Upload
lamnguyet
View
221
Download
3
Embed Size (px)
Citation preview
The shadow cost of education :How does private tutoring affect students’ well-being?
Junhee Lee
Master ThesisMaster in Applied Research in Economics and Business, Universitat Autònoma de Barcelona
Supervisor: Yarine Fawaz
Abstract Private supplementary tutoring has been highly popular for decades in
South Korea. Despite much of past studies focussed on determinants of private tu-
toring and its effect on academic achievements, its effect on subjective well being has
not been studied. This paper estimates the effect of private tutoring on students’ well
being, using Korea Youth Panel Survey(KYPS) data of 6,293 middle and high school
students over 2003-2008. Composite well being indices were constructed from taking
into consideration multidimensional domains of well being. Hausman-taylor estima-
tion was applied to address the potential endogeneity of private tutoring variable. Em-
pirical results indicates that more hours spent on private tutoring has negative impact
on overall well being and on satisfaction with family relationship, self esteem, study
related stress, depressive feeling and self reported physical health.
1. Motivation and Background
This paper aims to estimate the causal effect of private tutoring on individual youth’s
well being. Private tutoring(also called shadow education), a form of a private supple-
mentary class provided by individuals or commercial companies, has been tradition-
ally prosperous in Asian countries such as China, South Korea, Japan, Taiwan Singa-
pore, India and Kazakhstan and some European countries such as Turkey, and is ex-
Email address: [email protected] (Junhee Lee)
1 MOTIVATION AND BACKGROUND
panding and intensifying in other countries (Bray and Lykins (2012), Bray et al. (2013),
Tansel and Bircan (2006) ). It has been pronounced and persistent in particular in South
Korea (hereinafter “Korea”). A nationwide government survey on private tutoring
showed that 68.8% of students in primary and secondary education received private
tutoring in 2013. Main reasons of receiving private tutoring of academic subjects were
to “makeup for classes(72.4%)”, followed by “study in advance(41.2%)” and “prepare
for higher school level(23.6%)” (Statistics-Korea (2014)). The main objective of private
tutoring can thus be seen as raising academic grades for higher education, ultimately
for university admission. The Korean students usually spend long hours after school
taking classes in Hagwon(private academy or cram schools). It has been prevalent
among students in particular in concentrated metropolitan area, as we can see from
the fact that Seoul Metropolitan Education Office has imposed 10:00 pm curfew on hag-
wons in 2008 with an aim to secure health and leisure time of students(Constitutional
Court ruling, 2008). The drivers of private tutoring from a social viewpoint can be
Confucian traditions which value academic performance a main route to a socioeco-
nomic advancement(Bray and Lykins (2012)). This cultural factor is still prevalent as
academic credentialism;
The Korean government had strong commitment to expansion of public education.
As a result of national campaign of building schools, primary school enrollment in-
creased rapidly, from 1.37 million in 1945 to 4.94 million in 1965(Kim and Lee (2010)).
In 1965 number of pupils per class in elementary school was 65.4, and student-teacher
for elementary school ratio was 62.4, which more than halved in 40 years, by 2005 the
figures were 31.8 and 25.1 respectively(KEDI (2005)).(in the case of middle school, the
statistics are 60.7 and 39.4 in 1965, 35.3 and 19.4 in 2005) Public expenditure on edu-
cation per student for the same period increased approximately 22 times in real term
(from 139,171 KRW to 3,094,045 KRW). Enrollment rate to secondary school marked
below 20% in 1945, but reached 90% for primary, middle and high school in 1964, 1979
and 1993 respectively(OECD (2012)).
2
1 MOTIVATION AND BACKGROUND
Private tutoring has been dominant and of a persistent concern since the liberation
from the Japanese colonial rule and the establishment of the first Republic in 1945. It
has constantly increased despite various efforts to curb it. The Ministry of Education
even described the history of the Korean education policy as “history of war against
private tutoring”(Kim (2008)). Government policies to reduce private tutoring were
major pledge in presidential elections, and exhaustive, ranging from outright prohibi-
tion to indirect strategy of improving public school quality. From 1950s to 1968, mid-
dle schools(school for students with age from 12 to 14) accepted new students based
on their own entrance exams. As graduating from prestigious universities rather than
individual’s ability has been decisive in obtaining higher social and economic status,
elementary school students (from 6 to 11 years old) were forced to study intensively
for entrance exams. Private tutoring has become was considered an effective means
to obtain high grades in the entrance exam. In the late 1960s, approximately 60% of
elementary school 4th to 5th grade students and 90% of 6th grade students were al-
ready receiving private tutoring in Seoul (Lee and Jang (2008)). Higher education,
including high school(from age 15 to 17) and universities also had clear ranks and hi-
erarchy then. Therefore entering prestigious middle school was a key that ensured
a way to high ranking high schools and universities. Public teachers of elementary
schools gave private tutoring classes after school, which earned them lucrative gains.
The Park Jung-hee government responded to booming private tutoring by abolishing
middle school entrance exam in 1968, which was hailed as a “July 15th liberation(from
the date it was announced)”. The students were assigned to middle schools by lottery
system. Although private tutoring at elementary school seemed to have decreased,
it was still booming in the middle school to enter prestigious high schools. The gov-
ernment consequently abolished high school entrance exam and enacted “high school
equalization policy” in 1974. This again could not abolish private tutoring, but shifted
demand for it to high school students. The new military dictatorship government fol-
lowed similar but more radical solution: it prohibited all types of private tutoring in
3
1 MOTIVATION AND BACKGROUND
1980. The punishment was credible. A mayor of the province of Cheju island was
dismissed in 1986 for giving his child English private tutoring. As a result, however,
private tutoring became more clandestine and its cost rose as a consequence of higher
risk.
Following government after democratization shifted from hard-line stance to grad-
ual liberalization of the tutoring. In 1989 one-on-one private tutoring provided by
university students was legalized, and students were allowed to take courses in pri-
vate academies(cram schools: Hagwon) in school vacations. The policies focussing
on restricting private tutoring faced a definite end when in 2000, the Constitutional
Court ruled the law prohibiting private tutoring unconstitutional, on the grounds that
it “restricts the learning children’s and juvenile’s right to free development of person-
ality, parents’ right to educate children, and the occupational freedom and the right
to pursue happiness of the person who wishes to provide extracurricular lessons.”
However, the Court ruling also pointed out that private tutoring is “a social disease”
and “the socio-pathological phenomena”, and “Over-heated competition for extracur-
ricular lessons caused several undesirable side-effects other than economic burdens
on parents. They are the deficiency in students’ creativity and self-initiated studying
abilities, the impoverishment of school education due to the overheated race outside
schools, the disadvantages and the feeling of relative deprivation suffered by those
parents and children who cannot obtain extracurricular lessons for economic reasons,
and the undesirable impact on the national economy due to wasteful investment in
terms of human and physical resources”. Regarding detrimental effects of private tu-
toring on youth’s well-being, the constitutional judges also acknowledged that exces-
sive private tutoring “disrupts their emotional state and sound physical growth”, and
“the pressure about school grades has forced them into unstable emotional states and
then to juvenile delinquency. The accumulated mental and physical fatigue damages
children’s health”. (Constitutional Court of Korea (2000)). Since the ruling, the policy
agenda shifted to reducing burden of expenditure on private tutoring. The new focus
4
1 MOTIVATION AND BACKGROUND
was to improve quality of public education to compete with and reduce the demand
for private tutoring and reduce the effectiveness of private tutoring on university en-
trance exam. Regarding improvement of public education, the government provided
extracurricular classes in public schools and televised courses for preparation of stan-
dardized national university exam(CSAT : College Scholastic Aptitude Test) through
public television channel EBS (Educational Broadcasting System). These were aimed
to complement private tutoring, in particular for students from poor families and in
disadvantaged regions. In order to reduce the effectiveness of private tutoring on uni-
versity entrance, the government guided the universities to reduce the weight of na-
tionwide entrance exam and increase the weight of academic grades at high school.
The universities also diversified the channels of entrance. They increased the propor-
tion of students entering through special admission from 9.7% in 1997 to 23.4% in 2001
and 37.4% in 2006(Kim (2008)), which admitted students on the basis of talent and
knowledge in certain subjects such as foreign language and mathematics rather than
exam scores. The grades of the standardized university entrance exam was also sim-
plified to reduce its influence in university admission.
Despite government’s various efforts curb private tutoring, its expenditure has in-
creased constantly. Kim et al. (2012) estimated that the average household expenditure
on private tutoring increased annually from 1990 to 2010 by 12.5% and 5.5%, respec-
tively in nominal and real term, using national household survey data. This contrasts
with the annual decrease of 0.3% of household expenditure on public(regular) edu-
cation for the same period. The household expenditure on private tutoring relative
to income also increased from 2% in 1990 to 5.2% in 2008, and 10.7% in 2010(OECD
(2012)). For the same period, expenditure on public education relative to income re-
mained between 1% to 2%(1.7% in 1990 and 1.2% in 2008). Private tutoring expenditure
also increased in terms of proportion in total household expenditure, from 3% in 1990
to 7.7% in 2010. It is also notable that the gap in private tutoring expenditure between
low and high income household increased over time. The ratio of the private tutor-
5
2 RELATED LITERATURE
ing expenditure between1st and 10th decile in 1990 was 4.7, which increased to 14.6 in
2010. (Kim et al. (2012))
Strikingly high portion of youth suffering from emotional disorder hints that exces-
sive private tutoring has detrimental effect on well-being. Out of increasing concern
on mental health of the youth, the Ministry of Education carried out a survey of emo-
tional and behavioral characteristics on the 97% of the entire youth from age 6 to 18
in 2012 (approximately 6.5 million). The results showed that over a million students
(16.3%) needed psychiatric counseling, 223,000 (4.5%) needed intensive treatment, and
1.5% (97,500) were under imminent danger, such as committing suicide(Ministry of
Education, Science and Technology (2013)). Evidence imply that this mental health
problem is attributable to excessive studying. In a nationwide survey of 9,500 Ko-
rean youths, 68.7% of middle and high school students reported to be under stress
due to study (NYPI (2009)). 38.% of the same students thought of committing suicide,
of which 44.1% was due to academic grades at school. 60.8% reported that they are
not happy, and 58.1% of them attributed their reason of being unhappy to stress of
studying(32.6%) and anxiety about their future(25.5%). This shows how grave is the
psychological burden due to excessive competition in education.
The rest of the paper is organized as following. Section 2 reviews the literature
related with private tutoring and its implication for this paper. Section 3 and 4 explains
the data and the empirical methods applied. Section 5 analyses the empirical results,
and section6 concludes.
2. Related Literature
As literature on subjective well being is relatively new, there has yet been made
empirical research on the effect of private tutoring on well being. Most of the lit-
erature on private tutoring has focused on the determinants of private tutoring(Kim
and Lee (2010)) and the effect of private tutoring on academic achievement(Briggs
(2001), Cheo and Quah (2005), Dang (2007), Lee et al. (2004), Stevenson and Baker
6
2 RELATED LITERATURE
(1992), Suryadarma et al. (2006), Tansel and Bircan (2006), Zhang (2013)Briggs (2001),
Cheo and Quah (2005), Dang (2007), Lee et al. (2004), Stevenson and Baker (1992),
Suryadarma et al. (2006), Tansel and Bircan (2006), Zhang (2013)). The literature on
latter effect worth noting for the aim of this paper, because academic achievement can
also have considerable effect on individual youth’s well being especially for countries
with high participation in private tutoring. The empirical findings of the effect of pri-
vate tutoring on academic achievement are mixed. Dang (2007) found that the expen-
diture on private tutoring has positive effect on academic ranking in Vietnam. Studies
for other countries also found positive effect of tutoring on academic achievement,
such as Stevenson and Baker (1992) for Japan, Taiwan (Liu, 2012) and Kenya (Buch-
mann, 2002). On the contrary, Lee et al. (2004), Cheo and Quah (2005) found negative
effects of private tutoring on academic grades for Korea and Singapore respectively.
Briggs (2001) estimated the effect of “coaching(short term course or counseling)” on
SAT and ACT, which showed mixed effect depending on the subjects. Zhang (2013)
also found that private tutoring among students in Jinan province had positive im-
pact on National College Entrance Examination only for students in urban area with
lower achievement or low quality of school. The major limitations of the literature of
the effect of the private tutoring is insufficient control for endogeneity of private tutor
variables. Most of the literature applied cross section or pooled OLS but they do not
control for endogeneity of private tutoring. Others employed instrumental variables
such as official hourly fee of tutoring in a district in Dang (2007), and the number of pri-
vate tutoring participants among 5 closest friends and the distance between the nearest
private tutoring agency and home in Zhang (2013). However, these instruments can
be potentially endogenous. The official fee of tutoring inDang (2007) is decided by the
government based on mean income of districts. the participation rate of friends on
private tutoring in Zhang (2013) is positively correlated with the participation of the
individual, but individual’s participation can be the cause of having more friends re-
ceiving private tutoring. This reverse causality cannot be disregarded because Zhang
7
3 DATA
(2013) only used single period cross section data. The distance between home and pri-
vate tutoring academies can also be endogenous because for instance, houses close to
private academies can be correlated to higher house prices.
3. Data
This paper uses Korea Youth Panel Survey (KYPS) dataset collected by Korea’s Na-
tional Youth Policy Institute. This dataset contains 2 subgroups of youths, consisting
of a group which enters the survey from elementary school 4th grade(9 years old, here-
inafter elementary group) and another group that enters at middle school 2nd grade(13
years old, hereinafter middle group). Each subgroup consists of classes of students ran-
domly selected from provinces nationwide excluding Jeju island, proportional to stu-
dent population. The sample size is 2844 and 3449 respectively for elementary school
grade and middle school grade group. Therefore each individual belong to a class
in the initial year, and the entire class was selected if an individual is in the sample.
The panel dataset comprises of 5 and 6 years for elementary and middle group respec-
tively. Elementary group enters the data from 2004 to 2008, when they became middle
school 2nd grade, and the middle group entered from 2003 to 2008, when they became
university freshmen or high school graduates.
The advantage of this dataset is that it contains rich set of private tutoring and pub-
lic education variable and socioeconomic characteristics of individuals. For private
tutoring, it has variables for hours spent on private tutoring of each subjects including
academic, arts and physical subjects, expenditure on tutoring, and the types of tutor-
ing (which consists of 1-on-1 individual tutoring, group tutoring, tutoring at private
academies, correspondence papers, online courses, after school classes or overseas tu-
toring).
The dataset also contains rich set of variables related to subjective well being. These
questions can be grouped as various domains of well being, such as overall well being,
health, satisfaction with friends, family, stress due to study and exams, depressive
8
4 ESTIMATION METHOD
feelings, and self confidence(see below for complete list of these questions). Composite
well being indices are constructed in accordance with these groupings.
Other education variables that are considered important are hours spent on study-
ing by oneself, academic achievement in terms of self assessed academic grade(in Lik-
ert scale, from very poor to very good), ranking and self assessed grades of oneself in
class, school, and in nation(in national standardized practice exams or College Scholas-
tic Aptitude Test(CSAT, Suneung, is the test conducted at the end of high school, and
is the most important determinant in university admission process). All the rank vari-
ables are also available in percentile, so it allows to control for the difference in the size
of the class and quality of public schools.
Other variables that exhibit socioeconomic characteristics are income of household,
gender, years of education of parents, employment status of parents, binary dummy
variable for parents owning a house.
4. Estimation Method
For a academic achievement oriented society like Korea, it is reasonable to assume
that the subjective well being of youth depend much on academic grades. In partic-
ular, the effect of academic grade on subjective well being is expected to be more sig-
nificant if it is observed in terms of rank in cohorts, who are potential competitors in
university entrance, because if probability of entering prestigious universities is what
matters, only the rank will affect this probability, not the improvement in the academic
achievement per se without change in rank. Furthermore, the more hours spent on pri-
vate tutoring the greater would be the negative effect on well being. Therefore, as long
as more private tutoring raises rank in academic grade, it will have conflicting effect
on well being. In other words, the positive effect of private tutoring on well being from
rise in rank will be offset by increased psychological stress.
The empirical strategy is to estimate the effect of private tutoring on students’ well
being. Fixed effects regression and GLS instrumental variable regression are applied
to estimate the effect of private tutoring on well being. The regression equation can be
9
4 ESTIMATION METHOD
expressed as the following.
Wellbeingit = α + βRankit + δPrivateTutoringit + Controlsit + εit
Potential challenges with estimating the above equation is the following. Firstly the
variation in single questions such as overall life satisfaction(measured in scales from 1-
very unsatisfied to 5-very satisfied, phrased as “How satisfied are you with your life in
general? Circle the number most suitable for your level of satisfaction”) is small across
time and individuals. A limitation with the dataset is that the range of the possible re-
sponses are very narrow. Another possible reason could be that overall life satisfaction
consists of subdomains of life that can offset each other. Distinct domains of child well
being has been suggested in past literature such as Bronfenbrenner and Morris (1998)
and Fernandes et al. (2013) and their ecological model of human development. We
identified from the dataset the subdomains of well being : social relations(family and
friends), health, school, materiel well being, and personal characteristics. The effect of
private tutoring on overall life satisfaction can have such conflicting channels. For in-
stance, on one hand, more hours spent on private tutoring can reduce life satisfaction
by deteriorated physical health and more conflict with parents(given that parents force
their children to have tutoring). On the other hand, it can increase life satisfaction by
giving more friends(this would especially be plausible if many of his/her colleagues
receives private tutoring after school) and higher academic achievement. Therefore,
there is a need to desegregate the life satisfaction into different subdomains and esti-
mate the effect of private tutoring on different subdomains. I constructed the domains
as : physical health, satisfaction of interpersonal relationship(mainly with classmates
and friends), satisfaction of relationship with family members, study related stress, de-
pressive feelings, self confidence and aggregated overall well being. Composition of
variables in each domains are given below with descriptive statistics. i and t denote
individual i and year t(from 2003 to 2008). Each domain is created by summing up
relevant items. Each response is answered in Likert scale ranging from 1 to 5, which
mean “very untrue”, “somewhat untrue”, “neither true or untrue”, “somewhat true”,
10
4 ESTIMATION METHOD
and “very true”. All responses are recoded to correspond to the increase in well being
(increase means lower level of stress, higher satisfaction with friends, etc.). Each item
of responses is standardized before summed.
Table 1: Summary statistics for domains of well being : overall subjective well being
Variable Mean Std. Dev. Min. Max.composite WB 22.38 3.45 7.4 35overall life satisfaction 3.53 0.79 1 5
N 9083
Table 2: Summary statistics for domains of well being : family relationship(cronbach’s alpha:0.8914)
Variable Mean Std. Dev. Min. Max. NWB-family 3.95 0.41 1.98 5 9083My parents and I try to spend much time together 3.41 0.91 1 5 9083My parents always treat me with love and affection 3.75 0.85 1 5 9083My parents and I understand each other well 3.47 0.91 1 5 9083My parents and I candidly talk about everything 3.28 1.02 1 5 9083I frequently talk about my thoughtsand what I experience away from home with my parents
3.4 1.05 1 5 9083My parents and I have frequent conversations 3.55 0.93 1 5 9083When I go out, parents usually know where I am 3.58 0.97 1 5 9083When I go out, parents usually know whom I am with 3.49 0.98 1 5 9083When I go out, parents usually know what I am doing 3.38 0.98 1 5 9083When I go out, parents usually know when I return 3.32 0.98 1 5 9083I always get along well with brothers or sisters 3.51 0.98 1 5 9083I frequently see parents verbally abuse each other 4.04 1.04 1 5 9083I frequently see one of my parents beat the other one 4.41 0.88 1 5 9083I am often verbally abused by parents 4.37 0.89 1 5 9083I am often severely beaten by parents 4.4 0.9 1 5 9083I get stressed by disputes with my parents 3.31 1.01 1 5 9083I get stressed by my parents’ meddling into my life 3.38 1.04 1 5 9083I get stressed by communication difficulty with my parents 3.48 1.03 1 5 9083I get stressed by my parents’ concern on my sch. grades 3.12 1.05 1 5 9083I always get along well with brothers or sisters 3.51 0.98 1 5 9083
Another potential problem is the endogeneity of regressors. The above equation canalso be expressed as below.
Wellbeingit = βRankit + γ1PrivateTutoringi + γ2PrivateTutoring′it + δ1Controlsi +
δ2Controls′it + µi + υit
Controls i is time invariant and exogenous explanatory variables i.e. : gender and
parents’ years of education. Controls it is time variant exogenous variables, which
are income, income of sample households residing in the same district, age, school
11
4 ESTIMATION METHOD
Table 3: Summary statistics for domains of well being : friends relationship(cronbach’s alpha:0.7093)
Variable Mean Std. Dev. Min. Max. NWB-interpersonal relationship 4 0.61 0.27 5 9083I hope to maintain the close relationships for a long time 4.43 0.68 1 5 9083I am happy whenever I get together with them 4.43 0.66 1 5 9083I try to have the same thoughts and feelings with them 3.99 0.88 1 5 9083We can frankly talk about our troubles and worries 4.01 0.92 1 5 9083I get stressed by my friends’ teasing and ignoring me 4.09 0.88 1 5 9083I get stressed by lack of recognition from my frd. 4.05 0.89 1 5 9083I get stressed by sense of inferiority to my frd. 3.87 1 1 5 9083How often do you see your close friends in a week? 5.4 1.16 1 6 9083I am not in good terms with frd. at sch. 4.5 0.72 1 5 7770
Table 4: Summary statistics for domains of well being : school satisfaction(cronbach’s alpha:0.7021)
Variable Mean Std. Dev. Min. Max. NWB-school 3.1 0.70 0 5 9083I can talk about all my troubles and worriesto my teachers without reservation 2.52 1.04 1 5 9083Teachers treat me with love and affection 2.88 0.99 1 5 9083I hope to become a person just like my teacher 2.51 1.08 1 5 9083I am not interested in sch. work,and find it difficult to catch up 4.04 0.99 1 5 7771I am not in good terms with sch. teachers 4.37 0.82 1 5 7769I find it difficult to follow sch. rulesand regulations 4.29 0.93 1 5 7770
Table 5: Summary statistics for domains of well being : study related stress(cronbach’s alpha:0.7490)
Variable Mean Std. Dev. Min. Max. NWB-study related sress 2.28 0.99 0 5 9083I get stressed by poor sch. grades 2.98 1.11 1 5 9083I get stressed by home assignments or examinations 2.92 1.1 1 5 9083I get stressed because it is boring to study 2.98 1.07 1 5 9083WB str study (0-5 scale) 12.62 4.37 3 25 9083I get stressed by college preparation or job prospect 2.71 1.1 1 5 6577I am under great anxiety due to study 2.51 1.04 1 5 9083
Table 6: Summary statistics for domains of well being : self confidence(cronbach’s alpha:0.8126)
Variable Mean Std. Dev. Min. Max.WB-self esteem 3.04 0.71 0.15 5I think that I have a good character 3.28 0.87 1 5I think that I am a competent person 3.24 0.87 1 5I think that I am a worthy person 3.49 0.85 1 5Sometimes I think that I am a worthless person 3.37 1.01 1 5Sometimes I think that I am a bad person 3.29 1.01 1 5I generally feel that I am a failure in life 3.58 0.98 1 5I have confidence in my own decision 3.53 0.81 1 5I believe that I can deal with my problems by myself 3.57 0.81 1 5I am taking full responsibility of my own life 3.54 0.83 1 5
N 9083
12
4 ESTIMATION METHOD
Table 7: Summary statistics for domains of well being : depression(cronbach’s alpha:0.8411)
Variable Mean Std. Dev. Min. Max.WB-depression 3.08 0.99 0 5I am not interested in anything 3.76 0.89 1 5I worry about everything 3.01 1.07 1 5Sometimes I feel extremely anxious withno apparent reason 3.3 1.13 1 5Sometimes I feel extremely lonely with no apparent reason 3.34 1.13 1 5Sometimes I feel extremely sad and gloomy withno apparent reason 3.37 1.14 1 5Sometimes I feel suicidal with no apparent reason 3.91 1.08 1 5
N 9083
Table 8: Summary statistics for domains of well being : health
Variable Mean Std. Dev. Min. Max.I am not in good health 4.14 0.96 1 5
N 9083
grade, parents’ occupation and employment status, whether living in urban or ru-
ral area, whether parents are biological parents, whether the household owns their
house, and proximity of private academies. Academic achievement and private tutor
variables are likely to be correlated with unobserved factors(µi) such as innate ability,
motivation or aspiration of individual and parents or household’s wealth. Therefore
GLS (random effects) estimator is expected to be biased and inconsistent estimator be-
cause GLS assumes no correlation between µi and explanatory variables. The within
estimator(fixed effects) can eliminate endogeneity bias but it also makes impossible to
estimate the parameter of time invariant variables(γ1and δ1). Hausman and Taylor es-
timator uses averages and deviation from the averages of the exogenous explanatory
variables(Controlsi and Controls′it) and within variation in the time variant endogenous
variables(PrivateTutoring′it and Rankit) in the model as instruments(Baltagi (2014)). It
would be more efficient estimator than within estimator given that the model is robust
to overidentification test.
The variable of interest, private tutoring, is defined as supplementary education
of academic subjects provided by private for-profit individuals or institutions. It is
measured in following three terms. Firstly, hours spent on private tutoring is used. In
13
4 ESTIMATION METHOD
this case, private tutoring refers to all types; one-on-one tutoring(which usually takes
place at individuals’ homes), private academies, self-study sheets and online courses.
Secondly, private tutoring is also measured in terms of the total expenditure, regardless
of the types and hours. This expenditure variable is divided by hours spent on tutoring
to create hourly cost of tutoring(ptcosthr), which captures the quality of tutoring, if
controlled for the difference of cost of living across the regions. This cost variable is
introduced to estimate the effect of the quality of tutoring on well being. It is expected
that tutoring with higher cost will have greater effect on well being because it has
greater impact on academic grades. Also we have a variable of participation rate of
5 closest friends in private tutoring. As seen in the table below, on average 9% of the
five closest friends are taking the tutoring. This implies that private tutoring can be an
important means to socialize with peers. Since our dataset is based on extracting the
whole class in the initial year, we can extract the mean hours of tutoring received by the
class in the initial year. This variable can help analyze the effect of the peer pressure on
taking tutoring. The year 2008 for middle school group of the panel dataset is excluded
from the analysis because individuals graduated from high school and thus stopped
receiving private tutoring.
The control variables can be categorized into three types. One type is individual’s
academic achievement. They are self reported rank of overall academic grade in pre-
vious semester in individual’s class and self reported academic achievement of five
academic subjects. The rank variable is available for middle school and high school
grades. Second type of variables is school related variables, which are school grade
dummies for middle and high school and class size. The other type of controls are so-
cioeconomic characteristics of individual and family, measured by : gender(1 if male),
age, monthly average income of household, father’s year of education and occupa-
tion(1 if professional or administrative managers), mother’s employment status(1 if
employed), parents’ characteristic with regard to the child(“family”, which takes value
1 if both parents are biological parents and 0 otherwise), and whether parents owns a
14
4 ESTIMATION METHOD
house(1 if owns a house, 0 otherwise).
Table 9: summary statistics: private tutoring variables
Variable Mean Std. Dev. Min. Max. Ntutor cost per hour(ptcosthr) 3.3 7.59 0 360 8979hours of tutoring(PTHR) 7.8 8.57 0 75 9083received PT before elem. school(PT1) 0.29 0.45 0 1 9083participation rate of 5 closest friends in PT (PTFR) 0.09 0.4 0 5 9083proximity to private academy (DIST) .0000959 .0000648 2.56e-06 .0004184
Table 10: summary statistics: school related variables
Variable Mean Std. Dev. Min. Max.percentile rank in class, middle school(RANK1) 28.2 34.79 0 99percentile rank in class, high school(RANK2) 32.18 34.92 0 99hrs of studying by oneself(SUDY) 10.68 11.01 0 87class size (CLASS) 35.19 4.57 9 72
N 9083
Table 11: summary statistics: socioeconomic characteristics
Variable Mean Std. Dev. Min. Max.first born(BORDER) 0.46 0.5 0 1Born before March(MONTH) 0.18 0.39 0 1MALE 0.52 0.5 0 1log household income(INC) 5.69 0.52 0 8.60years of education-father(FEDUC) 13.41 2.79 0 20father’s occupation(=1 if high skilled)(FJOB) 0.15 0.36 0 1mother employed(MWRK) 0.51 0.5 0 1both parents are biological parents(FAMILY) 0.97 0.17 0 1own house(HOUSE) 0.78 0.41 0 1
N 9083
15
5 EMPIRICAL RESULTS
5. Empirical Results
Table 12 below reports the estimates of the conventional GLS and within(FE) esti-
mators, on different domains of well being index. The effects of academic grades and
its rank in class(hereinafter “rank”) and hours spent on private tutoring and school
grade dummy are significant in both models, whereas father’s occupation is signif-
icant in GLS and owning a house and mother’s employment status are significant
only in FE. The last two rows in the table reports chi-square statistics of the two mod-
els of Hausman test. The value is 16.56 so the test rejects the null hypothesis that
H0 : E(uit|Xit) = 0 at any conventional significance level, therefore the parameters of
GLS model does not satisfy the minimum asymptotic variance, leaving it biased and
inconsistent. Hausman test also confirms the same results that for other specifications
in the preceding columns. As Baltagi (2014) noted, FE model assumes endogeneity of
all the regressors, and GLS model assumes that all regressors are exogenous, so the two
models imposes “all or nothing” choice. It is clear that unobserved individual’s abil-
ity and socioeconomic background are correlated with some of the regressors, namely
rank, wealth and private tutoring, but other regressors are exogenous, such as gender,
month of birth, birth order and proximity to private academies. Therefore FE model is
excessively restrictive and inappropriate for our model. Another viable alternative to
GLS is Hausman Taylor specification.
The tables 13 to 15 shows the results of the Hausman-Taylor estimation. Column
(1) and (2) are estimations for a single dependent variable for overall life satisfaction.
(3) and (4) are regressions of Hausman Taylor specification(HT) and FE respectively
for composite index of well being. The following columns reports the estimations of
HT and FE for each domains : family and interpersonal relationship, school life, study
related stress, self esteem, depressive feelings and physical health. The last two rows
of each table shows Hausman test of overidentification over coefficients of HT and FE.
It fails to reject the null hypothesis that our regressors do not satisfy exogeneity, so
16
5 EMPIRICAL RESULTS
Table 12: Effects of private tutoring on well being : GLS and FE Estimation(1)
(1)GLS (2)FE (3)GLS (4)FE (5)GLS (6)FEWB_all WB_all WB_composite WB_composite WB_fam WB_fam
PTHR -0.00322*** -0.00361** -0.0164*** -0.0139*** -0.00104** -0.00129**(-2.71) (-2.48) (-4.17) (-3.07) (-2.25) (-2.45)
PT1 -0.0733*** . -0.300*** . -0.0241* .(-2.59) . (-2.72) . (-1.81) .
STUDY 0.00330*** 0.00185* 0.0110*** 0.00871*** 0.00151*** 0.000807**(3.84) (1.85) (3.89) (2.81) (4.57) (2.24)
RANK1 0.00566*** 0.00485*** 0.0256*** 0.0222*** 0.00268*** 0.00172***(10.16) (6.15) (13.51) (9.07) (11.98) (6.06)
RANK2 0.00409*** 0.00326*** 0.0171*** 0.0150*** 0.00171*** 0.00110***(7.09) (4.46) (8.86) (6.61) (7.54) (4.16)
PTHRM -0.00366 . 0.0418*** . 0.00403** .(-1.00) . (2.92) . (2.32) .
PTFR -0.0190 -0.0178 0.0168 0.00642 -0.00585 -0.00912(-0.84) (-0.69) (0.23) (0.08) (-0.68) (-0.97)
CLASS -0.00465** -0.00465 -0.00978 -0.0226** -0.00127 -0.00279**(-1.97) (-1.46) (-1.22) (-2.29) (-1.35) (-2.44)
BORDER -0.00697 . 0.0489 . 0.00408 .(-0.27) . (0.49) . (0.34) .
MONTH -0.000258 . 0.0969 . 0.00408 .(-0.01) . (0.75) . (0.26) .
HIGH -0.0673 -0.0198 0.780*** 0.378** 0.118*** 0.0735***(-1.36) (-0.32) (4.76) (1.97) (6.12) (3.31)
MALE 0.231*** . 0.318*** . -0.0860*** .(9.07) . (3.21) . (-7.16) .
INC 0.0539*** 0.000750 0.0413 0.0461 -0.00227 -0.0182*(2.60) (0.03) (0.59) (0.56) (-0.28) (-1.89)
EDU 0.0169*** 0.0138 -0.0155 1.061 0.00671*** 0.0968(3.32) (0.05) (-0.79) (1.25) (2.82) (0.98)
FJOB -0.0126 -0.138 -0.00294 -0.996** 0.0232 -0.0486(-0.35) (-0.99) (-0.02) (-2.31) (1.39) (-0.97)
MWRK -0.0451* 0.0355 -0.270*** 0.0185 -0.0195* 0.0269(-1.87) (0.51) (-2.97) (0.09) (-1.78) (1.08)
FAM 0.102* 0.0552 0.301 0.0857 0.0731*** -0.00432(1.72) (0.68) (1.49) (0.34) (3.08) (-0.15)
HOUSE 0.0836*** 0.101*** 0.168** 0.173* 0.0251** 0.0270**(3.37) (2.99) (1.99) (1.65) (2.53) (2.22)
DIST 114.2 292.1 208.0 6544.4*** -120.3 -43.09(0.59) (0.41) (0.29) (2.98) (-1.37) (-0.17)
N 9083 9083 9083 9083 9083 9083R2 0.0607 0.014 0.0536 0.026 0.0786 0.021χ32(p > χ32) 0.92(0.8217) 16.56(0.0009) 3.18(0.3644)t statistics in parentheses* p < 0.10, ** p < 0.05, *** p < 0.01
17
5 EMPIRICAL RESULTS
Table 13: Effects of private tutoring on well being : GLS and FE Estimation(2)
(7)GLS (8)FE (9)GLS (10)FE (11)GLS (12)FEWB_per WB_per WB_sch WB_sch WB_std WB_std
PTHR 0.000956 0.00137 0.00114 0.000761 -0.00978*** -0.00464***(1.22) (1.44) (1.28) (0.68) (-7.82) (-3.03)
PT1 -0.0394** . -0.00162 . -0.0443 .(-2.05) . (-0.08) . (-1.51) .
STUDY 0.00173*** 0.00134** 0.00416*** 0.00247*** -0.00440*** -0.000348(3.04) (2.05) (6.40) (3.22) (-4.85) (-0.33)
RANK1 0.00249*** 0.00194*** 0.00406*** 0.00383*** 0.00593*** 0.0105***(6.73) (3.76) (9.81) (6.32) (10.14) (12.74)
RANK2 0.00105*** 0.000993** 0.00597*** 0.00446*** 0.00212*** 0.00357***(2.74) (2.08) (13.89) (7.93) (3.50) (4.66)
PTHRM -0.00168 . 0.00218 . -0.000851 .(-0.67) . (0.85) . (-0.22) .
PTFR -0.0162 -0.00625 0.0148 -0.00812 0.00866 0.0303(-1.09) (-0.37) (0.87) (-0.41) (0.37) (1.11)
CLASS -0.00111 -0.00314 -0.00137 -0.00123 -0.00904*** -0.00403(-0.71) (-1.51) (-0.79) (-0.50) (-3.65) (-1.21)
BORDER -0.0176 . 0.0494*** . -0.0170 .(-1.02) . (2.76) . (-0.64) .
MONTH -0.0149 . -0.00898 . -0.00519 .(-0.66) . (-0.39) . (-0.15) .
HIGH 0.0970*** 0.103** -0.104*** -0.0267 -0.0479 0.141**(2.96) (2.56) (-2.80) (-0.56) (-0.92) (2.17)
MALE -0.132*** . 0.0324* . 0.225*** .(-7.64) . (1.81) . (8.50) .
INC 0.0145 -0.00461 -0.0221 -0.0180 0.0129 0.0395(1.05) (-0.26) (-1.43) (-0.87) (0.59) (1.41)
EDU 0.00380 0.103 -0.000520 0.197 -0.0262*** 0.174(1.10) (0.58) (-0.14) (0.94) (-4.95) (0.61)
FJOB 0.00921 -0.218** 0.0379 -0.0539 -0.0274 -0.154(0.37) (-2.39) (1.48) (-0.50) (-0.73) (-1.05)
MWRK -0.00710 0.0211 -0.0158 0.00473 -0.0356 0.00685(-0.44) (0.47) (-0.92) (0.09) (-1.41) (0.09)
FAM -0.0206 -0.0757 0.110** 0.0767 -0.0133 0.123(-0.52) (-1.42) (2.50) (1.23) (-0.21) (1.44)
HOUSE 0.0162 0.0212 -0.00747 0.0168 -0.00822 0.0172(0.98) (0.96) (-0.41) (0.65) (-0.32) (0.49)
DIST 194.9 888.8* 84.94 757.4 -139.0 2123.2***(1.50) (1.92) (0.62) (1.39) (-0.69) (2.86)
N 9083 9083 9083 9083 9083 9083R2 0.0193 0.010 0.0593 0.018 0.052χ32 9.25(0.0262) 3.25(0.3552) 11.24(0.0105)t statistics in parentheses* p < 0.10, ** p < 0.05, *** p < 0.01
18
5 EMPIRICAL RESULTS
Table 14: Effects of private tutoring on well being : GLS and FE Estimation(3)
(13)GLS (14)FE (15)GLS (16)FE (17)GLS (18)FEWB_conf WB_conf WB_dep WB_dep WB_health WB_health
PTHR -0.00137 -0.00174* -0.00278** -0.00356** -0.00348** -0.00382**(-1.59) (-1.72) (-2.26) (-2.42) (-2.25) (-2.03)
PT1 -0.0160 . -0.0776** . -0.0646* .(-0.72) . (-2.53) . (-1.73) .
STUDY 0.00474*** 0.00306*** 0.00103 -0.0000902 0.00118 0.000692(7.69) (4.40) (1.16) (-0.09) (1.06) (0.54)
RANK1 0.00486*** 0.00297*** 0.00210*** 0.00159** 0.00105 -0.000713(11.92) (5.41) (3.61) (2.00) (1.44) (-0.70)
RANK2 0.00304*** 0.00211*** 0.00215*** 0.00237*** -0.0000669 -0.0000390(7.24) (4.14) (3.59) (3.22) (-0.09) (-0.04)
PTHRM -0.000618 . -0.00873** . -0.00350 .(-0.21) . (-2.19) . (-0.72) .
PTFR -0.00445 0.00309 -0.0482** -0.0342 0.000690 0.0357(-0.28) (0.17) (-2.08) (-1.31) (0.02) (1.06)
CLASS -0.000189 -0.00287 -0.00659*** -0.00946*** 0.00287 0.00191(-0.11) (-1.30) (-2.68) (-2.95) (0.93) (0.46)
BORDER 0.0165 . -0.0329 . -0.0464 .(0.82) . (-1.19) . (-1.38) .
MONTH -0.0345 . -0.0345 . 0.0567 .(-1.33) . (-0.96) . (1.30) .
HIGH 0.143*** 0.102** -0.140*** -0.0650 0.0787 0.0131(4.00) (2.38) (-2.73) (-1.04) (1.22) (0.16)
MALE 0.0836*** . 0.277*** . 0.0351 .(4.18) . (10.03) . (1.04) .
INC 0.0450*** 0.0363* 0.0499** 0.0354 0.0378 -0.00402(2.98) (1.95) (2.32) (1.31) (1.40) (-0.12)
EDU 0.00992** 0.375** 0.00241 0.136 -0.000966 -0.180(2.49) (1.97) (0.44) (0.49) (-0.14) (-0.51)
FJOB 0.0138 -0.0975 -0.0386 -0.161 0.00702 -0.261(0.49) (-1.01) (-0.98) (-1.15) (0.15) (-1.45)
MWRK -0.0457** 0.0523 -0.0480* -0.0534 -0.0395 -0.0562(-2.45) (1.08) (-1.85) (-0.76) (-1.24) (-0.63)
FAM 0.0635 0.0448 0.128** 0.0878 -0.0523 -0.145(1.46) (0.79) (2.07) (1.07) (-0.68) (-1.37)
HOUSE 0.0515*** 0.0730*** 0.0133 -0.0156 0.0812** 0.0312(2.84) (3.11) (0.51) (-0.46) (2.51) (0.71)
DIST 374.6** 1510.6*** 495.3** 814.5 409.7 731.5(2.51) (3.07) (2.39) (1.14) (1.61) (0.80)
N 9083 9083 9083 9083 9083 9083R2 0.0633 0.025 0.0404 0.007 0.0068 0.004χ32(p > χ32) 10.87(0.0125) 1.26(0.7397) 2.73(0.4356)t statistics in parentheses* p < 0.10, ** p < 0.05, *** p < 0.01
19
5 EMPIRICAL RESULTS
Table 15: Effects of private tutoring on well being : Hausman-Taylor Estimation(1)
(1) (2) (3) (4) (5)WB_all WB_composite WB_fam WB_per WB_sch
PTHR -0.00355*** -0.0140*** -0.00130*** 0.00136* 0.000735(-2.72) (-3.97) (-3.15) (1.81) (0.84)
PT1 0.951 7.204 0.245 0.728 1.377(0.80) (0.31) (0.14) (0.36) (0.31)
STUDY 0.00192** 0.00865*** 0.000803*** 0.00134*** 0.00246***(2.15) (3.57) (2.85) (2.61) (4.10)
RANK1 0.00481*** 0.0222*** 0.00173*** 0.00195*** 0.00384***(6.84) (11.61) (7.76) (4.79) (8.10)
RANK2 0.00325*** 0.0150*** 0.00110*** 0.000997*** 0.00446***(4.98) (8.47) (5.33) (2.64) (10.15)
PTHRM -0.00797 0.0311 0.00357 -0.00484 -0.000960(-1.20) (0.18) (0.28) (-0.35) (-0.03)
PTFR -0.0185 0.00661 -0.00903 -0.00653 -0.00797(-0.84) (0.10) (-1.23) (-0.49) (-0.51)
CLASS -0.00563** -0.0224*** -0.00277*** -0.00307* -0.00119(-2.15) (-2.90) (-3.09) (-1.88) (-0.62)
BORDER -0.0531 -0.308 -0.00723 -0.0544 -0.0139(-0.76) (-0.20) (-0.06) (-0.41) (-0.05)
MONTH -0.0324 -0.0852 -0.00404 -0.0405 -0.0476(-0.54) (-0.06) (-0.04) (-0.32) (-0.17)
HIGH -0.0442 0.385** 0.0743*** 0.103*** -0.0259(-0.81) (2.58) (4.27) (3.25) (-0.70)
MALE 0.353** 1.241 -0.0583 -0.0338 0.201(2.29) (0.40) (-0.25) (-0.13) (0.35)
INC 0.0184 0.0450 -0.0181** -0.00431 -0.0181(0.82) (0.69) (-2.41) (-0.31) (-1.12)
EDU 0.0229*** 0.122 0.0147 0.0163 0.0195(2.61) (0.57) (0.89) (0.89) (0.48)
FJOB 0.0105 -0.948*** -0.0395 -0.154** -0.0455(0.18) (-2.86) (-1.05) (-2.40) (-0.56)
MWRK -0.0366 0.00839 0.0233 0.0161 0.00335(-1.07) (0.05) (1.23) (0.48) (0.08)
FAM 0.0652 0.0868 -0.00288 -0.0741* 0.0770(1.01) (0.44) (-0.13) (-1.78) (1.58)
HOUSE 0.0878*** 0.172** 0.0269*** 0.0206 0.0163(3.17) (2.11) (2.84) (1.19) (0.80)
DIST 358.5 6278.7*** -44.13 764.6** 720.7*(1.12) (3.72) (-0.23) (2.32) (1.75)
N 9083 9083 9083 9083 9083χ32(p > χ32) 1.41(0.7041) 1.38(0.7103) 0.79(0.8512) 1.32(0.7239) 0.77(0.8561)t statistics in parentheses* p < 0.10, ** p < 0.05, *** p < 0.01
20
5 EMPIRICAL RESULTS
Table 16: Effects of private tutoring on well being : Hausman-Taylor Estimation(2)
(6) (7) (8) (9)WB_std WB_conf WB_dep WB_health
PTHR -0.00469*** -0.00178** -0.00357*** -0.00379**(-3.92) (-2.24) (-3.03) (-2.56)
PT1 0.636 4.913 2.604 1.063(0.12) (0.62) (0.82) (0.22)
STUDY -0.000358 0.00304*** -0.0000956 0.000709(-0.44) (5.56) (-0.12) (0.70)
RANK1 0.0105*** 0.00297*** 0.00157** -0.000708(16.32) (6.89) (2.48) (-0.89)
RANK2 0.00357*** 0.00211*** 0.00237*** -0.0000411(5.98) (5.28) (4.01) (-0.06)
PTHRM -0.000715 -0.0148 -0.0199 -0.00830(-0.02) (-0.26) (-0.92) (-0.25)
PTFR 0.0300 0.00308 -0.0349* 0.0351(1.41) (0.22) (-1.67) (1.33)
CLASS -0.00406 -0.00283 -0.00934*** 0.00187(-1.56) (-1.63) (-3.66) (0.58)
BORDER -0.0602 -0.232 -0.163 -0.102(-0.17) (-0.43) (-0.79) (-0.32)
MONTH -0.0133 -0.176 -0.122 0.0208(-0.04) (-0.34) (-0.62) (0.07)
HIGH 0.143*** 0.103*** -0.0702 0.0118(2.84) (3.06) (-1.41) (0.19)
MALE 0.326 0.686 0.613 0.177(0.46) (0.65) (1.47) (0.28)
INC 0.0389* 0.0363** 0.0361* -0.00317(1.78) (2.48) (1.68) (-0.12)
EDU -0.0276 0.0968 0.0234 0.0144(-0.56) (1.40) (0.82) (0.33)
FJOB -0.144 -0.0956 -0.110 -0.202(-1.32) (-1.27) (-1.10) (-1.54)
MWRK 0.00546 0.0507 -0.0554 -0.0558(0.10) (1.34) (-1.07) (-0.84)
FAM 0.121* 0.0448 0.0883 -0.143*(1.82) (1.01) (1.36) (-1.74)
HOUSE 0.0165 0.0726*** -0.0159 0.0325(0.60) (3.94) (-0.59) (0.95)
DIST 1888.7*** 1498.0*** 873.0* 722.5(3.39) (3.90) (1.70) (1.08)
N 9083 9083 9083 9083χ32(p > χ32) 0.75(0.8608) 2.47(0.4813) 0.47(0.9264) 0.53(0.9113)t statistics in parentheses* p < 0.10, ** p < 0.05, *** p < 0.01
21
5 EMPIRICAL RESULTS
Table 17: Effects of private tutoring on well being : Hausman-Taylor Estimation(3)
(1) (2) (3) (4) (5)WB_all WB_composite WB_fam WB_per WB_sch
PTHR -0.00303** -0.0117*** -0.00112*** 0.00155** 0.00128(-2.02) (-3.27) (-2.71) (2.05) (1.45)
PT1 1.786 9.909 0.434 0.905 2.004(1.49) (0.41) (0.23) (0.43) (0.44)
STUDY 0.00155 0.00681*** 0.000652** 0.00114** 0.00238***(1.52) (2.81) (2.32) (2.23) (3.96)
PTHRM -0.00981 0.0257 0.00321 -0.00517 -0.00219(-1.51) (0.15) (0.25) (-0.36) (-0.07)
PTFR -0.0143 0.0299 -0.00715 -0.00416 -0.00575(-0.57) (0.47) (-0.97) (-0.31) (-0.37)
CLASS -0.00706** -0.0285*** -0.00325*** -0.00364** -0.00203(-2.42) (-3.68) (-3.63) (-2.24) (-1.06)
BORDER -0.0797 -0.377 -0.0116 -0.0577 -0.0322(-1.16) (-0.23) (-0.09) (-0.42) (-0.11)
MONTH -0.0569 -0.165 -0.00968 -0.0459 -0.0655(-0.97) (-0.10) (-0.08) (-0.35) (-0.22)
HIGH -0.147*** -0.0711 0.0347*** 0.0441*** 0.00815(-6.15) (-1.11) (4.71) (3.31) (0.52)
MALE 0.447*** 1.526 -0.0387 -0.0160 0.269(2.87) (0.47) (-0.16) (-0.06) (0.44)
INC 0.0310 0.0776 -0.0157** -0.00186 -0.0103(1.23) (1.19) (-2.07) (-0.14) (-0.63)
EDU 0.0316*** 0.168 0.0180 0.0197 0.0282(3.71) (0.76) (1.05) (1.04) (0.67)
FJOB 0.0468 -0.915*** -0.0372 -0.154** -0.0332(0.78) (-2.73) (-0.98) (-2.38) (-0.41)
MWRK -0.0532 0.0386 0.0256 0.0180 0.00985(-1.55) (0.23) (1.34) (0.54) (0.24)
FAM 0.0669 0.0677 -0.00442 -0.0757* 0.0737(0.93) (0.34) (-0.19) (-1.81) (1.50)
HOUSE 0.0780** 0.160* 0.0260*** 0.0197 0.0132(2.52) (1.93) (2.72) (1.14) (0.65)
DIST 492.9 6912.2*** 0.541 809.5** 901.1**(1.51) (4.06) (0.00) (2.44) (2.16)
N 9083 9083 9083 9083 9083χ32(p > χ32) 2.31(0.5098) 7.62(0.1783) 4.27(0.3705) 3.75(0.4416) 1.57(0.9052)t statistics in parentheses* p < 0.10, ** p < 0.05, *** p < 0.01
22
5 EMPIRICAL RESULTS
Table 18: Effects of private tutoring on well being : Hausman-Taylor Estimation(4)
(6) (7) (8) (9)WB_std WB_conf WB_dep WB_health
PTHR -0.00383*** -0.00146* -0.00330*** -0.00383***(-3.16) (-1.83) (-2.80) (-2.59)
PT1 1.526 5.324 2.872 1.020(0.27) (0.66) (0.88) (0.21)
STUDY -0.00168** 0.00280*** -0.0000622 0.000823(-2.04) (5.16) (-0.08) (0.82)
PTHRM -0.00245 -0.0157 -0.0204 -0.00822(-0.06) (-0.27) (-0.91) (-0.25)
PTFR 0.0445** 0.00610 -0.0345* 0.0339(2.07) (0.43) (-1.65) (1.29)
CLASS -0.00739*** -0.00364** -0.00962*** 0.00212(-2.81) (-2.09) (-3.77) (0.66)
BORDER -0.0759 -0.244 -0.170 -0.102(-0.20) (-0.45) (-0.80) (-0.32)
MONTH -0.0407 -0.188 -0.130 0.0222(-0.11) (-0.35) (-0.64) (0.07)
HIGH -0.287*** 0.0481*** -0.0235 0.0528**(-13.30) (3.36) (-1.13) (2.01)
MALE 0.415 0.730 0.642 0.173(0.56) (0.68) (1.49) (0.27)
INC 0.0502** 0.0407*** 0.0398* -0.00370(2.26) (2.78) (1.85) (-0.14)
EDU -0.00996 0.104 0.0270 0.0134(-0.19) (1.48) (0.91) (0.30)
FJOB -0.139 -0.0908 -0.105 -0.202(-1.24) (-1.20) (-1.04) (-1.55)
MWRK 0.0158 0.0549 -0.0532 -0.0563(0.28) (1.45) (-1.02) (-0.84)
FAM 0.112* 0.0422 0.0870 -0.142*(1.66) (0.95) (1.33) (-1.73)
HOUSE 0.0121 0.0709*** -0.0174 0.0327(0.43) (3.82) (-0.64) (0.96)
DIST 2059.5*** 1586.0*** 956.3* 719.6(3.63) (4.12) (1.84) (1.08)
N 9083 9083 9083 9083χ32(p > χ32) 50.09(0.0000) 4.26(0.5129) 1.11(0.8928) 0.81(0.9375)t statistics in parentheses* p < 0.10, ** p < 0.05, *** p < 0.01
23
5 EMPIRICAL RESULTS
Table 19: Effects of private tutoring on well being : Hausman-Taylor Estimation(5)
(1) (2) (3) (4) (5)WB_all WB_composite WB_fam WB_per WB_sch
ptcosthr -0.000654 0.000875 0.000402 0.00187*** 0.000312(-0.43) (0.26) (1.02) (2.61) (0.37)
PTHR -0.00327** -0.0121*** -0.00116*** 0.00175** 0.00121(-2.05) (-3.36) (-2.78) (2.30) (1.35)
PT1 2.254* 10.41 0.545 1.118 2.050(1.78) (0.41) (0.28) (0.50) (0.43)
STUDY 0.00158 0.00713*** 0.000673** 0.00114** 0.00237***(1.47) (2.92) (2.40) (2.22) (3.93)
PTHRM -0.0112 0.0240 0.00314 -0.00568 -0.00252(-1.59) (0.13) (0.23) (-0.36) (-0.07)
PTFR -0.0126 0.0378 -0.00706 -0.00448 -0.00560(-0.48) (0.59) (-0.96) (-0.33) (-0.36)
CLASS -0.00729** -0.0272*** -0.00313*** -0.00359** -0.00164(-2.35) (-3.48) (-3.49) (-2.19) (-0.85)
BORDER -0.102 -0.409 -0.0182 -0.0667 -0.0351(-1.39) (-0.24) (-0.14) (-0.45) (-0.11)
MONTH -0.0722 -0.193 -0.0136 -0.0524 -0.0659(-1.14) (-0.12) (-0.11) (-0.37) (-0.21)
HIGH -0.149*** -0.0872 0.0330*** 0.0453*** 0.00562(-5.86) (-1.35) (4.47) (3.37) (0.35)
MALE 0.505*** 1.582 -0.0257 0.0106 0.275(3.04) (0.47) (-0.10) (0.04) (0.44)
INC 0.0233 0.0810 -0.0161** 0.00173 -0.00880(0.86) (1.23) (-2.12) (0.12) (-0.54)
EDU 0.0328*** 0.176 0.0185 0.0204 0.0290(3.62) (0.77) (1.04) (1.01) (0.67)
FJOB 0.0584 -0.913*** -0.0375 -0.162** -0.0330(0.91) (-2.72) (-0.99) (-2.47) (-0.40)
MWRK -0.0539 0.0434 0.0215 0.0195 0.0129(-1.47) (0.26) (1.12) (0.57) (0.31)
FAM 0.0573 0.0200 0.00254 -0.0868** 0.0791(0.75) (0.10) (0.11) (-2.06) (1.59)
HOUSE 0.0732** 0.154* 0.0284*** 0.0171 0.0165(2.26) (1.85) (2.97) (0.98) (0.80)
DIST 582.4* 6808.2*** 26.46 837.1** 854.4**(1.67) (3.98) (0.14) (2.49) (2.04)
N 8979 8979 8979 8979 8979χ32(p > χ32) 2.43(0.4885) 1.36(0.7151) 0.80(0.8483) 1.09(0.7793) 0.77(0.8558)t statistics in parentheses* p < 0.10, ** p < 0.05, *** p < 0.01
24
5 EMPIRICAL RESULTS
Table 20: Effects of private tutoring on well being : Hausman-Taylor Estimation(6)
(1) (2) (3) (4)WB_std WB_conf WB_dep WB_health
ptcosthr -0.00269** 0.000193 0.000697 0.0000601(-2.33) (0.25) (0.62) (0.04)
PTHR -0.00414*** -0.00152* -0.00330*** -0.00399***(-3.38) (-1.88) (-2.77) (-2.67)
PT1 1.417 5.609 2.765 1.399(0.25) (0.67) (0.82) (0.28)
STUDY -0.00171** 0.00275*** 0.0000806 0.00100(-2.07) (5.05) (0.10) (0.99)
PTHRM -0.00252 -0.0166 -0.0211 -0.00908(-0.06) (-0.27) (-0.90) (-0.26)
PTFR 0.0458** 0.00591 -0.0358* 0.0412(2.12) (0.42) (-1.71) (1.57)
CLASS -0.00686*** -0.00357** -0.00958*** 0.00222(-2.60) (-2.05) (-3.75) (0.69)
BORDER -0.0690 -0.260 -0.164 -0.122(-0.18) (-0.46) (-0.74) (-0.37)
MONTH -0.0403 -0.201 -0.126 0.00680(-0.11) (-0.36) (-0.60) (0.02)
HIGH -0.290*** 0.0431*** -0.0261 0.0510*(-13.34) (3.00) (-1.24) (1.93)
MALE 0.396 0.768 0.630 0.221(0.52) (0.68) (1.41) (0.34)
INC 0.0464** 0.0397*** 0.0387* -0.00147(2.07) (2.69) (1.79) (-0.05)
EDU -0.00919 0.109 0.0254 0.0149(-0.17) (1.52) (0.83) (0.33)
FJOB -0.137 -0.0894 -0.109 -0.203(-1.23) (-1.18) (-1.07) (-1.55)
MWRK 0.0309 0.0619 -0.0591 -0.0642(0.55) (1.62) (-1.12) (-0.95)
FAM 0.102 0.0362 0.0669 -0.157*(1.49) (0.81) (1.02) (-1.90)
HOUSE 0.0113 0.0724*** -0.0140 0.0163(0.40) (3.90) (-0.52) (0.47)
DIST 2100.1*** 1585.4*** 897.7* 675.5(3.69) (4.11) (1.72) (1.01)
N 8979 8979 8979 8979χ32(p > χ32) 0.71(0.8711) 2.42(0.6584) 0.41(0.9379) 0.52(0.9145)t statistics in parentheses* p < 0.10, ** p < 0.05, *** p < 0.01
25
5 EMPIRICAL RESULTS
we can conclude that the instruments derived from explanatory variables within the
model are valid instruments, and HT estimator is consistent and more efficient. The
effects of hours spent on private tutoring support the hypothesis that well being is
multidimensional. Column (1) shows that the hours of tutoring on overall life satis-
faction is negative and significant at 1% significance level. Its coefficient is similar to
FE regarding academic achievement(rank) and wealth(owning a house). Column (3)
is HT estimation for constructed composite well being. The estimated parameters are
similar to (1) and (2) but it benefits from greater variation of dependent variable. We
can see from columns (5) to (18) the effect of private tutoring and other regressors are
different and in some cases contradictory. In particular, we can see that the effect of tu-
toring on interpersonal relationship domain is significant and positive, while the effect
is negative and significant for study related stress, family relationship, self confidence,
depressive feelings and health. The positive effect on interpersonal relationship is con-
sistent with what we observe in reality, that since the majority of students receives
private tutoring in private academies, more hours of tutoring functions as a means to
socialize with their peers. Since we control for temporal and cross individual change
in rank, it can also mean that students who receive tutoring but has no effect of it on
academic grades are instead satisfied with interpersonal relationship. Likewise, more
tutoring with no change in rank means that they are exerting low effort. It should also
be noted that tutoring has negative and significant effect on study related stress and
self reported physical health. Hours of studying by oneself is positive and significant
in most of the domains, which can imply that in contrast to tutoring that has more
involuntary and compulsory nature, studying by oneself comes from more motivation
and may bring a sense of self accomplishment. Nevertheless, its effect on study re-
lated stress in negative and significant. In general students have greater well being
in younger period compared to middle and high school periods. Proximity to private
academies are positively and significantly correlated with most of the domains but the
effect of mean income of households in the same district are not significant. This im-
26
5 EMPIRICAL RESULTS
plies that students benefit from better quality of education environment such as more
diverse choice of private academies, or less commuting time, other than higher income
of the district. Greater wealth (proxied by whether the household owns a house) is pos-
itively correlated with well being, but father’s education and occupation in high skilled
jobs are negatively correlated. Whether received tutoring before entering school shows
no correlation with the domains of well being.
Tables 17 and 18 above shows the results of Hausman Taylor estimation without
control for rank variable, to allow for the presumed positive effect of tutoring on aca-
demic achievement. Hausman test results of chi square statistics are reported in the
last two rows. It shows that HT estimators are consistent and efficient, except for in-
terpersonal relationship. However the coefficients have decreased in value compared
to previous specifications. We can see here the effect of rank variable working as a
channel of the effect of private tutoring, that private tutoring has positive impact on
raising the academic achievement which brings greater well being. This channel seems
to offset the negative effect of tutoring on well being when controlled for rank variable.
We introduced hourly price of tutoring(ptcosthr) as a proxy for differentiated effect
with the quality of the tutoring, as shown in tables 19 and 20. Its effect is positive
and significant for interpersonal relationship and negative and significant for study
related stress, but it is insignificant in other domains including the composite well
being. More expensive tutoring tend to be individual or group tutoring or private
academies that are considered more competitive, which can explain its negative impact
on study related stress. This could imply that students are more satisfied with their
relationship with peers because they suffer similarly higher level of stress, or it could be
that more expensive tutoring puts them in advantageous position vis a vis their peers.
impact on raising the academic achievement which brings greater well being. This
channel seems to offset the negative effect of tutoring on well being when controlled
for rank variable.
27
6 CONCLUSION
6. Conclusion
This paper applied panel data regressions to estimate the effect of hours spent on
private tutoring on multidimensional well being of 6,293 Korean students from age
10 to 16 over the period 2003-2008. Hausman and Taylor and FE models estimates
indicate that hours of tutoring has negative impact on domains of well being such as
study related stress, health and overall life satisfaction, while its effect has different
effects on different domains. The result shows that there clearly exists a shadow cost
of private tutoring, in the form of reduced well being of students. So far the cost of
private tutoring has been only assessed in terms of economic burden of households
and inefficiency in public education system, and its causal effect on well being has
been neglected. The results of this paper calls for more attention and resources to
measure the negative impact of excessive private tutoring prominent in some societies,
as negative effect on children’s well being can have more severe effect in the long run.
28
6 CONCLUSION
References
Baltagi, B. (2014). Econometric analysis of panel data, volume 1. John Wiley & Sons.
Bray, M. and Lykins, C. R. (2012). Shadow education: Private supplementary tutoring and its implications forpolicy makers in Asia. Asian Development Bank Mandaluyong City, Philippines.
Bray, M., Mazawi, A. E., and Sultana, R. G. (2013). Private tutoring across the Mediterranean: power dynamicsand implications for learning and equity. Sense.
Briggs, D. C. (2001). The effect of admissions test preparation: Evidence from nels: 88. Chance, 14(1):10–18.
Bronfenbrenner, U. and Morris, P. A. (1998). The ecology of developmental processes. Handbook of childpsychology, 1:993–1028.
Cheo, R. and Quah, E. (2005). Mothers, maids and tutors: An empirical evaluation of their effect onchildren’s academic grades in singapore. Education Economics, 13(3):269–285.
Constitutional Court of Korea (2000). Extracurricular Lesson Ban Case. 12-1 KCCR 427,:1–47.
Dang, H.-A. (2007). The determinants and impact of private tutoring classes in vietnam. Economics ofEducation Review, 26(6):683–698.
Fernandes, L., Mendes, A., and Teixeira, A. (2013). A weighted multidimensional index of child well-being which incorporates children’s individual perceptions. Social indicators research, 114(3):803–829.
KEDI (2005). Analysis on the Growth of Korean Education for 60 Years. pages 1–221.
Kim, S. and Lee, J.-H. (2010). Private tutoring and demand for education in South Korea. Economicdevelopment and cultural change, 58(2):259–296.
Kim, Y.-B., Yang, S.-K., and Park, S.-H. (2012). An Analysis of Private Educational Expenditure Transi-tions and Trends: Analysis of the “Household Survey” Data. The Journal of Korean Education, 39(1):261–284.
Kim, Y. C. (2008). A Study on the history of education policy of the Republic of Korea. Technical report,Ministry of Education of Korea.
Lee, C. J. and Jang, H. M. (2008). An analysis of patterns of the government policy to shadow education.Asian Journal of Education, 9(4):173–200.
Lee, J.-T., Kim, Y.-B., and Yoon, C.-H. (2004). The effects of pre-class tutoring on student achievement:Challenges and implications for public education in korea. KEDI Journal of Educational Policy, 1(1):25–42.
Ministry of Education, Science and Technology (2013). Announcement of major student mental healthprojects of 2013(Press Release).
NYPI (2009). Human Rights Condition of Korean Youth and Children In Comparison to InternationalStandards IV: Right of Survival and Protection(in Korean). Technical report.
OECD (2012). OECD Economic Surveys: Korea 2012. OECD Publishing.
Statistics-Korea (2014). Private Education Expenditure 2013 (Press Release). Technical report.
Stevenson, D. L. and Baker, D. P. (1992). Shadow education and allocation in formal schooling: Transitionto university in japan. American Journal of Sociology, 97(6):1639.
29
6 CONCLUSION
Suryadarma, D., Suryahadi, A., Sumarto, S., and Rogers, F. H. (2006). Improving student performancein public primary schools in developing countries: Evidence from indonesia. Education Economics,14(4):401–429.
Tansel, A. and Bircan, F. (2006). Demand for education in Turkey: A tobit analysis of private tutoringexpenditures. Economics of Education Review.
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.
30