0.052 -0.252 0.786 -0.285 1.500 -2.302 4.131
(S.E.) (0.001) (0.004) (0.001) (0.005) (0.008) (0.004) (0.010) (0.014) (0.017) (0.183) (3.375) (2.123)
Kor. Cram 0.002 0.027 0.000 0.028 0.010 -0.272 0.140 0.067 0.156 0.060 0.548 -0.098
(S.E.) (0.001) (0.004) (0.001) (0.004) (0.003) (0.032) (0.006) (0.005) (0.011) (0.015) (16.246) (1.185)
Wald test 8.62 (0.999) 1.04 (1.000)
Note: Bootstrapped standard errors are shown in brackets. Estimates in the boldface are twice greater than its standard error. For the Wald test
statistics in the last row, the number in parenthesis is p-value at the 95% critical value of
2
(25).
67
Table 2.13: Own- and cross-price elasticities: Mother's employment
Unemployed mother Employed mother
English Math Korean English Math Korean
Small Cram Small Cram Small Cram Small Cram Small Cram Small Cram
Unconstrained estimates Unconstrained estimates
Eng. Small -0.851 0.004 0.005 0.003 -0.025 0.005 -0.825 0.006 0.005 0.006 -0.006 0.003
(S.E.) (0.005) (0.002) (0.001) (0.001) (0.010) (0.002) (0.007) (0.002) (0.001) (0.002) (0.002) (0.001)
Eng. Cram 0.002 -0.240 0.002 0.016 -0.018 0.015 0.006 -0.070 0.002 0.080 -0.005 0.048
(S.E.) (0.003) (0.034) (0.002) (0.003) (0.024) (0.004) (0.002) (0.041) (0.002) (0.012) (0.004) (0.007)
Math Small 0.004 0.002 -0.825 0.002 -0.017 0.003 0.005 0.002 -0.799 0.003 -0.008 0.002
(S.E.) (0.001) (0.002) (0.007) (0.001) (0.008) (0.001) (0.001) (0.002) (0.009) (0.002) (0.003) (0.002)
Math Cram 0.002 0.017 0.001 -0.290 -0.011 0.013 0.006 0.089 0.003 -0.029 0.000 0.055
(S.E.) (0.002) (0.003) (0.002) (0.036) (0.019) (0.003) (0.002) (0.011) (0.002) (0.044) (0.005) (0.007)
Kor. Small -0.029 -0.040 -0.024 -0.025 -2.718 -0.036 -0.003 -0.003 -0.004 0.000 -1.405 0.000
(S.E.) (0.008) (0.011) (0.006) (0.009) (0.111) (0.012) (0.003) (0.007) (0.005) (0.009) (0.018) (0.006)
Kor. Cram 0.003 0.017 0.002 0.015 -0.018 -0.355 0.003 0.050 0.002 0.052 0.000 -0.245
(S.E.) (0.003) (0.003) (0.002) (0.002) (0.023) (0.027) (0.001) (0.007) (0.002) (0.008) (0.003) (0.034)
Symmetry-constrained estimates Symmetry-constrained estimates
Eng. Small -0.845 0.004 0.005 0.003 -0.025 0.005 -0.807 0.006 0.005 0.006 -0.006 0.003
(S.E.) (0.025) (0.593) (0.012) (0.208) (21.119) (0.326) (0.007) (0.002) (0.001) (0.002) (0.706) (0.040)
Eng. Cram 0.001 -0.165 0.001 0.011 -0.012 0.011 0.006 -0.060 0.003 0.069 0.023 0.042
(S.E.) (0.614) (15.262) (0.313) (5.320) (591.047) (8.400) (0.004) (0.036) (0.005) (0.011) (4.681) (0.227)
Math Small 0.004 0.002 -0.800 0.002 -0.016 0.003 0.005 0.002 -0.771 0.003 -0.007 0.002
(S.E.) (0.015) (0.379) (0.010) (0.134) (13.488) (0.208) (0.001) (0.002) (0.008) (0.002) (0.947) (0.059)
Math Cram 0.001 0.010 0.001 -0.169 -0.007 0.008 0.005 0.079 0.003 -0.025 0.000 0.049
(S.E.) (0.267) (6.482) (0.141) (3.330) (254.021) (3.633) (0.004) (0.010) (0.004) (0.040) (4.631) (0.202)
Kor. Small -0.029 -0.040 -0.024 -0.025 -2.711 -0.036 -0.030 0.318 -0.017 -0.101 -1.396 1.206
(S.E.) (0.739) (18.401) (0.377) (6.528) (638.929) (10.096) (0.093) (0.006) (0.095) (0.008) (61.405) (5.914)
Kor. Cram 0.003 0.016 0.002 0.013 -0.017 -0.321 0.007 0.048 0.007 0.049 0.140 -0.233
(S.E.) (0.145) (3.596) (0.074) (1.274) (136.477) (1.983) (0.027) (0.006) (0.028) (0.007) (53.431) (1.683)
Wald test 4.87 (0.999) 6.75 (0.999)
Note: Bootstrapped standard errors are shown in brackets. Estimates in the boldface are twice greater than its standard error. For the Wald test statis-
tics in the last row, the number in parenthesis is p-value at the 95% critical value of
2
(25).
68
Table 2.14: Own- and cross-price elasticities: Baseline test score
Bottom 50% of baseline test score Top 50% of baseline test score
English Math Korean English Math Korean
Small Cram Small Cram Small Cram Small Cram Small Cram Small Cram
Unconstrained estimates Unconstrained estimates
Eng. Small -0.990 0.000 0.000 0.000 -0.001 0.000 -0.656 0.005 0.028 -0.003 -0.060 0.002
(S.E.) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.016) (0.003) (0.005) (0.004) (0.019) (0.003)
Eng. Cram 0.000 -0.271 0.000 0.044 0.002 0.037 0.003 0.056 0.002 0.045 -0.024 0.032
(S.E.) (0.000) (0.033) (0.000) (0.009) (0.003) (0.008) (0.005) (0.047) (0.004) (0.009) (0.021) (0.005)
Math Small 0.000 0.000 -0.977 0.000 -0.002 0.000 0.026 0.003 -0.594 0.005 -0.050 0.002
(S.E.) (0.000) (0.000) (0.001) (0.000) (0.000) (0.000) (0.006) (0.002) (0.019) (0.003) (0.017) (0.003)
Math Cram 0.000 0.049 0.001 -0.241 0.010 0.039 -0.002 0.049 0.003 0.130 -0.006 0.035
(S.E.) (0.000) (0.008) (0.000) (0.038) (0.003) (0.008) (0.007) (0.008) (0.005) (0.055) (0.021) (0.006)
Kor. Small 0.000 0.001 0.000 0.002 -1.256 0.002 -0.051 -0.040 -0.045 -0.009 -1.688 -0.022
(S.E.) (0.000) (0.012) (0.000) (0.013) (0.013) (0.015) (0.022) (0.013) (0.018) (0.014) (0.067) (0.012)
Kor. Cram 0.000 0.043 0.000 0.041 0.010 -0.357 0.001 0.034 0.001 0.035 -0.014 -0.033
(S.E.) (0.000) (0.007) (0.000) (0.008) (0.003) (0.031) (0.005) (0.005) (0.005) (0.006) (0.019) (0.045)
Symmetry-constrained estimates Symmetry-constrained estimates
Eng. Small -0.939 0.000 0.000 0.000 0.000 0.000 -0.657 0.005 0.028 -0.003 -0.061 0.002
(S.E.) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.018) (0.002) (0.008) (0.003) (0.133) (0.003)
Eng. Cram 0.000 -0.231 0.000 0.037 0.002 0.032 0.009 0.046 0.006 0.037 0.018 0.026
(S.E.) (0.000) (0.029) (0.000) (0.008) (0.398) (0.007) (0.008) (0.039) (0.006) (0.007) (0.027) (0.005)
Math Small 0.000 0.000 -0.955 0.000 -0.002 0.000 0.026 0.003 -0.592 0.005 -0.050 0.002
(S.E.) (0.000) (0.000) (0.001) (0.000) (0.001) (0.000) (0.007) (0.002) (0.020) (0.002) (0.204) (0.003)
Math Cram 0.000 0.042 0.000 -0.206 0.009 0.034 0.001 0.039 0.004 0.105 0.008 0.028
(S.E.) (0.000) (0.007) (0.000) (0.034) (0.352) (0.008) (0.008) (0.007) (0.006) (0.045) (0.024) (0.005)
Kor. Small 0.000 0.001 0.000 0.002 -1.254 0.002 -0.064 0.466 -0.055 0.139 -1.724 0.139
(S.E.) (0.000) (0.030) (0.000) (0.027) (0.025) (0.025) (0.018) (0.010) (0.024) (0.011) (0.070) (0.011)
Kor. Cram 0.000 0.041 0.000 0.039 0.009 -0.343 0.006 0.033 0.005 0.033 0.017 -0.031
(S.E.) (0.000) (0.006) (0.000) (0.007) (0.190) (0.030) (0.007) (0.004) (0.006) (0.005) (0.020) (0.042)
Wald test 4.94 (0.999) 21.58 (0.660)
Note: Bootstrapped standard errors are shown in brackets. Estimates in the boldface are twice greater than its standard error. For the Wald test
statistics in the last row, the number in parenthesis is p-value at the 95% critical value of
2
(25).
69
Chapter 3
Eect of Private Tutoring on
Academic Achievement: Evidence
from South Korea
3.1 Introduction
A new term, \edu-poor," has appeared in South Korea, which indicates households that
undergo overspending on private tutoring even though they are saddled with debt and
decits. While the average educational expenditure of households is 18% of total house-
hold expenditure, the edu-poor group spends 28.5% of their total expenditure on edu-
cation. From the Household Consumption Survey conducted by the Korean National
70
Statistics Oce (KNSO), 13% of households that spend on children's education are es-
timated as the edu-poor group in 2011. The existence of high-stakes exams and similar
curriculum for both public and private schools due to Korean educational policy has made
tutoring a very attractive way to enhance students' performance without considering the
actual eectiveness.
Education production function has been used when analyzing the eect of tutoring
on educational outcomes. Although most past works ignore the possibility that inputs
applied in the past in
uence the current educational outcomes, this paper reduces assump-
tion on the depreciation of the past input. Based upon education production function,
the main estimation strategy is the xed eect and dynamic panel method to eliminate
the components of unobserved school district or individual in
uences. Dynamic panel
approach allows the eect estimated from students who make a change in educational
input while other strategies, such as the xed eects, provide only the average impact.
Despite school district characteristics in Korea, many researches have not accounted
for them due to the lack of data. While students are randomly assigned to junior high
schools within their school district due to Korea's secondary school equalization policy,
which keeps students from selecting schools within a school district, they can still exercise
selectivity on school districts. Since parents are still able to choose which district to
reside in, the link of household decisions to private tutors and school districts and its
impact on academic achievement would be necessary to address. Choi (2012) also points
out the endogenous choice of school districts in the eect of school inputs on educational
71
outcomes in Korea. Even though the Korean Educational Longitudinal Survey (KELS)
that this study uses is a very rich educational data, it does not provide accurate location
information of each sample. In order to control school district eects, the KELS is linked
to the school lists from each city and the provincial Oce of Education in Korea using
summary statistics of schools, and identies school district information. This paper uses
this matched dataset to enable the detection of school district samples.
This paper nds a positive relation between private tutoring and academic achieve-
ment of English and mathematics. Results using dynamic panel data methods not only
present the average eect of private tutoring but also identify the impact estimated from
the dierences in test scores for students who have changed investment in private coach-
ing. The result is, however, that only English tutoring can exert a signicant eect on
English academic scores. Tutoring eects may vary, depending on students' initial char-
acteristics, for example, their ability. In order to examine this possibility, an estimation
is performed on two dierent groups, based on the students' baseline test scores as an
indication of their academic ability. When this is done, English and mathematics tutoring
turns out to play an enrichment role for more able students.
The next section presents a review of the previous literature for evaluating the impact
of private tutoring. This paper is organized as follows. Section 3 introduces the analytical
approach used to evaluate the eectiveness of tutoring services. The data and sampling are
described in Section 4. Section 5 is devoted to the estimation results of private tutoring for
English and mathematics. Finally, Section 6 concludes and discusses for future research.
72
3.2 Previous Literature
Generally cautious perspectives on private tutoring across the world are provided by
Bray's research (Bray, 1999; Bray and Kwok, 2003; Bray, 2011). Dang and Rogers (2008)
also reviews the literature on the eect of private tutoring and shows that results about
the eectiveness are not consistent. Buchmann et al. (2010) indicates that private coach-
ing can increase the probability of students going on to the next grade in Kenya. Liu
(2012) also nds a positive eect on students' performance in Taiwan. Thapa (2011)
nds insignicant impacts in Nepal.
Quantitative analyses of the impacts on academic achievements are performed in order
to address endogeneity issues. Dang (2007), Suryadarma et al. (2006), and Zhang (2013)
use an instrumental variable (IV) approach. Dang (2007) investigates the impact of pri-
vate tutoring on primary and lower-secondary education in Vietnam. Ocial tuition fees,
which are determined based on village wealth level by the authorities, are regarded as
an IV for private tutoring, which is demonstrated to have a positive impact. Households
living in wealthier districts of cities need to pay more for private tutoring per hour due to
the higher ocial tuition, and simultaneously they might be willing to spend more money
to provide a better quality of education. The ocial fee, the IV, may not be exogenous.
Suryadarma et al. (2006) nds an insignicant eect of extracurricular classes on mathe-
matics and dictation scores for primary school students in Indonesia. The proportion of
classmates taking an extracurricular class is considered an IV of extracurricular courses.
73
However, extracurricular class is not well dened in the paper, so it cannot be considered
a type of private tutoring as we dene it.
Zhang (2013) shows the inconsistent impact of private tutoring on National College
Entrance Exam (NCEE) scores in China. Two kinds of IVs are chosen in the paper: the
number of friends who participated in private tutoring among the nearest ve students
for peer eect and the distance between tutoring service agency and home. As IVs are
individual-level variables, the endogenous issues associated with neighborhood or school
district level cannot be addressed. On the other hand, Briggs (2001) examines the eect
on private coaching in SAT (Scholastic Assessment Test) and ACT (American College
Testing) in the U.S. A linear regression model, which is what he uses, cannot correct
self-selection bias of participation in tutoring compared to instrumental variable (IV)
estimation, and also, he does not control school level input, which can aect students'
academic performance.
Even though Korea secures relatively abundant data sets on private tutoring, their
results vary with the type of educational outcomes, the measure of tutoring, or the sam-
ples. Byun (2014) and Ryu and Kang (2013) consider propensity score matching method
to estimate the eect of private tutoring, and both nd a positive but modest eect of
tutoring expenditure on test scores. Except participation in cram school, other forms
of tutoring activities such as one-to-one private tutor, subscription to correspondence
courses, or Internet courses do not aect academic achievement. Choi (2012) uses the
Seoul Education Longitudinal Study and nds the positive eect on English and math
74
test scores. Lee et al. (2004) nds the insignicant eect of pre-class tutoring on the aca-
demic achievement by using a simple mean dierence of rankings of test scores between
two periods. However, the data covers only Seoul, and it is a one-time survey for 2000 and
2001, depending on the retrospective questions of pre-class tutoring for the last ve years.
Moreover, this does not carry on an additional method to control for the endogeneity or
selection problem.
Few quantitative studies on the causal eect of private tutoring and academic achieve-
ment address the endogeneity issue due to data constraint. In particular, selection issue
on school-district has been ignored. Using a unique matched data, this paper can deal
with the unobserved school-district eect. While most studies in Korea only focus on
the eect of private tutoring in academic achievement \level", this paper analyzes the
tutoring eect on \gain" of test scores. Previous researches of the eect of educational
input on academic performance level assume that there is no in
uence of the past inputs
in education production function, which implies that time-discount rate on the previous
input is zero. This paper investigates the private tutoring eect on test scores and with
no assumption on the depreciation rate of the past inputs. In addition, this paper pro-
vides a rst glimpse at whether private tutoring is an eective way to promote student's
performance in comparison with self-study hours. Also, I will show how the eect of
tutoring activities would be expected to vary among students with dierent individual or
household features. The gap between more- and less-able students, based on their ability,
75
can help us understand the two primary roles of private tutoring: a remedy for the less
able students and enrichment for the more able students
3.3 Specication and Estimation
In this section, the model specication and its estimation strategy are introduced. The
main interest of this paper is to examine whether time or money investment in private
tutoring can raise students' academic achievement. The key issue to estimate it arises
from the possibility that students who choose private tutoring are self-selected and tend
to be dierent in unobserved abilities from students who do not choose private tutoring.
In order to address the challenge in the absence of random assignment, this paper follows
the strategy of Hanushek et al. (2007).
As educational production function, equation (3.1), describes academic achievement
variable A for student i in time t and in school district s:
A
its
=PT
it
+X
it
+S
its

+
t1
X
=1

t
PT
i
+
t1
X
=1

t
X
i
+
t1
X
=1

t
S
is
(3.1)
+ (
i
+
t1
X
=1

t

i
) + (
s
+
t1
X
=1

t

s
) +e
its
;
where the dependent variable A
its
corresponds to academic achievement variable of the
individual i of year t in school-district s. PT indicates a measure of private tutoring
76
and X is a vector for individual and family background variables including gender, self-
study hours for each academic subject, parental education background, household income
level, and so on. Time-varying and time-invariant school inputs, such as the number
of students per class or operation of an extracurricular class or level-dierentiated class
relevant to academic subjects are a subset of X. S includes the measure of school-district
characteristics, such as the proportion for tuition exemption of each district. Each
i
and
s
represents unobserved individual and school-district eects. e
its
is stochastic error
term. And, past explanatory variables are assumed that their eects declines at a constant
rate (1), where 0 1.
Based on the education production framework, equation (3.2) and (3.3) present the
case of = 0. It implies no eect of prior inputs for current test score.
A
its
=PT
it
+X
it
+S
its
+
i
+e
its
(3.2)
A
its
=PT
it
+X
it
+S
its
+
s
+e
its
(3.3)
The rst estimation strategy this paper considers is two dierent xed eects estima-
tion for consistent estimator because of reducing two types of unobserved characteristics

i
and
s
: one is individual xed eect (FE) estimation regarding (3.2) and the other
is school-district FE regarding (3.3). As the dependent variable of (3.2) and (3.3) is the
level of test scores, it is denoted as `level' model in this paper. In addition to unob-
served variables, endogeneity of the right hand side regressors causes inconsistency and
needs instrument variable (IV) approach such as two-stage least squares (2SLS) to attain
77
consistent estimates (Baltagi, 2008). One possible endogeneity problem is students with
higher achievement level resulting in more investment in private tutoring or more hours
spent on study as educational inputs in order to enrich their learning. Hence, the xed
eect with 2SLS approach is also employed and the measure of private tutoring and self-
study hours are instrumented with four variables: Average test scores within a school
district except for the current observation, average teacher's education and experience
years within a school district excluding the observation, and average unit values within a
school district excluding the observation.
On the other hand, the condition of = 1 means no depreciation of the eect of past
variables. Equation (3.1) can be expressed as
A
its
=PT
it
+X
it
+S
its

+
t1
X
=1
PT
i
+
t1
X
=1
X
i
+
t1
X
=1
S
is
+ (
i
+
t1
X
=1

i
) + (
s
+
t1
X
=1

s
) +e
its
(3.4)
=PT
it
+X
it
+S
its
+A
i(t1)s
+
i
+
s
+e
its
Appropriate estimation strategy for the equation (3.4) is xed eect for consistent
parameter estimates. For student xed eect,
A
its
=A
its
A
i(t1)s
=PT
it
+X
it
+S
its
+
i
+e
its
:
(3.5)
78
Similarly, for school district eect,
A
its
=A
its
A
i(t1)s
=PT
it
+X
it
+S
its
+
s
+e
its
:
(3.6)
In this case, (3.5) and (3.6) are called `gain' model as the dependent variable A
its
is
the change in test score over year. The FE with IV approach is also adopted to deal with
the endogeneity.
However, if 0 < < 1, the equation (3.1) can be presented as education production
function with lagged dependent variable on the right-hand side:
A
its
=PT
it
+X
it
+S
its
+A
i(t1)s
+
i
+"
its
"
its
=e
its
e
i(t1)s
:
(3.7)
The equation (3.7), called a lagged achievement model in this paper, can be the most
preferred model because of no necessity to assume the time-discount rate, . While the
lagged achievement model (3.7) takes an advantage from allowing previous test score to
control the cumulative eects of the observed past inputs, as Hsiao (2003) points out, the
inclusion of lagged dependent variable in FE model requires the use of an instrument for
lagged dependent variable to achieve the consistency. Specically, one lagged academic
achievement variable is likely to be related with time-invariant unobserved variable, such
as
i
. For example, a person of low intelligence is more likely to have low academic
achievement level, which can lead to less interests to study and attend tutoring, which
79
may subsequently cause lower test scores. Hence, this endogenous issue may lead upward
bias in the estimation. First-dierencing can remove the unobserved variable and may
gain the consistent estimate to solve the endogeneity.
1
However, the endogeneity issue
still remains after rst-dierencing. The rst lagged achievement is instrumented with
twice lagged achievement as a valid instrument as Arellano and Bond (1991) suggested.
Given only three waves panel data of this paper, this reduces initial period observation
for each student and it restricts the use of the baseline characteristics to investigate
heterogeneous eect of private tutoring. In addition, since tutoring measures and self-
study hours variables may be endogenous, four external instruments are used: Test scores
mean within a school district excluding the current observation, teacher's education and
experience years means within a school district excluding the current observation, and
average unit values within a school district except for the observation.
3.4 Data
The data used in this paper is the rst three waves of the Korean Educational Longitudinal
Study (KELS) launched by the Korean Educational Development Institute (KEDI) in
2005. Questionnaires about family backgrounds, teacher and school characteristics, and
students' academic performance, were answered by students, parents, teachers, school
1
First-dierencing transforms the equation (3.7) into Aits =PTit +Xit+Sits+A
i(t1)s
+
"its
80
principals and school administrators. The sample in the KELS was drawn following two-
stage stratied cluster sampling. In the rst stage, all Korean middle schools into four
types of regions by city size were stratied and the KEDI randomly chose 150 sampling
schools out of of 2,929 middle schools in Korea. Fifty students in the 7th grade of each
middle school were randomly selected in the second stage. The survey started with 6,908,
and 6,502 (94%) answered all the questionnaires. Attrition occurred due to emigration or
from dropping out of the survey even though this survey tracked transferred students.
2
5,943 respondents in the 2nd wave of 2006 and 5,594 respondents in the 3rd wave of 2007
answered all questionnaires.
This paper limits the sample to students who consistently responded to the survey for
three years (2005 to 2007), and it covers the rst, second, and third grades of junior high
school. Since variables in academic achievement and private tutoring play important roles
in the analysis, observations with missing values for those variables are eliminated. The
number of the nal sample used in this study is 2,330 students who stayed in the survey
until 2007, and the balanced panel data have a total of 6,990 person-year observations on
these students.
2
In 2006, 204 students transferred to the school from the sample schools and eight students transferred
to the sample schools. Both are tracked, even those out of the sample schools. In 2007, 175 students
transferred to the school from the sample schools and two students transferred to the sample schools.
81
Academic achievement The KEDI conducts achievement tests in Korean, English, and
mathematics at the end of each year (December) for junior high school students. The raw
test scores are scaled from 0 to 100. Instead of the raw test score, the standardized test
score considers academic achievement variables, which become the primary dependent
variable in the estimation. The standardized test scores, called Z-score, are calculated
from the normalization of the raw test scores for the subject to have a mean of 0 and
standard error of 1.
Private tutoring The KELS provides very detailed information on private tutoring.
Students' parents answered the relevant questions for each private tutoring separately
for English and mathematics: \Did your child participate in private tutoring during this
year?"; \How many hours did the student spend on private tutoring per week during this
year on average?"; and \During this year, how much money did you spend on private
tutoring for the student per month during on average?" The calculated unit value of
private tutoring is also considered as a measure of private tutoring. The unit value is
constructed from the expenditure of private tutoring per hour from the monthly spending
on private tutoring divided by the hours of private tutoring.
Households and school characteristics The survey delivers individual and household
characteristics of observations, and summary statistics of schools that students attend.
Parents reported their ages, educational backgrounds, and the number of siblings at the
baseline survey. Originally, the KELS grouped parents' education level into seven types:
primary (6-year), middle school (3-year), high school (3-year), 2-year junior college, 4-year
82
university, master degree, and doctorate. This paper reorganized it into four categories:
junior high school graduates or less, high school graduates, 2-year junior college graduates,
and 4-year university graduates or more.
In the original data, the information about a student's location is provided based on
the region sizes: Seoul (the capital city in Korea), six other large cities (the metropolitan
areas), medium-sized cities, and small towns or rural areas (called up or myeon). In order
to identify more specic information, a unique data was used in the rst essay. This data
was created by matching the KELS with school lists because both data sets have the same
summary statistics of schools. The nal sample indicates 131 school districts.
School inputs include the average years of education and career of teachers within
a school, the average number of hours per week that teachers work, the average num-
ber of students per class and per teacher, the school's use of English or mathematics
dierentiated-classes by achievement level, the school's advanced class for each subject,
its co-educational school, whether private or public, and its observance of the equalization
policy.
Summary Statistics The summary statistics of the socio-economic variables and school
characteristics is presented in Table 3.1. These variables are categorized into two parts:
time-invariant and time-variant variables. The father and mother's education variables
are categorized into four groups: junior high school or lower, high school, two-year college,
and four-year university graduate or more. While only 20% of students have a mother
83
with a four-year university degree, more than one-third of the total sample have fathers
who graduated with a four-year college degree or higher.
49% of the sample are male students and 46% live in urban areas. 63% of all students
attend schools in
uenced by the equalization policy, and 66% are in the co-education
school. About 80% of the sample are in public school. Since the monthly household
income is 10,000 Korean Won (KRW) and the currency rate is almost 1,100 KRW/$, the
average monthly household income in 2005 (3,726,700 KRW) is equivalent to about 3,381
US dollars. Each observed family unit averagely involves about 1.27 siblings.
The summary statistics of the school characteristics are reported. The average number
of years of education of teachers from entire sample schools are almost 16.7 years, and
the average teaching experience of teachers is about 15.5 years. The percentage of schools
oering a level-dierentiated class or advanced class for students becomes less as they
move to a higher grade. On average, 8% of the total number of students in each school
are exempted from paying tuition.
Table 3.2 describes the summary statistics of four dierent private tutoring variables
measured by participation rate, expenditure, hours spent, and unit values. Unit values
of tutoring activities are calculated by the hours spent for private tutoring. All tutoring
activity variables for English and mathematics slightly increased with higher grades. Both
subjects have similar sizes within and between variations. Furthermore, within variation
of private tutoring is slightly lesser than between variation of private tutoring. That is,
the cross-sectional variation is larger than the time-variation of private tutoring. But,
84
the standard deviation (SD) of test scores across time is much less than the SD across
observations. Hence, the households' decisions for private tutoring are varied over time,
as much as across observations, unlike the test scores.
3.5 Results
Estimation results depending on the assumption of are presented based on dierent
estimation strategies. Specication for (3.2) and (3.3) assuming = 0 results from the
pooled OLS, hereafter POLS, school-district and student FE, and FE with IV estimation.
Similarly, the gain models, (3.5) and (3.6), are estimated by the POLS, school-district FE,
student FE and FE with 2SLS. Without assumption on depreciation rate in education
production equation (3.7), the Arellano-Bond (AB) approach is used.
The dependent variable for academic achievement is the standardized test scores for
English and mathematics. There are four dierent measures of private tutoring: (a)
Whether the student participates in private tutoring for each academic subject (b) how
much the household spends monthly on the student's private tutoring (c) how many hours
the student spends on private tutoring per week, and (d) how much money the household
spends on the student's tutoring per hour (unit values). Hence, specications for the
eect of private tutoring include eight cases (two academic subjects eight measures of
private tutoring).
The top panel (a) in each table describes the eects of participation in tutoring ser-
vices. The two middle panels, (b) and (c), present the eects of the logarithm of monthly
85
expenditure and weekly hours in private tutoring respectively, and the bottom panel
(d) shows the impact of the unit value. All specications include a full set of school
characteristic variables and school district covariates, such as the proportion of tuition
exemption; household characteristics, such as income and the number of siblings; and in-
dividual characteristics, such as self-study hours for each subject. The POLS estimation
includes time-invariant variables for the gender of students, parents' education (middle
school graduate or less, high-school graduate, 2-year college, and 4-year university or
more), location (capital city, six metropolitan cities, medium and small sized cities, and
rural area), equalization policy, public school, and co-education school.
3.5.1 Eect of private tutoring on test score
Table 3.3 and 3.4 report the results of a series of specications based upon the assumption
of time-discount factor,. The rst three columns present parallel estimates from POLS,
school-district FE, students FE, and FE with IVs when assuming = 0. Estimation re-
sults of gain model are reported in the next three columns, and the last column represents
the AB estimates.
In the rst three columns of the table, most private tutoring measures have signicant
positive association with students' academic achievement of English and math respec-
tively, except the unit values of English tutoring in the panel (d). Further, the dierence
in students' unobserved ability is likely to bias the relation between the private tutoring
and Z-score. The magnitude of student FE estimates is less than that of the POLS esti-
mates, suggesting positive selection bias into private tutoring. The gap between the POLS
86
and district FE estimates is smaller than the dierence between the POLS and individual
FE estimates. Hence, the unobserved district characteristics relatively less aect the use
of private tutoring and Z-scores than the unobserved student characteristics. In order to
deal with the additional endogenous problem from the regressors even after eliminating
an unobserved individual characteristics, the FE with 2SLS approach is employed and
the measure of private tutoring and self-study hours are instrumented. Students who
participated in English tutoring tend to have lower test scores, but it is insignicant from
the estimates of panel (a).
Similarly, panel (b) and (c) of Table 3.3 and 3.4 represent the time and money in-
vestment in private tutoring and can be positively associated with the standardized score
of English and mathematics even after the district and individual xed eect and have
smaller FE estimates than POLS result. Specically, the FE with IV approach for hours
for English tutoring results in larger magnitude of estimate than that of the FE. It im-
plies the endogeneity issue related to the private tutoring and self-study hour variables.
Since four tutoring measures have dierent scales, standardized coecients help to com-
pare them. In the specications of = 0, the logarithm of tutoring expenditure shows a
higher standardized coecient than the tutoring hour. If tutoring expenditure increases
by a standard deviation, the test score changes by 0.028 in English.
In the panel (d), the coecient on expenditure per hour for English is insignicant
at 5% signicance level in student FE and signicant at 10% level in FE with IVs that
the signicance is weak. Students whose parents are more concerned with their children's
87
education have better help, which results in better test scores. At the same time, this
type of parent is willing to choose tutoring services at a higher price (unit value) and
is able to oer nancial support under the belief that tutoring at a higher price helps
students' learning. Controlling student FE removes the confounding eect that might
induce insignicant estimates. Interestingly, math Z-scores are signicantly positively
correlated with unit values of tutoring, even after FE with 2SLS. This implies that learning
English may depend on students' unobserved abilities and the use of tutoring services has
less impact on test scores than it does in mathematics.
The estimation result of the gain model (3.4) is in the middle columns of Table 3.3
and 3.4. Estimates for all tutoring measures appear insignicant when English test score
gain is the dependent variable. It shows that the students, who spend the same amount
of money or hours on English tutoring, hardly raise the test scores. On the other hand,
mathematics tutoring is signicantly related to the change in test score and has a larger
estimate of individual FE than the POLS estimate. Unlike in English, time and money
investments in mathematics tutoring produces an increase in scores, even for students,
who stay on the same status in private tutoring. The impact of tutoring services on
mathematics test scores may be less likely to be persistent and can be an instant treatment
to implement academic achievement. The result from FE with IVs becomes signicant
for mathematics tutoring that implies the endogeneity of math tutoring hours and self-
study hours variables. Both log expenditure and tutoring hours in mathematics tutoring
88
have similar standardized coecients. When time or money investment is higher by one
standard deviation, students tend to have a larger test score gain by 0.095-0.096.
All previous specication results are based on the assumption of in education pro-
duction function. Assuming = 0 for the achievement level model and = 1 for the
achievement gain model, the POLS, the FE, and the FE with 2SLS are considered an
estimation strategy to obtain consistent estimates. However, without restriction on ,
the Arellano-Bond (AB) approach are used to estimate the equation (3.7), which shows
the dynamics of the dependent variable. Since the AB method employs instrument vari-
able approach, the bottom row of each panel shows the test results for the validity of
instruments. The Sargan-Hansen (SH) statistics, which tests for over-identifying restric-
tions, should be used because the AB estimation is one kind of over-identied model that
adopts an IV strategy. So, if the SH statistics rejects the null hypothesis, the validity of
instruments cannot be guaranteed.
The implication of private tutoring eect relies upon dierent identication approaches.
In comparison with the POLS, the FE estimate identies the private tutoring eect ad-
justed for unobserved individual characteristics and the estimates of FE with 2SLS implies
the average tutoring eect accounting for the endogeneity issue of explanatory variables.
However, the AB estimator identies the eect of private tutoring from changes in test
scores for students who made a dierent decision about private tutoring from the one in
the prior period. In other words, while the AB estimator detects the impact of tutoring
from the dierence in the Z-scores of students who change to private tutoring, accounting
89
for student unobserved features, the estimates of FE and FE with IV describe the average
private tutoring relation, controlling for student xed eects.
As a result of AB method, an increase in English private tutoring can raise a student's
English test score, but math private tutoring cannot. A 10% increase in expenditure
on English tutoring can change English test scores by 0.012%. One more hour spent
on English tutoring service can raise standardized-test scores by 0.031. The result of
spending on English tutoring is similar to the nding of Ryu and Kang (2013) that a 10%
increase in spending can change the English test scores 0.008%-1.26% (varying over the
estimation strategy). Despite the positive association between math private tutoring and
math test scores, the actual eect of private tutoring appears only in English. Assuming
that language ability is important to learn English, English tutoring can supplement with
the lack of ability and then make an actual change in test scores. Byun (2014) detects the
positive eect of expenditure on math cram schools, which is a certain type of tutoring, on
mathematics test scores for junior high school. Although hiring an individual tutor oers
more personalized learning to the student, tutors in the cram-schools are easy to analyze
the pattern of the school's exam and develop their own teaching resources. Hence, while
Byun (2014) nds the positive eect of attending in cram schools but insignicant eect
of hiring an individual tutor, the impact of tutoring including both types of tutoring turns
out to be insignicant in this paper.
90
3.5.2 Heterogeneous eect of private tutoring
Table 3.5-3.24 presents the heterogeneous impact of private tutoring on academic achieve-
ment from samples divided by baseline characteristics: student's gender, location, parental
education, and baseline test score. Socio-economic status is identied as an important
determinant of tutoring activities, as previous literature conrms (Dang, 2007; Ryu and
Kang, 2013; Byun, 2014). Their ndings may imply that tutoring can prevent intergener-
ational mobility, since family background can be strongly related to the choice of tutoring
services being able to improve children's educational outcomes. This subsection investi-
gates whether individual or household characteristics make a dierence in the eectiveness
of tutoring activities. Since it makes it possible to detect the association between family
background and the eect of tutoring on academic achievement, the result may provoke
the possibility of educational inequality caused by socio-economic characteristics through
supplementary education.
Baseline test score The purpose of tutoring can vary with students' academic ability.
The high-performing students rely on private tutoring to complement the knowledge pro-
vided by the school and enable them to reach the advanced course. However, to students
who do not perform well in test scores, supplementary tutoring is a remedial treatment
to catch up with the school curriculum. This part investigates whether tutoring can ac-
complish these two purposes, by dividing samples into two groups based on the baseline
test score. Since the KELS data does not provide the direct measurement of student
academic ability, baseline test scores in the survey (2005) become the standard to divide
91
students into two groups. Students are assigned to two dierent groups based on their
percentile scores. A students is included in the top group if his or her percentile is more
than fty; the students in the other group, the bottom group, has percentiles that are less
than fty. Overall, both groups have positive associations with academic achievement and
tutoring activities (Table 3.5 to 3.8). Although both groups have signicant AB estimates
on English tutoring, the top 50% group has more signicance on a dummy for attending
tutoring and expenditure in tutoring. In mathematics tutoring, more hours spent can
increase test scores for able students. Hence, private tutoring may play an enriching role
in academic performance for more able students.
Gender Let us suppose that tutoring plays an enrichment role for the more able students.
Since girls in middle school tend to perform better than boys in Korea, I speculate that
the girls will be more in
uenced by tutoring activities. For both groups, English private
tutoring is positively related to test scores, and the changes in private tutoring increase
English achievement from the AB estimates in Table 3.9 and 3.11. However, Table 3.12
shows that an increase in hours spent on mathematics tutoring might improve academic
achievement only for girls. Both groups can take advantage of English tutoring, but
only girls are aected by math tutoring. It implies that the supposition I made can be
satised, and math tutoring can more heavily perform an enrichment role for the more
able students group, which is the girls group.
Location Educational investment of students who live in the urban sector may depend
to a great extent on relatively easier access to tutoring activities and higher peer pressure
92
without considering the actual eectiveness. Under this belief, students in the urban
area may have a smaller impact of private tutoring on test scores. In this paper, the
urban sector includes the capital city and six metropolitan cities, and the rural area has
small- or medium-sized cities and rural area categorized by the KELS. In Table 3.13-3.16,
only more hours spent on English tutoring can raise test scores in the urban sector while
just attending English tutoring makes a signicant change in test scores in the rural area.
However, only students in urban areas can be in
uenced by mathematics tutoring. Hence,
students in the urban sector show similar patterns to the more able students group. As
the test for the hypothesis that the Z-scores for urban and rural sectors are the same
is rejected (t-statistics is 14.25 and p-value is 0.00), the result of the urban group can
resemble the result of the more able group. Another possible explanation is that since
more qualied tutors tend to be located in urban areas, students in rural areas might be
comparatively disadvantaged.
Parents' education Since the higher SES family has the means to choose a better qual-
ity of education for their children, students whose parents are highly educated may excel
in educational outcomes. Hence, students with highly educated parents may have similar
characteristics to more able students. In order to detect the heterogeneous eect de-
pending on parental education, samples are divided into two groups. The lower-educated
father/mother group indicates students whose father's/mother's educational attainment
is high school or less. Higher-educated parents have a 2- or 4-year college degree or more.
The dierence in tutoring eects between the two parental groups varies according to
93
the academic subject. While the results in mathematics of the higher-educated father
group coincide with those of the more able group, the eect of English tutoring on the
higher-educated parent group shows dierent patterns for students who perform better
in test scores. In Table 3.17-3.20, although the AB estimates on English tutoring are
signicant, the less-educated father group has a stronger signicance on English tutoring.
On the other hand, only the higher-educated father group shows a signicant impact of
math tutoring hours on test scores. Groups divided by maternal education are similar to
the groups based on the father's education (Table 3.21 to 3.24). If parental education in-
dicates a household's wealth, children with higher-educated parents are more likely to be
exposed to early childhood English education. Since these children have already reached
a higher level of English ability, tutoring activities during junior high school may not im-
prove their test scores. Hence, the higher-educated-parents group may have (relatively)
insignicant estimates of English tutoring.
3.5.3 Comparison with self-study hours
In addition to the eect on private tutoring, Table 3.26 represents the test results showing
whether hours of private tutoring of English and mathematics have the same impact on
achievement as hours of self-study, respectively. Since the impact of private tutoring turns
out to be modest or insignicant, the comparison between tutoring and self-study hours
may assist in understanding the eectiveness of tutoring on test scores. Self-study hours
are surveyed and determined according to how many hours students study, excluding
hours spent on preparation or assignments from tutors. The middle row of each table
94
displays the p-value of the joint signicant test for the hypothesis that coecients on
tutoring hours and self-study hours are the same. The results of total tutoring hours and
total self-study hours on the average test score over academic subjects are shown in Table
3.25. As a result, both tutoring and self-study hours have signicant positive eects and
are no signicant dierence between them at 5% level.
Regarding English tutoring, the eect of self-study hours on private tutoring has no
signicant dierence with that of tutoring hours. Despite the signicant eect of English
tutoring on test scores, the eectiveness of hours spent on English tutoring activities is
similar to the impact of self-study hours. Furthermore, while hours spent on math tutoring
has an insignicant impact, the eect of self-study hours for mathematics is positive and
signicantly dierent from the eect of math tutoring. Hence, private tutoring may not
be the best, most eective way to increase a student's achievement, and just raising hours
spent on study can be enough to better academic performance. This result can also be
interpreted as the importance of supervision to make students concentrate on their study.
Although parents may expect to enhance children's academic ability when they hire tutors
or send their children to institutions, tutoring may only help children to force to study,
not increase their ability. Hence, we can expect that if parents or even grandparents, who
stay at home and keep children studying, can be supervisors and tutors. Of course, private
tutoring measures only how many hours students spend, with tutors, and it excludes the
number of hours that the students spend in preparing for the private tutoring. Then, the
estimates of private tutoring hours would be underestimated.
95
In order to detect the heterogeneity, the joint test also performs with dierent de-
mographic groups. Most groups show that there is no dierent impact between tutoring
and self-studying in English. Regarding the mathematics AB estimates, the rural area
and lower educated father groups are likely to have the dierence between tutoring hours
and self-studying hours. Self-studying hour estimates in those groups also have a larger
magnitude than tutoring service estimates. This may correspond to the result of the
heterogeneous eect that mathematics tutoring tends to be an enrichment for the more
able students. That is, mathematics tutoring may relatively aect able students more.
3.6 Conclusions and Discussion
This article analyzes the eect of private tutoring on junior high school students' aca-
demic achievement in English and mathematics. Empirical specications, containing the
dynamic panel model, are employed using the rst three waves of the KELS (2005-2007).
Private tutoring is signicantly positively associated with academic achievement. With
respect to the comparison of results from POLS, school-district FE, individual FE and FE
with 2SLS estimation, the unobserved individual characteristics cause relatively stronger
selectivity issues related to achievement and private tutoring than the unobserved school-
district variable. While these approaches measure the average eect, the AB procedure
estimates the change in the test scores of students who make a change in tutoring in-
vestment, controlling for the unobserved variables. POLS, FE, and FE with IV results
establish a positive relation between tutoring and test scores, and AB estimates show the
96
signicant eect of English tutoring. This paper examines the hypothesis that private
tutoring eects vary depending on individual or family characteristics. Students in top
50th percentile have a stronger eect of English and math tutoring on the test scores than
on students in the bottom 50th percentile. The two main purposes of tutoring can be
associated to interpret these results. Tutoring activities may be more likely to provide
enriching learning for able students.
While the results of this study show that private tutoring enhances student test scores,
it is not the best way to improve test scores, especially as compared to self-study hours.
There is no signicant dierence between the eect of tutoring and that of self-study
hours on English test scores. Even self-study hours for mathematics will provide better
educational outcomes than tutoring hours. Why does not private tutoring always improve
academic achievement? One explanation is that the quality of teaching is standardized
downward in Korean tutoring industry. Since the private tutoring market is growing
rapidly in Korea, less-qualied tutors may be
owing into the market and, therefore, the
teaching quality that is provided by the tutoring market may not be enough to improve
educational outcomes. On the other hand, the educational policy that will reduce the
reliance on private supplementary education may improve teaching quality in schools and
provide a subsidy for tutoring services in the public school system.
3
These eorts can
3
The number of students per teacher in middle schools rapidly decreased from 42.3 in 1970 to 19.1 in
2007. The school teacher career is very popular because of the work duties and job security; so, being a
teacher in Korea is highly competitive and teaching quality in schools has been improved. Only 15.3%
of those who graduated from four-year secondary teacher education institutes with teaching certicates
97
also relatively weaken the eect of tutoring. Furthermore, attending tutoring activities
in Korea may sometimes be attributed to social pressure by peer group. As admission to
prestigious universities has become dicult with high competition, parents and students
may not want to fall behind their peers. Hence, ignoring the purpose of private tutoring
as remedy and enrichment, students just stay in the private tutoring market and invest
more in tutoring services because of irrational concerns. In addition, even though parents
expect tutors to augment their children's learning, their actual function may be only that
of supervisor to keep the focus on study. Therefore, more time spent on private tutoring
cannot guarantee the enhancement of academic performance we expect.
While ndings from past literature are based on the assumption of time-discount
factors in the educational production function and indicate average treatment eects to
account for an individual unobserved variable, this paper does not assume time-discount
rates on the past inputs and then make it possible to identify the actual impact con-
sidering an individual unobserved characteristics. Moreover, even though school-district
heterogeneity is important to be considered in the education production function in Korea
because households choose the district (Choi et al., 2014), previous literature has ignored
this issue due to the lack of the data. Since the data that I built up is able to identify loca-
tion information, school-district characteristics are included in the estimation. Although
had passed the competitive employment exam, administered by each metropolitan or provincial oce of
education, and were hired by schools in 2009. Additionally, several projects of school have been provided
as a subsidy of tutoring, such as after-school programs or classrooms by achievement level (Kim et al.,
2010).
98
Byun (2014) adopts propensity score matching approach to address selectivity issues, a
limitation he suggests is the quality variation of tutoring. Since his private tutoring mea-
sure is only a dummy for attending tutoring, a unit value of tutoring that is considered
as a proxy for prices in this paper can assist the eect of quality variation of tutoring on
academic achievement.
Despite the control to district xed eects, there exists another limitation in terms
of peer eects in this paper. Without information at the classroom level or friends to
cause the choice of private tutoring and aect academic achievement, this study only
depends on the school district level variables. Furthermore, although I consider a unit
value as a proxy for quality of tutoring, a more accurate tool is required. If information
on the supply side of private tutoring can be collected simultaneously, we may consider a
better measure of tutoring quality such as the tutor's educational background, experience,
or institutions' facilities. Future research to deal with these limitations will be more
benecial to understand the impact of private tutoring on educational outcome.
99
Table 3.1: Summary statistics
Mean S.D Mean S.D
Between Within
Mean S.D
Between Within
S.D S.D S.D S.D
Gender 0.49 0.50 Household income Advanced Class
in 2005 372.67 217.58 of English in 2005 0.84 0.36
Father's Education in 2007 379.87 230.67 317.92 329.19 of English in 2006 0.57 0.50 0.28 0.42
Middle school grad or less 0.07 0.25 in 2008 472.94 722.21 of English in 2007 0.18 0.39
High school grad 0.45 0.50
2-year college 0.13 0.33 Teacher's Education of math in 2005 0.86 0.35
4-year university or more 0.36 0.48 in 2005 16.66 0.33 of math in 2006 0.21 0.41 0.26 0.42
Mother's Education in 2006 16.70 0.32 0.27 0.12 of math in 2007 0.70 0.46
Middle school grad or less 0.08 0.27 in 2007 16.63 0.23
High school grad 0.62 0.48 Teacher's Experience Proportion for tuition exemption
2 years university 0.10 0.30 in 2005 15.12 3.79 in 2005 0.06 0.10
4 years university or more 0.20 0.40 in 2006 15.47 3.91 3.82 0.83 in 2006 0.09 0.14 0.10 0.09
in 2007 15.74 3.98 in 2007 0.09 0.14
Number of siblings 1.27 0.58
Level dierentiated Class Average Z-score in a school district
Location of English in 2005 0.43 0.49 of English in 2005 0.14 0.44
Seoul 0.18 0.38 of English in 2006 0.55 0.50 0.30 0.31 of English in 2006 0.16 0.41 0.39 0.18
6 Metropolitan cities 0.28 0.45 of English in 2007 0.30 0.46 of English in 2007 0.25 0.42
Medium and small size cities 0.46 0.50
Rural area 0.08 0.27 of math in 2005 0.50 0.50 of math in 2005 0.13 0.37
of math in 2006 0.55 0.50 0.31 0.39 of math in 2006 0.14 0.41 0.37
Equalization policy 0.63 0.48 of math in 2007 0.34 0.47 of math in 2007 0.17 0.44 0.17
Co-education school 0.66 0.48
Public school 0.80 0.40
Observation 6990
100
Table 3.2: Summary statistics for private tutoring and academic achievement
Mean S.D
Between Within
Mean S.D
Between Within
S.D S.D S.D S.D
Participation on private tutoring Unit value of private tutoring (+)
of English in 2005 0.66 0.47 of English in 2005 1.80 0.75
of English in 2006 0.60 0.49 0.37 0.32 of English in 2006 1.88 0.81 0.77 0.50
of English in 2007 0.62 0.49 of English in 2007 2.12 0.98
of math in 2005 0.67 0.47 of math in 2005 1.72 0.68
of math in 2006 0.61 0.49 0.37 0.31 of math in 2006 1.84 0.72 0.74 0.54
of math in 2007 0.64 0.48 of math in 2007 2.15 1.03
Expenditure in private tutoring (+) Standardized test score (Z-score)
of English in 2005 11.99 10.33 of English in 2005 0.14 0.96
of English in 2006 13.73 10.49 10.58 8.23 of English in 2006 0.16 0.97 0.88 0.43
of English in 2007 19.37 16.77 of English in 2007 0.25 1.00
of math in 2005 10.93 9.31 of math in 2005 0.13 0.96
of math in 2006 13.70 10.27 10.79 8.60 of math in 2006 0.14 0.99 0.87 0.49
of math in 2007 19.68 17.44 of math in 2007 0.17 1.05
Hours for private tutoring (+) Self-study hours
of English in 2005 4.39 2.17 of English in 2005 1.89 1.97
of English in 2006 4.80 2.63 2.23 1.94 of English in 2006 2.01 2.00 0.77 0.50
of English in 2007 5.21 3.52 of English in 2007 2.13 2.14
of math in 2005 4.42 2.15 of math in 2005 1.98 2.00
of math in 2006 4.88 2.71 2.21 1.87 of math in 2006 2.08 1.99 0.74 0.54
of math in 2007 5.16 3.28 of math in 2007 2.25 2.17
101
Table 3.3: Eect of private tutoring on English Z-score
Dep. Variable: English Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.271*** 0.258*** 0.062*** -0.930 0.024 0.031 0.066 6.114 0.160**
(10.04) (11.82) (3.01) (-0.82) (1.16) (1.32) (1.31) (0.87) (2.05)
(b) Monthly expenditure in private tutoring
Log expenditure 0.111*** 0.104*** 0.021*** 0.176 0.006 0.009 0.025 1.335 0.120**
(11.20) (12.52) (2.67) (1.40) (0.84) (1.08) (1.30) (0.89) (2.01)
(c) Weekly hours for private tutoring
Hours 0.037*** 0.036*** 0.007** 0.156** 0.004 0.005 0.011* 0.993 0.031**
(8.05) (9.58) (2.33) (2.31) (1.22) (1.53) (1.71) (1.28) (2.56)
(d) Expenditure per hour for private tutoring
Log unit value 0.244*** 0.222*** 0.031 -1.040* -0.005 0.004 0.036 -3.402 0.017
(6.64) (7.67) (1.27) (-1.78) (-0.19) (0.13) (0.61) (-1.25) (0.14)
R-squared for (a) 0.25 0.31 0.81 0.02 0.01 0.06 0.31 0.001
R-squared for (b) 0.26 0.31 0.81 0.16 0.01 0.06 0.31 0.000
R-squared for (c) 0.25 0.31 0.81 0.12 0.01 0.06 0.31 0.001
R-squared for (d) 0.25 0.30 0.81 0.02 0.01 0.06 0.31 0.001
p-value of Hansen test for (a) 0.10
p-value of Hansen test for (b) 0.73
p-value of Hansen test for (c) 0.47
p-value of Hansen test for (d) 0.10
Mean of Dependant Variable 0.18 0.05 0.18
Observation 6990 4660 6990
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
102
Table 3.4: Eect of private tutoring on mathematics Z-score
Dep. Variable: math Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.311*** 0.283*** 0.100*** 2.473*** 0.093*** 0.075*** 0.135** 1.827 -0.307
(11.04) (11.74) (4.06) (3.53) (3.79) (2.63) (2.43) (1.06) (-0.79)
(b) Monthly expenditure in private tutoring
Log expenditure 0.112*** 0.102*** 0.037*** 0.402** 0.039*** 0.033*** 0.060*** 0.546 -0.088
(9.79) (10.96) (4.16) (2.12) (4.20) (3.16) (2.98) (0.99) (-0.67)
(c) Weekly hours for private tutoring
Hours 0.038*** 0.036*** 0.011*** 0.231*** 0.013*** 0.011*** 0.025*** 0.507** 0.021
(7.78) (9.25) (3.24) (2.97) (3.67) (2.79) (3.46) (2.30) (1.55)
(d) Expenditure per hour for private tutoring
Log unit value 0.227*** 0.194*** 0.081*** 0.897* 0.096*** 0.081** 0.084 0.628 -0.042
(5.51) (6.09) (2.84) (1.85) (3.30) (2.38) (1.29) (0.21) (-0.37)
R-squared for (a) 0.17 0.24 0.76 0.07 0.01 0.04 0.29 0.004
R-squared for (b) 0.17 0.24 0.76 0.02 0.02 0.04 0.29 0.004
R-squared for (c) 0.17 0.23 0.76 0.01 0.01 0.04 0.29 0.001
R-squared for (d) 0.16 0.23 0.76 0.03 0.01 0.04 0.29 0.002
p-value of Hansen test for (a) 0.88
p-value of Hansen test for (b) 0.99
p-value of Hansen test for (c) 0.38
p-value of Hansen test for (d) 0.31
Mean of Dependant Variable 0.15 0.02 0.15
Observation 6990 4660 6990
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
103
Table 3.5: Eect of private tutoring on English Z-score: Bottom 50% of percentile baseline score
Dep. Variable: English Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.135*** 0.129*** 0.041 0.246 0.061** 0.065* 0.074 2.197 0.123
(4.83) (5.29) (1.39) (0.15) (2.03) (1.87) (1.03) (1.57) (1.39)
(b) Monthly expenditure in private tutoring
Log expenditure 0.049*** 0.047*** 0.025** 0.488 0.022* 0.023* 0.027 0.735 0.065*
(4.47) (4.65) (2.03) (1.03) (1.86) (1.66) (0.92) (1.40) (1.86)
(c) Weekly hours for private tutoring
Hours for private tutoring 0.015*** 0.017*** 0.007 0.128 0.005 0.006 0.005 0.729 0.023
(3.02) (3.76) (1.42) (0.50) (0.93) (0.95) (0.44) (1.13) (1.63)
(d) Expenditure per hours for private tutoring
Log unit value 0.127*** 0.102*** 0.061 2.971 0.049 0.047 0.044 1.751 0.053
(3.59) (3.12) (1.59) (1.21) (1.22) (0.96) (0.47) (1.04) (0.27)
R-squared for (a) 0.14 0.23 0.60 0.04 0.03 0.09 0.30 0.004
R-squared for (b) 0.14 0.23 0.60 0.05 0.03 0.09 0.30 0.004
R-squared for (c) 0.14 0.22 0.60 0.04 0.02 0.09 0.30 0.000
R-squared for (d) 0.14 0.22 0.60 0.04 0.03 0.09 0.30 0.003
p-value of Hansen test for (a) 0.21
p-value of Hansen test for (b) 0.69
p-value of Hansen test for (c) 0.57
p-value of Hansen test for (d) 0.41
Mean of Dependant Variable -0.60 0.18 -0.60
Observation 2985 1990 2985
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
104
Table 3.6: Eect of private tutoring on mathematics Z-score: Bottom 50% of percentile baseline score
Dep. Variable: math Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.154*** 0.122*** 0.050 -1.075 0.090** 0.078* 0.066 0.133 0.048
(5.01) (4.41) (1.38) (-0.87) (2.33) (1.83) (0.85) (0.05) (0.56)
(b) Monthly expenditure in private tutoring
Log expenditure 0.059*** 0.050*** 0.070*** 1.699* 0.040*** 0.038** 0.038 -0.004 0.008
(4.37) (4.38) (5.00) (1.75) (2.72) (2.34) (1.27) (-0.00) (0.15)
(c) Weekly hours for private tutoring
Hours for private tutoring 0.016*** 0.012*** 0.016*** 0.916 0.014** 0.012* 0.025** 0.150 0.001
(3.58) (2.68) (3.05) (1.29) (2.48) (1.92) (2.41) (0.63) (0.08)
(d) Expenditure per hours for private tutoring
Log unit value 0.136*** 0.116*** 0.188*** 4.071* 0.092** 0.089* -0.006 -0.845 -0.045
(3.01) (2.94) (4.13) (1.72) (2.05) (1.66) (-0.06) (-0.27) (-0.22)
R-squared for (a) 0.12 0.22 0.56 0.03 0.06 0.12 0.31 0.001
R-squared for (b) 0.12 0.22 0.57 0.05 0.07 0.12 0.31 0.001
R-squared for (c) 0.12 0.22 0.56 0.04 0.06 0.12 0.31 0.000
R-squared for (d) 0.12 0.22 0.57 0.04 0.06 0.12 0.31 0.001
p-value of Hansen test for (a) 0.13
p-value of Hansen test for (b) 0.90
p-value of Hansen test for (c) 0.67
p-value of Hansen test for (d) 0.37
Mean of Dependant Variable -0.60 0.18 -0.60
Observation 2898 1932 2898
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
105
Table 3.7: Eect of private tutoring on English Z-score: Top 50% of percentile baseline score
Dep. Variable: English Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.181*** 0.177*** 0.090*** -6.746 0.046* 0.049 0.068 -8.144 0.234**
(7.75) (6.87) (3.25) (-1.36) (1.76) (1.48) (0.98) (-1.03) (2.46)
(b) Monthly expenditure in private tutoring
Log expenditure 0.080*** 0.079*** 0.021** -0.246 0.012 0.013 0.021 -1.238 0.080**
(9.19) (8.77) (2.19) (-0.65) (1.34) (1.16) (0.84) (-1.06) (2.20)
(c) Weekly hours for private tutoring
Hours for private tutoring 0.028*** 0.028*** 0.007** -0.023 0.007** 0.007** 0.013* 0.356 0.016
(6.14) (6.95) (2.08) (-0.15) (2.25) (2.00) (1.78) (1.61) (1.31)
(d) Expenditure per hours for private tutoring
Log unit value 0.157*** 0.157*** 0.023 -1.385 -0.008 -0.000 0.012 -4.666* -0.261
(4.99) (4.89) (0.73) (-1.20) (-0.27) (-0.01) (0.17) (-1.86) (-1.40)
R-squared for (a) 0.14 0.20 0.68 0.04 0.04 0.09 0.33 0.002
R-squared for (b) 0.15 0.20 0.68 0.02 0.04 0.09 0.33 0.002
R-squared for (c) 0.14 0.20 0.68 0.02 0.04 0.09 0.33 0.000
R-squared for (d) 0.13 0.19 0.68 0.02 0.04 0.09 0.33 0.000
p-value of Hansen test for (a) 0.27
p-value of Hansen test for (b) 0.65
p-value of Hansen test for (c) 0.24
p-value of Hansen test for (d) 0.49
Mean of Dependant Variable 0.77 -0.04 0.77
Observation 4005 2670 4005
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
106
Table 3.8: Eect of private tutoring on mathematics Z-score: Top 50% of percentile baseline score
Dep. Variable: math Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.234*** 0.211*** 0.145*** 3.907* 0.166*** 0.164*** 0.214*** 2.369 0.066
(8.31) (7.33) (4.43) (1.93) (5.40) (4.28) (2.88) (1.01) (0.52)
(b) Monthly expenditure in private tutoring
Log expenditure 0.084*** 0.076*** 0.020* 1.801 0.060*** 0.058*** 0.080*** 0.401 0.035
(7.94) (7.16) (1.76) (0.72) (5.16) (4.25) (3.06) (0.49) (0.47)
(c) Weekly hours for private tutoring
Hours for private tutoring 0.031*** 0.029*** 0.009** 0.583 0.019*** 0.019*** 0.025*** 0.857 0.031**
(6.16) (6.84) (2.25) (1.11) (4.16) (3.77) (2.65) (0.71) (2.20)
(d) Expenditure per hours for private tutoring
Log unit value 0.185*** 0.152*** 0.022 5.493 0.139*** 0.125*** 0.153* -0.520 -0.054
(5.18) (4.16) (0.62) (0.63) (3.66) (2.75) (1.84) (-0.14) (-0.43)
R-squared for (a) 0.10 0.16 0.64 0.001 0.06 0.10 0.31 0.02
R-squared for (b) 0.11 0.16 0.64 0.011 0.06 0.10 0.31 0.02
R-squared for (c) 0.10 0.16 0.64 0.009 0.06 0.10 0.31 0.00
R-squared for (d) 0.10 0.16 0.64 0.016 0.06 0.10 0.31 0.01
p-value of Hansen test for (a) 0.44
p-value of Hansen test for (b) 0.52
p-value of Hansen test for (c) 0.79
p-value of Hansen test for (d) 0.35
Mean of Dependant Variable 0.68 -0.09 0.68
Observation 4092 2728 4092
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
107
Table 3.9: Eect of private tutoring on English Z-score: Boys
Dep. Variable: English Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.237*** 0.212*** 0.086*** 1.111 0.045 0.054 0.085 9.626 0.137
(6.03) (6.37) (2.68) (0.25) (1.39) (1.43) (1.07) (1.17) (1.06)
(b) Monthly expenditure in private tutoring
Log expenditure 0.100*** 0.090*** 0.026** -0.418 0.011 0.014 0.028 2.521 0.108**
(7.08) (7.21) (2.15) (-0.68) (0.93) (1.02) (0.92) (1.18) (2.52)
(c) Weekly hours for private tutoring
Hours 0.030*** 0.029*** 0.009* -0.052 0.002 0.004 0.009 0.516 0.031**
(4.42) (4.84) (1.93) (-0.50) (0.41) (0.85) (0.91) (1.29) (2.17)
(d) Expenditure per hours for private tutoring
Log unit value 0.236*** 0.193*** 0.040 -2.680 0.022 0.017 0.024 13.886 0.066
(4.37) (4.43) (0.99) (-0.95) (0.52) (0.33) (0.23) (0.62) (0.36)
R-squared for (a) 0.23 0.30 0.79 0.11 0.03 0.08 0.31 0.001
R-squared for (b) 0.23 0.30 0.79 0.02 0.03 0.08 0.31 0.001
R-squared for (c) 0.22 0.30 0.79 0.06 0.03 0.08 0.31 0.010
R-squared for (d) 0.22 0.30 0.79 0.01 0.03 0.08 0.31 0.001
p-value of Hansen test for (a) 0.39
p-value of Hansen test for (b) 0.31
p-value of Hansen test for (c) 0.39
p-value of Hansen test for (d) 0.36
Mean of Dependant Variable 0.03 0.04 0.03
Observation 3414 2276 3414
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
108
Table 3.10: Eect of private tutoring on mathematics Z-score: Boys
Dep. Variable: math Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.296*** 0.264*** 0.089** 6.945 0.055 0.042 0.042 11.299 0.025
(7.01) (7.38) (2.41) (1.15) (1.55) (0.96) (0.52) (0.82) (0.27)
(b) Monthly expenditure in private tutoring
Log expenditure 0.097*** 0.087*** 0.023* 0.467** 0.017 0.016 0.020 1.067 -0.043
(6.12) (6.45) (1.74) (2.16) (1.43) (1.01) (0.69) (1.13) (-0.57)
(c) Weekly hours for private tutoring
Hours 0.030*** 0.030*** 0.008* 0.220** 0.008 0.009 0.014 0.210 -0.017
(4.36) (5.18) (1.74) (2.12) (1.60) (1.50) (1.31) (1.04) (-0.61)
(d) Expenditure per hours for private tutoring
Log unit value 0.213*** 0.175*** 0.026 1.228** 0.021 0.008 -0.038 3.661 0.065
(4.02) (3.73) (0.61) (2.17) (0.48) (0.15) (-0.39) (1.37) (0.25)
R-squared for (a) 0.17 0.26 0.75 0.078 0.02 0.06 0.30 0.001
R-squared for (b) 0.17 0.26 0.75 0.001 0.02 0.06 0.30 0.001
R-squared for (c) 0.16 0.26 0.75 0.000 0.02 0.06 0.30 0.000
R-squared for (d) 0.16 0.25 0.75 0.004 0.02 0.06 0.30 0.000
p-value of Hansen test for (a) 0.02
p-value of Hansen test for (b) 0.85
p-value of Hansen test for (c) 0.72
p-value of Hansen test for (d) 0.96
Mean of Dependant Variable 0.15 0.03 0.15
Observation 3414 2276 3414
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
109
Table 3.11: Eect of private tutoring on English Z-score: Girls
Dep. Variable: English Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.294*** 0.293*** 0.036 -0.716 -0.010 0.015 0.042 0.261 0.163***
(8.39) (10.21) (1.38) (-0.37) (-0.35) (0.50) (0.67) (0.30) (2.74)
(b) Monthly expenditure in private tutoring
Log expenditure 0.120*** 0.116*** 0.015 -0.539 -0.004 0.005 0.016 0.084 0.062*
(9.56) (10.63) (1.53) (-0.58) (-0.43) (0.42) (0.69) (0.28) (1.87)
(c) Weekly hours for private tutoring
Hours for private tutoring 0.044*** 0.044*** 0.005 -0.090 0.003 0.004 0.010 0.265 0.019**
(7.91) (9.71) (1.28) (-0.19) (0.83) (1.03) (1.22) (1.62) (2.38)
(d) Expenditure per hours for private tutoring
Log unit value of private tutoring 0.252*** 0.235*** 0.019 -2.687 -0.038 -0.004 0.031 -0.146 0.066
(6.04) (6.19) (0.65) (-0.90) (-1.18) (-0.12) (0.48) (-0.15) (0.36)
R-squared for (a) 0.27 0.35 0.83 0.09 0.02 0.08 0.33 0.000
R-squared for (b) 0.27 0.35 0.83 0.10 0.02 0.08 0.33 0.001
R-squared for (c) 0.26 0.35 0.83 0.08 0.02 0.08 0.33 0.001
R-squared for (d) 0.26 0.34 0.83 0.09 0.02 0.08 0.33 0.010
p-value of Hansen test for (a) 0.10
p-value of Hansen test for (b) 0.37
p-value of Hansen test for (c) 0.19
p-value of Hansen test for (d) 0.36
Mean of Dependant Variable 0.33 0.07 0.33
Observation 3576 2384 3576
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
110
Table 3.12: Eect of private tutoring on mathematics Z-score: Girls
Dep. Variable: math Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.322*** 0.304*** 0.110*** 0.662 0.120*** 0.104*** 0.230*** 0.355 0.154
(9.29) (9.42) (3.35) (0.49) (3.75) (2.66) (3.05) (0.15) (1.27)
(b) Monthly expenditure in private tutoring
Log expenditure 0.127*** 0.119*** 0.051*** -0.031 0.056*** 0.049*** 0.098*** 0.002 0.039
(9.02) (9.45) (4.25) (-0.10) (4.47) (3.38) (3.54) (0.00) (0.95)
(c) Weekly hours for private tutoring
Hours for private tutoring 0.044*** 0.042*** 0.013*** 0.082 0.018*** 0.014*** 0.038*** 0.506 0.041**
(7.00) (7.92) (2.95) (0.67) (3.79) (2.64) (3.94) (1.46) (2.57)
(d) Expenditure per hours for private tutoring
Log unit value 0.255*** 0.227*** 0.131*** -0.212 0.158*** 0.143*** 0.181** -2.884 0.065
(4.95) (5.34) (3.50) (-0.21) (4.18) (3.19) (2.05) (-0.82) (0.25)
R-squared for (a) 0.20 0.28 0.77 0.10 0.03 0.06 0.29 0.006
R-squared for (b) 0.20 0.28 0.78 0.07 0.03 0.07 0.30 0.004
R-squared for (c) 0.19 0.28 0.77 0.09 0.03 0.06 0.30 0.001
R-squared for (d) 0.18 0.27 0.77 0.07 0.03 0.06 0.29 0.001
p-value of Hansen test for (a) 0.03
p-value of Hansen test for (b) 0.53
p-value of Hansen test for (c) 0.83
p-value of Hansen test for (d) 0.96
Mean of Dependant Variable 0.14 0.02 0.14
Observation 3576 2384 3576
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
111
Table 3.13: Eect of private tutoring on English Z-score: Urban sector
Dep. Variable: English Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.258*** 0.248*** 0.093*** -2.858 0.033 0.046 0.059 -2.453 -0.052
(6.87) (7.35) (3.02) (-1.58) (1.02) (1.31) (0.80) (-1.44) (-0.37)
(b) Monthly expenditure in private tutoring
Log expenditure 0.105*** 0.101*** 0.033*** -0.206 0.013 0.018 0.025 -0.876 0.082
(7.41) (8.28) (2.99) (-0.86) (1.16) (1.46) (0.90) (-1.42) (1.53)
(c) Weekly hours for private tutoring
Hours for private tutoring 0.041*** 0.039*** 0.012*** -0.076 0.008* 0.009* 0.008 -0.270 0.038*
(7.07) (7.61) (2.86) (-0.75) (1.69) (1.93) (0.78) (-1.24) (1.71)
(d) Expenditure per hours for private tutoring
Log unit value 0.246*** 0.243*** 0.072** -0.936 0.018 0.030 0.065 -4.426 -0.254
(4.45) (5.65) (2.01) (-0.91) (0.47) (0.69) (0.75) (-1.11) (-1.14)
R-squared for (a) 0.22 0.27 0.81 0.08 0.03 0.06 0.31 0.001
R-squared for (b) 0.22 0.27 0.81 0.06 0.03 0.06 0.31 0.000
R-squared for (c) 0.22 0.27 0.81 0.05 0.03 0.06 0.31 0.001
R-squared for (d) 0.21 0.26 0.81 0.06 0.03 0.06 0.31 0.000
p-value of Hansen test for (a) 0.11
p-value of Hansen test for (b) 0.08
p-value of Hansen test for (c) 0.66
p-value of Hansen test for (d) 0.63
Mean of Dependant Variable 0.32 0.01 0.32
Observation 3237 2158 3237
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
112
Table 3.14: Eect of private tutoring on mathematics Z-score: Urban sector
Dep. Variable: math Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.295*** 0.283*** 0.151*** 0.125 0.113*** 0.106** 0.169* -0.436 0.299***
(7.88) (7.71) (3.92) (0.13) (2.85) (2.36) (1.95) (-0.30) (2.83)
(b) Monthly expenditure in private tutoring
Log expenditure 0.104*** 0.099*** 0.054*** 0.087 0.051*** 0.047*** 0.072** 0.002 0.106**
(5.98) (7.36) (4.03) (0.49) (3.37) (2.96) (2.32) (0.00) (2.04)
(c) Weekly hours for private tutoring
Hours for private tutoring 0.040*** 0.041*** 0.016*** 0.068 0.017*** 0.016** 0.033*** 0.363* 0.054**
(5.15) (7.01) (3.14) (0.70) (2.79) (2.56) (2.89) (1.67) (2.54)
(d) Expenditure per hours for private tutoring
Log unit value 0.198*** 0.177*** 0.112*** -0.071 0.125** 0.109** 0.066 -1.102 -0.180
(3.33) (3.87) (2.62) (-0.14) (2.66) (2.15) (0.65) (-0.78) (-1.01)
R-squared for (a) 0.15 0.20 0.75 0.005 0.02 0.04 0.29 0.000
R-squared for (b) 0.15 0.20 0.75 0.010 0.02 0.04 0.29 0.002
R-squared for (c) 0.15 0.20 0.74 0.004 0.02 0.04 0.29 0.010
R-squared for (d) 0.14 0.19 0.74 0.013 0.02 0.04 0.28 0.000
p-value of Hansen test for (a) 0.18
p-value of Hansen test for (b) 0.16
p-value of Hansen test for (c) 0.97
p-value of Hansen test for (d) 0.45
Mean of Dependant Variable 0.22 -0.03 0.22
Observation 3237 2158 3237
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
113
Table 3.15: Eect of private tutoring on English Z-score: Rural sector
Dep. Variable: English Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.283*** 0.259*** 0.035 2.658 0.015 0.015 0.079 5.442** 0.159**
(7.52) (9.01) (1.26) (0.54) (0.51) (0.45) (1.14) (2.52) (2.36)
(b) Monthly expenditure in private tutoring
Log expenditure 0.119*** 0.106*** 0.008 0.804 0.000 -0.001 0.026 2.035** 0.065
(8.94) (9.31) (0.76) (0.87) (0.01) (-0.04) (0.96) (2.38) (1.30)
(c) Weekly hours for private tutoring
Hours for private tutoring 0.035*** 0.033*** 0.003 0.481 0.001 0.001 0.014* 0.684*** 0.015
(5.11) (6.17) (0.82) (1.22) (0.26) (0.28) (1.75) (2.64) (1.48)
(d) Expenditure per hours for private tutoring
Log unit value 0.245*** 0.199*** -0.012 -1.952 -0.025 -0.025 -0.001 4.695 0.066
(5.14) (5.09) (-0.35) (-0.55) (-0.76) (-0.60) (-0.01) (0.98) (0.45)
R-squared for (a) 0.27 0.33 0.81 0.12 0.02 0.07 0.32 0.000
R-squared for (b) 0.28 0.34 0.81 0.13 0.02 0.07 0.32 0.000
R-squared for (c) 0.27 0.33 0.81 0.11 0.02 0.07 0.32 0.002
R-squared for (d) 0.27 0.32 0.81 0.06 0.02 0.07 0.32 0.000
p-value of Hansen test for (a) 0.26
p-value of Hansen test for (b) 0.24
p-value of Hansen test for (c) 0.29
p-value of Hansen test for (d) 0.38
Mean of Dependant Variable 0.06 0.09 0.06
Observation 3753 2502 3753
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
114
Table 3.16: Eect of private tutoring on mathematics Z-score: Rural sector
Dep. Variable: math Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.321*** 0.270*** 0.056* 2.672*** 0.074** 0.044 0.110 3.623* 0.029
(7.68) (8.44) (1.75) (2.92) (2.33) (1.20) (1.52) (1.79) (0.34)
(b) Monthly expenditure in private tutoring
Log expenditure 0.121*** 0.100*** 0.021* 0.396*** 0.029** 0.017 0.049* 1.723* 0.007
(8.14) (7.77) (1.68) (2.84) (2.43) (1.26) (1.86) (1.64) (0.14)
(c) Weekly hours for private tutoring
Hours for private tutoring 0.036*** 0.030*** 0.005 0.243*** 0.009** 0.005 0.018** 0.500* 0.001
(5.88) (5.78) (1.26) (3.37) (2.11) (1.09) (1.97) (1.77) (0.10)
(d) Expenditure per hours for private tutoring
Log unit value 0.258*** 0.203*** 0.051 1.147** 0.082** 0.053 0.102 8.683 0.040
(4.57) (4.59) (1.31) (2.24) (2.18) (1.16) (1.19) (0.91) (0.21)
R-squared for (a) 0.20 0.28 0.78 0.110 0.02 0.06 0.31 0.004
R-squared for (b) 0.20 0.28 0.78 0.005 0.02 0.06 0.31 0.003
R-squared for (c) 0.19 0.28 0.78 0.003 0.02 0.06 0.31 0.000
R-squared for (d) 0.19 0.27 0.78 0.009 0.02 0.06 0.31 0.000
p-value of Hansen test for (a) 0.00
p-value of Hansen test for (b) 0.76
p-value of Hansen test for (c) 0.83
p-value of Hansen test for (d) 0.76
Mean of Dependant Variable 0.09 0.07 0.09
Observation 3753 2502 3753
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
115
Table 3.17: Eect of private tutoring on English Z-score: Lower educated father group
Dep. Variable: English Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.334*** 0.318*** 0.092*** -0.529 0.074*** 0.082** 0.153** 4.489 0.195***
(10.16) (10.94) (3.32) (-0.47) (2.81) (2.44) (2.22) (1.33) (2.81)
(b) Monthly expenditure in private tutoring
Log expenditure 0.138*** 0.128*** 0.031*** 0.135 0.026** 0.028** 0.053* 1.744 0.094***
(10.37) (10.56) (2.70) (0.71) (2.53) (2.09) (1.86) (1.00) (2.67)
(c) Weekly hours for private tutoring
Hours for private tutoring 0.045*** 0.043*** 0.010** -0.033 0.005 0.005 0.015 0.763 0.025**
(6.51) (7.76) (2.18) (-0.34) (1.24) (1.14) (1.48) (1.13) (2.31)
(d) Expenditure per hours for private tutoring
Log unit value 0.331*** 0.292*** 0.056 -0.511 0.077** 0.085* 0.143 5.797 0.020
(7.25) (6.76) (1.51) (-0.65) (2.09) (1.73) (1.53) (0.80) (0.09)
R-squared for (a) 0.17 0.27 0.78 0.05 0.02 0.08 0.32 0.001
R-squared for (b) 0.18 0.27 0.78 0.01 0.02 0.08 0.32 0.001
R-squared for (c) 0.17 0.26 0.78 0.03 0.02 0.08 0.31 0.000
R-squared for (d) 0.16 0.26 0.78 0.03 0.02 0.08 0.31 0.001
p-value of Hansen test for (a) 0.04
p-value of Hansen test for (b) 0.09
p-value of Hansen test for (c) 0.15
p-value of Hansen test for (d) 0.21
Mean of Dependant Variable -0.13 0.07 -0.13
Observation 3630 2420 3630
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
116
Table 3.18: Eect of private tutoring on mathematics Z-score: Lower educated father group
Dep. Variable: math Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.329*** 0.297*** 0.069** 1.003 0.080** 0.060 0.071 -0.428 0.023
(8.23) (9.36) (2.13) (1.23) (2.45) (1.53) (0.99) (-0.04) (0.35)
(b) Monthly expenditure in private tutoring
Log expenditure 0.124*** 0.110*** 0.030** -0.145 0.039*** 0.032** 0.055* -0.242 0.082
(7.42) (8.30) (2.28) (-0.58) (2.89) (2.10) (1.95) (-0.06) (1.43)
(c) Weekly hours for private tutoring
Hours for private tutoring 0.043*** 0.039*** 0.012** 0.027 0.016*** 0.014** 0.030*** 0.130 0.011
(6.05) (6.83) (2.50) (0.31) (2.99) (2.42) (3.05) (0.09) (0.53)
(d) Expenditure per hours for private tutoring
Log unit value 0.283*** 0.230*** 0.068 -0.839 0.108** 0.080 0.064 -1.144 0.082
(5.00) (4.78) (1.52) (-0.83) (2.36) (1.53) (0.68) (-0.08) (0.39)
R-squared for (a) 0.12 0.22 0.74 0.08 0.02 0.06 0.31 0.001
R-squared for (b) 0.11 0.21 0.74 0.04 0.02 0.06 0.31 0.001
R-squared for (c) 0.11 0.21 0.74 0.05 0.02 0.06 0.31 0.000
R-squared for (d) 0.10 0.20 0.74 0.04 0.02 0.06 0.30 0.001
p-value of Hansen test for (a) 0.02
p-value of Hansen test for (b) 0.57
p-value of Hansen test for (c) 0.26
p-value of Hansen test for (d) 0.09
Mean of Dependant Variable -0.12 0.01 -0.12
Observation 3630 2420 3630
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
117
Table 3.19: Eect of private tutoring on English Z-score: Higher educated father group
Dep. Variable: English Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.181*** 0.192*** 0.025 -4.588 -0.037 -0.028 -0.034 -4.533 0.134
(4.33) (5.70) (0.83) (-0.49) (-1.15) (-0.78) (-0.45) (-0.90) (1.14)
(b) Monthly expenditure in private tutoring
Log expenditure 0.084*** 0.083*** 0.011 0.296 -0.012 -0.009 -0.001 -1.720 0.072*
(5.91) (7.13) (1.07) (1.29) (-1.12) (-0.72) (-0.05) (-0.67) (1.84)
(c) Weekly hours for private tutoring
Hours for private tutoring 0.029*** 0.031*** 0.005 0.419** 0.003 0.004 0.007 0.051 0.027*
(5.11) (6.16) (1.29) (1.97) (0.80) (1.03) (0.88) (0.10) (1.94)
(d) Expenditure per hours for private tutoring
Log unit value 0.168*** 0.153*** 0.009 -0.305 -0.074** -0.070* -0.052 -2.423 0.074
(3.23) (3.85) (0.28) (-0.56) (-2.25) (-1.79) (-0.68) (-0.82) (0.42)
R-squared for (a) 0.17 0.26 0.80 0.06 0.02 0.07 0.31 0.000
R-squared for (b) 0.18 0.26 0.80 0.09 0.02 0.07 0.31 0.000
R-squared for (c) 0.18 0.26 0.80 0.06 0.02 0.07 0.31 0.001
R-squared for (d) 0.17 0.26 0.80 0.05 0.02 0.07 0.31 0.000
p-value of Hansen test for (a) 0.23
p-value of Hansen test for (b) 0.95
p-value of Hansen test for (c) 0.92
p-value of Hansen test for (d) 0.88
Mean of Dependant Variable 0.51 0.04 0.51
Observation 3360 2240 3360
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
118
Table 3.20: Eect of private tutoring on mathematics Z-score: Higher educated father group
Dep. Variable: math Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.280*** 0.259*** 0.143*** 3.637** 0.103*** 0.095** 0.212** 5.293 0.005
(7.19) (6.88) (3.75) (2.50) (2.75) (2.09) (2.43) (1.17) (0.03)
(b) Monthly expenditure in private tutoring
Log expenditure 0.097*** 0.092*** 0.044*** 0.491** 0.036*** 0.033** 0.064** 1.810 0.033
(6.91) (7.08) (3.58) (2.29) (2.98) (2.22) (2.21) (1.45) (0.51)
(c) Weekly hours for private tutoring
Hours for private tutoring 0.032*** 0.033*** 0.009** 0.274*** 0.011** 0.008 0.019* 0.457** 0.033**
(4.69) (6.24) (2.05) (2.60) (2.27) (1.40) (1.78) (1.99) (2.01)
(d) Expenditure per hours for private tutoring
Log unit value 0.175*** 0.161*** 0.089** 1.237** 0.080** 0.086* 0.099 1.537 -0.194
(3.45) (3.73) (2.42) (2.03) (2.17) (1.82) (1.09) (0.64) (-0.94)
R-squared for (a) 0.11 0.21 0.75 0.03 0.02 0.06 0.28 0.002
R-squared for (b) 0.11 0.21 0.75 0.01 0.02 0.06 0.28 0.003
R-squared for (c) 0.11 0.21 0.75 0.00 0.02 0.06 0.28 0.000
R-squared for (d) 0.10 0.20 0.75 0.02 0.02 0.06 0.28 0.003
p-value of Hansen test for (a) 0.64
p-value of Hansen test for (b) 0.85
p-value of Hansen test for (c) 0.38
p-value of Hansen test for (d) 0.96
Mean of Dependant Variable 0.43 0.03 0.43
Observation 3360 2240 3360
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
119
Table 3.21: Eect of private tutoring on English Z-score: Lower educated mother group
Dep. Variable: English Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.301*** 0.289*** 0.081*** 3.840 0.053** 0.064** 0.097 4.088 0.195***
(9.41) (11.31) (3.34) (0.80) (2.21) (2.25) (1.64) (1.41) (3.06)
(b) Monthly expenditure in private tutoring
Log expenditure 0.128*** 0.120*** 0.026*** 0.405 0.023** 0.025** 0.038* 1.294 0.105**
(10.52) (11.68) (2.70) (0.81) (2.48) (2.32) (1.65) (1.31) (2.07)
(c) Weekly hours for private tutoring
Hours for private tutoring 0.041*** 0.040*** 0.009*** 0.184 0.005 0.006 0.012 0.878 0.034***
(7.02) (8.80) (2.59) (0.89) (1.31) (1.63) (1.43) (1.47) (3.38)
(d) Expenditure per hours for private tutoring
Log unit value 0.307*** 0.278*** 0.048 -0.210 0.067** 0.067* 0.091 1.986 0.016
(7.21) (7.67) (1.56) (-0.12) (2.38) (1.79) (1.27) (0.68) (0.11)
R-squared for (a) 0.19 0.27 0.80 0.08 0.02 0.07 0.31 0.001
R-squared for (b) 0.20 0.28 0.80 0.08 0.02 0.08 0.31 0.001
R-squared for (c) 0.19 0.27 0.80 0.08 0.02 0.07 0.31 0.000
R-squared for (d) 0.18 0.26 0.80 0.06 0.02 0.07 0.31 0.001
p-value of Hansen test for (a) 0.10
p-value of Hansen test for (b) 0.51
p-value of Hansen test for (c) 0.62
p-value of Hansen test for (d) 0.20
Mean of Dependant Variable -0.01 0.06 -0.01
Observation 4896 3264 4896
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
120
Table 3.22: Eect of private tutoring on mathematics Z-score: Lower educated mother group
Dep. Variable: math Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.332*** 0.307*** 0.080*** 1.862*** 0.091*** 0.073** 0.106 1.440 0.031
(9.60) (10.88) (2.83) (2.88) (3.19) (2.15) (1.64) (0.43) (0.53)
(b) Monthly expenditure in private tutoring
Log expenditure 0.128*** 0.115*** 0.037*** 0.212* 0.041*** 0.033*** 0.052** 0.427 0.031
(8.96) (10.17) (3.35) (1.78) (3.75) (2.64) (2.14) (0.37) (0.76)
(c) Weekly hours for private tutoring
Hours for private tutoring 0.039*** 0.037*** 0.012*** 0.175*** 0.014*** 0.012** 0.024*** 0.306 0.009
(6.54) (7.65) (2.86) (2.80) (3.14) (2.42) (2.79) (0.80) (0.54)
(d) Expenditure per hours for private tutoring
Log unit value 0.307*** 0.257*** 0.100*** 0.323 0.109*** 0.082** 0.055 1.585 0.075
(6.39) (6.53) (2.86) (0.95) (3.15) (1.97) (0.71) (0.30) (0.33)
R-squared for (a) 0.13 0.22 0.75 0.07 0.02 0.05 0.30 0.002
R-squared for (b) 0.13 0.22 0.75 0.01 0.02 0.05 0.30 0.002
R-squared for (c) 0.12 0.21 0.75 0.00 0.02 0.05 0.30 0.000
R-squared for (d) 0.12 0.21 0.75 0.02 0.02 0.05 0.29 0.002
p-value of Hansen test for (a) 0.00
p-value of Hansen test for (b) 0.07
p-value of Hansen test for (c) 0.11
p-value of Hansen test for (d) 0.41
Mean of Dependant Variable -0.01 0.02 -0.01
Observation 4896 3264 4896
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
121
Table 3.23: Eect of private tutoring on English Z-score: Higher educated mother group
Dep. Variable: English Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.168*** 0.176*** 0.012 -3.880 -0.063 -0.078 -0.028 -2.507 0.038
(3.21) (4.06) (0.31) (-1.23) (-1.30) (-1.53) (-0.29) (-0.88) (0.33)
(b) Monthly expenditure in private tutoring
Log expenditure 0.072*** 0.068*** 0.010 0.049 -0.030** -0.030* -0.009 -1.096 0.037
(4.20) (4.71) (0.81) (0.19) (-2.00) (-1.82) (-0.27) (-0.70) (1.04)
(c) Weekly hours for private tutoring
Hours for private tutoring 0.028*** 0.028*** 0.002 0.103 0.000 -0.000 0.006 -0.003 0.015
(3.89) (4.28) (0.39) (0.98) (0.07) (-0.07) (0.62) (-0.01) (1.61)
(d) Expenditure per hours for private tutoring
Log unit value 0.115* 0.096** 0.002 -0.981 -0.141*** -0.137** -0.083 -1.810 0.096
(1.94) (1.98) (0.06) (-1.09) (-3.03) (-2.57) (-0.80) (-0.60) (0.47)
R-squared for (a) 0.19 0.28 0.79 0.04 0.02 0.09 0.32 0.000
R-squared for (b) 0.19 0.29 0.79 0.00 0.02 0.09 0.32 0.001
R-squared for (c) 0.19 0.29 0.79 0.05 0.02 0.09 0.32 0.000
R-squared for (d) 0.18 0.28 0.79 0.00 0.03 0.10 0.32 0.001
p-value of Hansen test for (a) 0.20
p-value of Hansen test for (b) 0.34
p-value of Hansen test for (c) 0.46
p-value of Hansen test for (d) 0.49
Mean of Dependant Variable 0.62 0.03 0.62
Observation 2094 1396 2094
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
122
Table 3.24: Eect of private tutoring on mathematics Z-score: Higher educated mother group
Dep. Variable: math Z-score = 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(a) Participation on private tutoring
Participation 0.245*** 0.222*** 0.162*** 2.886** 0.094* 0.084 0.238** 1.060 0.259
(4.50) (4.60) (3.26) (2.30) (1.91) (1.41) (2.16) (0.63) (1.33)
(b) Monthly expenditure in private tutoring
Log expenditure 0.074*** 0.067*** 0.039** 0.412 0.033* 0.030 0.088** 0.531 0.107**
(3.77) (4.05) (2.55) (1.57) (1.90) (1.55) (2.43) (0.83) (1.97)
(c) Weekly hours for private tutoring
Hours for private tutoring 0.034*** 0.033*** 0.009 0.257* 0.010* 0.010 0.027** 0.447 0.033**
(3.91) (4.68) (1.60) (1.92) (1.74) (1.39) (2.14) (1.53) (2.18)
(d) Expenditure per hours for private tutoring
Log unit value 0.074 0.058 0.047 1.330 0.076 0.069 0.170 -2.194 -0.220
(1.08) (1.04) (0.95) (1.40) (1.44) (1.11) (1.39) (-0.61) (-1.07)
R-squared for (a) 0.13 0.23 0.74 0.01 0.02 0.08 0.29 0.001
R-squared for (b) 0.12 0.23 0.74 0.02 0.02 0.08 0.29 0.002
R-squared for (c) 0.13 0.23 0.74 0.00 0.02 0.08 0.29 0.000
R-squared for (d) 0.12 0.22 0.74 0.02 0.02 0.08 0.29 0.002
p-value of Hansen test for (a) 0.85
p-value of Hansen test for (b) 0.94
p-value of Hansen test for (c) 0.74
p-value of Hansen test for (d) 0.39
Mean of Dependant Variable 0.51 0.02 0.51
Observation 2094 1396 2094
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
123
Table 3.25: Comparison between eects of private tutoring hours and self-study hours on average test score
Dep. Variable Average Z-score
= 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(c) Weekly hours for private tutoring
Hours for private tutoring 0.022*** 0.021*** 0.005*** 0.163*** 0.004*** 0.004*** 0.010*** 0.413 0.020***
(11.08) (10.88) (3.49) (4.56) (3.01) (2.64) (3.26) (0.88) (2.78)
Hours for self-study 0.039*** 0.036*** 0.005** -0.053 0.002 0.003 0.008 2.444 0.039***
(14.25) (13.33) (2.13) (-0.42) (0.87) (0.99) (1.36) (0.60) (3.32)
Joint Test 22.79 18.83 0.00 2.41 0.46 0.19 0.15 0.28 3.43
(p-value) 0.00 0.00 0.95 0.12 0.50 0.66 0.70 0.60 0.06
Mean of Dependant Variable 0.16
Observation 6990
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
124
Table 3.26: Comparison between eects of private tutoring hours and self-study hours
Dep. Variable English Z-score
= 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(c) Weekly hours for private tutoring
Hours for private tutoring 0.037*** 0.036*** 0.007** 0.156** 0.004 0.005 0.011* 0.993 0.031**
(8.05) (9.58) (2.33) (2.31) (1.22) (1.53) (1.71) (1.28) (2.56)
Hours for self-study 0.077*** 0.072*** 0.009* 0.179 0.007 0.009 0.016 2.320 0.037**
(11.55) (13.44) (1.86) (0.99) (1.22) (1.64) (1.28) (0.94) (2.04)
Joint Test 22.30 27.14 0.21 0.02 0.24 0.49 0.12 0.47 0.12
(p-value) 0.00 0.00 0.65 0.90 0.63 0.48 0.73 0.49 0.73
Mean of Dependant Variable 0.18 0.05 0.18
Observation 6990 4660 6990
mathematics Z-score
Hours for private tutoring 0.038*** 0.036*** 0.011*** 0.231*** 0.013*** 0.011*** 0.025*** 0.507** 0.021
(7.78) (9.25) (3.24) (2.97) (3.67) (2.79) (3.46) (2.30) (1.55)
Hours for self-study 0.078*** 0.072*** 0.019*** -0.495* 0.013** 0.012* 0.029** -0.245 0.086***
(11.95) (12.80) (3.58) (-1.92) (2.47) (1.83) (2.15) (-0.16) (4.27)
Joint Test 23.62 24.90 1.73 5.71 0.00 0.02 0.06 0.21 8.12
(p-value) 0.00 0.00 0.19 0.02 0.98 0.88 0.80 0.64 0.00
Mean of Dependant Variable 0.15 0.02 0.15
Observation 6990 4660 6990
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels. 125
Chapter 4
Conclusion
Policy makers and scholars have paid attention to private tutoring issues because tutoring
services may cause educational inequality between students with dierent demographic
characteristics. Despite several policies to counteract this, household spending on private
tutoring has been increasing and tutoring activities are still very popular in South Korea.
In spite of the popularity of supplementary education and its potential in
uence on inter-
generational mobility, there has been relatively little focus on it in the research area. This
dissertation examines both the demand for tutoring and the eects of private tutoring on
academic achievement.
Chapter 2 investigates the demand system for private tutoring while most previous
literature has been focused on determinants of demand for private tutoring. Since price
data do not exist, unit values, which is expenditure per hour, are an attractive proxy
126
for the price. However, quality issues associated with the unit values need to be consid-
ered. I also nd positive quality elasticities for tutoring activities with respect to total
educational expenditure indicating that better-o households are willing to pay more per
hour. Hence, this paper adopts Deaton's demand analysis to address the quality eect.
Deaton's demand analysis models the determination of quantity and quality related to
unit values and develops an appropriate estimation strategy.
This essay nds that both the small-group tutoring, namely, to hire an individual
tutor, and the cram school, namely, to attend a large-size class, are normal goods with
positive elasticities of demand with respect to total educational expenditure, and that
they have negative own-price elasticities. Small-group tutoring is almost three times as
elastic as large-scale class tutoring. Own-price elasticities for cram schools in the urban
areas are insignicant, because students who live in urban areas can more easily nd
other cram schools, oering the same tuition. More educated parents are likely to have
smaller elasticities of demand with respect to prices because parents with a higher level
of educational attainment are more likely to generate better-o households and can more
easily aord to maintain the same level of consumption, corresponding to an increase in
tutoring tuition. Students who perform better in tests in the baseline year are less likely
to change their demand for tutoring services if there is an increase in the price.
Chapter 3 detects the eectiveness of private tutoring for English and Mathematics.
I investigate not only the average association between tutoring and test scores through
the POLS, FE, and FE with 2SLS approach, controlling unobserved characteristics, but
127
also whether the change in private tutoring can raise academic achievement through the
AB estimator, a kind of dynamic panel estimation. With the POLS, FE, and FE with IV
methods, tutoring activities have signicantly positive relation with academic achievement
in terms of four dierent measures of private tutoring: participation in, time and money
investment in private tutoring, and unit values. However, the AB estimator results in the
positive eect of only English tutoring on test scores.
While the result of this study shows that English private tutoring enhances student
test scores, there is no signicant dierence in eects on test results between hours spent
on self-study and English tutoring hours. From the results of the heterogeneous eects,
this essay identies that the eects of tutoring services vary with socio-economic char-
acteristics. Students who are in top 50% group of percentile score benet from English
and math tutoring on test score while students in bottom 50% group benet only from
English tutoring. Two main purposes of tutoring can be associated to interpret this re-
sults. English and mathematics tutoring can achieve enrichment purpose for more able
students.
128
References
Arellano, M., and Bond, S. (1991). Some tests of specication for panel data: Monte
Carlo evidence and an application to employment equations. The Review of Economic
Studies, 58(2): 277{297.
Aurini, J., and Davies, S. (2004). The Transformation of Private Tutoring: Education in
a Franchise Form. The Canadian Journal of Sociology / Cahiers canadiens de sociologie,
29(3): 419{438.
Baker, D. P., and LeTendre, G. K. (2005). National dierences, global similarities: World
culture and the future of schooling. Stanford Social Science.
Baltagi, B. (2008). Econometric analysis of panel data. John Wiley & Sons.
Baltagi, B. H., Bresson, G., and Pirotte, A. (2003). Fixed eects, random eects or
Hausman-Taylor?: A pretest estimator. Economics Letters, 79(3): 361{369.
Boumahdi, R., and Thomas, A. (2008). The Econometrics of Panel Data : Fundamentals
and Recent Developments in Theory and Practice. chapter 4. Endogenous Regressors
and Correlated Eects, pages 89{112. Springer-Verlag Berlin Heidelberg, third edition.
Bray, M. (1999). The shadow education system. UNESCO: International Institute for
Educational Planning.
Bray, M. (2009). Confronting the Shadow Education System: What Government Poli-
cies for What Private Tutoring? United Nations Educational, Scientic and Cultural
Organization.
Bray, M. (2011). The Challenge of Shadow Education: Private tutoring and its implica-
tions for policy makers in the European Union. European Commission.
Bray, M., and 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.
129
Brehm, W. C., Silova, I., et al. (2014). Hidden privatization of public education in
Cambodia: Equity implications of private tutoring. Journal for educational research
online, 6(1): 94{116.
Briggs, D. C. (2001). The eect of admissions test preparation: Evidence from NELS:
88. Chance, 14(1): 10{18.
Buchmann, C. (2002). Getting ahead in Kenya: Social capital, shadow education, and
achievement. Schooling and social capital in diverse cultures, pages 133{159.
Buchmann, C., Condron, D. J., and Roscigno, V. J. (2010). Shadow education, American
style: Test preparation, the SAT and college enrollment. Social Forces, 89(2): 435{461.
Byun, S.-y. (2014). Shadow Education and Academic Success in Republic of Korea. In
Korean education in changing economic and demographic contexts, pages 39{58. Springer.
Choi, E. J., Moon, H. R., and Ridder, G. (2014). Estimation of an Education Production
Function under Random Assignment with Selection. American Economic Review, 104(5):
206{211.
Choi, J. (2012). Unequal Access to Shadow Education and Its Impacts on Academic
Outcomes: Evidence from Korea. Working Paper.
Cohn, E., and Geske, T. (1986). Benet-cost analysis of investment in higher education.
chapter Values in con
ict: Funding priorities for higher education, pages pp. 183 { 215.
Ballinger, Cambridge.
Cohn, E., and Geske, T. (1990). Economics of Education. Mason: South Western College
Publishing, 3rd edition.
Cornwell, C., and Rupert, P. (1988). Ecient Estimation with Panel Data: An Empirical
Comparison of Instrumental Variables Estimators. Journal of Applied Econometrics,
3(2): 149{155.
Crawford, I., Laisney, F., and Preston, I. (2003). Estimation of household demand
systems with theoretically compatible Engel curves and unit value specications. Journal
of Econometrics, 114(2): 221{241.
Dang, H. A. (2007). The determinants and impact of private tutoring classes in Vietnam.
Economics of Education Review, 26(6): 683{698.
Dang, H.-A., and 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: Re
ections of
inequality in Japan, South Korea, and Cambodia. Asia Pacic Education Review, 11(1):
14{24.
130
de Castro, B. V., and de Guzman, A. B. (2014). From scratch to notch understanding
private tutoring metamorphosis in the Philippines from the perspectives of cram school
and formal school administrators. Education and Urban Society, 46(3): 287{311.
Deaton, A. (1987). Estimation of own- and cross-price elasticities from household survey
data. Journal of Econometrics, 36(1-2): 7{30.
Deaton, A. (1988). Quality, quantity, and spatial variation of price. The American
Economic Review, pages 418{430.
Deaton, A. (1990). Price elasticities from survey data: extensions and Indonesian results.
Journal of Econometrics, 44(3): 281{309.
Deaton, A. (1997). The analysis of household surveys: a microeconometric approach to
development policy. World Bank Publications.
Deaton, A., Friedman, J., and Alatas, V. (2004). Purchasing power parity exchange
rates from household survey data: India and Indonesia. Princeton Research Program in
Development Studies Working Paper.
Deaton, A., Grimard, F., and Mundial, B. (1992). Demand analysis and tax reform in
Pakistan. 85. World Bank.
Deaton, A., and Parikh, a. S. S., K (1994). Food Demand Pattern and Pricing Policy
in Maharashtra: An Analysis Using Household Level Survey Data. Sarvekshana, 17(4):
11{34.
Domingue, B. W., and Briggs, D. C. (2009). Using linear regression and propensity
score matching to estimate the eect of coaching on the SAT. Multiple Linear Regression
Viewpoints, 35(1): 12{29.
Friedman, J., and Levinsohn, J. (2002). The distributional impacts of Indonesia's -
nancial crisis on household welfare: A "rapid response" methodology. The World Bank
Economic Review, 16(3): 397{423.
Gibson, J., and Rozelle, S. (2005). Prices and unit values in poverty measurement and
tax reform analysis. The World Bank Economic Review.
Gurun, A., and Millimet, D. L. (2008). Does private tutoring payo? IZA Discussion
paper, (3637).
Hanushek, E. A. (1979). Conceptual and empirical issues in the estimation of educational
production functions. Journal of human Resources, pages 351{388.
Hanushek, E. A., Kain, J. F., Markman, J. M., and Rivkin, S. G. (2003). Does peer
ability aect student achievement? Journal of applied econometrics, 18(5): 527{544.
131
Hanushek, E. A., Kain, J. F., Rivkin, S. G., and Branch, G. F. (2007). Charter school
quality and parental decision making with school choice. Journal of Public Economics,
91(5): 823{848.
Hartmann, S. (2008). The Informal Market of Education in Egypt: private tutoring and
its implication. Mainez: Institutfur Ethnologie and Afrikastudien, Working Paper No.99.
Hausman, J. A., and Taylor, W. E. (1981). Panel data and unobservable individual
eects. Journal of Econometrics, 16(1): 155{155.
Hsiao, C. (2003). Analysis of panel data. Cambridge university press, second edition
edition.
Jung, J. H., and Lee, K. H. (2010). The determinants of private tutoring participation
and attendant expenditures in Korea. Asia Pacic Education Review, 11(2): 159{168.
Kenayathulla, H. B. (2013). Household expenditures on private tutoring: emerging
evidence from Malaysia. Asia Pacic Education Review, 14(4): 629{644.
Kim, D. I. (2008). The Eects of Children in School on Married Women's Labor Supply.
Korean Journal of Labor Economics, 31(2): 73{102.
Kim, J.-H. (2007). The determinants of demand for private tutoring in South Korea.
Working paper.
Kim, J.-H., and Park, D. (2010). The determinants of demand for private tutoring in
South Korea. Asia Pacic Education Review, 11(3): 411{421.
Kim, K., Kim, G., Kim, S., Kim, J., Kim, J., and Park, J. (2010). OECD Review on
Evaluation and Assessment Frameworks for Improving School Outcomes: Country Back-
ground Report for Korea. Discussion paper, Korean Educational Development Institute.
Kim, K. K. (2010). Sixty years of Korean education, chapter Educational equality, pages
285{325. Seoul: Seoul National University Press.
Kim, S., and Lee, J. (2008). Private Tutoring and the Demand for Education in South
Korea. Unpublished manuscript.
Kim, S., and Lee, J.-H. (2010). Private Tutoring and Demand for Education in South
Korea. Economic Development and Cultural Change, 58(2): 259{296.
Kim, Y., and Yang, S. (2011). The Prediction for The Trend and the Size of Private
Tutoring Expenditure. Position Paper 8(21), Korean Education Dvelopment Institute.
[In Korean].
Kwok, P. L. Y. (2010). Demand intensity, market parameters and policy responses
towards demand and supply of private supplementary tutoring in China. Asia Pacic
Education Review, 11(1): 49{58.
132
Lee, J.-T., Kim, Y.-B., and Yoon, C.-H. (2004). The Eects of Pre-Class Tutoring
on Student Achievement: Challenges and Implications for Public Education in Korea.
KEDI Journal of Educational Policy, 1: 25{42.
Liu, J. (2012). Does cram schooling matter? Who goes to cram schools? Evidence from
Taiwan. International Journal of Educational Development, 32(1): 46{52.
McKelvey, C. (2011). Price, unit value, and quality demanded. Journal of Development
Economics, 95(2): 157{169.
Moon, A. (2015). The eect of private tutoring on academic achievement: Evidence
from South Korea. Working paper.
Myles, G. D. (1995). Public economics. Cambridge university press.
Nath, S. R. (2008). Private supplementary tutoring among primary students in
Bangladesh. Educational Studies, 34(1): 55{72.
Ono, H. (2007). Does examination hell pay o ? A cost - benet analysis of "ronin" and
college education in Japan. Economics of Education Review, 26(3): 271{284.
Park, C. (2011). A Study on the Determinants of Parental Satisfaction with Public
Education and Demand for Prvate Tutoring for the Child. Journal of Korean Economic
Analysis, 17(1): 73{132.
Psacharopoulos, G., and Papakonstantinou, G. (2005). The real university cost in a free
higher education country. Economics of Education Review, 24(1): 103 { 108.
Ramsey, F. P. (1927). A Contribution to the Theory of Taxation. The Economic Journal,
pages 47{61.
Rivkin, S. G., Hanushek, E. A., and Kain, J. F. (2005). Teachers, schools, and academic
achievement. Econometrica, 73(2): 417{458.
Ryu, D., and Kang, C. (2013). Do Private Tutoring Expenditures Raise Academic
Performance? Evidence from Middle School Students in South Korea. Asian Economic
Journal, 27(1): 59{83.
Sargan, J. D. (1958). The Estimation of Economic Relationships using Instrumental
Variables. Econometrica, pages 393{415.
Sullivan, P. (2010). As Private Tutoring Booms, Parents Look at the Returns. New York
Times.
Suryadarma, D., Suryahadi, A., Sumarto, S., and Rogers, F. H. (2006). Improving
Student Performance in Public Primary Schools in Developing Countries: Evidence from
Indonesia. Education Economics, 14(4): 401{429.
133
Tansel, A., and Bircan, F. (2006). Demand for education in Turkey: A tobit analysis of
private tutoring expenditures. Economics of Education Review, 25(3): 303{313.
Tansel, A., and Bircan, F. (2007). Private Supplementaru Tutoring in turkey: recent
evidence on its various aspects. In The IIEP Policy Forum. Paris: IIEP-UNESCO.
Taylor, D. (2007). Making Good Progress: the use of one to one tuition in improving
rates of pupil progression. In The IIEP Policy Forum. Paris: IIEP-UNESCO.
Thapa, A. (2011). Does Private School Competition Improve Public School Performance?
The Case of Nepal. Dissertations, Ph.D., Columbia University.
Willis, R. J., and Rosen, S. (1979). Education and Self-Selection. Journal of Political
Economy, 87(5): pp. 7{36.
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.
134
Appendix
Table A1.1: Cause of deletion in re-sampling process: the number of missing values
Cause of deletion 2005 2006 2005 or 2006
Missing values of time invariant
Father's education 201 402
Mother's education 202 404
Age of HH 45 90
baseline score for Eng 20 40
baseline score for math 71 142
baseline score for Kor 70 140
Missing values of time varying
Income 377 433 700
Educational expenditure 828 686 1236
House 1 1 1
Marriage status 51 51 51
z-score for Eng. 18 139 157
z-score for math 12 54 66
z-score for Kor. 30 150 180
No live with parents 412 409 525
Missing values even though participated in tutoring
Hours spent in Eng. small-group 129 136 255
Expenditure in Eng. small-group 154 149 293
Hours spent in math small-group 118 151 261
Expenditure in math small-group 141 157 290
Hours spent in Kor. small-group 90 98 182
Expenditure in Kor. small-group 101 103 198
Hours spent in Eng. cram school 419 397 768
Expenditure in Eng. cram school 586 478 970
Hours spent in math cram school 414 406 773
Expenditure in math cram school 577 481 967
Hours spent in Kor. cram school 366 376 698
Expenditure in Kor. cram school 508 469 895
Note: This table shows the process to re-sampling in order to drop the missing values and
make the nal sample this paper makes use of. The rst two columns indicate the number
of missing values of variables by year. And the last column shows the number of deletion of
observations from pooled samples because of missing values over years. If one observation
has missing value at least in one year, that observation is deleted. Students who do not live
with their parents are dropped. Since this study requires both hours spent and expenditure,
the observations that either hours or spending are not reported, even if students attended
tutoring, are also deleted.
135
Table A1.2: Attrition of the KELS
2006 2007 2006-2007
Baseline Mean Mean di. (abs.t) Mean Mean di. (abs.t) Mean(e) di. (abs.t)
Characteristics (a) (b) (a)-(b) (c) (d) (c)-(d) (b)+(d) (c)-(e)
GENDER 0.52 0.57 -0.05* (2.28) 0.52 0.52 0.00 (0.04) 0.55 -0.03 (1.70)
ln(INC) 5.55 5.56 -0.01 (0.21) 5.55 5.60 -0.05 (1.02) 5.58 -0.03 (0.76)
Father's age 44.06 43.77 0.29 (1.39) 44.05 44.23 -0.18 (0.65) 43.94 0.11 (0.64)
Mother's age 40.97 40.72 0.24 (1.18) 40.97 40.92 0.05 (0.18) 40.80 0.17 (1.01)
Number of siblings 1.18 1.31 -0.13** (3.31) 1.18 1.09 0.09* (2.57) 1.22 -0.04 (1.37)
Father's edu - Middle 0.10 0.09 0.01 (0.60) 0.09 0.11 -0.01 (0.79) 0.10 -0.00 (0.10)
High school 0.48 0.42 0.05* (2.46) 0.48 0.44 0.04 (1.41) 0.43 0.05** (2.79)
2-years college 0.12 0.10 0.02 (1.52) 0.12 0.10 0.02 (1.25) 0.10 0.02 (1.96)
4-years university 0.31 0.39 -0.08*** (3.80) 0.30 0.35 -0.05 (1.75) 0.38 -0.07*** (4.05)
Mother's edu - Middle 0.10 0.08 0.01 (1.07) 0.10 0.11 -0.02 (0.91) 0.10 0.00 (0.14)
High school 0.64 0.58 0.05* (2.36) 0.64 0.61 0.03 (0.97) 0.60 0.04* (2.43)
2-years college 0.10 0.08 0.02* (2.06) 0.10 0.07 0.03* (2.06) 0.07 0.03** (2.86)
4-years university 0.16 0.26 -0.09*** (4.66) 0.16 0.20 -0.04 (1.81) 0.24 -0.07*** (4.79)
Mother's employment 0.51 0.47 0.04 (1.81) 0.51 0.50 0.00 (0.10) 0.48 0.03 (1.45)
Location- Seoul 0.18 0.20 -0.02 (1.02) 0.18 0.22 -0.05* (2.07) 0.21 -0.03* (2.15)
6 Metropolitan cities 0.28 0.29 -0.00 (0.12) 0.28 0.31 -0.03 (1.18) 0.30 -0.01 (0.87)
Medium and small cities 0.44 0.44 -0.00 (0.07) 0.44 0.40 0.04 (1.50) 0.43 0.02 (0.91)
Rural area 0.10 0.07 0.02 (1.86) 0.10 0.06 0.04** (2.78) 0.07 0.03** (3.13)
z-score of Eng -0.05 -0.15 0.10* (2.24) -0.05 -0.11 0.06 (1.08) -0.14 0.09* (2.41)
z-score of math -0.04 -0.18 0.14** (2.95) -0.04 -0.13 0.09 (1.57) -0.16 0.12** (3.27)
z-score of Kor. -0.04 -0.23 0.19*** (4.09) -0.03 -0.18 0.15* (2.55) -0.21 0.18*** (4.78)
Participants in Eng. Tutoring 0.73 0.71 0.02 (0.88) 0.73 0.70 0.03 (1.04) 0.71 0.02 (1.35)
math tutoring 0.72 0.69 0.03 (1.50) 0.72 0.73 -0.00 (0.06) 0.71 0.02 (1.11)
Korean tutoring 0.55 0.50 0.05* (2.33) 0.55 0.57 -0.02 (0.70) 0.53 0.02 (1.31)
n 5943 559 5594 349 908
Eng. tutoring
Budget share 0.37 0.38 -0.01 (0.58) 0.37 0.35 0.02 (0.94) 0.37 0 (0.19)
Expenditure 11.04 12.81 -1.78** (2.81) 10.98 11.67 -0.69 (1.14) 12.24 -1.27** (2.77)
Hours 4.2 4.22 -0.02 (0.14) 4.2 4.16 0.04 (0.28) 4.19 0.01 (0.09)
Unit value 0.85 0.95 -0.09 (1.78) 0.85 0.88 -0.03 (0.55) 0.91 -0.06 (1.62)
n
+
3169 283 2998 171 454
Math tutoring
Budget share 0.33 0.35 -0.02 (0.76) 0.33 0.34 -0.01 (0.59) 0.34 -0.02 (0.95)
Expenditure 10.15 11.90 -1.75* (2.23) 10.08 11.28 -1.19 (1.58) 11.65 -1.57** (2.75)
Hours 4.23 4.39 -0.15 (0.89) 4.24 4.07 0.18 (1.06) 4.26 -0.02 (0.15)
Unit value 0.74 0.81 -0.07 (1.32) 0.74 0.83 -0.09 (1.64) 0.82 -0.08* (2.05)
n
+
3126 272 2949 177 449
Kor. tutoring
Budget share 0.26 0.26 0.00 (0.04) 0.26 0.25 0.01 (0.50) 0.26 0.01 (0.28)
Expenditure 7.16 7.33 -0.17 (0.35) 7.11 8.01 -0.90 (1.29) 7.62 -0.51 (1.21)
Hours 3.48 3.45 0.02 (0.11) 3.47 3.59 -0.12 (0.47) 3.51 -0.04 (0.23)
Unit value 0.63 0.71 -0.07 (1.34) 0.63 0.68 -0.05 (0.68) 0.69 -0.06 (1.43)
n
+
2213 176 2086 127 303
Note: (a) actual response at 2006, (b) attrition from 2005 to 2006, (c) actual response at 2007, (d) attrition from
2006 to 2007, (e) attrition from 2005 to 2007.: The last three panels describe average measures of private tutoring
only considering participants in tutoring.: Budget share= expenditure on tutoring / total educational expenditure,
Expenditure = monthly expenditure on private tutoring, Hours = weekly hours spent for tutoring,
unit value = expenditure / (hours * 4).: n
+
implies the number of observations who make a purchase on
certain type of tutoring.
136
Table A1.3: Quality elasticity with respect to total educational expenditure: the rst
stage results of unit values
English Math Korean
Small Cram Small Cram Small Cram
ln (Edu. Expenditure per capita) 0.213*** 0.121*** 0.164** 0.125*** 0.163** 0.106***
(0.06) (0.03) (0.06) (0.02) (0.05) (0.03)
Gender (boys=1) -0.077** -0.027 -0.004 -0.029* -0.035 -0.000
(0.03) (0.01) (0.02) (0.01) (0.04) (0.01)
Number of HH. Members 0.007 0.022 0.032 -0.008 0.003 -0.013
(0.04) (0.02) (0.04) (0.02) (0.05) (0.02)
Ratio of older siblings 0.106 -0.070 0.037 0.002 0.255 0.076
(0.20) (0.07) (0.16) (0.08) (0.16) (0.09)
Ratio of younger siblings 0.145 -0.004 0.161 0.045 0.202 0.118
(0.20) (0.07) (0.15) (0.08) (0.17) (0.08)
Age of HH head -0.046 -0.011 -0.045 -0.009 0.038 -0.001
(0.05) (0.02) (0.05) (0.01) (0.09) (0.01)
Age of HH head squared 0.000 0.000 0.000 0.000 -0.001 -0.000
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
Father's education - high school 0.005 0.074* -0.071 0.063* -0.125 0.033
(0.06) (0.03) (0.06) (0.03) (0.07) (0.03)
2-year college 0.006 0.065 -0.145* 0.073** -0.158 0.026
(0.07) (0.03) (0.06) (0.03) (0.08) (0.03)
4-year university 0.008 0.069 -0.123 0.061* -0.139 0.037
(0.07) (0.04) (0.06) (0.03) (0.08) (0.03)
Mother's education - high school 0.005 -0.013 0.084 0.008 0.039 -0.021
(0.08) (0.03) (0.07) (0.03) (0.05) (0.03)
2-year college 0.040 -0.019 0.147* 0.015 0.056 -0.031
(0.09) (0.04) (0.07) (0.03) (0.08) (0.04)
4-year university 0.045 0.030 0.178* 0.077* 0.068 0.022
(0.08) (0.04) (0.07) (0.04) (0.07) (0.04)
Mother's employment -0.004 -0.010 -0.022 -0.007 -0.083* -0.014
(0.02) (0.01) (0.02) (0.01) (0.03) (0.01)
Home owned 0.019 0.028* 0.010 0.016 -0.015 0.027
(0.02) (0.01) (0.02) (0.01) (0.04) (0.01)
z-score of Eng. -0.014 0.004 0.008 -0.008 -0.034 -0.014
(0.02) (0.01) (0.02) (0.01) (0.03) (0.01)
z-score of Math 0.008 0.008 -0.007 0.008 0.011 0.010
(0.02) (0.01) (0.02) (0.01) (0.02) (0.01)
z-score of Kor 0.005 -0.017 0.018 -0.017 -0.009 -0.014
(0.02) (0.01) (0.02) (0.01) (0.02) (0.01)
Observations 779 3120 836 3133 223 2120
Note: Standard errors are shown in brackets. * p<0.05, ** p<0.01, *** p<0.001
137
Table A1.4: Quantity elasticity with respect to total educational expenditure: the rst
stage results of budget shares
English Math Korean
Small Cram Small Cram Small Cram
ln (Edu. Expenditure per capita) 0.041*** 0.027** 0.048*** 0.035*** 0.010*** 0.005
(0.01) (0.01) (0.01) (0.01) (0.00) (0.01)
Gender (boys=1) -0.003 0.002 -0.010 0.008 0.002 0.016**
(0.00) (0.01) (0.01) (0.01) (0.00) (0.01)
Number of HH. Members 0.008 -0.001 0.008 -0.006 -0.000 -0.001
(0.01) (0.01) (0.01) (0.01) (0.00) (0.01)
Ratio of older siblings 0.018 0.011 0.038* 0.012 0.005 0.030
(0.02) (0.03) (0.02) (0.03) (0.01) (0.02)
Ratio of younger siblings 0.039 0.083** 0.063*** 0.072* 0.005 0.068***
(0.02) (0.03) (0.02) (0.03) (0.00) (0.02)
Age of HH head -0.003 -0.017* -0.004 -0.016* -0.003 -0.008
(0.00) (0.01) (0.00) (0.01) (0.00) (0.01)
Age of HH head squared 0.000 0.000 0.000 0.000 0.000 0.000
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
Father's education - high school -0.001 -0.001 -0.013 -0.002 -0.006 0.001
(0.01) (0.01) (0.01) (0.01) (0.00) (0.01)
2-year college 0.009 0.005 -0.008 0.012 -0.004 -0.003
(0.01) (0.02) (0.01) (0.02) (0.00) (0.01)
4-year university 0.000 -0.010 -0.010 -0.012 -0.005 -0.007
(0.01) (0.01) (0.01) (0.01) (0.00) (0.01)
Mother's education - high school 0.005 0.017 0.013* 0.019 0.000 0.003
(0.01) (0.01) (0.01) (0.01) (0.00) (0.01)
2-year college 0.019* 0.016 0.021* 0.020 0.001 0.002
(0.01) (0.02) (0.01) (0.02) (0.00) (0.01)
4-year university 0.021* 0.033* 0.030** 0.041** 0.002 -0.004
(0.01) (0.02) (0.01) (0.02) (0.00) (0.01)
Mother's employment 0.000 -0.003 -0.004 0.000 0.001 0.008*
(0.00) (0.01) (0.00) (0.01) (0.00) (0.00)
Home owned 0.006 0.019** 0.014*** 0.009 0.002 0.004
(0.00) (0.01) (0.00) (0.01) (0.00) (0.00)
z-score of Eng. 0.005 0.021*** 0.007* 0.011* -0.003* -0.002
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
z-score of Math -0.000 0.024*** -0.007* 0.026*** -0.001 0.008**
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
z-score of Kor -0.007* -0.007 -0.005 -0.007 -0.002 0.000
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
Observations 5820 5820 5820 5820 5820 5820
Note: Standard errors are shown in brackets. * p<0.05, ** p<0.01, *** p<0.001
138
Table A2.1: Comparison between eects of private tutoring hours and self-study hours: Boys
Dep. Variable English Z-score
= 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(c) Weekly hours for private tutoring
Hours for private tutoring 0.030*** 0.029*** 0.009* -0.052 0.002 0.004 0.009 0.516 0.031**
(4.42) (4.84) (1.93) (-0.50) (0.41) (0.85) (0.91) (1.29) (2.17)
Hours for self-study 0.072*** 0.067*** 0.010 0.365** 0.004 0.005 0.015 1.451* 0.047*
(7.73) (8.36) (1.36) (2.31) (0.57) (0.58) (0.88) (1.67) (1.86)
Joint Test 12.10 13.69 0.03 3.41 0.08 0.01 0.11 1.31 0.31
(p-value) 0.00 0.00 0.87 0.07 0.78 0.93 0.74 0.25 0.58
Mean of Dependant Variable 0.03 0.04 0.03
Observation 3414 2276 3414
mathematics Z-score
Hours for private tutoring 0.030*** 0.030*** 0.008* 0.220** 0.008 0.009 0.014 0.210 -0.017
(4.36) (5.18) (1.74) (2.12) (1.60) (1.50) (1.31) (1.04) (-0.61)
Hours for self-study 0.064*** 0.060*** 0.016** -0.307* 0.009 0.007 0.019 0.081 0.056*
(6.68) (7.50) (2.15) (-1.95) (1.27) (0.77) (1.00) (0.15) (1.84)
Joint Test 7.86 8.38 0.69 5.29 0.01 0.01 0.06 0.07 4.16
(p-value) 0.01 0.00 0.41 0.02 0.91 0.93 0.81 0.79 0.04
Mean of Dependant Variable 0.15 0.03 0.15
Observation 3414 2276 3414
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
139
Table A2.2: Comparison between eects of private tutoring hours and self-study hours: Girls
Dep. Variable English Z-score
= 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(c) Weekly hours for private tutoring
Hours for private tutoring 0.044*** 0.044*** 0.005 -0.090 0.003 0.004 0.010 0.265 0.019**
(7.91) (9.71) (1.28) (-0.19) (0.83) (1.03) (1.22) (1.62) (2.38)
Hours for self-study 0.084*** 0.073*** 0.008 -1.500 0.011 0.014* 0.017 -0.314 0.056***
(9.62) (10.25) (1.26) (-0.89) (1.47) (1.95) (1.06) (-0.91) (3.87)
Joint Test 13.81 10.41 0.21 1.08 0.79 1.36 0.19 3.20 4.49
(p-value) 0.00 0.00 0.65 0.30 0.38 0.24 0.66 0.07 0.03
Mean of Dependant Variable 0.33 0.07 0.33
Observation 3576 2384 3576
mathematics Z-score
Hours for private tutoring 0.044*** 0.042*** 0.013*** 0.082 0.018*** 0.014*** 0.038*** 0.506 0.041**
(7.00) (7.92) (2.95) (0.67) (3.79) (2.64) (3.94) (1.46) (2.57)
Hours for self-study 0.094*** 0.083*** 0.023*** 0.927 0.018** 0.016* 0.040** -0.396 0.129***
(9.47) (10.29) (2.98) (1.05) (2.31) (1.77) (2.10) (-0.44) (4.36)
Joint Test 17.00 15.44 1.19 0.78 0.00 0.01 0.00 0.58 8.28
(p-value) 0.00 0.00 0.28 0.38 0.97 0.91 0.95 0.45 0.00
Mean of Dependant Variable 0.14 0.02 0.14
Observation 3576 2384 3576
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
140
Table A2.3: Comparison between eects of private tutoring hours and self-study hours: Urban
Dep. Variable English Z-score
= 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(c) Weekly hours for private tutoring
Hours for private tutoring 0.041*** 0.039*** 0.012*** -0.076 0.008* 0.009* 0.008 -0.270 0.038*
(7.07) (7.61) (2.86) (-0.75) (1.69) (1.93) (0.78) (-1.24) (1.71)
Hours for self-study 0.056*** 0.050*** -0.000 -0.475** 0.005 0.010 0.025 -0.528 0.010
(6.36) (6.52) (-0.03) (-2.31) (0.52) (1.23) (1.38) (-0.87) (0.45)
Joint Test 1.77 1.13 2.28 2.90 0.09 0.01 0.82 0.23 0.86
(p-value) 0.19 0.29 0.13 0.09 0.77 0.90 0.37 0.63 0.35
Mean of Dependant Variable 0.32 0.01 0.32
Observation 3237 2158 3237
mathematics Z-score
Hours for private tutoring 0.040*** 0.041*** 0.016*** 0.068 0.017*** 0.016** 0.033*** 0.363* 0.054**
(5.15) (7.01) (3.14) (0.70) (2.79) (2.56) (2.89) (1.67) (2.54)
Hours for self-study 0.065*** 0.058*** 0.018** -0.214 0.010 0.011 0.027 -0.069 0.027
(8.69) (7.23) (2.31) (-1.09) (1.25) (1.11) (1.38) (-0.21) (0.85)
Joint Test 5.41 2.83 0.06 1.15 0.52 0.23 0.07 0.96 0.41
(p-value) 0.02 0.09 0.81 0.28 0.47 0.63 0.79 0.33 0.52
Mean of Dependant Variable 0.22 -0.03 0.22
Observation 3237 2158 3237
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
141
Table A2.4: Comparison between eects of private tutoring hours and self-study hours: Rural
Dep. Variable English Z-score
= 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(c) Weekly hours for private tutoring
Hours for private tutoring 0.035*** 0.033*** 0.003 0.481 0.001 0.001 0.014* 0.684*** 0.015
(5.11) (6.17) (0.82) (1.22) (0.26) (0.28) (1.75) (2.64) (1.48)
Hours for self-study 0.098*** 0.093*** 0.019*** 1.061 0.011 0.008 0.005 -0.367 0.043
(10.97) (12.41) (2.68) (1.33) (1.66) (1.07) (0.28) (-0.41) (1.29)
Joint Test 31.26 37.35 3.80 0.32 1.41 0.62 0.26 1.10 0.84
(p-value) 0.00 0.00 0.05 0.57 0.24 0.43 0.61 0.30 0.36
Mean of Dependant Variable 0.06 0.09 0.06
Observation 3753 2502 3753
mathematics Z-score
Hours for private tutoring 0.036*** 0.030*** 0.005 0.243*** 0.009** 0.005 0.018** 0.500* 0.001
(5.88) (5.78) (1.26) (3.37) (2.11) (1.09) (1.97) (1.77) (0.10)
Hours for self-study 0.091*** 0.086*** 0.021*** -0.220 0.018** 0.014 0.034* 0.133 0.059*
(8.65) (10.94) (2.92) (-0.78) (2.56) (1.63) (1.92) (0.22) (1.68)
Joint Test 20.27 30.29 3.16 2.13 1.08 0.73 0.57 0.28 3.27
(p-value) 0.00 0.00 0.08 0.14 0.30 0.39 0.45 0.60 0.07
Mean of Dependant Variable 0.09 0.07 0.09
Observation 3753 2502 3753
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
142
Table A2.5: Comparison between eects of private tutoring hours and self-study hours: Lower educated father
Dep. Variable English Z-score
= 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(c) Weekly hours for private tutoring
Hours for private tutoring 0.045*** 0.043*** 0.010** -0.033 0.005 0.005 0.015 0.763 0.025**
(6.51) (7.76) (2.18) (-0.34) (1.24) (1.14) (1.48) (1.13) (2.31)
Hours for self-study 0.094*** 0.084*** 0.013* -0.389 0.009 0.015* 0.017 1.564 0.041*
(9.35) (9.83) (1.73) (-0.96) (1.21) (1.69) (1.08) (1.04) (1.88)
Joint Test 13.69 14.24 0.16 0.78 0.17 0.80 0.02 0.34 0.58
(p-value) 0.00 0.00 0.69 0.38 0.68 0.37 0.89 0.56 0.45
Mean of Dependant Variable -0.13 0.07 -0.13
Observation 3630 2420 3630
mathematics Z-score
Hours for private tutoring 0.043*** 0.039*** 0.012** 0.027 0.016*** 0.014** 0.030*** 0.130 0.011
(6.05) (6.83) (2.50) (0.31) (2.99) (2.42) (3.05) (0.09) (0.53)
Hours for self-study 0.094*** 0.085*** 0.021*** 0.581 0.013 0.011 0.033* 6.418 0.043
(9.78) (9.61) (2.63) (1.55) (1.47) (1.07) (1.74) (0.28) (1.29)
Joint Test 15.92 16.14 0.93 1.72 0.06 0.07 0.02 0.07 0.61
(p-value) 0.00 0.00 0.34 0.19 0.81 0.79 0.88 0.79 0.44
Mean of Dependant Variable -0.12 0.01 -0.12
Observation 3630 2420 3630
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
143
Table A2.6: Comparison between eects of private tutoring hours and self-study hours: Higher educated father
Dep. Variable English Z-score
= 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(c) Weekly hours for private tutoring
Hours for private tutoring 0.029*** 0.031*** 0.005 0.419** 0.003 0.004 0.007 0.051 0.027*
(5.11) (6.16) (1.29) (1.97) (0.80) (1.03) (0.88) (0.10) (1.94)
Hours for self-study 0.066*** 0.059*** 0.007 0.496 0.004 0.006 0.014 -1.303 0.023
(8.68) (8.61) (1.07) (1.17) (0.57) (0.84) (0.81) (-0.76) (0.95)
Joint Test 15.98 10.21 0.10 0.04 0.02 0.06 0.14 1.04 0.03
(p-value) 0.00 0.00 0.75 0.85 0.88 0.81 0.70 0.31 0.86
Mean of Dependant Variable 0.51 0.04 0.51
Observation 3360 2240 3360
mathematics Z-score
Hours for private tutoring 0.032*** 0.033*** 0.009** 0.274*** 0.011** 0.008 0.019* 0.457** 0.033**
(4.69) (6.24) (2.05) (2.60) (2.27) (1.40) (1.78) (1.99) (2.01)
Hours for self-study 0.064*** 0.062*** 0.017** -0.457** 0.012 0.012 0.025 0.273 0.076***
(7.51) (8.20) (2.41) (-2.22) (1.64) (1.27) (1.37) (0.40) (2.74)
Joint Test 7.51 8.68 0.83 8.17 0.04 0.13 0.08 0.06 1.86
(p-value) 0.01 0.00 0.36 0.00 0.85 0.72 0.78 0.80 0.17
Mean of Dependant Variable 0.43 0.03 0.43
Observation 3360 2240 3360
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
144
Table A2.7: Comparison between eects of private tutoring hours and self-study hours: Lower educated mother
Dep. Variable English Z-score
= 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(c) Weekly hours for private tutoring
Hours for private tutoring 0.041*** 0.040*** 0.009*** 0.184 0.005 0.006 0.012 0.878 0.034***
(7.02) (8.80) (2.59) (0.89) (1.31) (1.63) (1.43) (1.47) (3.38)
Hours for self-study 0.091*** 0.082*** 0.011* 1.157 0.010 0.011 0.017 1.542 0.046**
(10.63) (11.54) (1.70) (1.27) (1.48) (1.57) (1.19) (1.42) (2.20)
Joint Test 18.63 21.89 0.03 1.15 0.30 0.29 0.10 0.48 0.29
(p-value) 0.00 0.00 0.86 0.28 0.59 0.59 0.75 0.49 0.59
Mean of Dependant Variable -0.01 0.06 -0.01
Observation 4896 3264 4896
mathematics Z-score
Hours for private tutoring 0.039*** 0.037*** 0.012*** 0.175*** 0.014*** 0.012** 0.024*** 0.306 0.009
(6.54) (7.65) (2.86) (2.80) (3.14) (2.42) (2.79) (0.80) (0.54)
Hours for self-study 0.085*** 0.077*** 0.020*** -0.166 0.014** 0.013 0.034** 1.808 0.081***
(10.84) (10.57) (2.95) (-0.75) (2.02) (1.51) (2.07) (0.59) (3.42)
Joint Test 19.91 18.54 1.03 1.73 0.00 0.01 0.23 0.21 6.33
(p-value) 0.00 0.00 0.31 0.19 0.96 0.94 0.63 0.64 0.01
Mean of Dependant Variable -0.01 0.02 -0.01
Observation 4896 3264 4896
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
145
Table A2.8: Comparison between eects of private tutoring hours and self-study hours: Higher educated mother
Dep. Variable English Z-score
= 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(c) Weekly hours for private tutoring
Hours for private tutoring 0.028*** 0.028*** 0.002 0.103 0.000 -0.000 0.006 -0.003 0.015
(3.89) (4.28) (0.39) (0.98) (0.07) (-0.07) (0.62) (-0.01) (1.61)
Hours for self-study 0.058*** 0.051*** 0.008 -0.021 0.002 0.007 0.011 -0.810 0.033**
(5.74) (6.29) (0.91) (-0.15) (0.23) (0.70) (0.50) (-1.03) (2.11)
Joint Test 6.63 4.83 0.37 1.03 0.03 0.43 0.04 2.53 0.89
(p-value) 0.01 0.03 0.55 0.31 0.87 0.51 0.84 0.11 0.35
Mean of Dependant Variable 0.62 0.03 0.62
Observation 2094 1396 2094
mathematics Z-score
Hours for private tutoring 0.034*** 0.033*** 0.009 0.257* 0.010* 0.010 0.027** 0.447 0.033**
(3.91) (4.68) (1.60) (1.92) (1.74) (1.39) (2.14) (1.53) (2.18)
Hours for self-study 0.066*** 0.058*** 0.018** -0.543* 0.011 0.010 0.023 -0.060 0.067***
(6.02) (6.30) (2.07) (-1.81) (1.16) (0.91) (1.00) (-0.07) (2.84)
Joint Test 4.11 4.58 0.77 4.70 0.01 0.00 0.03 0.33 1.27
(p-value) 0.05 0.03 0.38 0.03 0.92 0.95 0.87 0.57 0.26
Mean of Dependant Variable 0.51 0.02 0.51
Observation 2094 1396 2094
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
146
Table A2.9: Comparison between eects of private tutoring hours and self-study hours: Bottom 50 % of percentile baseline
score
Dep. Variable English Z-score
= 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(c) Weekly hours for private tutoring
Hours for private tutoring 0.015*** 0.017*** 0.007 0.128 0.005 0.006 0.005 0.729 0.023
(3.02) (3.76) (1.42) (0.50) (0.93) (0.95) (0.44) (1.13) (1.63)
Hours for self-study 0.053*** 0.049*** 0.016* 0.931** 0.018* 0.016 -0.001 1.577 0.024
(6.35) (6.62) (1.76) (2.07) (1.85) (1.37) (-0.05) (1.24) (1.03)
Joint Test 13.80 12.98 0.76 1.77 1.45 0.58 0.06 0.50 0.00
(p-value) 0.00 0.00 0.38 0.18 0.23 0.45 0.81 0.48 0.96
Mean of Dependant Variable -0.60 0.18 -0.60
Observation 2985 1990 2985
mathematics Z-score
Hours for private tutoring 0.016*** 0.012*** 0.016*** 0.916 0.014** 0.012* 0.025** 0.150 0.001
(3.58) (2.68) (3.05) (1.29) (2.48) (1.92) (2.41) (0.63) (0.08)
Hours for self-study 0.053*** 0.045*** 0.015 1.328 0.012 0.007 -0.012 -1.068 0.028
(5.98) (5.71) (1.64) (0.97) (1.16) (0.62) (-0.54) (-0.54) (0.95)
Joint Test 13.97 12.60 0.01 0.24 0.01 0.11 2.12 0.41 0.52
(p-value) 0.00 0.00 0.94 0.62 0.91 0.74 0.15 0.52 0.47
Mean of Dependant Variable -0.60 0.18 -0.60
Observation 2898 1932 2898
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
147
Table A2.10: Comparison between eects of private tutoring hours and self-study hours: Top 50 % of percentile baseline
score
Dep. Variable English Z-score
= 0 = 1 No
POLS
District Student Student
POLS
District Student Student
AB
FE FE FE-IV FE FE FE-IV
(c) Weekly hours for private tutoring
Hours for private tutoring 0.028*** 0.028*** 0.007** -0.023 0.007** 0.007** 0.013* 0.356 0.016
(6.14) (6.95) (2.08) (-0.15) (2.25) (2.00) (1.78) (1.61) (1.31)
Hours for self-study 0.033*** 0.032*** 0.009 -0.791** 0.015** 0.015** 0.019 -0.051 0.018
(4.93) (5.78) (1.49) (-2.25) (2.27) (2.19) (1.28) (-0.09) (1.13)
Joint Test 0.36 0.22 0.04 4.26 0.98 0.98 0.12 0.91 0.02
(p-value) 0.55 0.64 0.83 0.04 0.33 0.32 0.73 0.34 0.89
Mean of Dependant Variable 0.77 -0.04 0.77
Observation 4005 2670 4005
mathematics Z-score
Hours for private tutoring 0.016*** 0.012*** 0.016*** 0.916 0.019*** 0.019*** 0.025*** 0.857 0.031**
(3.58) (2.68) (3.05) (1.29) (4.16) (3.77) (2.65) (0.71) (2.20)
Hours for self-study 0.053*** 0.045*** 0.015 1.328 0.024*** 0.023*** 0.036** -2.091 0.058***
(5.98) (5.71) (1.64) (0.97) (3.61) (2.86) (2.26) (-0.44) (2.67)
Joint Test 2.99 3.12 2.63 1.65 0.36 0.16 0.31 0.25 1.19
(p-value) 0.09 0.08 0.11 0.20 0.55 0.69 0.58 0.62 0.28
Mean of Dependant Variable 0.68 -0.09 0.68
Observation 4092 2728 4092
Robust t-statistics in parentheses, clustered at school district level. *, **, and *** indicate signicance at the 10, 5 and 1 percent levels.
148

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Essays on development and health economics

PDF

Essays on development and health economics: social media and education policy

PDF

Essays on the empirics of risk and time preferences in Indonesia

PDF

Essays on education and institutions in developing countries

PDF

Essays on education: from Indonesia to Los Angeles, CA

PDF

Essays on health economics

PDF

Essays on health insurance programs and policies

PDF

Three essays in education and urban economics

PDF

Essays in political economy and mechanism design

PDF

Three essays on cooperation, social interactions, and religion

PDF

Essays on labor and development economics

PDF

Essays on family and labor economics

PDF

Essays in health economics: evidence from Medicare

PDF

Essays in health economics and provider behavior

PDF

Essays on human capital accumulation -- health and education

PDF

Three essays on health & aging

PDF

Essays on economics of education

PDF

Essays on political economy of privatization

PDF

Essays on family planning policies

PDF

Essays in international economics

Asset Metadata

Creator Moon, Ahram (author)

Core Title Essays on economics of education and private tutoring

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Economics

Publication Date 07/11/2016

Defense Date 04/27/2016

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Academic Achievement,Deaton,education demand,GMM,OAI-PMH Harvest,price elasticity,private tutoring

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Strauss, John (committee chair), Nugent, Jeffrey (committee member), Painter, Gary (committee member)

Creator Email ahrammoo@usc.edu,ahrammoon@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-267599

Unique identifier UC11281223

Identifier etd-MoonAhram-4536.pdf (filename),usctheses-c40-267599 (legacy record id)

Legacy Identifier etd-MoonAhram-4536.pdf

Dmrecord 267599

Document Type Dissertation

Format application/pdf (imt)

Rights Moon, Ahram

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA