Content uploaded by Linxiu Zhang
Author content
All content in this area was uploaded by Linxiu Zhang on Sep 06, 2015
Content may be subject to copyright.
This article was downloaded by: [202.93.10.41]
On: 26 August 2015, At: 16:51
Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered
office: 5 Howick Place, London, SW1P 1WG
Click for updates
Asia Pacific Journal of Education
Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/cape20
Do poor students benefit from China's
Merger Program? Transfer path and
educational performance
Xinxin Chena, Hongmei Yib, Linxiu Zhangb, Di Moc, James Chud &
Scott Rozelled
a School of Economics, Zhejiang Gongshang University, Hangzhou,
China
b Center for Chinese Agricultural Policy, Institute of Geographical
Sciences and Natural Resource Research, Chinese Academy of
Sciences, Beijing, China
c LICOS Centre for Institutions and Economic Performance,
University of Leuven (KUL), Leuven, Belgium
d Freeman Spogli Institute, Stanford University, Stanford, CA, USA
Published online: 25 Aug 2013.
To cite this article: Xinxin Chen, Hongmei Yi, Linxiu Zhang, Di Mo, James Chu & Scott Rozelle
(2014) Do poor students benefit from China's Merger Program? Transfer path and educational
performance, Asia Pacific Journal of Education, 34:1, 15-35, DOI: 10.1080/02188791.2013.790781
To link to this article: http://dx.doi.org/10.1080/02188791.2013.790781
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the
“Content”) contained in the publications on our platform. However, Taylor & Francis,
our agents, and our licensors make no representations or warranties whatsoever as to
the accuracy, completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content
should not be relied upon and should be independently verified with primary sources
of information. Taylor and Francis shall not be liable for any losses, actions, claims,
proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or
arising out of the use of the Content.
This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
Do poor students benefit from China’s Merger Program? Transfer
path and educational performance
Xinxin Chen
a
*, Hongmei Yi
b
, Linxiu Zhang
b
,DiMo
c
, James Chu
d
and Scott Rozelle
d
a
School of Economics, Zhejiang Gongshang University, Hangzhou, China;
b
Center for Chinese
Agricultural Policy, Institute of Geographical Sciences and Natural Resource Research, Chinese
Academy of Sciences, Beijing, China;
c
LICOS Centre for Institutions and Economic Performance,
University of Leuven (KUL), Leuven, Belgium;
d
Freeman Spogli Institute, Stanford University,
Stanford, CA, USA
(Received 26 September 2011; final version received 25 January 2012)
Aiming to provide better education facilities and improve the educational attainment of
poor rural students, China’s government has been merging remote rural primary
schools into centralized village, town, or county schools since the late 1990s. To
accompany the policy, boarding facilities have been constructed that allow (mandate)
primary school-aged children to live at school rather than at home. More generally,
there also have been efforts to improve rural schools, especially those in counties and
towns. Unfortunately, little empirical work has been available to evaluate the impact of
the new merger and investment programmes on the educational performance of
students. Drawing on a unique dataset that records both the path by which students
navigate their primary school years (i.e., which different types of schools did students
attend) as well as math test scores in three poverty-stricken counties, we use descriptive
statistics and multivariate analysis (both Ordinary Least Squares (OLS) and covariate
matching) to analyse the relationship between different transfer paths and student
educational performance. This allows us to examine the costs and benefits of the school
merger and investment programmes. The results of the analysis show that students who
attend county schools perform systematically better than those who attend village or
town schools. However, completing primary school in town schools seems to have
no effect on students’ academic performance. Surprisingly, starting primary education
in a teaching point does not hurt rural students; on the contrary, it increases their test
scores in some cases. Finally, in terms of the boarding effect, the neutral estimate in
OLS and the negative estimate in covariate matching results confirm that boarding at
school does not help the students; in some cases it may even reduce their academic
performance.
Keywords: Merger Program; transfer path; educational performance; rural China
Introduction
Aiming to improve the quality of rural education and reduce educational disparities
between urban and rural areas, China’s State Council implemented the Rural Primary
School Merger Program in the late 1990s. Especially during early and mid 2000s, many
one-room schoolhouses offering schooling from Grade 1 to 4 – so called teaching points
(jiaoxuedian) – were shut down and merged into centralized schools in larger villages and
towns. Resources were channelled towards larger schools in selected towns and the county
seat, and the role of smaller village schools was downgraded. Indeed, the number of
q2013 National Institute of Education, Singapore
*Corresponding author. Email: chenxx@mail.zjgsu.edu.cn
Asia Pacific Journal of Education, 2014
Vol. 34, No. 1, 15–35, http://dx.doi.org/10.1080/02188791.2013.790781
Downloaded by [202.93.10.41] at 16:51 26 August 2015
primary schools in rural China fell by 50% between 1999 and 2009 (National Bureau of
Statistics, 2000,2010).
Given the broad scope of the Merger Program (and future plans to continue the
Program), a significant question is whether the initiative has had any net benefit for poor
children in rural areas. In teaching points a single teacher is responsible for teaching every
student in every grade from Grade 1 to 4. The teacher typically teaches all students – in
Grade 1 to 4 – in a single classroom. With insufficient resources, teachers in teaching
points often only teach math and Chinese and little in the way of other courses (such as,
art, science or music). In contrast, centralized town or county schools have specialized
teachers, better facilities and more curricular offerings (Zhuo, 2006). If access to these
better facilities, teachers and curriculum positively affects the educational performance of
children, we can say that there is a positive resource effect.
On the other hand, there are aspects of school mergers that may have a number of
adverse effects on students. First, transferring may reduce the student’s level of comfort
and familiarity (associated with going to school in one’s own village – as it typically is in a
teaching point or a village primary school), thus negatively affecting educational
performance. Second, students who transfer to a new school usually live far away and must
board at the school. Third, the lack of parental care (because children live away from
home) might also lead to psychological problems, especially for young students from
Grade 1 to 3 (Luo, Shi, et al., 2009; Pang, 2006). Fourth, students who board at centralized
schools have been shown to have poorer nutrition and health relative to the students who
live at home (Luo, Shi, et al., 2009; Luo, Zhang, et al., 2010). In turn, poor nutrition and
psychological problems almost certainly detract from student learning. We call these
adverse effects the disruption effect.
Finding the net benefit of the Merger Program requires an analysis of both its benefits
(the resource effects) and costs (the disruption effects). To date, we know of only one
research team that has published an empirical paper disaggregating costs and benefits to
determine the net effect of the Merger Program on students. Using data from a large
sample in Shaanxi province in the early 2000s, Liu, Zhang, Luo, Rozelle, and Loyalka
(2010) found that the overall effect of transferring students from a village school or
teaching point closed under the Merger Program to a larger, more central school is neutral;
that is, the positive benefits from the improved resource effect were similar in magnitude
to the negative costs.
However, specific policies to merge schools differ across provinces, prefectures, and
even within counties (Liu et al., 2010). Henan Province merged all village schools into one
for each village if the village has a population above a certain threshold. Yunnan Province
shut down teaching points with only one teacher and merged them into village schools.
Qinghai Province established town and county boarding schools to receive students from
nearby villages and teaching points. These are but three province level policies; counties
and prefectures also adjusted these policies in practice (Liu et al., 2010). In addition, in our
interviews in China’s poor northwest region in 2009 and 2010, we often have found that
two or more of these different policies and resulting transfer paths can exist in a single
county. As such, the Merger Program does not only shift students from teaching points to
county schools. In fact, primary school students transfer from school to school in a variety
of permutations: teaching point to village school; teaching point to town school; village
school to another village school; village school to town school; village or town school to
the county seat school; and more. In the rest of the paper we term the different paths taken
by different students through primary school transfer paths.
16 X. Chen et al.
Downloaded by [202.93.10.41] at 16:51 26 August 2015
In summary, specific transfer paths can and often do differ by student. Moreover,
because each transfer path has its own unique set of benefits (resource effects vary across
types of schools) and costs (disruption effects also can vary across types of schools), it is
possible that different transfer paths will have different impacts on the educational
performance of students. This fact further complicates the debate about merger costs and
benefits and must be addressed to develop a comprehensive argument.
To our knowledge, no published empirical study in development economics exists that
evaluates the costs and benefits of the Merger Program while accounting for different
transfer paths of students. In the Liu et al. paper (2010), the authors compare students who
transfer to schools (guest students) with students who were originally at the school (host
students) to examine the effect of mergers. However, they only measure the effect of
switching schools due to mergers and do not account for the different kinds of schools that
students can transfer out from or transfer into. Certainly a rigorous analysis of the costs
and benefits associated with the programme could suggest potential adjustments to
policymakers. Given the scope of the mergers today (and plans for continuing in the
future), such a study is overdue.
The overall goal of this paper is to evaluate the effect of different transfer paths on
student educational performance. At the broadest level we ask, what are the net benefits
related to the different transfer paths for poor, rural students? The key questions we
attempt to address in our study include: (a) What transfer paths are students taking as a
result of the Merger Program and other educational policy shifts? (b) How are student test
scores affected by transfer paths? (c) Are there any negative impacts of the Merger
Program? (d) Is educational performance affected by whether or not students live at home
or board at school?
In order to answer these questions, we draw on a dataset we collected in three poverty-
stricken counties in China. We begin by categorizing types of transfer paths among
students in the sample counties. We then compare standardized math test scores between
students who have taken different transfer paths. We also use different estimation
strategies – Ordinary Least Squares (OLS) estimation and covariate matching – to
examine the impact of transfer paths and boarding statuses of the students on their
educational performance. These estimation strategies attempt to control for the fact that
both educational performance and transfer paths may be affected by the characteristics of
students and their families. Finally, based on the empirical results, we discuss the net
potential benefits of the Merger Program and its impact on poor, rural students.
Data collection
Our data come from a survey we fielded in three counties (Ningshan, Shiquan and Hanyin)
in the south part of Shaanxi Province, one of the nation’s poorest provinces. These
counties are well-suited to answering our research question. First, these counties were
nationally – or provincially – designated poor counties.
1
They thus broadly represent the
rural poor in China. Second, county officials launched the Merger Program in these
counties at the end of the 1990s, and sufficient time has passed for the policy to take effect.
Third, the transfer paths of the students in these sample counties are diverse, such that
there is sufficient variation to study the relationship of transfer paths and educational
performance.
Our sample was drawn from 36 junior high schools in the three study counties. In
Ningshan County, all the seventh and eighth grade classes in all of the junior high schools
were selected. In Shiquan County and Hanyin County, a subset of seventh and eighth grade
Asia Pacific Journal of Education 17
Downloaded by [202.93.10.41] at 16:51 26 August 2015
classes in every junior high school was randomly selected. Because of the size of Shiquan
and Hanyin, it would have been impossible, given our budget, to survey all classes in each
school in these two counties.
In every sample class, we surveyed all the students. The total sample of 5,700 students
consists of 2,798 seventh graders and 2,902 eighth graders in 2009. The sample of students
appears to be similar in nature to what would be expected in a rural, poor setting. For
example, we find 6% more boys than girls, a ratio similar to that cited in the Ministry of
Education’s 2010 Annual Yearbook: 7% more boys than girls. Approximately 98% of the
seventh graders are aged between 11 and 15 years and about 99% of the eighth graders are
aged between 12 and 16 years.
Our measure of educational performance, the key dependent variable in the study, was
based on a 30-minute standardized math test that we administered ourselves. Since the test
was administered at the beginning of the school year, the test is measuring the
accumulated math ability from each student’s elementary schooling for seventh graders,
and for eighth graders it is, in part, measuring the accumulated math ability from students’
elementary schooling (but, only in part, since they had already studied in junior high for
one year). This math test was scored on a scale from zero to 100. The results that we obtain
closely approximate a normal distribution with a mean score of 54 points and a standard
deviation of 17 points for seventh graders and a normal distribution of mean score of 57
points and a standard deviation of 19 points for eighth graders. We keep the scores without
any further manipulation (that is, we do not normalize them as is done in some educational
studies) for the ease of interpretation.
To measure our main independent variable, the transfer path of each student, we
collected detailed information on the schooling histories of each student. We asked
students when and where they attended each grade during their primary school years in
order to create a variable to characterize each student’s transfer path from Grade 1 to 6. In
addition, we also asked questions about school type and boarding status (either living at
home or living at school) in each grade. Based on the answers to these questions, we
created variables for student transfer path, boarding status and each student’s primary
school educational experience.
In addition to educational performance and transfer path status, we also collected
information on each student’s personal and family characteristics to use as controls.
Variables included each student’s age, gender, household registration status (either urban
or rural, also called hukou), and ethnicity; each family member’s age, educational
attainment, and employment status; the household’s land holdings; and the total number of
household members. The answers to detailed questions about household assets were used
to generate a variable measuring the value of the household durable assets to represent
household socioeconomic status or wealth. All of the controls in the study’s empirical
model are produced from the above information.
Transfer paths and educational performance
In part because of the closing and/or merging of a large number of rural schools, 2,062 of
the 5,700 students in our sample (nearly 36% of our sample) transferred from one school to
another at some point during their primary school years. Our data contain many unique
starting and ending points for the transfer experiences of students, which we use to identify
a variety of student transfer paths. In this section, we provide a general picture of these
transfer paths, and describe the relationship between these paths and educational
performance.
18 X. Chen et al.
Downloaded by [202.93.10.41] at 16:51 26 August 2015
Student transfer paths
Our data confirm that there are indeed different transfer paths among the 36% of students
in our sample who started school in one type of school and finished in another (Table 1).
Generally speaking, more students transfer to town and county schools than transfer from
them (row 3; row 4). Likewise, more students transfer from teaching points and village
schools than transfer to them (row 1; row 2). To be specific, only 28% of all students
started primary education in town schools, but 45% of students finished primary education
in town schools. At the same time, the percentage (41%) of the students who finished
primary education in village schools is less than the percentage (44%) of the students who
started primary education in village schools. Furthermore, the percentage (14%) of
students who finished in county schools is a little more than the percentage (10%) of
students who started primary education in county schools. These patterns suggest that, in
our sample, students are being encouraged to transfer from teaching points or village
schools to more centralized town and county schools. Moreover, town schools are the
main destination schools for rural students.
There are some types of schools that play specific roles in different transfer paths.
Among the students who started primary education in teaching points (18% of all
students), none of them transferred to a teaching point (row 1). The fact that no student
finished primary education in a teaching point reflects the fact that teaching points, by
definition, do not provide education beyond the fourth grade. Moreover, no student
transferred from one teaching point in his/her village to a teaching point in another village.
Again, this finding reflects the fact that teaching points are designed to allow younger
students to go to school in his/her own village under the tutelage of a teacher who can get
to know the students.
Our data also show that transfer paths differ substantially even among students who
start primary education in the same type of school (Table 2). For example, of the 1,039
students (316 þ655 þ68) who started in teaching points (column 1 – rows 1 to 3),
around 63% (655 of them) eventually transferred to town schools. Another 30% (316 of
them) transferred to village schools. The remaining 7% (68 of them) transferred to county
schools. As such, our data shows that students from teaching points were transferring to
more centralized but different kinds of schools.
The same is true for students who started their primary school years in village schools
(Table 2). Of the 761 students (417 þ98 þ246) who started in village schools(column 1,
Table 1. Distribution of sample students by the type of school in which they started primary school
and by the type of school in which they finished primary school, in three study counties in Northwest
China, 2009.
Starting school
a
Ending school
b
No. % No. %
Type of schools (1) (2) (3) (4)
1 Teaching point schools 1,039 18
2 Village schools 2,480 44 2,340 41
3 Town schools 1,587 28 2,544 45
4 County schools 594 10 816 14
5 Total 5,700 100 5,700 100
Data source: Authors’ survey.
a
Starting school means the school where the students started primary school.
b
Ending school means the school where the students finished primary school.
Asia Pacific Journal of Education 19
Downloaded by [202.93.10.41] at 16:51 26 August 2015
rows 4 to 6), 55% (417 of them) transferred to town schools and 13% (98 of them)
transferred to county schools. The remaining 32% (246 of them) transferred to another
village school. This movement further reveals that, in our sample, more students
transferred to town schools than village/county schools under the Merger Program.
Taken together, the data show that few students transferred from town schools to
village schools (or other combination of “reverse flows”). Only 48 students transferred
from a town school to a village school (Table 2, row 7). Even fewer (26 ¼11 þ15)
transferred from a county school to either a town or village school (rows 10 and 11).
Finally, a subset of students transferred between schools, but stayed in the same
administrative level. For example, there were some students (82) who transferred between
town schools (from town to town – row 8). There were 20 students in our sample who
transferred between county schools (county to county – row 12).
Educational performance across transfer paths
We now turn to consider how various student transfer paths are correlated with the mean
math scores, which are used as proxies for educational performance (Table 2). This
analysis helps us identify and isolate several of the effects associated with different
transfer paths.
First, it appears that the higher the level of the school (village versus town versus
county), the higher the test scores. According to our data, among students who did not
switch schools, students in county schools score higher than students in town or village
schools. To be specific, the test scores of students who spent all six years of primary school
in the same county school (66.0 – row 15) are greater than students who stayed in the same
town school (52.1 – row 14). The scores of students who stayed in the same town school
for all six years, in turn, are higher than students who spent all six years in the same village
school (51.7 – row 13). Likewise, when comparing students who transferred from one
Table 2. The maths scores of sample students by the transfer path that the students took during their
primary school years in three study counties in China, 2009.
Obs.
No. % Score
Transfer paths (1) (2) (3)
1 Teaching point to village school 316 6 56.2
2 Teaching point to town school 655 11 58.7
3 Teaching point to county school 68 1 63.6
4 Village to village school 246 4 51.0
5 Village to town school 417 7 53.9
6 Village to county school 98 2 66.8
7 Town to village school 48 1 55.1
8 Town to town school 86 2 55.4
9 Town to county school 82 1 67.7
10 County to village school 11 0.2 46.8
11 County to town school 15 0.3 60.0
12 County to county school 20 0.4 72.3
13 In the same village school 1719 30 51.7
14 In the same town school 1371 24 52.1
15 In the same county school 548 10 66.0
Total 5,700 100 55.1
Data source: Authors’ survey.
20 X. Chen et al.
Downloaded by [202.93.10.41] at 16:51 26 August 2015
school to another within the same level of schooling (that is: county to county; town to
town; or village to village – rows 4, 8 and 12), scores are higher in the case of county
schools (72.3) relative to those in town (55.4) and village schools (51.0). From these
numbers it would appear that the Merger Program, to the extent that schooling is being
concentrated in towns and counties, is benefitting rural students.
The propensity to score higher on the standardized test of the schools in the county seat
can also be seen in our data. The scores of students who either spent all six years in county
schools (transferring from county to county or staying in the same county school – rows 12
and 15) or finished their primary school years in a county school (teaching point to county;
village to county; town to county – rows 3, 6 and 9) all exceed 63 points. When students
did not attend a county school (with the rare exception of the few students who started in
county schools and finished in either a village or town school – rows 10 and 11), the scores
are all 59 points or fewer. The mean scores of students who attended county schools (rows
3, 6, 9, 12 and 15) are 66.2 points; the mean scores of students who did not attend county
schools (rows 1, 2, 4, 5, 7, 8, 13, 14) are 53.3 points. Taken together, this evidence suggests
that county schools foster the best academic performance among all types of schools.
Furthermore, using the same data, but examining different combinations of transfer
paths, the disruption effect of switching schools (that is, holding the level of schooling
constant) is not particularly evident. Specifically, when we compare the scores of students
who transferred from one county school to another county school, their scores are higher
(72.3 – row 12) than students who stayed in county schools for all six years of primary
school (66.0 – row 15). The town to town transferees also scored higher (55.4 – row 8)
than students who stayed in one town school for all six years (53.1 – row 14). The scores
of village to village transferees are almost identical to scores of students who stay in the
same village school (51.7 versus 51.0 – rows 4 and 13). From this descriptive analysis
there is even more support for the efficacy of the Merger Program; there are no empirical
grounds in our sample that suggests the disruption effect is serious.
2
However, somewhat surprisingly, the descriptive data in Table 2 may point to a
weakness in the Merger Program. One of the main targets of the Merger Program is to shut
down teaching points, but students may benefit when they start their primary schooling in
teaching points versus other types of schools. In the case of transfers from teaching points to
village schools (row 1), the scores of the students (56.2) are higher than when students are
either in the village to village (51.0) or village only (51.7) transfer paths (rows 4 and 13). In
the case of transfers from teaching points to town schools (row 2), the scores of the students
(58.7) also are higher than when students are in the town to town (55.4) or town only (52.1)
transfer paths (rows 8 and 14). One interpretation of these comparisons is that, in fact, there
may be some benefit to having young children (younger than fourth grade) stay in their own
village to go to school, rather than attending a school outside of the familiar environment of
one’s own home community. It could also be that the attention/care paid to students in a
teaching point up to Grade 4 is able to offset lower quality in teaching ability and/or less
broad course offerings. Our study, of course, can never tell us why. However, this evidence
raises the hypothesis that teaching points have strengths, and those strengths might need to
be preserved in teaching points or replicated in larger, centralized schools.
Boarding or living at home
One of the key components of China’s Merger Program is to allow students to board at school
as a way to enjoy the positivebenefits of greater resources incentralized schools. However, it
is possible that living and taking meals away from home is one of the large parts of the cost
Asia Pacific Journal of Education 21
Downloaded by [202.93.10.41] at 16:51 26 August 2015
associated with the disruption effect. Luo, Shi, et al. (2009) have documented the deficient
nature of boarding facilities. Luo, Zhang, et al. (2010) show that students in boarding facil ities
are less well nourished and have lower scores on standardized tests. Because of this fact, it is
important to try to isolate the boarding effect, given equal levels of the resource effect.
We can begin isolating the boarding effect by comparing the scores of boarding and
non-boarding students who went through the same transfer path. When looking at the data
in this way, we can see that 10 of the 15 different transfer paths can be used to make
comparisons. Observations on five of the transfer paths (teaching point to county; village
to county; town to county; county to county; and in the same county only) could not be
used since none of these students in the transfer path live at school (because in our three
sample counties, there were no boarding facilities in the county seat schools). Among the
other 10 transfer paths (those that involved schools in villages and towns and not schools
in the county seat), some students boarded whereas others lived at home.
Using this subset of data, descriptive statistics provide a mixed picture of the
relationship between boarding
3
and test scores (Table 3). There are six transfer paths
(teaching point to village school; teaching point to town school; village to village school;
village to town school; county to village school; and in the same town school only) in
which the scores of boarding school students are higher than the non-boarding school
students (rows 1, 2, 4, 5, 10 and 14). In the other four transfer paths (town to village school;
town to town school; county to town school; in the same village school only) the scores of
the non-boarding students are higher than the boarding students (rows 7, 8, 11 and 13).
Since there are more observations in the subset of six transfer paths in which the boarding
school students outperform the non-boarding school students, one might be inclined to
suggest that there is no detrimental effect to living at school.
Table 3. The maths scores of sample students who live at home and in boarding schools by the
transfer path that the students took during their primary school years in three study counties in China,
2009.
Boarding
a
Non-boarding
No. % Score No. % Score
Transfer paths (1) (2) (3) (4) (5) (6)
1 Teaching point to village school 163 8.5 57.9 145 4.0 54.9
2 Teaching point to town school 471 24.6 59.8 169 4.6 56.5
3 Teaching point to county school 0 45 1.2 62.1
4 Village (home) to village (away) school 46 2.4 53.9 180 4.9 50.7
5 Village to town school 191 10.0 55.0 201 5.5 53.2
6 Village to county school 0 91 2.5 66.5
7 Town to village school 11 0.6 49.1 37 1.0 56.5
8 Town(home) to town (away) school 29 1.5 53.6 50 1.4 57.5
9 Town to county school 0 72 2.0 68.1
10 County to village school 2 0.1 50.0 9 0.2 46.1
11 County to town school 7 0.4 57.14 8 0.2 62.5
12 County (home) to county (away) school 0 16 0.4 75.0
13 In the same village school 472 24.7 51.2 1235 33.6 51.9
14 In the same town school 479 25.0 52.7 885 24.1 51.8
15 In the same county school 0 532 14.5 66.2
16 Total 1,871 100 54.24 3675 100 56.0
Data source: Authors’ survey.
a
The boarding dummy variable equals one if the student boarded at the school where he/she finished primary
education, otherwise it equals zero.
22 X. Chen et al.
Downloaded by [202.93.10.41] at 16:51 26 August 2015
When looking at the entire sample, however, a different story emerges (Table 3, row 16).
Out of the total number of students in our dataset (5,546), 34% of them (or 1,871 students)
boarded at school in the year that they were finishing their primary school (that is, typically
during Grade 6). The other 66% (or 3,675 students) lived at home during their sixth year in
primary school. When comparing these two groups, the scores of the non-boarders (56.0) are
greater than the scores of the boarders (54.2). Hence, given the mixed nature of the results
using descriptive statistics, it is important to examine the results froma multivariate analysis.
Other effects and descriptive analysis summary
The educational performance of the students might also be affected by other
characteristics in addition to the transfer paths and boarding status. According to the
literature (Chen, Huang, Rozelle, Zhang, & Shi, 2009; Linnemayr, Alderman, & Ka, 2008;
Liu et al., 2010; Shariff, 1998, etc.) individual student characteristics, such as gender, age,
hukou identity, pre-school experience and number of siblings may also affect educational
performance. Parental characteristics, like age, education and occupation; and household
characteristics like household size and wealth also have been shown to affect educational
performance. These findings underline the importance of conducting multivariate analysis
and including parental and household characteristics in the analysis as control variables,
since they may also be correlated with student transfer paths.
In summary, then, most of the descriptive analysis appears to support the need to put
more resources into schools, as has been done in the town and county schools. In most
cases, students who have gone to schools in the higher levels of schooling have higher
scores. Furthermore, transferring per se also does not have any obvious, large disruption
effects. Hence, the resource effect seems to dominate. The one (potentially important)
exception is that the quality of schooling in teaching points in our sample counties may not
be so bad that teaching points should be shut down at any cost. In fact, students who
attended teaching points as part of their transfer paths often performed better in terms of
test scores than many of their counterparts. Overall, boarding appears to reduce
educational performance, but there are many transfer paths in which the boarding effect is
positive. Such findings in the descriptive statistics provide a rich empirical basis for
proceeding with the multivariate analysis in the rest of the paper.
Econometric estimation strategy
In this section we describe the estimation strategy for the econometric analysis (which will
be used to further examine the impact of student transfer paths on academic performance).
In the first subsection we present different estimators and specifications. In the second
subsection we discuss how we intend to perform robustness and sensitivity checks.
Basic estimator – OLS
In order to estimate the impact of student transfer paths and boarding status on math test
scores, we use OLS controlling (at least in part) for selection bias and endogeneity due to
observed heterogeneity. We do so by including a set of observable covariates in the
regression of key independent variables on math scores:
Yi¼
a
þ
b
0Piþ
g
Biþ
d
0Xiþ1ið1Þ
where the dependent variable Y
i
indicates the math score of student i;P
i
is a vector of our
variables of interest that includes 14 student transfer path dummy variables.
4
The variables
Asia Pacific Journal of Education 23
Downloaded by [202.93.10.41] at 16:51 26 August 2015
are the transfer paths that include students transferring: (a) from a teaching point school to a
village school; (b) from teaching point school to a town school; (c) from a teaching point
school to a county school; (d) from a village school to another village school; (e) from a
village school to a town school; (f) from a village school to a county school; (g) from a town
school to a village school; (h) from a town school to another town school; (i) from a
town school to a county school; (j) from a county school to a village school; (k) from a
county school to a town school; (l) from a county school to another county school; (m) stay
in the same town school and (n) stay in the same county school. The vector of parameters,
b
0,
contains measures of the effects of the transfer path that we are interested in. The
comparison group in this specification includes the students who studied in the same village
school for all six years of primary school.
In the rest of the equation, the symbol, B
i
, the boarding status indicator variable, is also
one of the other variables of interest. In Equation (1),B
i
is a dummy variable which equals
one if the student boarded at the school where he/she finished primary education and zero
if he/she did not board. Finally, the term, X
i
is a vector of covariates included to capture the
effect of the characteristics of the student, his/her parents, and household on the dependent
variable (see discussion in the previous paragraph)
5
.
Analysed in this way, we will be able to test more rigorously some of the observations
that were made in the descriptive analysis. Specifically, we can see, ceteris paribus, if math
scores of students in county schools are systematically higher than the scores of students in
town and village schools. The model can be used to assess the relative performance of
students who transferred compared to those who did not transfer (all other things being
equal). Additional tests can be carried out to measure the differences in scores of students
who attended teaching points and those who did not. Finally, holding the nature of the
transfer path constant, we will also be able to examine if living at school or home is
associated with higher test scores.
Alternative estimation approach – covariate matching
In place of controlling for the covariates by adding additional regressors as we do in the
OLS regression (described in the subsection immediately above), we can also use
covariate matching as an alternative method to estimate the effect of transfer path on
student educational performance. The main idea behind covariate matching is to select a
treatment group and comparison or control group with identical observables, X, and
compare the outcomes of these two sets of groups (Rubin, 1974). Given a certain set of
assumptions, covariate matching helps to correct the bias in estimation due to observables,
because the source of the bias is the difference of observables in the treated group and
comparison group. Matching on covariates by definition will remove this difference and
hence the bias (Zhao, 2004).
One of the main advantages of covariate matching over regression adjustment is that
this method highlights areas of the covariate distribution where there is insufficient
overlap between the treatment and control groups, such that the resulting treatment effect
estimates would rely heavily on extrapolation (Stuart, 2010). Regression models have
been shown to perform poorly when there is insufficient overlap, but their standard
diagnostics do not involve checking this overlap (Dehejia & Wahba, 1999; Glazerman,
Levy, & Myers, 2003). Another advantage of covariate matching is that it does not require
assumptions about linearity or constant treatment effects, and thus improves bias
correction (Zhao, 2004). Studies have shown that methods such as linear regression
adjustment can actually increase bias in the estimated treatment effect when the true
24 X. Chen et al.
Downloaded by [202.93.10.41] at 16:51 26 August 2015
relationship between the covariate and outcome is even moderately non-linear. Linear
regression can increase bias especially when there are large differences in the means and
variances of the covariates in the treated and control groups (Heckman, Ichimura, & Todd,
1998; Rubin, 1974). Thus, covariate matching is becoming a more general method than
standard linear regression. In our setup, the treatments are defined to be the different
student transfer paths and boarding statuses and we can use matching to examine the
difference in test scores among students who are in different subsets of transfer paths.
In summary, we are ultimately interested in estimating the average treatment effects on
the treated (ATT) – measured in test score differences – of attending a county school
(versus not attending county schools); of transferring to a new school (versus staying
within the same school); of starting primary education in a teaching point school (versus
not starting in a teaching point); and of living at school as a boarder (versus living at
home).
Econometric results
The estimation results of the basic estimator (OLS) using Equation (1) are presented in
Table 4. Columns (1) and (2) of Table 4 differ in the independent variables that are
included in estimation: column (1) only includes the student transfer path variables (with
no covariates); in column (2) we include the boarding status variable and all of the
covariates. In addition, columns (3) and (4) are almost the same as column (2) except that a
different comparison group of transfer path dummy variables are used in order to get the
disruption effect of switching schools directly. That is, in column (2), the comparison
group consists of students who studied in the same village school throughout primary
school. The comparison group in model (3) consists of students who spent all six years in
the same town school, and the comparison group in model (4) consists of students who
stayed in the same county school for all six years. The model performs better as we move
from column (1) to columns (2), (3) and (4) as the R-square grows and covariates are
shown to effectively capture more of the variation in math scores. Therefore, in the rest of
our discussion we mostly focus on the results in columns (2), (3) and (4).
Based on Table 4, there are four main results. First, consistent with the descriptive
analysis, the higher the level of the school the higher the test scores of students. Our results
show that math scores of students in county schools are systematically higher than scores
of students in town or village schools. Among students who never transferred to a new
school throughout primary education, students who stay in the same county school
perform better than students who stay in the same town or village schools. To be specific,
everything else held constant, a student who stayed in the same county school scores 9.4
(column 2, row 15) and 11.2 points (column 3, row 15) higher than his or her peers who
spent all six years in the village or town school, respectively.
Likewise, among students who transferred to a new school, students who transferred to
county schools perform better than those who started in the same school and later
transferred to village or town schools, regardless of whether students started primary
education in village or town schools. Specifically, among students who started primary
education in village schools, students who later transferred to county schools score 9.0
points (8.5 2[20.5] – column 2, rows 6 and 4) and 7.8 points (8.5– 0.8 – column 2, rows
6 and 5) higher than students who transferred to village or town schools.
6
This is almost
the same case for students who started primary education in village or town schools (rows
8 and 9).
7
In short, students attending county schools perform better than students
Asia Pacific Journal of Education 25
Downloaded by [202.93.10.41] at 16:51 26 August 2015
Table 4. Multivariate regression results analysing the effect of transfer path on the educational performance of all the sample students in three study counties in
China, 2009
a
.
Dependent variable: Standardized Maths score (0– 100 points)
Transfer path dummy variable (1)
a
(2)
b
(3)
c
(4)
d
1. Transfer from teaching point to village school, 1 ¼yes 4.591 2.325 4.078 27.116
(4.46)*** (2.26)** (3.90)*** (4.91)***
2. Transfer from teaching point to town school, 1 ¼yes 7.063 3.061 4.814 26.381
(9.13)*** (3.58)*** (5.64)*** (4.89)***
3. Transfer from teaching point to county school, 1 ¼yes 11.948 5.836 7.588 23.606
(5.73)*** (2.77)*** (3.62)*** (1.60)
4. Transfer from village school to village school, 1 ¼yes 20.679 20.533 1.219 29.975
(0.59) (0.40) (0.89) (5.90)***
5. Transfer from village school to town school, 1 ¼yes 2.263 0.776 2.528 28.666
(2.46)** (0.86) (2.75)*** (6.48)***
6. Transfer from village school to county school, 1 ¼yes 15.182 8.451 10.203 20.991
(8.67)*** (4.19)*** (5.07)*** (0.46)
7. Transfer from town school to village school, 1 ¼yes 3.032 22.244 20.492 211.686
(1.23) (0.81) (0.18) (4.00)***
8. Transfer from town school to town school, 1 ¼yes 3.694 20.374 1.379 29.816
(1.98)** (0.24) (0.90) (5.59)***
9. Transfer from town school to county school, 1 ¼yes 16.028 12.731 14.484 3.290
(8.41)*** (7.33)*** (8.39)*** (1.80)*
10. Transfer from county school to village school, 1 ¼yes 24.837 210.979 29.226 220.420
(0.95) (2.50)** (2.11)** (4.64)***
11. Transfer from county school to town school, 1 ¼yes 8.345 21.171 0.582 210.612
(1.91)*(0.29) (0.14) (2.61)***
12. Transfer from county school to county school, 1 ¼yes 20.595 17.114 18.867 7.673
(5.43)*** (3.62)*** (4.00)*** (1.62)
13. Study in the same village school 1.753 29.442
(2.72)*** (7.99)***
14. Study in the same town school 0.427 21.753 211.194
(0.70) (2.72)*** (9.86)***
15. Study in the same county school 14.321 9.442 11.194
(17.32)*** (7.99)*** (9.86)***
26 X. Chen et al.
Downloaded by [202.93.10.41] at 16:51 26 August 2015
Board dummy variable
16. Boarding status, 1 ¼boarded in the ending primary school 20.208 20.208 20.208
(0.36) (0.36) (0.36)
Student characteristics
17 Boy ¼1 1.624 1.624 1.624
(3.50)*** (3.50)*** (3.50)***
18. Age, year 23.476 23.476 23.476
(13.47)*** (13.47)*** (13.47)***
19. Hukou identity, 1 ¼rural 20.286 20.286 20.286
(0.47) (0.47) (0.47)
20. Kindergarten, 1 ¼attended 21.179 21.179 21.179
(1.40) (1.40) (1.40)
21. Having no sibling, 1 ¼yes 20.326 20.326 20.326
(0.49) (0.49) (0.49)
22. Grade dummy, 1 ¼grade 8 6.727 6.727 6.727
(13.26)*** (13.26)*** (13.26)***
Parental characteristics
23. Age of father, year 20.086 20.086 20.086
(1.15) (1.15) (1.15)
24. Age of mother, year 0.056 0.056 0.056
(0.71) (0.71) (0.71)
25. Father holding middle school diploma or above, 1 ¼yes 1.310 1.310 1.310
(2.37)** (2.37)** (2.37)**
26. Mother holding middle school diploma or above, 1 ¼yes 0.758 0.758 0.758
(1.13) (1.13) (1.13)
27. Father working in agriculture, 1 ¼yes 20.931 20.931 20.931
(1.77)*(1.77)*(1.77)*
28. Mother working in agriculture, 1 ¼yes 0.870 0.870 0.870
(1.63) (1.63) (1.63)
Household characteristics
29. Household size 20.418 20.418 20.418
(1.73)*(1.73)*(1.73)*
30. Poor dummy, 1 ¼yes 20.619 20.619 20.619
(Continued)
Asia Pacific Journal of Education 27
Downloaded by [202.93.10.41] at 16:51 26 August 2015
Table 4 – (Continued.)
Dependent variable: Standardized Maths score (0– 100 points)
Transfer path dummy variable (1)
a
(2)
b
(3)
c
(4)
d
(1.24) (1.24) (1.24)
31. Shiquan County, 1 ¼yes 6.306 6.306 6.306
(9.42)*** (9.42)*** (9.42)***
32. Hanyin County, 1 ¼yes 24.918 24.918 24.918
(7.58)*** (7.58)*** (7.58)***
Constant (not reported)
Observations 5,700 4,850 4,850 4,850
R-squared 0.08 0.20 0.20 0.20
Data source: Authors’ survey.
a
tstatistics in parentheses; *significant at 10%; **significant at 5%; ***significant at 1%.
b
The comparison group in model (1) and (2) is the students who studied in the same village school throughout the primary school.
c
The comparison group in model (3) is the students who studied in the same town school throughout the primary school.
d
The comparison group in model (4) is the students who studied in the same county school throughout the primary school.
28 X. Chen et al.
Downloaded by [202.93.10.41] at 16:51 26 August 2015
attending village or town schools. These findings may be interpreted as the positive
resource effect in county schools: better teachers and higher quality facilities.
8
However, although town schools are the main destination schools to which students
transferred under the Merger Program, our results show that attending town primary school
seems to have no positive effect on students’ academic performance. Specifically, among
students who spend all six years of primary education in a same school, students who study
in town schools for all six years score 1.8 points lower (column 2, row 14) than students
studying in village schools. Among students who started primary education in village
schools and later transferred to another school, students who finished in a town school score
1.3 points higher (0.8 2[20.5] – column 2, rows 4 and row 5) than those who finished in
another village school. However, the joint test of the coefficients of village to village school
transfer and village to town school transfer are not significantly different from zero. These
findings suggest that students who transferred to town schools actually do not benefit from
the Merger Program in terms of the test score and are consistent with Liu et al. (2010).
Second, our OLS results show that the disruption effect of switching schools on
students’ educational performance is neutral. That is, according to our results, the
coefficient of the dummy variable of transfers from village schools to another village school
is negative but statistically insignificant (20.533 – column 2, row 4). Compared with
students who studied in the same village school for all six years, students who transferred
from village schools to other village schools score almost the same. When we further look at
the pure impact of transferring to another town school, the coefficient of the town to town
school transfer dummy variable is positive but also statistically insignificant (1.379 –
column 3, row 8). The result also holds when we compare the scores of students who
transferred from a county school to another county school with the scores of students who
stayed in the same county school throughout their primary education (column 4, row 12). In
short, there is no evidence showing that there is a serious disruption effect of transferring to a
new school on poor rural primary students in terms of math test scores. This result holds
regardless of whether students transfer to another village, town or county school.
Third, although teaching points have long been considered to have the least resources
available, students who started primary education in teaching points perform better than
those who started primary education in village or town schools. That is, among students
who transfer to village schools, students who start primary education in teaching points
score 2.8 points higher than those who started primary education in village schools
(2.3 2[20.5] – column 2, rows 1 and 4).
9
Likewise, other things being equal, compared
to students on the town to town school transfer path, students who transfer from teaching
points to town schools score 3.5 points higher (3.1 2[20.4] – column 2, rows 2 and 8).
10
As such, students who transferred from teaching points seem to benefit from their
experience. A number of reasons may explain this finding. First, students in the teaching
points live close to home and are thus familiar with the surroundings, spend less time
going to and back from the school, and receive more care from their family. Second,
students in the teaching points are in a much smaller class than that in more centralized
schools. On average, in Ningshan County, there are only three students per class in
teaching points, but there are more than 12, 16 and 18 students in village, town, and county
schools respectively (Ningshan Education Bureau, 2010). Small class sizes may mean that
students in teaching points receive more care from the teachers. Especially because
younger students may need more attention than older students, the benefits of starting in
teaching points might offset and even outweigh the disadvantages of the lower quality of
teachers in teaching points. Finally, the quality gap in our sample counties between
teachers in teaching points and other schools have narrowed substantially since 2000. In
Asia Pacific Journal of Education 29
Downloaded by [202.93.10.41] at 16:51 26 August 2015
2000, a number of teachers were fired after failing some teaching examinations (Wang,
2008). Since most of the sample students started their primary education in teaching points
in 2001 and 2002, they benefitted from this measure. In addition, before 2000, the teachers
from the teaching points often attended special training courses. For example, during 1998
and 2000, every summer teachers from the teaching points received training on teaching
method for about a week in Ningshan County (Ningshan Education Bureau, 2010). Thus,
in terms of teacher quality, the gap between teaching point schools and village or town
schools may have narrowed in recent years.
Fourth, our results reveal that the effect of boarding is neutral on student educational
performance (Table 4). Holding other factors constant, the effect of boarding at school is
negative but statistically insignificant (row 16)
11
. Some reasons suggest why the neutral
effect of boarding status might be true. Although the main disadvantage of boarding at
school is poorer nutrition and health in boarding schools (relative to the home
environment) and less care from parents (Luo, Shi, et al., 2009), parents are migrating
from rural areas. As such, children who live at home may be receiving minimal care
anyway. Thus, compared to students who live at home, the cost of boarding at school in
terms of reduced parent care might not be high.
In summary, our results from basic estimation show that although transfer per se does
not have a serious effect on student educational performance, transferring to county
schools systematically benefits rural students in terms of test scores. Second, transferring
to town schools seems not to have any positive effects on student educational performance
as the Merger Program expected. Third, studying in teaching points through Grade 1 to 4
does seem to improve rural students’ educational performance. Fourth, there is no
evidence in our analysis that boarding at school improves student educational
performance. Finally, our covariates affect student educational performance as expected.
Covariate matching
Importantly (since they show our results are robust), the results of covariate matching
analysis are similar to the OLS results (Table 5). First, consistent with OLS results, the
covariate matching results reveal the dominance of county schools. Compared to students
who remain in their village or town schools throughout primary education, students who
study in the same county school for all six years or transfer from village to county schools
score 14.6 points and 7.6 points higher, respectively (Panel A, columns 1 and 2). Village
school students who transfer to a county school score 11.8 points higher than village
school students who transfer to another village school (Panel A, column 3). Likewise,
town school students who transferred to a county school score 9.0 points higher than town
school students who transferred to another town school (Panel A, column 4). As such, the
resource effect in county schools is positive.
However, as found in the OLS analysis, the resource effect in town schools is not
evident. Students who spent all six years in town schools scored 1.7 points lower than
students who spent all six years in village schools (Panel B, column 1). Although students
transferring from village to town schools score 1.5 points higher than students taking
village to village school transfer path, the coefficient is not statistically significant (Panel
B, column 2). That is, students attending town schools do not appear to benefit from the
resource effect compared to students attending village schools.
Second, the result from covariate matching also reveals that the effect of transferring
between schools of the same type is neutral. That is, compared to students who study in the
same village, town, or county school throughout primary education, students who transfer
30 X. Chen et al.
Downloaded by [202.93.10.41] at 16:51 26 August 2015
Table 5. Covariate matching results analysing the effect of transfer path and boarding status on
student educational performance (Shaanxi Province, China 2009)
a,b
.
Average treatment effect on the treated (ATT)
Panel A. The effect of attending the primary school in a county school
(1) (2) (3) (4)
Students in the same
county school
vs.
Students in the same
village school
Students in the same
county school
vs.
Students in the same
town school
Village to county
school students
vs.
Village (home) to village
(away) school students
Town to county
school students
vs.
Town (home) to town
(away) school students
14.57 7.58 11.84 8.98
(4.79) *** (3.55) *** (3.52) *** (3.33) ***
Panel B. The effect of attending the primary school in a town school
Students in the same
town school
vs.
Students in the same
village school
Village to town
school students
vs.
Village (home) to village
(away) school students
21.69 1.51
(2.16) ** (0.86)
Panel C. The pure disruption effect of transfer
Village (home) to village
(away) school students
vs.
Students in the same
village school
Town (home) to town
(away) school students
vs.
Students in the same
town school
County (home) to county
(away) school students
vs.
Students in the same
County school
20.65 0.65 6.65
(0.43) (0.31) (1.36)
Panel D. The effect of starting the primary education in a teaching point
Teaching point
to village school students
vs.
Students in the same
village school
Teaching point to
town school students
vs.
Students in the
same town school
1.42 5.64
(0.68) (2.10) **
Panel E. The effect of boarding at school
Boarding students
vs.
Non-boarding students
Boarding students
who studied in the
same village schools
vs.
Non-Boarding students who
studied in the same
village schools
Boarding students
who studied in the
same town schools
vs.
Non-Boarding students
who studied in the
same town schools
22.13 23.32 20.55
(3.17) *** (2.97) *** (0.43)
Data source: Authors’ survey.
a
zstatistics are reported in parentheses; *significant at 10%; **significant at 5%; ***significant at 1%, and
nearest neighbour matching was used in matching.
b
In each model, the treatment group is students before the term of “vs.” and the comparison group is students after
“vs.” For example, in model (1) in Panel A, the treatment group is students who studied in the same county school
for all six years throughout their primary school and the comparison group is students who studied in the same
village school for all six years throughout their primary school.
Asia Pacific Journal of Education 31
Downloaded by [202.93.10.41] at 16:51 26 August 2015
to a new village, town, or county school at the same administrative level perform about the
same (Panel C, columns 1, 2, and 3).
Third, the covariate matching result reveals that starting primary education from a
teaching point positively affects student educational performance, regardless when the
student later transferred to a village or town school (Panel D, columns 1 and 2). In the case
of students who transfer to a town school, students who start primary education in a
teaching point score 5.6 points higher than students who start primary education in a town
school. The coefficient is statistically significant (Panel D, column 2).
Fourth and finally, our results show that boarding at school does not help student
educational performance. Departing slightly from OLS findings, covariate matching
results show that on average, students boarding at school score 2.1 points lower than
students who lived at home. This result is statistically significant (Panel E, column 1).
Among students staying in their village schools for six years, students boarding at school
score much lower (3.3 points) than students living at home. This result is significant as
well (Panel E, column 2). In addition, among students staying in town schools for six
years, students boarding at school score 0.55 points lower than students living at home.
However, this result is insignificant (Panel E, column 2). Taken together, these findings
suggest that boarding does not help students improve their educational performance. In
some cases boarding clearly reduces students’ academic performance.
Discussion and conclusions
In this paper we have attempted to understand whether poor students in rural China have
benefitted from the Merger Program by analysing a set of transfer paths that students have
taken during their primary education. Despite controversies about the benefits and costs of
the Merger Program, our results show that students who attend county schools perform
systematically better than those who attend village or town schools. However, completing
primary school in town schools seems to have no effect on students’ academic
performance. Surprisingly, starting primary education in a teaching point does not hurt the
rural students; on the contrary, it increases their test scores in some cases. Finally, in terms
of the boarding effect, the neutral estimate in OLS and the negative estimate in covariate
matching results confirm that boarding at school does not help the students; in some cases
it may even reduce their academic performance.
Although there may be good reasons (fiscally or pedagogically) for the changes
(mergers/building up centralized county schools), our results imply that poor students are
being systematically hurt by the rural China’s educational system reforms. First, we find
that county schools foster the best academic performance among all the schools in rural
areas. However, there are no boarding facilities in these county primary schools, so poor
students have no real way of attending. If they want to attend, families must rent rooms;
students’ guardians may have to quit their jobs and live with them. These arrangements are
usually too costly for poor households, who thus opt to send their children to town or
village schools. However, the effect of completing primary education in town or village
schools is neutral. That is, students do not benefit from transferring to such schools.
Furthermore, although rural poor students do well when starting in teaching points, these
teaching points are being shut down. The students in the teaching points leave their
familiar circumstances and most of them now study in village or town schools further
away. Finally, for poor rural students, village and town schools are mostly far from home
and they have to board at school. However, our results show that boarding does not help
the poor rural students.
32 X. Chen et al.
Downloaded by [202.93.10.41] at 16:51 26 August 2015
In terms of policy, our paper has several implications. First, the results confirm that
transferring to the county schools helps poor, rural students. However, these are largely
unavailable to poor students. Policies that improve access to county schools (or schools
with resources similar to county schools) might improve the education quality for poor
rural students. Second, it seems that the positive resource effect that the government was
hoping to achieve by the Merger Program does not apply to most rural students, who are
transferring to village or town schools. If the government continues merging village
schools (including teaching points) into town schools, it should invest more in town
schools to increase the teaching quality (and facilities) in these schools. Third, village
schools like teaching points may have redeeming qualities and should not be closed
without further investigation of their benefits. Finally, boarding at school does not help
poor rural students; if a way could be made to address the negative effects associated with
boarding – poor nutrition, lack of familiarity, and less personal care – students might
benefit from increased access to learning resources and facilities.
Notes
1. In 1994, China’s government launched a poverty-reduction initiative under the “8 – 7 Plan” and
designated 592 counties as national designated poor counties. Provinces followed with their own
initiatives.
2. Of course, it is possible that there still is a negative disruption effect, but, that the gain in test
scores is due to some sort of selection effect (i.e., better students – who have higher test scores –
were the ones who sought to move from poorer schools to better schools) and that this selection
effect was high enough to more than offset any disruption effect.
3. In later analysis, the boarding dummy variable equals one if the student boarded at the school
where he/she finished primary education, and it equals zero if the student did not board at the
school where he/she finished primary education. In rural areas, some students might rent rooms
near school due to the unavailability of dorms in the school, and they are regarded as those who
did not board at school. We also tried defining boarding status as the boarding dummy variable
equals one if the student ever boarded in primary school and it equals zero if he/she never
boarded. The results are more or less the same.
4. It is possible that some students may have transferred more than once. We try to correct for this
possibility by controlling the student, parental and household characteristics that may affect
transfer decision. We do not account for it explicitly because any extra transfer as a result of the
Merger Program is also part of the transfer effect we intend to estimate.
5. In the analysis, we are not able to control for the “reasons of moving”. Therefore, the reader
needs to be aware of this fact and exercise caution in interpreting the results. In fact, we are
aware that this is a problem, because it may be that parents who moved for better educational
opportunities are overachievers. We did not control (and it is almost impossible to control for)
things such as overachieving. Hence, we admit we can never be sure that we fully controlled for
these types of factors. However, this is precisely why we use covariate matching. Under the right
circumstances (key unobservables are correlated with observables), our approach will account
for factors such as these. This approach would also help us control for the selectivity of
differential rates of dropping out of elementary school. Fortunately, for China (and our analysis),
few children (elementary school students) in the areas of our study are dropping out.
6. The joint test of coefficients of village to village school transfer dummy and village to county
school transfer dummy shows that the two coefficients are significantly different from each other
at 1% level. And, the joint test of the coefficient of village to town school transfer dummy and
the coefficient of village to county school transfer dummy shows that the two coefficients are
significantly different from each other at 1% level.
7. The joint test of coefficients of town to town school transfer dummy and town to county school
transfer dummy shows that the two coefficients are significantly different from each other at 1%
level.
8. It should be pointed out that the results do not fully hold for students who started primary
education in teaching points. Although students transferring from teaching points to county
Asia Pacific Journal of Education 33
Downloaded by [202.93.10.41] at 16:51 26 August 2015
schools score 3.5 points (5.8– 2.3 – column 2, rows 3 and 1) and 2.7 points (5.8 – 3.1 – column
2, rows 3 and 2) higher than students transferring from teaching points to village or town
schools, the joint test of the coefficients shows that they are not significantly different from each
other. Specifically, the joint test of coefficients of teaching point to village school transfer
dummy variable and teaching point to county school transfer dummy variable shows that the two
coefficients are not significantly different from each other. And the joint test of coefficients of
teaching point to town school transfer dummy variable and teaching point to county school
transfer dummy variable also shows that these two coefficients are not significantly different
from each other.
9. The joint test of coefficients of teaching point to village school transfer dummy variable and
village to village school transfer dummy variable shows that the two coefficients are
significantly different from each other at 10% level.
10. The joint test of coefficients of teaching point to town school transfer dummy variable and town
to town school transfer dummy variable shows that the two coefficients are significantly
different from each other at 5% level.
11. While the negative coefficient on the boarding school variable may suggest that the poor
conditions of the boarding school (or emotional stress of living away from home) may lead to
poor educational performance, there are other possibilities. In some places in our sample (and
elsewhere in China), parents may rent rooms near the school and live with their children instead
of putting their children into boarding schools. Sometimes it is because there is not enough room
in the boarding schools (though this is becoming less of a problem, at least in our study area);
sometimes it is by choice of the parent. If students choose to rent instead of boarding, it is
possible that such students may be overachievers, and such selectivity may influence our results.
Hence, the negative coefficient may be measuring this additional effect. Unfortunately, we did
not ask such questions in our survey so we cannot control for this directly. We do not believe this
to be a major part of the effect: in our study area, it is uncommon for parents to rent rooms near
the school. However, the possibility does exist.
References
Chen, X., Huang, Q., Rozelle, S., Zhang, L., & Shi, Y. (2009). Effect of migration on children’s
educational performance in rural China. Comparative Economic Studies,51(3), 323 – 343.
Dehejia, R. H., & Wahba, S. (1999). Causal effects in nonexperimental studies: Re-evaluating the
evaluation of training programs. Journal of the American Statistical Association,94(448),
1053– 1062.
Glazerman, S., Levy, D. M., & Myers, D. (2003). Nonexperimental versus experimental estimates of
earnings impacts. The Annals of the American Academy of Political and Social Science,589(1),
63– 93.
Heckman, J. J., Ichimura, H., & Todd, P. E. (1998). Matching as an econometric evaluation
estimator. Review of Economic Studies,65(2), 261– 294.
Linnemayr, S., Alderman, H., & Ka, A. (2008). Determinants of malnutrition in Senegal: Individual,
household, community variables, and their interaction. Economics & Human Biology,6(2),
252– 263.
Liu, C., Zhang, L., Luo, R., Rozelle, S., & Loyalka, P. (2010). The effect of primary school mergers
on academic performance of students in rural China. International Journal of Educational
Development,30(6), 570– 585.
Luo, R., Shi, Y., Zhang, L., Liu, C., Rozelle, S., & Sharbono, B. (2009). Malnutrition in China’s rural
boarding schools: The case of primary schools in Shaanxi province. Asia Pacific Journal of
Education,29(4), 481–501.
Luo, R., Zhang, L., Liu, C., Rozelle, S., Sharbono, B., & Adams, J. (2010). Behind before they start:
The challenge of early childhood education in rural China. REAP working paper No. 209,
CCAP, Beijing, China.
Ministry of Education. (2010). China educational statistics yearbook. Beijing: China education press.
National Bureau of Statistics. (2000). China statistics yearbook. Beijing: China statistics press.
National Bureau of Statistics. (2010). China statistics yearbook. Beijing: China statistics press.
Ningshan Education Bureau. (2010). The annual statistics report on education. Report, Ningshan.
Pang, L. (2006). Primary and high school consolidation in rural China: Issues, reasons, and solutions.
Education Development Research,2,1–6.
34 X. Chen et al.
Downloaded by [202.93.10.41] at 16:51 26 August 2015
Rubin, D. B. (1974). Estimating causal effects of treatments in randomized and nonrandomized
studies. Journal of Educational Psychology,66(5), 688– 701.
Shariff, A. (1998). Women’s status and child health. In M. Krishnaraj, R. M. Sudarshan & A. Shariff
(Eds.), Gender, population and development (pp. 185 – 219). New Delhi: Oxford University
Press.
Stuart, E. A. (2010). Matching methods for causal inference: A review and a look forward. Statistical
Science,25(1), 1–21.
Wang, Y. (2008). Current composition of teachers in primary and junior high schools in Ningshan
and policy implications. Retrieved from http://rd.ankang.gov.cn/Article/Class11/200801/776.
html
Zhao, Z. (2004). Using matching to estimate treatment effects: Data requirements, matching metrics,
and Monte Carlo evidence. Review of Economics and Statistics,86(1), 91 – 107.
Zhuo, M. (2006). Primary school consolidation in rural China: Advantages, disadvantages and
solutions. Education Forum, the Education Department of Fujian Province.
Asia Pacific Journal of Education 35
Downloaded by [202.93.10.41] at 16:51 26 August 2015