Secondary School Madrasas in Bangladesh:

Incidence, Quality, and Implications for Reform

Mohammad Niaz Asadullah
Department of Economics, University of Reading &
Department of Education, University of Oxford

Nazmul Chaudhury
The World Bank

Syed Rashed Al-Zayed Josh
The World Bank





Draft: March 15, 2009









Human Development Sector
South Asia Region
ii
Acknowledgements


This study was made possible by generous funding from the EPDF Trust Fund. We also
express our appreciation for additional funding and institutional support from the Education
Unit of South Asia Human Development, World Bank as well as the Bangladesh Country
Management Unit. The authors would like to highlight the services provided by the survey
firm, DATA, who conducted this complex survey with utmost professional competence.
Furthermore we are grateful to DATA for providing valuable statistical input in preparing
this report. The authors are grateful from valuable comments and feedback received from the
Ministry of Education and Economic Relations Division, Government of Bangladesh. The
authors would also like to thank Rafiqul Huda Chaudhury, Markus Goldstein, Michelle
Riboud, and Halsey Rogers for their valuable comments.

This report does not necessarily reflect the views of the World Bank or the Governments they
represent.
iii
Executive Summary

1. A unique feature of the Bangladeshi secondary education sector is the large presence of
Islamic institutions of religious learning, commonly known as madrasas. However, unlike
other countries in the region with large Muslim populations, the religious education sector
comprises of both state regulated private madrasas as well as independent, private madrasas.
The former are popularly known as Aliyah madrasas where alongside Islamic education,
modern general education is also provided. Given that majority of these private registered
madrasas operate with state funding, they are regulated in terms of curriculum content and
teacher recruitment policy under a unified state recognized Madrasa Education Board. On the
other hand, an unknown number of private, traditional madrasas exists outside the state
sector. These seminaries specialize in religious education and are popularly known as
Quomi madrasas.

2. Bangladesh has a long history of pragmatic reforms of madrasas at the secondary level
which sets it apart from other countries in South Asia, and elsewhere. First it introduced
fiscal incentives to traditional unregistered all-male madrasa high-schools to register and
include modern subjects such as mathematics. Then it introduced another financial incentive
to registered madrasa high schools to start admitting female students. Most madrasa
secondary schools in Bangladesh are now registered, follow a modern curriculum alongside
traditional religious subjects, and have become coeducational (50% of the enrolment in
Madrasa high-schools are now females) an unparalleled development in South Asia so far.

3. Despite having made considerable progress in increasing access to secondary education
over the past decade, particularly for girls (gender parity has now been reached in secondary
schooling), achieving further progress under the prevalence of widespread poverty requires
reaching out to difficult-to-reach children. However, if there is a positive link between
poverty and religious education this may further increase the share of pupils enrolled in
religious institutions. Therefore, a clear understanding of the religious school sector is
necessary if this institution is to serve as an instrument for the rural population in Bangladesh
to escape future poverty traps. Nonetheless, for at least three reasons, there is a serious policy
concern regarding the provision of education through the institution of madrasas. First, these
schools may promote skills that are incompatible with a modern economy. The academic
standard attained is popularly perceived to be much lower than in general education.
Consequently, it is believed that madrasa graduates fail to pursue a productive economic life
and therefore add to the pool of educated unemployed. With little learning taking place,
current attendance can only reinforce the curse of poverty in the future. Second, they may not
promote civic values that are essential for a functioning democracy, thereby causing concerns
among policy makers with an interest in the relationship between education and citizenship.
Third, an unknown fraction of religious schools still remains completely outside the state
system, making it a daunting task to regulate curriculum content and pedagogic practices.

4. For the above reasons, many hold non-state provision of education by Islamist groups and
the expansion of madrasas as a contributing factor to radicalization. These concerns are
extremely relevant in the age of globalization as Bangladesh today competes with other
developing countries in the international market. Education system should not only create a
skilled workforce it also needs to reduce social polarization in order to guarantee a stable
political environment. Therefore, it is vital for the government to identify the strengths and
weakness of the educational structure of the country in order to improve the skill set of the
workforce on one hand, while improve social integration on the other.
iv

5. Given certain deficiencies in the existing nationally representative household and
administrative data, the World Bank (WB) initiated a detailed survey to integrate both
demand and supply side information on secondary schooling to explicitly focus on the
incidence and quality of madrasa schooling in Bangladesh. To this end, in the year 2008 the
World Bank commissioned a multi-purpose sample survey Quality of Secondary School
Madrasa Education in Bangladesh (QSSMEB) - to gather detailed information on all types of
secondary institutions in rural Bangladesh. A full census of all educational institutions was
carried out in sample unions (primary sampling units) where secondary school quality was
assessed by employing a battery of cognitive tests. Apart from detailed survey of secondary
school and madrasa students, complete household census was carried out in sample villages,
followed by a detailed survey of households with children of secondary school-going age.
The research effort therefore combines data on educational institutions, teachers, students,
households, parents, and children, from multiple inter-linked sources census of primary and
secondary schools, survey of secondary schools, household survey and census in order to
provide a detailed assessment of the secondary education with an emphasis on the institution
of madrasa.

6. The two primary objectives of this report were to address the issues of:

Incidence (how large is the madrasa sector?)
Quality of educational institutions operating in rural Bangladesh (are
learning outcomes poorer in madrasas compared to secular schools?)
This is the first ever comprehensive survey on the size, structure and quality of religious
schools in a country with a large Muslim population using data from Bangladesh. The
comprehensive nature of the data collected allows us to not only present a detailed picture of
the madrasa education system in the country, but also to draw comparison with the
mainstream secular state and state-aided education sectors. Given the diversity of the
Bangladeshi education sector, the data collected, will help to dispel or confirm many popular
myths about madrasa education. Household based survey data will be used to address issues
in both the primary and secondary sector, while facility-based survey data will focus on the
secondary sector. Given the vastness of the data collected at different levels (ranging from
learning assessment to social and political preferences), it is not possible for one report to
contain all the possible descriptions and analysis. Thus, in this initial report we focus on the
incidence of madrasa enrolment and quality of public-aided madrasas (relative to public-
aided schools).

Findings from QSSMEB Survey

Incidence

7. The number and share of Quomi madrasas in both the primary and secondary
sector is much lower than what is portrayed in the popular press. Consistent with
existing nationally representative household data, we find that even in terms of primary
enrolment share, incidence of Quomi madrasa is not large in rural Bangladesh - they account
for only 1.9% of the total primary enrolment (Figure E2). These numbers are very small
when compared to enrolment in a similar non-religious, non-state school that also caters to
children from poor families Non-Government Organization (NGO) schools account for
8.2% of the primary enrolment in our study. Share of Aliyah madrasa, however, is
v
significant: They account for 8.4% of the total primary enrolment. Overall madrasa enrolment
as a share of total primary enrolment is large (i.e. 13.8%) when we take into account
enrolment in other madrasas that are non-formal in nature and offer exclusively religious
education. Thus, while as a group the number of madrasas might be less than the number of
NGOs, the share of enrolment in madrasas is significant and larger than the NGO share.
Turning to the secondary education segment, Quomis have a slightly higher share, accounting
for 2.2% of the total enrolment. This is also true for Aliyahs (who have an enrolment share of
almost 19%).

Source: Authors calculation based on data from QSSMEB

Quality of Learning from the Institutional Survey

8. Patterns in Mathematics and English tests in public-aided secondary schools and
madrasas show that:
Learning achievement is poor across institution types
Schools outperform Aliyahs
Girls systematically score lower than boys
9. We administered four distinct cognitive tests to 8
th
grade students during the institutional
survey. First, a Math test instrument was constructed by using 25 items previously used in the
Trends in International Mathematics and Science Study (TIMSS). On the basis of the national
curriculum we devised a test to assess proficiency in English (20 items), General Knowledge
(6 items) and Islamic studies (10 items). The test on Islamic studies, however, was restricted
to Muslims only. Given that Quomi education is structured in a pedagogic structure that is
very different from schools and Aliyahs, it was difficult to identify 8
th
grade equivalence in
Quomis, and hence Quomis were excluded from the analysis of learning outcomes.
10. Figure E3 reports % of question items correctly answered by our sample students across
institution type, gender and tests. For presentational ease, we use the following groupings:
schools (government schools, private schools, private aided-schools) and Aliyah madrasa
(aided and unaided). On average, only 35% of the math test questions could be correctly
answered by students in our sample. This number is even lower (i.e. 33%) for English.
Interestingly, the performance is far from satisfactory even if we consider performance in the
general knowledge test: only 39% of the questions were correctly answered on average.
However, when assessed in matters related to religion (i.e. Islam), the performance was very
satisfactory: students on average correctly answered 75% of all the test items.
Figure E2a: Distribution of primary enrolment share of
educational institutions by type
8.4
1.9
3.5
59.7
18.3
8.2
Aliyah
Quomi
Other madrasa
Government
Non-government
NGO
Figure E2b: Distribution of secondary enrolment
share of educational institutions by type
18.5
2.2
1.4
6.9
70.9
0.2
Aliyah
Quomi
Other madrasa
Government
Non-government
NGO
vi
Figure E3: Distribution of test score by school type and gender
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
M
a
t
h
s
M
a
t
h
s

(
M
a
l
e
)
M
a
t
h
s

(
F
e
m
a
l
e
)
E
N
G
E
N
G

(
M
a
l
e
)
E
N
G

(
F
e
m
a
l
e
)
G
K
G
K

(
M
a
l
e
)
G
K

(
F
e
m
a
l
e
)
I
s
l
a
m
I
s
l
a
m

(
M
a
l
e
)
I
s
l
a
m

(
F
e
m
a
l
e
)
Test score by type & gender
F
r
a
c
t
i
o
n

o
f

a
n
s
w
e
r
s

c
o
r
r
e
c
t
General school
Aliyah Madrasa
Source:
Authors calculation based on data fromQSSMEB. Data corresponds to grade 8 students fromsample secondary schools and
madrasas.

11. The figure above only reports the fraction of correct answers in different subjects. As
such, it is not obvious whether the questions have been set at a reasonable degree of
difficulty. The state of poor-quality becomes even more evident if we consider few examples
of student performance in individual questions where a majority of students were unable to
answer. For instance, let us focus on two simple questions related to division and fractions.
Only 62% of school students and 54% of Aliyah students could correctly answer the
following question: Divide:
15
4
35
8

. By 8
th
grade, only 11% of school children and 8% of
Aliyah children can correctly divide 24.56 by 0.004. A simple measurement test Which of
these is the longest time? {(a) 15000 seconds (b) 1500 minutes (c) 10 hours (d) 1 day} is too
difficult for 46% of grade 8 children in our sample. When disaggregated by institution type,
only 56% and 51% of the school and Aliyah students, respectively, could correctly answer
the question.
Distribution of Test Scores
12. There is considerable variation in quality within schools and Aliyahs, even greater
than the difference between types of institution. We then take a closer look at the full
distribution of test scores across students instead of institutional averages. Even within
institution type, there might be considerable variation in learning outcomes. Figure E4 below
plots individual-level test score by institution type across the entire distribution of the test
score. For the sake of brevity, our discussion focuses on two extreme points of the test score
distribution, the bottom and top deciles. We tested over 9,311 pupils in Grade 8 where 73%
and 27% of the students belong to school and Aliyah madrasa, respectively. If there are not
major differences in the distribution of test scores between institution-type, then the fraction
of a given institution-type in a particular decile should be similar to its sample fraction.
vii

Source: (a) Authors calculation based on data fromQSSMEB. (b) Data corresponds to grade 8 students fromsample
secondary schools and madrasas. (c) For each decile/quartile, mean score is reported in the parenthesis.

13. First note that 12% of our sample students belong to the bottom 10% (with a mean score
of 2.9 out of 25) of the Math score distribution while only 7% belong to top 10% (with a
mean score of 16.8 out of 25). This means that on average, the bottom 10% could correctly
answer only 12% of the Maths questions while even the top 10% could only correctly answer
66% of the questions. Now, the relative share of school and Aliyah students in the bottom
10% population is 60% and 40%, respectively. This suggests that students of Aliyah
madrasas are somewhat excessively represented in the lowest performing declie. On the other
hand, the relative share of school and Aliyah students in the top 10% population is 73% and
27% respectively - this suggests that both institutions are proportionally represented in the
top decile (relative to their sample fraction).
14. The above analysis suggests that the distribution of individual Math test score is slightly
skewed to the left for Aliyahs and normal for general schools. The non-normal nature of the
test score distribution of Aliyah students is more evident if we look at the English test score
data. The relative share of school and Aliyah students in the top 10% student population is
90% and 10%, respectively. This suggests that almost all of the top scorers in English come
from schools, while Aliyah students are grossly under-represented in the top decile (relative
to their sample fraction). The distribution of student performance in English test is therefore
skewed to the right for schools and to the left for Aliyah.


Figure E4b: English score (decile distribution by institution type)
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
1 (7) 2 (15) 3 (20) 4 (25) 5 (30) 6 (38) 7 (45) 8 (50) 9 (57) 10 (73)
Test scor e deciles
D
e
c
i
l
e

s
a
m
p
l
e

f
r
a
c
t
i
o
n
School
Aliyah madrasa
Overall sample
share: school
Overall sample
share: Madrasa
Fi gure E4a: Math score (deci l e di stri buti on by i nsti tuti on type)
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1 (12.1) 2 (20) 3 (24) 4 (28) 5 (32) 6 (36) 7 (40) 8 (46) 9 (54) 10 (66)
Test scor e deciles
D
e
c
i
l
e

s
a
m
p
l
e

f
r
a
c
t
i
o
n
School
Aliyah madrasa
Overall sample
share: school
Overall sample
share: Madrasa
viii
Correlates of Test Scores
15. Irrespective of the regression specifications, the type of control variables used,
there is a clear statistically significant learning disadvantage associated with madrasa
education. The madrasa penalty is largest (i.e. 10 percentage points) in case of English. Apart
from test score gaps by institution type, girls suffer a significant learning disadvantage -
average gender gap in Math score is much larger than the average negative gap associated
with any type of madrasa schooling.
16. Institutions matter in educational production in rural Bangladesh but little is
known about what separates a good institution from a bad institution. A large part
of the test score variation is attributed to between-institution differences. Even within the
madrasa and school sectors, most of the variation in test score is a between-institution
phenomenon. Yet, only a small fraction of the institution effect can be explained by
conventional indicators of quality such as institutional type, teacher characteristics and
resources available to the institution for cognitive production.
Findings from national household surveys and institutional census
17. Irrespective of demand and supply, there appears to be a modest link between
Aliyah madrasa education and poverty. On the supply-side, Aliyah madrasas are found to
concentrate in poor areas although there is no evidence that they arise in regions which are
under-provided by the state in terms of availability of mainstream schools. On the demand
side, children from poor households and poorer regions are more likely to enrol in madrasas.
The income effect is, however, small in terms of magnitude indicating that there are other
major factors besides household wealth that influences madrasa enrolment.
18. Regardless of what factors ultimately shape household decision to send their
children to madrasas, the incidence in terms of enrolment share of unregistered
traditional madrasas in primary and secondary education is very small. According to data
fromrecent round of national household survey, 1.2% and 1.3% of primary and secondary
aged population attend Quomi madrasa in Bangladesh respectively. This conclusion is also
supported by evidence available for other sources for South Asian countries: available
estimates for India and Pakistan range between 1% and 3%.
Box 1: Summary of Key Findings

Madrasas are more likely to be found in poor rural areas
While poorer households are more likely to send their child to a madrasas, there are other factors
more important than income which shape household preferences for religious schooling
Madrasa enrollment has grown rapidly, particularly for girls
Despite this growth, the overall share of enrollment in traditional madrasas is very small;
however, the share of enrollment in registered pubic-aided Aliyah madrasas is significant
Quality of learning (in terms of Math and English) is relatively lower in Aliyah madrasas
Overall quality of learning in terms of Math and English scores is low across the board
reflecting the poor quality of all types of education service providers in rural Bangladesh
Girls do worse than boys, particularly in Math; this gender-gap in learning outcomes is,
however, more pronounced in Aliyah madrasas.
There is strong between and within institutional influence despite overall low learning base.
Within the madrasa/school sector, the difference in average Math score of top 10% and bottom
10% students is 42 percentage points. The top-bottom difference is also sizable in case of
English score -- 28 and 25 percentage points in case of school and madrasa sectors respectively.
ix
Recommendations
19. Some key recommendations arising out of the study are:
There is need for a more nuanced policy discussion regarding madrasas reform we must
distinguish between Aliyah and Quomi madrasas not only for syntactic reasons but also
for policy relevance. Private Aliyah madrasas, funded by the government and regulated
by a state-sanctioned Board, have already reformed their curriculum to include modern
subjects similar to those taught in secular schools. On the other hand, private unaided
traditional Quomi madrasas still have a major difference in curriculum content and
pedagogic structure. While Aliyah madrasas have a significant presence in both the
primary and secondary sector, the share of children studying in traditional Quomi
madrasas is still miniscule in comparison.
There is need to reduce the quality gap between public-aided secondary schools and
Aliyah madrasas. This finding also highlights the shortcomings of only relying upon
Secondary School Certificate (SSC) pass rates as an indicator of quality. The SSC pass
rate is higher for aided-Aliyah madrasas compared to aided-schools. However, we find
that in terms of assessment of actual numerical and literacy skills, Aliyahs fare worse than
schools. Currently the SSC examination system is under different regional and
institutional boards making it difficult to compare quality across providers using SSC
exam scores.
The discussion on quality should not only be about Aliyahs raising their standards to
match schools both have to be held accountable to improving basic numeracy and
literacy skills. The punch line of this report remains that quality of schooling in rural
Bangladesh is low regardless of institution type. The Government should attempt to find
more innovative ways to link substantial public resources that it gives these aided private
institutions, religious or otherwise, with concurrent improvements in numeracy and
literacy skills.
There is need to empirically anchor the debate about madrasa reform in the metrics of
learning outcomes. Currently there is no systematic assessment of basic literacy and
numeracy skills. Learning assessments will not only help to shed light on the madrasa
debate, but help to address other important issues such as gender and regional disparities
as well. For example, we find that the average gender gap in Math score is much larger
than the average negative gap associated with any type of madrasa schooling. This
gender gap is most pronounced in madrasas. Thus, while Aliyah madrasas have played
an instrument role in reducing gender inequality in access, the Government should ensure
that they are also not inadvertently increasing gender inequality in learning.
Quomi madrasas are structurally different than schools and Aliyah madrasas (e.g.,
curriculum, gender composition, classroom organization). Initiatives to reform traditional
Quomis will be a challenging task given that they are unregistered, source of financing is
unknown, and many are organized informally under numerous federations/boards.
Despite this complex challenge, the Government should engage with this sector to discuss
how students can best be imparted skills that are relevant to the needs of the modern
economy.
1
Chapter 1: Introduction
1. There is tremendous diversity in delivery of education services worldwide ranging
from public schools financed and operated by the state to purely private schools. Within that
spectrum one finds a cornucopia of financing and delivery arrangements, such as public-
aided private schools, school vouchers to subsidize household expenses for private schooling,
and non-formal schools operated by NGOs. Often this blend is found even within education
sub-sectors of the same country. What shapes the incidence and nature of these education
delivery systems is a complex confluence of history, idiosyncrasies, and above all, politics.
The public discourse about how the state should include or exclude the private sector rages
vociferously across both rich and poor countries, and within that debate, the role of private
religious providers is particularly contentious. For example, despite the constitutionally
ingrained separation between Church and state in the USA and the constitutionally mandated
financing of Catholic schooling in Canada, in both countries there is lively debate between
political parties, religious groups, teachers unions, and parents about public financing of
private religious providers (school voucher eligibility for parochial private schools in the
USA; extending public-funding to all religious schools in Ontario, Canada).
2. While mass schooling of any type is a modern phenomenon, several Europe countries
and the United States relatively have the longest experience with provision of schooling
1
.
Particularly in Europe, historically schools were often founded and operated by Christian
Churches to simultaneously serve as institutions of religious and worldly learning. Currently
several European countries such as Ireland, Netherlands, and UK still have a significant
numbers of state-financed religious schools as part of that historical legacy. For example,
almost one third of schools in UK operate under a religious orientation. Most of these
European schools operated by religious establishments and financed by the state are supposed
to be open to all eligible pupils, however, like in the case of UK, these schools can ration
based on denomination/religion in cases of over-subscription. Recently the public discourse
of the role of state financing has been amplified due to concerns about the fact that these
religious schools might be inculcating their graduates with values fundamentally different
from the mainstream. For example, there is a major debate in Sweden about banning
creationism from biology classes in some Christian schools which receive public funding.
The major policy concern in many European countries is, however, with public financing of
Islamic schools (madrasas), reflecting the inter-twined concerns about extremism and
integration given that most children attending Islamic schools tend to be from immigrant
households (e.g., polarizing public debate in Netherlands to change Article 23 of the
Constitution to bar all public funding to private religious schools, with the subtext of stopping
funding to Islamic schools). Since the terrorist attacks on September 11, 2001, madrasas
have been of particular interest to the foreign policy formulation of many Western
governments.
3. Muslim-majority countries with a large state-funded religious school sector such as
Bangladesh, Indonesia, and Turkey are grappling with similar issues, reflecting recent geo-
political events as well as intrinsic domestic concerns. For example, the contestation of the
state between secular forces and political parties with a Muslim identity has been at the
forefront of the current Turkish political debate. While graduates of government-funded
Islamic schools in Turkey have been characterised as being disproportionately partisan to the

1
Even compared to Europe, United States was a league ahead it had mass provision of secondary schooling
by early the 1900s (Golding and Katz 2008).
2
agenda of Muslim parties, some graduates of madrasas in Bangladesh and Indonesia have
been characterized with a more nefarious identification - terrorism. In the absence of
evidence-based research, much of the current discourse on madrasas in Bangladesh (and
South Asia is general) is predicated on anecdotal accounts. The rise of religious extremism
and intolerance in South Asia has been traced to the growth of the madrasa education system
(e.g. Stern, 2001; Bandyopadhyay, 2002; Lintner, 2003; Ahmed, 2005; Griswold, 2005).
Some extremist madrasa school graduates in Bangladesh have indeed been involved in
despicable acts such as killing of judges and bombing traditional musical festivals i.e.,
targeting symbols of the state and secular society. Unsurprisingly, the local and international
media has singled out madrasas as the key to the radicalization of countries like Bangladesh
and Pakistan. Furthermore there has indeed been a steady increase in the number of religious
schools in these countries. Madrasas in Bangladesh are the fastest growing sub-sector of the
education sector, allegedly making up for the lack of government provision and as a response
to poverty. A similar pattern is also observed in Pakistan, although, in terms of absolute and
relative measures, Bangladesh hosts the largest number of registered religious seminaries in
the sub-continent, 2
nd
largest in the world (after Indonesia). It is rarely noted that Islamic
extremists associated with violence in Bangladesh tend to be associated with certain non-
regulated private madrasas, rather than graduates of public-aided private madrasas. Long
before 9/11 Bangladesh had embarked on a pragmatic direction to reform madras as at the
secondary level which sets it apart from other countries in the region. First it introduced fiscal
incentives to traditional unregistered predominantly all-male madrasa high-schools to register
and include modern subjects such as mathematics. Then it introduced another financial
incentive to registered madrasa high-schools to start admitting female students. Most madrasa
secondary schools in Bangladesh are now registered, follow a modern curriculum alongside
traditional religious subjects, and have become coed (50% of the enrollment in Madrasa high-
schools are now females) an unparalleled development in South Asia so far.
4. Bangladesh has achieved impressive progress in certain human development outcomes.
Gross primary enrolment rate is above 100 percent, secondary enrolment has more than
doubled since independence, and the Millennium Development Goal (MDG) gender parity
target has already been achieved at both the primary and secondary education level. These are
all remarkable feats when compared to countries at similar levels of income in the region.
Much of the achievement in the three-fold increase in overall secondary enrolments and
seven-fold increase in female enrolments since 1980 has been a result of successful
expansion of the private sector via supply-side incentives and gender-targeted demand-side
incentives program since the early 1990s. However, despite these achievements, key
challenges remain. These include: (a) low completion rates (only 50% of children complete
primary schooling); (b) poor quality of learning outcomes, particularly in rural areas; and (c)
many poor children are still left out of the secondary school system. The policy dialogue on
these key issues is hampered by the fact that there is a dearth of rigorous studies on
determinants of learning outcomes. In particular, we know very little about the religious
segments of the education sector. This reflects the overall lack of rigorous documentation of
learning quality of rural educational institutions in Bangladesh regardless of school-type
(religious or secular). Thus, any discussion of relative school quality has to encompass the
full spectrum of school-types operating in rural Bangladesh.
5. A unique feature of the Bangladeshi secondary education sector is the large presence of
religious schools. However, unlike other countries in the region with large Muslim
populations, the religious education sector comprises of both state regulated private madrasas
as well as independent, private madrasas. The former are popularly known as Aliyah
3
madrasas where alongside Islamic education, modern general education is also provided.
Given that these madrasas operate with state funding, they are well-regulated in terms of
curriculum content and teacher recruitment policy. On the other hand, an unknown number of
private, traditional madrasas exists outside the state sector. These seminaries specialize in
religious education and are popularly known as Quomi madrasas.
6. Given that net secondary enrolment rate still remains low, achieving further progress
under widespread poverty requires that we attract hitherto difficult-to-reach children from
poor families to schools. However, if there is a positive link between poverty and religious
education this will further increase the share of religious schooling. Therefore, a clear
understanding of the religious school sector is necessary if this institution is to serve as an
instrument for poverty reduction in rural Bangladesh. For at least three reasons, there is a
serious policy concern regarding the provision of education through the institution of
madrasas. First, these schools may not provide skills necessary a modern economy. The
academic standard attained is popularly perceived to be much lower than in general
education. Consequently, it is believed that madrasa graduates fail to pursue a productive
economic life and therefore add to the pool of educated unemployed. With little learning
taking place, current attendance can only reinforce the curse of poverty in the future. Indeed,
existing analysis of Bangladeshi labour market earnings data reveals a negative correlation
between madrasa attendance and wages (Asadullah, 2009; Asadullah, 2006). Second, they
may not promote civic values that are essential for a functioning democracy, thereby causing
concerns among policy makers with an interest in the relationship between education and
citizenship. Third, a segment of religious schools still remains completely outside the state
system, making it a daunting task to regulate the curriculum content and practices.
7. For the above reasons, many hold non-state provision of education by Islamist groups
and the expansion of madrasas as a contributing factor to radicalization. These concerns are
extremely relevant in the age of globalization as Bangladesh today competes with other
developing countries in the international market. Education system should not only create a
skilled workforce it also needs to reduce social polarization in order to guarantee a stable
political environment. Therefore, it is vital for the government to identify the strengths and
weakness of the educational institutional structure of the country in order to improve the skill
set of the workforce on one hand, while improve social integration on the other.
8. Unfortunately, none of the available national datasets provide reliable information on
enrolment in madrasas. While the latest round of the nationally representative Household
Income and Expenditure Survey (HIES) collected information on enrolment in both
recognized and unrecognized madrasas
2
, reliable information is not available on quality of
madrasas. Therefore, any assessment of secondary education with a focus on both the
incidence and quality of faith schools in Bangladesh requires purposefully designed yet
nationally representative survey data combining both demand and supply side information on
schooling. To this end, in the year 2008 the World Bank commissioned a multi-purpose
sample survey to gather detailed information on all types of secondary schools in rural
Bangladesh. A full census of all educational institutions was carried out in sample unions (an
administrative unit bigger than village but smaller than sub-district) where secondary school
quality was assessed by employing a battery of cognitive tests. Apart from a detailed survey

2
Similar information is also available from another survey dataset compiled by CAMPE (2005), an NGO apex
body in the education sector. Despite the comprehensive nature of the survey, it was not designed to shed light
on the incidence of unregistered, Quomi madrasas, nor quality of education provided therein. For further details,
see http://www.campebd.org/content/EW_2005.htm
4
of secondary school and madrasa students, a complete household census was carried out in
sample villages, followed by a detailed survey of households with children of secondary
school-going age. The research effort therefore combines data on education institutions,
teachers, students, households, parents, and children students, from multiple inter-linked
sources census of primary and secondary schools, survey of secondary schools, household
survey and census in order to provide a detailed assessment of the secondary education with
an emphasis on the institution of madrasa.
9. This is the first ever comprehensive survey on the size, structure and quality of religious
schools in a Muslim majority country using data from Bangladesh. The comprehensive nature
of the data collected allows us to not only present a detailed picture of the madrasa education
system in the country, but also to draw comparison with the mainstream secular state and
state-aided education sectors. Given the diversity of the Bangladeshi education sector, the
data collected will help to dispel or confirm many popular myths about madrasa education.
Household based survey data will be used to address issues in both the primary and
secondary sector, while facility based survey data will focus on the secondary sector. Given
the vastness of the data collected at different levels (ranging from learning assessment to
social and political preferences), it is not possible for one report to contain all the possible
descriptions and analysis. The two primary objectives of this report is to address the issue of
incidence (how large is the madrasa sector?), and touch upon the issue of quality of education
institutions operating in rural Bangladesh (are learning outcomes poorer in madrasas
compared to secular schools?).
10. In order to highlight various specific aspects of the madrasa education system, we
structure the report along the following lines. Chapter 2 provides background information on
the secondary education system in Bangladesh and discusses various reform initiatives in the
madrasa sector. In Chapter 3 we provide new evidence from available national (survey and
census) datasets on the incidence of religious schools. This is followed by a presentation of
results from quantitative analysis where we use linked secondary school/madrasas census and
household survey data. Using this dataset, we achieve two purposes: (a) shed light on the
spatial distribution of registered madrasas across the country, and (b) provide a basic
description of household demand for madrasa education in Bangladesh. We pay particular
attention to the role of poverty in decisions to send children to madrasas. Chapter 4 describes
the survey design and methodology. First, we provide an introduction to the WB survey on
"Quality of Secondary School Madrasa Education in Bangladesh (QSSMEB). We then
explain the survey design, sampling and sample selection, sample size and finally the survey
contents. Main findings based on the WB survey are presented in Chapter 5. Here, we begin
by documenting the question of size of the madrasa sector. Then, we describe the structure
briefly commenting on selected institutional features of secondary rural schools and
madrasas. In the remaining two sub-sections, we present evidence on quality, measured in
terms of level of student achievement in four cognitive tests. We conclude in Chapter 6 by
providing a summary of the main findings, implication for policy reforms, and outlining an
agenda for future research.
5
Chapter 2: Background

2.1 Overview of the education sector
11. The Bangladeshi education sector has a complex structure where the composition of the
sector changes as one moves from pre-primary to secondary levels. Primary education in
Bangladesh spans grades 1 to 5, and falls under the purview of the Ministry of Primary and
Mass Education (MOPME). While most primary school children study in government and
registered aided non-government schools, there is a significant share of NGOs and non-
registered madrasas. In 2005 the primary school NER in Bangladesh was 62.9% compared to
54.8% and 50.5% respectively for India and Pakistan during the same period. While the NER
is high compared to neighboring countries, there is considerable variation by income groups
within the country. The NER of children from the poorest quintile is only 56.8% compared to
a NER of 77.3% for children from the richest quintile.
12. There has been a significant improvement in primary school completion rate, from 67%
to 74% between 2000 and 2005. Further, unlike low female participation in neighboring
countries like India and Pakistan, there is not only gender parity in the primary sector,
however, girls also have higher completion rates than boys. Even though there is no national
primary school leaving exam, nor do most secondary school require entrance examinations,
the transition rate to secondary schooling is low, particularly for children from poor
households. The transition rate to secondary education is relatively low for the poorest
children (58%) compared to that of children from the richest quintile (83%). In other words,
about 42% of children from the poorest households who complete grade 5 do not enter
secondary school.
13. Secondary education in Bangladesh spans grades 6 to 10, and is under the purview of the
Ministry of Education (MOE). Bangladesh has pursued a service delivery strategy in the
secondary education sector which combines a Public-Private-Partnership (PPP) model by
combining public financing with private provision. Given the historical continuity of this
policy strategy (through British and Pakistan periods as well), it is not surprising that the
secondary education sector is dominated by registered aided non-government schools. These
aided non-government schools accounted for 98% of the secondary enrollment share in 2003
(BANBEIS census 2004). There is also a strong historical precedence of engaging various
types of non-state providers in provision of secondary education, notably, madrasas. While
81 percent of the share of secondary enrollment is in secular aided non-government schools
(henceforth referred to as schools), 17 percent of the enrollment share is in registered
reformed aided non-government madrasas. Government schools are fully funded and only
charge nominal monthly fees to pupils who qualify for admission to those schools. In
registered non-government schools and madrasas, public financing is provided primarily in
the form of teacher salary where the government funds at least 9 teachers per institution.
Since teacher salary is essentially covered by the government, government aided schools are
supposed to only charge a nominal tuition fee to pupils.
14. Religious schools in Bangladesh are almost entirely in the non-state sector, operating in
both primary and secondary levels. Moreover, majority of the state-recognized schools also
benefit fiscally from the state under the PPP model. Little systematic information is available
on those religious schools that remain outside the state sector and are unrecognized by the
state. There is a perceived difference among the various madrasas in terms of in-school
conduct and curriculum content. If so, lumping all Islamic schools together and calling them
madrasas is seriously misleading. Indeed, madrasas operating at primary or secondary
6
levels in Bangladesh are classified in two broad categories: Aliyah madrasas and Quomi
madrasas. Outside the primary/secondary sector, there are large number of pre-primary
Islamic education institutions which are of two types: Maktab (or Nourani madrasa) and
Furqania/Hafizia madrasas
3
.
15. The establishment of Aliyah madrasas and their activities follow government regulations
as prescribed by the Madrasa Education Board. The Board approves curriculum for all
registered madrasas from primary (i.e. Ebtidai) to masters (i.e. Kamil) level. For a madrasa
operating at any level, there are clearly prescribed requirements for state recognition and
support (for details, see Sattar (2004) and Abdalla, Raisuddin and Hussein (2004)). In terms
of curriculum contents, Aliyah madrasas offer a hybrid education where students are taught in
both religious education and modern general education. Students in government recognized
(i.e. Madrasa Education Board affiliated) schools are taught subjects similar to those taught in
general education primary schools in addition to a curriculum of religious and Arabic studies.
In other words, in addition to Bengali, general mathematics, social science, general science,
English and Arabic, students are taught Quran, Hadith and Aqaid and Fiqh (for a detailed
description of curriculum followed in Aliyah sector, see Appendix Table A).
16. The secondary level of Aliyah stream is known as Dakhil which spans grades 6-10.
Similar to mainstream education system, students appear in a public examination at the end
of Dakhil education which is organized by the Bangladesh Madrasa Education Board. The
Dakhil certificate is equivalent to the Secondary School Certificate (SSC) obtained after
successful completion of secondary education from mainstream schools. There are four
Dakhil Madrasa curriculum groupings: (a) General group where the focus is on subjects such
as languages, mathematics, religious studies, and social studies; (b) Science group where in
addition to languages, mathematics, and religious studies, students are examined in computer
science, physics, chemistry, and biology; (c) Muzabbid group where students are taught in
religious studies such as Quran, hadith, and fiqh; (d) Hifzul Quran group. Similarly, students
in mainstream secondary schools can opt for different group specializations such as science,
arts and commerce. However, they cannot specialise in religious studies. This is despite the
fact that in up to grade 8, Muslim students in mainstream schools are taught Islamic studies
on a compulsory basis. Therefore, in comparison to mainstream schools, Aliyah secondary
madrasas use a hybrid curriculum, giving students the option to specialise in general, secular
stream as well as Islamic theology (for a detailed of the grade-equivalent structure followed
in Quomi sector, see Appendix Table B).
17. Lastly, Quomi madrasas operate completely outside the state sector. The final stage of
the Quomi Madrasa education is called Dawrah Hadith, in which Hadith certificates are
awarded to the students after successful completion of the "Dawrah" class. This is popularly
perceived as being equivalent to a Kamil degree in Aliyah madrasas. The number of Quomi
madrasas is unknown and has been subject to wild speculations. According to one source,
there are about 8,000 madrasas of this type in Bangladesh (Mercer et al. 2006), while using
information collected from officials from one Quomi Madrasa Association, Sattar (p. 352,
2004) puts the figure at 4,000. However, in Ahmeds (2005) expert testimony prepared on
behalf of the International Crisis Group for the US Senate Foreign Relations Committee
Hearing on 'Combating Terrorsm through Education', she claims:

3
Abdalla, Raisuddin and Hussein (2004) provide an excellent discussion of the curriculum and structure of
primary and pre- primary religious and secular schools within the state regulated sector.
7
"Bangladesh's madrasa sector has mushroomed, reaching an estimated 64,000 madrasas from
roughly 4,100 in 1986, with little if any government oversight."
18. The author does not disclose the source of this statistics. As per published government
education statistics for the year 2005, there were approximately 6,800 (primary) Ebtidai
madrasas and 9,200 (post-primary) Aliyah madrasas recognized by the state. Therefore, if the
number quoted in Ahmed (2005) is correct, there are at least 48,000 madrasas outside the
state-sector in Bangladesh. In other words, the existing estimates of the number of Quomi
madrasas in the country ranges from 4,000 to approximately 48,000. However, these numbers
are entirely anecdotal - they are not based on any systematic evidence on the incidence of
traditional madrasas. Several decades ago, most of the Islamic seminaries in Bangladesh
were indeed of traditional types, operating with little state oversight. As pointed out already,
although privately managed, these primary and secondary madrasas today are largely
regulated by the state and this has been owing to a number of important reforms. We discuss
this issue next. Given that we are primarily interested in the secondary education sector, our
discussion excludes Ebtidai madrasas.
Brief Historical Background of Madrasas in Bangladesh
19. The establishment of madrasas in South Asia goes back to ascendency of the Delhi
sultanate in the 13
th
century (however, there might have been some madrasas associated with
earlier influx of Muslims from the Middle East and Central Asia). These madrasas under the
patronage of the Mughals were institutions of both religious and worldly learning given that
elite graduates mostly severed the needs of the Mughal courts. Despite the patronage of the
Mughals, the overwhelming majority of Muslims in the sub-continent (as well as non-
Muslims) never attended any type of school given that the supply of schooling (of any form,
level, or location) was extremely limited. Modern mass education was only introduced in the
Sub-Continent during the British Raj (and even then supply of schooling, particularly in rural
areas was negligible). The British Raj was instrumental in changing the nature of madrasa
orientation in South Asia in direct and indirect ways. First, the British stopped madrasas
from receiving state land-based revenues which had been a major source of public financing
during the Mughals (to maximize tax revenues going to the coffers of the Raj). Madrasas then
increasingly turned to private sources of donations (Metclaf 1979). Also in response to the
ascendency of the Raj, there was a movement among certain Muslims to de-link themselves
from any association with the usurped state. Madrasas supposedly began to remove subjects
such as mathematics and science, and focusing solely on religion. The Dar-ul-Uloom
Deoband madrasa was set upon in 1867 in the Indian state of Uttar Pradesh as a response to
the British Raj. The focus was on introspection and in cultivating a religious identity.
20. Anecdotal evidence suggests that the majority of Qoumi madrasa today in Bangladesh
are allegedly run along the Deobandi line. This is the same ideological pedigree of most
madrasas in India and Pakistan. Historically, these madrasas have relied on own assets and
charities to finance day to day activities. This is because, financial autonomy concerns five of
eight fundamental principles laid down by Maulana Muhammad Qasim at the founding of the
Deoband madrasa in 1867 (Ladbury, 2004):
As long as the madrassah has no fixed source of income, it will, God willing, operate as
desired. And if it gain any fixed source of income, like jagir holdings, factories, trading
interests or pledges from nobles, then the madrassah will lose the fear and the hope which
inspire submission to God and will lose His hidden help. Disputes will begin among the
8
workers. In matters of income and buildings let there be a sort of deprivation (Quoted in
Metcalf, 1978).
21. Traditional madrasas in the sub-continent, therefore, are alleged to exclusively rely on
external finance and private donations made for religious purposes. A majority of these
madrasas also supposedly house and educate orphans and children from poor families. But it
is not known exactly how they differ in their interpretations of religious doctrine. No single
curriculum applies to these madrasas so that some divisions prevail in terms of course
contents. These madrasas focus on reciting the Koran and learning the duties of the Maulvi
(pastor) in order to prepare students for running the mosques day-to-day operations.
22. As opposed to Deobandi madrasas which were shaped by their defiance to the state, the
root of Aliyah madrasas goes back to a unique event during the British Raj as well. Governor
Warren Hasting of Bengal established the first Aliyah madrasa, initially known as the
Calcutta Madrasa in 1781. The madrasa taught both religious and secular subjects, and was
supposed to be a model for madrasa reform for the rest of India which never really took off
(even in Bengal). After the India-Pakistan partition, the Calcutta Madrasa was transferred to
Dhaka. Even then there were only a handful of Aliyah madrasas throughout the East Pakistan
period it was only in the early 1980s that there was a proliferation of Aliyah madrasas in
response to a series of reform measures initiated by the government.
2.2 Secondary madrasa reform initiatives
23. In the late 1970s and early 1980s the government introduced financial incentives to
madrasas to modernize if the madrasas would register and introduce modern courses
alongside religious subjects, then the state would cover a significant portion of teacher salary.
Then it provided another financial incentive tied to increasing the share of female students
(this incentive was offered to both secular and religious schools). Drawing upon
administrative data, Asadullah and Chaudhury (2008a) show that the impact of financial
incentives offered by the government was significant. A large number of madrasas opted to
modernise in response to the government scheme. In terms of numbers, almost half of all
modern madrasas today comprise of formerly traditional religious schools. It is also
noteworthy that a significant number of them opted to convert following introduction of the
female stipend scheme. This highlights the hitherto overlooked fact that the stipend scheme
along with governments modernisation scheme succeeded in converting orthodox, all-boys
religious seminaries into modern madrasas that opened up to students of both sex.
24. The positive impact of the government reform programs on female secondary school
participation rate today is an undisputed fact. The role played by modernised madrasa in
promoting female education is unique. Asadullah and Chaudhury (2008b) show that
Bangladeshi regions which have greater number of religious modernised schools experienced
greater growth in female enrolment in recent years. Despite the mushrooming of government
subsidized madrasa secondary school sector, we know very little about the quality of these
schools. Previous field-based research (Asadullah et al. 2007) on learning outcomes in
secular and modernised religious high schools provides partial answers to this question.
Several findings are noteworthy:
Religious school students have lower test scores in mathematics when compared to
their peers in secular schools.
Once we account for selection into a given school type, there is no significant learning
differences across religious and secular schools.
9
Level of learning (measured in terms of maths skills) in general is very low;
Mathematics competency is low even when measured in terms of primary standard
maths test.
Gender-gap in test score prevails at the beginning of the secondary schooling cycle
(grade 6) and prevails through grade 8.
25. The parallel expansion of the religious and secular education sub-sectors brought about
by modernisation reform has given rise to an additional concern. Education of large number
of girls in religious schools has implications for demographic outcomes (e.g. fertility choices)
and labour market participation. Once again new research has documented differences in
socio-economic attitudes among female graduates of religious and secular schools (see
Asadullah and Chaudhury 2006). Key findings are:
Even after controlling for differences in individual characteristics and family
backgrounds, there are attitudinal differences by school type. Religious school
students have perverse fertility choices- they are more likely to rely on God on the
question of desired number of children. Additionally, madrasa graduates are less
favourable to higher education for female than for male.
Exposure to female and younger teachers helps close the attitudinal gaps between
madrasa and school graduates. This therefore highlights the importance of the policy
of recruiting more female teachers at the secondary level.
26. Success of the Bangladeshi experiment of reforming religious schools using financial
incentives needs to be better documented, given the wider implication this may have for other
countries with a large religious education sector, and for recalibrating reform measures in
Bangladesh itself. On the basis of aggregate data, we have suggestive evidence that financial
incentives can be an important instrument in modernising traditional religious schools
(Asadullah and Chaudhury, 2008a; 2008b). However, we need clearer causal evidence of the
supply-side response to the provision of financial incentives from independent sources of
micro-level data as well.
27. It should be noted that a reform scheme similar to that in Bangladesh was attempted in
India in 1986 when the Indian government proposed to introduce subjects like science,
mathematics, English, and Hindi in the madrasa curriculum. Modernized madrasas were
eligible to apply for government financial aid. The government also promised to arrange for
recognition of certain madrasas by certain state-funded universities
4
. However, the success of
this initiative has so far, been limited. Not surprising, the nature of the incentive was quite
insignificant. Only one teacher would have been financed by the government to teach all
modern subjects (Khan et al., 2003). Additionally, wary of the state's interventions in
religious instruction and their academic and administrative freedom, most of Indias 30,000
madrasas
5
have refused to participate in the scheme (Fahimuddin, 2004). Moreover, the
financial incentive to modernize remains considerably weak. The Indian state only promises
to provide financial assistance to madrasas to teach secular subjects whereas in Bangladesh,
all teachers of a recognized madrasa qualifies for salary payments from state exchequer.

4
In contrast, Aliyah madrasa degrees are recognised by all universities in Bangladesh.
5
This figure is from Kennedy (2004).
10
Consequently, most madrasas in India today are traditional and independent of the state for
funds
6
.

6
At present, only seven Indian states (Assam, Bihar, Madhya Pradesh, Orissa, Rajasthan, Uttar Pradesh and
West Bengal) have government-sponsored madrasa education boards, to which a number of madrasas are
affiliated.
11
Chapter 3. Incidence of religious schools: New evidence from available national
datasets
28. The Bangladesh Bureau of Statistics (BBS) carries out a nationally representative
household survey which forms the basis of consumption based poverty measures and other
important welfare metrics know as the Household Income and Expenditure Survey (HIES).
This survey also includes information on education characteristics of households. On the
other hand, BANBEIS, the Statistical arm of MOE conducts a census
7
of registered
secondary educational institutions which provide administrative information on enrolment by
school type. As pointed out earlier, however, existing efforts to gather educational data have
so far remained limited to schools that are recognized by the state. Nonetheless, given the
nationwide coverage, it is useful to document key trends in enrolment rates by school type.
The following two sections describe findings from these two data sources in a descriptive
manner. In addition, we shed some light on specific aspects of religious education, namely,
who attends madrasas and where do religious schools operate. Answer to the latter two
questions will help verify the veracity of the popular claim that madrasas are primarily
attended by children from poor households and/or madrasas tend to locate in regions that are
relatively poorer and under served by mainstream (private and state) schools.
3.1 Trends in enrolment in religious schools

3.1.1 Evidence from administrative census data
29. As per the latest census carried out by BANBEIS in 2005, there were 27,714 registered
secondary educational institutions in Bangladesh of which 33% were Aliyah madrasas (i.e.
recognized Islamic schools). Therefore, in terms of numbers, the religious education sector is
quite large. Its noteworthy that most of these madrasas are not only registered with the
government, they also receive fiscal support from the state to pay for teacher salary. Despite
being managed by private bodies, most of the Aliyah madrasas in the country are actually
regulated by the state in terms of curriculum and teacher recruitment policy. The Madrasa
Education Board is the apex body that oversees functioning of these madrasas.
30. Figure1 plots time-series data on the relative share of registered madrasas in secondary
education in a variety of ways. In terms of numbers, share of madrasas have increased
significantly up to 1990 after which growth in the number of schools outstripped madrasas.
However, in terms of share in total secondary enrolment, madrasa sector grew faster than its
secular counterpart since 1980. This growth is partly explained by the introduction of salary
subsidy scheme by the state under which teachers of registered madrasas (like teachers of
aided secular schools) would be eligible for monthly salary subsidies. However, in post-1980
period, secondary education sector in Bangladesh went through another important cash
transfer program Female Stipend Program (actually a collection of state-funded and donor-
funded stipend projects) which also led to the growth of enrolment in madrasas. Like
schools, registered madrasas were eligible to participate in the scheme. However, given the
gender targeted nature of the intervention, exposure to the scheme led to rise in enrolment via
feminization of registered madrasas in the country. This is obvious from Figure1 as well: in
post-1990 period, share of females in total madrasa enrolment rose astronomically from less
than 10% in 1990 to nearly 50% by 2005.

7
In most years the census consists of self-reported data provided by the headmaster of the institution; in some
years it is a census carried out by visits to the institutions by enumerators.
12
Figure 1: Relative size of Aliyah Madrasa
sector
0
10
20
30
40
50
60
1970 1975 1980 1985 1990 1995 2000 2005
% of madrasa students being girls
Madrasas as % of total #of secodnary educational institutions
Madrasas as % of total secondary enrolment
Madrasa as % of total secondary female enrolment
% of secondary school students being girls

Source: Authors calculation based on BANBEIS data.

31. To sum up, the size of the registered madrasa sector is large irrespective of how we
measure it. And when measured in terms of enrolment share (in total as well as female
enrolment), the madrasa education is even a growing phenomenon. However, most of the
enrolment growth is explained by the rising participation of females in the post-1990 period.
3.1.2 Evidence from household data
32. While the estimates of the share of registered madrasas in total secondary enrolment are
large, we dont know from the administrative data how large this share is as a percentage of
total school age population
8
. To this end, we turn to an alternative source of enrolment data
by school type as available from household based HIES data. While HIES does not include
any metric on learning outcomes, an advantage of HIES is that it includes socio-economic
information on the childs household. A further added advantage of using HIES data is that
we are able to compare the relative share of madrasa enrolment with that of another non-state
school which is known to target kids from poor households, namely NGO schools. In
Bangladesh, a large number of non-formal schools are operated by NGOs (mainly BRAC)
targeting drop-outs from public primary schools and out-of-school children from poor
households. In other words, both NGO and madrasa schools arguably attract children from
poorer households. Therefore, by means of comparing relative share of madrasas with NGO
schools, we are also able to assess the veracity of the claim that poverty is the main correlate
of demand for religious schools in Bangladesh. While the HIES survey data is available for
various years since 1983, only from 2000 onwards does the dataset provide comparable
information on enrolment by school type. Therefore, we restrict our analysis to HIES 2000
and 2005 data only. It should be pointed out that the 2005 HIES round was the first time data
was collected specifically on Quomi madrasas (in 2000, the distinction was only between
school and Aliyah madrasa). We should also point out that regardless of type of institution
there is a serious problem with missing data. For example, for almost 34% of currently
enrolled children in HIES 2005 there is no data on type of institution that the child is enrolled
in. This further underscores our point that existing data sources are not comprehensive

8
BANBEIS does collect age data however, the credibility of that data is of serious concern (most headmasters
just fill in the age that the child should be for that grade instead of the actual age of the child).
13
enough for a serious examination of madrasa education in Bangladesh. This should be kept
in mind when reviewing the findings presented below.
33. The rise in madrasa enrolment holds even when we exclude enrolment in Quomi
madrasas (as documented only in 2005). This indicates that the share of Quomi madrasas in
school enrolment is very low in Bangladesh across both primary and secondary levels (Table
1). Table 2 shows data on enrolment for secondary and primary madrasas as well as primary
NGO schools. Data is presented separately for primary and secondary age cohorts.
Table 1: Enrollment share of currently enrolled in HIES 2005
Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 All
Government 57.3 53.0 43.6 38.7 36.0 43.4
Private Subsidized 27.4 34.4 47.8 53.6 56.1 46.9
Private (Not Sub) 2.0 2.0 1.7 1.8 4.2 2.6
NGO/Other 4.5 2.1 1.3 1.0 0.3 1.5
Madrasa (A
1
) 6.9 7.6 4.7 4.4 2.8 4.8
Madrasa (Q
2
) 1.8 1.0 1.0 0.5 0.7 0.9
Total 100.0 100.0 100.0 100.0 100.0 100.0
*Quintile groupings are from consumption-based poverty measure with Quintile 1 being the poorest and
Quintile 5 being the richest.
1: Aliyah; 2: Quomi
Source: Authors calculation using HIES 2005 data.

Table 2: Madrasa Enrolment (%) by Madrasa Type, Education Level and Year
Quintile
*
1 Quintile 2 Quintile 3 Quintile 4 Quintile 5
Primary NGO 2000 8.9 7.5 1.7 0.8 0.4
Primary Madrasa 2000 3.7 1.4 2.1 2.7 2.5
Secondary Madrasa 2000 6.0 3.3 5.4 5.3 5.3
Primary NGO 2005 6.0 2.7 2.7 2.3 1.2
Primary Madrasa (A
1
) 2005 3.5 4.2 3.8 2.3 0.5
Primary Madrasa (Q
2
) 2005 1.7 0.6 1.4 0.9 0.7
Secondary Madrasa (A) 2005 9.2 9.0 4.9 5.0 3.8
Secondary Madrasa (Q) 2005 2.0 1.3 0.8 0.4 0.7
*Quintile groupings are from consumption-based poverty measure with Quintile 1 being the poorest and
Quintile 5 being the richest.
1: Aliyah; 2: Quomi
Source: Authors calculation using HIES 2000 and 2005 data.
34. Even when compared to enrolment in NGO schools, share of primary Quomi madrasas
is miniscule. Given that both are known to target the poor, this comparison suggests that the
incidence of Quomi madrasas is likely to be less significant than is otherwise suggested in the
popular media. However, Quomi and Aliyah combined, madrasas dominate NGO schools in
terms of enrolment shares in primary education
9
.
35. As pointed out earlier, female schooling in Bangladesh has made tremendous progress in
the last two decades and much of this has been due to the fact that registered madrasas
opened their gates to girls, particularly at the secondary level. HIES data shows a rise in
madrasa enrolment during the period 2000-2005. This growth in madrasa enrolment is
observed across gender groups, rural and urban areas and well across income quintiles.
However, it is more pronounced for girls, poorer households, and rural areas (see Appendix
Tables 1.a Tables 1.d). The rise is also evident if we compare enrolment rates in madrasas

9
We revisit this issue later using data from the World Bank survey.
14
across various age-cohorts. However, the rise is more prominent for secondary school age
groups which are consistent with the fact that most registered Aliyah madrasas in the country
are in the secondary sector.
3.2 Analysis of linked secondary school/madrasas census and household survey data
36. While the latest round of HIES data provides information on enrolment of children in all
types of secondary educational institutions, there is no supply side information on these
schools. As previously mentioned, this information is available in the 2005 administrative
census dataset of BANBEIS (at least for registered schools). However, BANBEIS data by
itself contains no demand-side information. Existing attempts to explain supply decisions of
madrasas have so far remained partial, limiting the focus on either demand or supply-side
factors only. To circumvent this data problem, we created a cross-walk between HIES and
BANBEIS census data at the thana/upazila level
10
. A Thana geo-code was created which is
unique across both datasets. This facilitated linking of BANBEIS data with HIES 2005 data
spanning a total of 344 upazilas
11
. The linked sample contained 20515 registered secondary
educational institutions of which 33% were madrasas. Figure2 below presents a division-wise
breakdown of these secondary schools and madrasas in our linked sample (as a % of sample
total). The same graph also plots % share of each division in the total number of schools as
well as madrasas in the country (using BANBEIS census database).
Figur e2: Division-wise dist ribut ion of secondary regi stered
schools and madrasas in BANBEIS census & l inked BANBEIS-HIES
sample
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
Barisal Chittagong Dhaka Khulna Rajshahi Sylhet
Division
%

s
h
a
r
e
#of madrasa (Linked sample)
#of school (Linked sample)
#of madrasa (BANBEIS census)
#of school (BANBEIS census)

37. Comparison with divisional share of schools and madrasas in the census database shows
that the number of secondary educational institutions in our HIES-linked sample is fully
representative of the division-wise national averages figures. This is reassuring although not
surprising since HIES sample itself is representative of the country. Therefore, sub-districts
that comprise our linked HIES-BANBEIS sample give a representative picture of educational
infrastructure across the country. We carry out two separate analyses using this linked

10
Thana/upazila is a sub-district; Bangladesh is administratively broken down into 6 divisions and 64 districts
(Zilas).
11
We could successfully match 96% of all HIES 2005 round sample sub-districts to BANBEIS 2005 census
dataset.
15
sample dataset. First, we explore the spatial patterns in the distribution of madrasas across
HIES sample thanas and ask what factors lead to higher concentration of madrasas in a sub-
district. Second, we turn to the demand-side by estimating child-level regression model and
study household decisions to enrol a child in madrasa. The rationale for such regression
analysis is as follows.
38. As pointed out earlier, the current international popular debate on madrasa education
maintains that: (a) madrasas are of poor quality little learning takes place there; (b) they
arise as a response to under provision of education by the state; and (c) they primarily cater to
the poor. In the light of this characterization, any analysis of madrasa education is
incomplete without a discussion of the underlying demand and supply decisions. In other
words, before we address the question of quality and incidence of traditional, unrecognized
madrasas, we must attempt to have an understanding of how their registered, modernized
counterparts operate. We also have to have an understanding of who attends Aliyah madrasas
in Bangladesh.
3.2.1 Geography of religious schools: Analysis of spatial distribution of madrasas
39. In order to explore the geographic concentration of madrasas, we estimate a simple
regression model using log of number of registered secondary madrasas as the dependent
variable. The unit of our analysis is thana or sub-district. There are 344 sub-districts in our
linked sample spread across 6 divisions. As explanatory variables, we use (a) log of number
of secondary schools, (b) mean per capita household expenditure in the thana (in logs and
constant price), (c) fraction of thana population being religious minority, (d) thana being
predominantly a rural one, and (e) division dummies.
40. Using this regression framework, we can formally test whether madrasas are largely set
up in locations where theres a scarcity of mainstream schools and/or which are mostly
inhibited by poor households. In other words, rise of religious schools is explained by the
governments failure to provide public schools in rural Bangladesh. Table 3 below reports
results from the regression analysis. Column 1 reports a parsimonious model while column 2
adds controls for division dummies.
16
Table 3: OLS regression of thana-level correlates of # of Aliyah madrasas (in logs)
Number of secondary schools (in logs) 0.76*** 0.73***
(11.27) (10.26)
Mean per capita household expenditure -0.70*** -0.63***
(-3.899) (-3.293)
Fraction of population being Hindu -0.37* -0.27
(-1.916) (-1.393)
Rural area 0.23** 0.21*
(1.973) (1.765)
Division: Barisal 0.51**
(2.223)
Division: Chittagong 0.25
(1.278)
Division: Dhaka 0.23
(1.227)
Division: Khulna -0.04
(-0.204)
Division: Rajshahi 0.36*
(1.78)
Constant 4.75*** 4.12***
(3.543) (2.915)
N 338 338
R-squared 0.352 0.379
Notes: (a) *** p<0.01, ** p<0.05, * p<0.1; (b) t-stats in parentheses. (c) Authors calculations based on linked
HIES-BANBEIS sample data. (d) Omitted (category) division is Sylhet.

41. A number of findings follow from column 1 of Table 3. First, the coefficient on
number of schools is large, positively signed and highly significant. This implies that sub-
districts with more mainstream schools are also more likely to have large number of Aliyah
madrasas. Note that the measure of availability of mainstream secular schools in our data
captures all government, private and state-aided (registered as well as unregistered) schools.
Therefore, our estimate of the influence of availability of mainstream schools is not biased
owing to measurement errors related issues.
42. Second, areas that are more rural have more madrasas. This is consistent with our earlier
analysis which shows that madrasas are more of a rural phenomenon in Bangladesh. Third,
regions with more non-Muslims have significantly less madrasas. This is also consistent with
our prior- areas with more non-Hindus will see less demand for Islamic education. Fourth,
high-income areas have fewer madrasas. This finding is important and lends support to the
popular belief that madrasas tend to operate in poor areas. Therefore, while we find no
evidence in support of the hypothesis that madrasas necessarily arise in educationally under-
provided areas, we find a strong correlation between local area poverty and concentration of
madrasas in the locality. The latter finding suggests that (Aliyah) madrasas target poor areas.
Since poor households have low income and also tend to live in poorer areas, we may expect
children from poor households to have a higher incidence of madrasa attendance. We will
revisit this issue in the next section when we analysis household decisions to enrol children in
madrasas.
43. It may be pointed out that these findings do not change when we add controls for
division dummies in column 2 an exception is the Hindu variable which becomes
17
insignificant. The coefficients on division dummies, however, are interesting in their own
right as well. Compared to the omitted base-category, only Barisal and Rajshahi have
significantly more madrasas. The other 3 districts Dhaka, Chittagong and Khulna no
longer have a statistical advantage in number of madrasas. This latter finding is interesting
because in the raw data, these three divisions have 3 to 5 folds more madrasas than Sylhet
division. The numerical advantage enjoyed by these 3 divisions in the raw data is potentially
explained by the between-division difference in poverty and stock of mainstream schools.
Net of the effects of these two correlates, therefore, we find no statistical difference in the
stock of madrasas between these 3 divisions and Sylhet region. If true, what explains
significantly higher stock of madrasas in Rajshahi and Barisal needs further examination.
44. Lastly, as an additional robustness test, we also controlled for district fixed-effects. This
specification led to more precise estimates of the effect of availability of mainstream schools
and local area poverty. However, both of these two variables retained positively and
negatively signed (respectively) and retained their level of statistically significant, even when
we fully controlled for district dummies
12
.
3.2.2 Household demand for madrasa education
45. The analysis of the geographic concentration of madrasas presented in the previous
section points towards a correlation between poverty and demand for religious education.
Because madrasas tend to be located more in areas that are poor and the poor also tend to
concentrate in under-developed regions, we may find a similar negative correlation between
household income and childrens participation in madrasas. This sub-section therefore
examines the determinants of madrasa enrolment using HIES data.
46. As pointed out earlier, the latest round of HIES collects enrolment information on both
traditional and recognized madrasas. Our earlier analysis of HIES data in section showed that
the incidence of enrolment in traditional unregistered (Quomi) madrasas is an insignificant
phenomenon. Nonetheless, we define madrasa attendance broadly, treating enrolment in
modernised and traditional madrasa as one group.
47. In HIES 2005, we have a total of 7,830 children aged between 6 and 18 years, 5.48% of
whom were enrolled in madrasas
13
. If disaggregated by school type, 4.52% and 0.86% were
reported to be enrolled in recognized (Aliyah) and unrecognized (Quomi) madrasas
respectively. After ignoring cases where data on other regression controls are missing, we
had a sample size of 6,492 of which 5.64% were madrasa enrolees.
48. The probability of (ever) enrolment in madrasas for children is analysed in a regression
framework where we regress the binary dependent variable (1 if the child is/was in madrasa;
0 otherwise) controlling for socio-economic background of the student, parental educational
profile, and household income. A further innovation in our analysis is that we bring into the
HIES data thana level controls for the number of secondary schools and madrasas on one
hand and control for sub-district level per capita expenditure on the other. Therefore, on the
basis of the estimated regression model, we can test two specific hypotheses: (a) is there a
correlation between madrasa attendance and household poverty? (b) Is madrasa attendance
related to local poverty level?

12
Results not shown, however, available from authors.
13
Because our analysis is about decision to enrol in Islamic schools, we have excluded all non-Muslim
households from the regression sample.
18
49. The results are reported in Table 4. In addition, we estimate an extended enrolment
regression model where we hold differences in the availability of secondary schools and
madrasas constant at sub-district level. Given the small fraction of children in HIES 2005
who reported to have even enrolled in madrasas, we do not repeat our analysis separately for
primary and secondary enrolment decisions. For the same reason, we restrict our analysis to
ever enrolment decision (instead of current enrolment status). We indeed find a positive
correlation between poverty and madrasa enrolment. Children from poor households are more
likely to be enrolled in madrasas. However, the magnitude of that effect is very small, being
close to zero. At the mean, a 10% increase in "log of per capita household expenditure"
reduces the probability of madrasa enrolment by only 0.3% (see column 1). The effect is even
smaller if we further control for a full set of district dummies (see column 2). Note however
that this income effect is much lower than what's been found previously in current enrolment
regressions using HIES data. For example, in Asadullah and Chaudhury (2009,
Forthcoming), found that a 10% increase in log of per capita household expenditure increased
the probability of current enrolment by 1.8% and 2.9 % in rural and urban Bangladesh,
respectively. Therefore, while household wealth plays a significant role in children's school
attendance decision in Bangladesh, the effect appears to be modest when it comes to religious
school attendance.
50. The above finding holds even when we replace the continuous expenditure variable by a
set of 4 expenditure quintile dummies (leaving out dummy for the poorest 20% as the base
category); enrolment probability declines as we move to higher quantiles although the
coefficient on the highest quintile dummy still remains very small (see columns 3 and 4). The
finding suggests that poverty only partly explains the incidence of religious education in
Bangladesh and is consistent with the descriptive analysis of enrolment data by school type
and expenditure quintiles presented earlier. Overall, our finding is consistent with that of
Andrabi et al. (2006) who using survey data from Pakistan also find an association between
lower-income households and madrassa enrollment, but at the same time, report the
magnitude of this association to be small. As pointed out earlier, only 10% of ever enrolled
children from poorest quintile attend madrasas, and even some children from well-off
households are enrolled in secondary madrasas. Second, poor children from the same
household are not equally likely to be enrolled in madrasas. In other words, HIES data
suggests that madrasa enrolment is not entirely a between-household phenomenon in
Bangladesh.
19
Table 4: Probit regression of madrasa enrolment (matched HIES 2005-BANBIES 2005
data)
(1) (2) (3) (4)
Childs age 0.02*** 0.02*** 0.02*** 0.02***
(3.07) (2.87) (3.09) (2.86)
Childs age square -0.00*** -0.00*** -0.00*** -0.00***
(2.85) (2.71) (2.87) (2.70)
Child is female -0.01 -0.01 -0.01 -0.01
(1.26) (1.60) (1.26) (1.57)
Education of the household head 0.01*** 0.01*** 0.01*** 0.01***
(3.48) (3.91) (3.44) (3.89)
Education of the heads spouse -0.00 -0.00* -0.00* -0.00*
(1.56) (1.65) (1.66) (1.74)
Log of per capita household expenditure -0.03*** -0.02***
(4.29) (3.86)
Household expenditure quintile 2 0.01 0.01
(0.04) (0.37)
Household expenditure quintile 3 -0.02** -0.01*
(2.23) (1.92)
Household expenditure quintile 4 -0.03*** -0.02***
(4.01) (2.75)
Household expenditure quintile 5 -0.03*** -0.03***
(3.91) (3.62)
Mean Household expenditure in sub-district -0.05*** -0.04* -0.05*** -0.04**
(3.59) (1.90) (3.73) (2.11)
Number of madrasas in sub-district in 2000 (in logs) 0.01* 0.02** 0.01* 0.02**
(1.80) (2.14) (1.84) (2.12)
Number of schools in sub-district in 2000 (in logs) -0.01 -0.01 -0.01 -0.01
(0.78) (0.69) (0.80) (0.66)
Rural area 0.02*** 0.02*** 0.02*** 0.02***
(2.97) (3.15) (2.98) (3.13)
N 6492 6012 6492 6012
Division dummies Yes No Yes No
District dummies No Yes No Yes
Pseudo R
2
0.06 0.11 0.06 0.11
Note: (a) Sample is restricted to Muslim households and children aged 6-18 years old only. (b) Per capita
expenditure data is in logs, deflated using regional CPI. (c) Marginal effects reported instead of coefficients. (d)
The regressions also control for age-squared of the child. (e) Dependent variable is 1 if the child ever enrolled in
Madrasa; 0 otherwise. (f) The number of madrasas and schools in the year 2000 was calculated by using
information on the year in which an institution was established. Data on all other variables correspond to the
year 2005.

51. We also formally tested this in a regression framework, adding a dummy indicator if a
sibling also attends madrasa on the right had side of the model. We carry out an additional
test by comparing enrolment pattern of children within the household by schools type. Even
after holding household and local poverty constant, we find a significant albeit small
correlation between siblings madrasa attendance and the index childs enrolment in a
madrasa
14
. In other words, while poverty is certainly a strong correlate of madrasa
enrollment, household poverty by itself cannot be the only explanation for the growth of
religious school enrolment in Bangladesh. In the country madrasa enrollment has been
increasing while overall poverty level has been falling.
52. Our finding that household poverty only partly explains the incidence of madrasa
education does not capture the effect of living in poor areas where there is under provision of

14
Results not presented, however, available from authors upon request.
20
public goods by the state. Note that our model already controls for the stock of educational
institutions in the sub-district. In addition, we hold constant sub-district income level constant
as well. However, increases in mean per capita expenditure of the sub-district indeed reduces
the probability of enrolment in madrasa, the effect (-0.04) being twice that of household per
capita expenditure (-0.02). This therefore once again confirms a modest effect of poverty on
madrasa enrolment.
53. Turning to supply-side factors, our model includes two measures of availability of
secondary educational institutions (brought in from BANBEIS data). One may argue that
these correlates suffer from reverse causality problem: madrasas may be set up in areas where
demand for religious education is high. We partially circumvent this problem by using lag
measures of supply side variables. Consistent with the common belief, indeed we find that
greater availability of mainstream schools in the sub-district reduces the probability of
enrolment in madrasas. But the effect is small and insignificant. On the contrary, availability
of registered secondary madrasas has an equally small but statistically significant effect on
the decision to attend madrasas
15
.
54. Lastly, note that all the models control for regional influence on household enrolment
decision. Models 1 and 3 control for division of residence whilst models 2 and 4 holds the
effect of district of residence constant. Although suppressed in Table 4, controls for region
effect were always significant in all specifications. Of all the 6 divisions in Bangladesh,
residence in Barisal raises the probability of madrasa attendance (relative to the omitted
division, Sylhet). This result is consistent with our earlier analysis of geographic
concentration of madrasas: after controlling for local poverty and stock of schools, we found
Barisal to have the largest number of religious schools in the country (see Table 3).
55. Note that the very small size of the madrasa sub-sample in HIES data has constrained
our analysis in important ways. If poverty is indeed the only explanation for madrasa
enrolment, the correlation between poverty and madrasa enrolment should hold true for all
children from same (poor) households. Owing to limited sample size and missing data
problem, we did not carry out further investigation into the sibling story. Even if we take the
largest possible sub-sample (by not discarding observations where other data is missing) of
429 children who ever attended madrasas in HIES 2005 data, they come from as many as 328
households yielding only 1.27 madrasa children per household where at least one child is
enrolled in madrasas. More importantly, the school type data in HIES 2005 is missing for
38% currently enrolled school children (aged between 6 and 18 years). Therefore, it is
difficult to shed light on intra-household incidence of madrasa attendance using the latest
round of HIES data. This issue is instead left to future work where we will use household
census data from the World Bank survey which has a larger household sample of madrasa
students.


15
Note that such an analysis is common plagued by concern over the direction of causality. We addressed this
issue by replacing our measures of school and madrasa availability in 2005 by those in 2000. Even using lagged
measures (which are arguably more exogenous regressors in the context of our model), our earlier findings
remain unchanged.
21
Chapter 4: Survey design and methodology
56. A specifically designed large education institution and household based sample survey,
Quality of Secondary School Madrasa Education in Bangladesh (QSSMEB) was initiated
by the World Bank in 2008 to primarily address the quality of education in public-aided
secondary school madrasas (the survey was executed by the firm DATA). We randomly
selected 12 districts from 6 divisions (highest administrative unit in Bangladesh). The
probability proportional to size (PPS) method of random sampling was used, based on
division/district level secondary school going age population data from the 2001 national
population census and the concentration of secondary schools and madrasas based on
BANBEIS website 2007. Two upazilas (sub-districts) were randomly selected using PPS
from each of the selected 12 districts. Then 2 unions were randomly selected with PPS from
each of the selected 24 upazilas. Again, the population weight was union level population
data from the 2001 national population census.
Fielding the Institutional Survey
Census all pre-primary, primary, and secondary education institutions in the Union
Survey of all secondary educational institutions, which included:
Detailed information about the institution
Survey of all Teachers
Testing all children from one randomly selected section of Grade 8 (Math, English,
General Knowledge test given to all of those pupils; Islamic knowledge test given to only
Muslim pupils)
Collecting self-reported household information from tested children
Basic information from the students tested in Grade 8
Basic information about the Math class and homework
Basic information about the English class and homework
Anthropometric data of tested students
Specific math teacher survey
Specific English teacher survey
Math Teachers evaluation about those students in grade 8 who were tested
English Teachers evaluation about those students in grade 8 who were tested
Socio-Economic Attitudes of Pupils and Students
Fielding the Household Survey
57. In each of the 96 villages randomly selected using with PPS based on village-level
population data from the 2001 national census:
A complete census of all households in the village
From the census, 25 households that have at least one school-aged child (aged 618
years) regardless of current enrollment status, was randomly selected from each village.
i. Detailed multi-module household survey was administered (e.g., complete
roster, education history, employment status, assets, consumption, religious
practices)
22
ii. A numeracy and literacy competency test was given to all primary and
secondary school-aged child in the household
iii. A RAVEN progressive-matrix test was given to all primary and secondary
school-aged children in the household
iv. Anthropometric data collected on children and mothers

A village-level survey on availability of other services, price-levels, etc, was also
administered
58. It is noteworthy to mention that in the rare event when institutions attended by children
were outside our original sample union, we also covered those secondary schools and
madrasas located in neighbouring unions. Below in Table 5 we present a snap-shot of some
selected QSSMEB statistics:
Table 5
Characteristics No. Remarks
#of divisions 6
#of districts 12
#of upazilas 24
#of unions 48
#of villages 96
#of institutions in school census 2,229
#of pre-primary institutions 585
#of primary institutions 1,224
#of secondary institutions in detailed survey 403
#of households in census 24,841
#of households in detailed survey 2,400
Gross primary enrolment from census 92.3 excluding children living away from household
Net primary enrolment from census 72.6 excluding children living away from household
Gross secondary enrolment from census 72.7 excluding children living away from household
Net secondary enrolment from census 63.3 excluding children living away from household
% share of primary enrolment madrasa (census) 13.8 8.14% is in Aliyah, 1.9% is in Quomi, and 43.5%
is in Other madrasa that includes Nurania/
Quiratia/Furkania madrasa
% share of secondary enrolment madrasa (census) 22.1 18.65% is in Aliyah, 2.32% is in Quomi, and
1.34% is in Other madrasa that includes Nurania/
Quiratia/Furkania madrasa
Gross primary enrolment from hh survey 108.4 excluding children living away from the
household
Net primary enrolment from hh survey 76.7 excluding children living away from the
household
Gross secondary enrolment from hh survey 81.1 excluding children living away from the
household
Net secondary enrolment from hh survey 72.2 excluding children living away from the
household
% share of primary enrolment madrasa from survey 15.7 9.48% is in Aliyah, 2.3% is in Quomi, and
43.76% is in Other madrasa that includes Nurania/
Quiratia/Furkania madrasa
% share of secondary enrolment madrasa from survey 24.5 20.53% is in Aliyah, 2.56% is in Quomi, and
1.76% is in Other madrasa that includes Nurania/
Quiratia/Furkania madrasa
#of 8
th
grade pupils tested in Math 9,436
#of 8
th
grade pupils tested in English 9,286
#of Math teachers surveyed 413
#of English teachers surveyed 508
#of children given household based numeracy and literacy
tests
3,244
#of children given Ravens tests 3,232
23
Chapter 5: Preliminary Findings
59. The survey designed for this study is among one of the most detailed educational
surveys in South Asia and the first where alongside secular schools, complete information is
available on both state and non-state schools providing secular and religious education. The
findings presented in this report represent the first-cut exploratory analysis restricted to the
incidence of religious schools and the relative quality of education provided therein. These
findings are presented separately in two sections. In addition, we briefly highlight key
institutional differences between secular and religious schools in Bangladesh. Analyzing the
entire volume of the data collected will take considerable time and specific studies will have
to the tailored to address different issues (and audiences). We discuss major research outputs
plans from this survey in section 6.2.
5.1 Incidence- Size of the madrasa sector
60. Using information on both the relative share of religious schools in the total number of
educational institutions and total number of students enrolled, this section provides first-hand
evidence on the size of the traditional madrasa education sector in Bangladesh. For this
purpose, we not only draw upon three types of datasets (school census, household census and
household surveys), we report the statistics separately for primary and secondary education
sectors. The key objective of this section therefore is to establish whether enrolment rates in
madrasas are high in Bangladesh. To this end, we also comment on trends in enrolment over
time. While we cannot comment on enrolment patterns in traditional madrasas over time, it is
possible to shed some light on student enrolment in registered madrasas. We have
commented earlier on whether enrolment rates in madrasas are on the rise using national
representative household sample surveys and administrative data.
61. By means of a complete census of educational institutions in 48 unions, we could locate
a total of 2,229 schools of which 26% and 54% of were pre-primary and primary educational
institutions, respectively. The rest were secondary educational institutions. A crude measure
of the numeric size of madrasa sector is its relative sample share in pre-primary, primary and
secondary sectors. To this end, Figure3 below plots data on distribution of our sample
educational institutions by type and levels of education
16
.


16
There were no private primary schools in our rural sample unions given that purely for-profit private
schooling is still an urban phenomenon in Bangladesh. In the secondary sector there are private schools and
private-aided schools. Most schools in rural Bangladesh are established as private schools before they apply for
public financing. Hence most private-schools in rural Bangladesh are public-aided schools in waiting. In
Figure 7 and subsequent graphs/tables we have lumped them with non-government schools unless noted
otherwise.
24
Figure 3: Distribution of # of sample educational institutions by
type and level
0.00
10.00
20.00
30.00
40.00
50.00
60.00
70.00
80.00
Pre-Primary Primary Secondary
Level of education
%

s
h
a
r
e

i
n

a

g
i
v
e
n

l
e
v
e
l
Aliyah
Quomi
Other madrasa
Govt sch
Non gov sch
NGO sch

Source: Authors calculation based on data from QSSMEB

62. Starting with the pre-primary sector, madrasas retain the largest share accounting for
71.28% of all educational institutions. However, as little as 1.2% of these are formally
attached to a Quomi madrasa; the remaining are other madrasas such as Nurania, Quiratia,
Furkania and Hafizia madrasas. Very little is know about these madrasas, however, the
common perception is that at the pre-primary level they are essentially teaching elementary
religion in a non-formal setting. In our sample only 10% of the currently enrolled children
attend a pre-primary institution. Turning to the primary level, the sector is represented by all
types of institutions government, government-aided and NGO schools. Only 2.48% of all
the primary madrasas are of Quomi stream while 3.02% of them are Ebtidai (i.e. they
belong to Aliyah stream). However, a sizable fraction (6.78%) belongs to other madrasa
type as well. At the primary level, a useful benchmark to ascertain the relative size of the
madrasa is the relative share of NGO schools which also target the poor. In the primary
sector, all three types of madrasas together (Aliyah, Quomi and others) constitute 12.57% of
the sample primary educational institutions, while NGO schools account for 19.48%. In other
words, even when taking into account all types of madrasas, their numerical share in the
primary sector is less than that of NGO run schools. Numerical share of the madrasa sector is
largest if we restrict attention to secondary level. Aliyah and Quomi together account for 49%
of the total secondary educational institutions in our study area. However, the largest share
belongs to Aliyah madrasas (30% of the sample total).
63. Assessing the madrasa sector simply in terms of the number of existing madrasas can be
misleading. A more relevant metric is the enrolment share of madrasas. For this purpose, we
exploit data from complete population census carried out in 96 villages in our study area
which covered approximately 24,000 households. If madrasa sector is sizable, as indicated by
their sample share in the distribution of total educational institutions in our study area,
25
households should also report a large fraction of children being enrolled in madrasas in our
sample villages.
64. Figure 4 below plots data on enrolment share of different schools by levels of education
in our survey area. Once again, even in terms of primary enrolment share, incidence of
Quomi madrasa is not large in rural Bangladesh: They account for only 1.9% of the total
primary enrolment. These numbers are very small when compared to enrolment in a similar
non-religious, non-state school that also caters to children from poor families NGO schools
account for 8.2% of the primary enrolment in our study. Share of Aliyah madrasa, however,
is significant: They account for 8.4% of the total primary enrolment. Overall madrasa
enrolment as a share of total primary enrolment (13.8%) is large when we take into account
enrolment in other madrasas that are non-formal in nature and offer exclusively religious
education. Thus, as a group the share of enrolment in madrasas is significantly larger than the
NGO share.
65. Turning to the secondary education segment, Quomis have a slightly higher share,
accounting for 2.2% of the total enrolment. This is also true for Aliyahs (who have an
enrolment share of almost 18.5%). Interestingly enough, these estimates are consistent with
enrolment figures reported in another large-scale secondary school survey (see CAMPE,
2005) where of those enrolled in secondary level, 76% students were found to be in non-
government secondary schools while secondary registered madrasas accounted for 14% of
enrolment. Only 2% of the enrolled children were in Quomi madrasas.
66. To summarize, the number and share of Quomi madrasas in both the primary and
secondary sector is much lower than what is portrayed in the popular press. As discussed
earlier in this report, nationally representative HIES survey dataset for the year 2005 indicates
even a smaller share of these madrasas in rural Bangladesh (1.2% and 1.3% of primary and
secondary aged population, respectively). Overall though, madrasas as a group (registered
and non-registered) do constitute a significant share of the educational institutions and share
of enrolments in rural Bangladesh.
Source: Authors calculation based on data from QSSMEB
67. One of the challenges facing estimation of enrolment share in madrasas using household
records is that of missing data. A fraction of the madrasas is residential and hence enrolled
children reside away from home. Consequently, enrolment figures based on household
surveys such as HIES are likely to under-estimate the actual share of religious schools. To
8.4
1.9
3.5
59.7
18.3
8.2
Figure 4a: Distribution of
pri mary enrol ment share of
educational institutions by
type
Aliyah
Quomi
Figure 4b: Distribution of secondary enrolment
share of educational institutions by type
18.5
2.2
1.4
6.9
70.9
0.2
Aliyah
Quomi
Other madrasa
Government
Non-government
NGO
Figure 4a: Distribution of primary enrolment share of
educational institutions by type
8.4
1.9
3.5
59.7
18.3
8.2
Aliyah
Quomi
Other madrasa
Government
Non-government
NGO
Figure 4b: Distribution of secondary enrolment
share of educational institutions by type
18.5
2.2
1.4
6.9
70.9
0.2
Aliyah
Quomi
Other madrasa
Government
Non-government
NGO
26
circumvent this problem, we also gathered data on children of the household head in two
additional ways. We completed the household roster separately for all non-resident children
in addition to asking the head the reasons for why child is living away from home. We
therefore re-estimated the share of madrasas in enrolment on the basis of household records
accounting for both resident and non-resident children. This exercise did not alter the relative
share of madrasas as reported above. Even if we go as far as including all the non-resident
siblings who are currently living elsewhere for educational purposes and/or for other reasons,
our conclusion remains unaltered. This is true both irrespective of whether we focus on
primary as well as secondary schools.
68. Lastly, our estimate of madrasa sector size based on enrolment share does not change
significantly if we disaggregate the enrolled child sample by gender (see Figure 5). Two
notable differences are as follows: (a) Quomi madrasas have a larger share in male enrolment
(3.5%) than female enrolment (1.4%); (b) other madrasas largely educate boys. In sum, the
share of un-recognized, traditional religious schools (i.e. Quomis) depends on how we
measure it (share of educational institutions or total enrolment) and in which sector we focus
on (pre-primary, primary or secondary). In the secondary sector, Quomis have a large share
when assessed in terms of number of educational institutions. However, given that the
average student population in these Quomis is small, the Quomi share is almost insignificant
when assessed in terms of student enrolment share. The debate over religious schools in the
secondary sector in Bangladesh therefore is largely about the state of recognized pubic-aided
Aliyah madrasas which have a large share not only in terms of the numbers of educational
institutions but also in terms of share of total secondary enrolment.
Figure 5: Enrolment share of secondary educational institutions
by gender and type
0
10
20
30
40
50
60
70
80
Secondary (Male) Secondary (Female)
Gender
E
n
r
o
l
m
e
n
t

s
h
a
r
e

(
i
n

%
)
Aliyah
Quomi
Other madrasa
Government
Non-government

Source: Authors calculation based on data from QSSMEB
69. The other finding from the preceding discussion is the hitherto overlooked role played
by state recognized secondary madrasas in educating female students. In our study area, these
madrasas account for 19% (18%) of the total secondary female (male) enrolment. This
finding is consistent with patterns observed in national level secondary school enrolment
27
figures (e.g. see Asadullah and Chaudhury, 2008a). The feminization
17
of the Aliyah system
was arguably induced by the secondary school scholarship scheme which was initiated in
1994. This is also evident in our sample schools. From enrolment records we collected
retrospective information on the share of female students in total enrolment spanning the
period 1993-2007. Figure 6 below plots this data separately for secular schools, Aliyah and
Quomi madrasas.
Figure 6: Trends in Female enrolment by school type
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
1993 1999 2003 2007
% of students enrol l ed bei ng femal e
School_g
Madrasa_Aliya_g
Madrasa_Qum_g

Source: Authors calculation based on data from QSSMEB
70. Indeed, there has been a steep rise in the share of females in total enrolment in Aliyah
madrasas since 1993 so much so that by the year 2003, share of girls in total enrolment in
these madrasas exceeded that of boys. More importantly, by the year 2007, Aliyah madrasas
in our study unions have become more feminized than main stream secular schools. For the
sake of comparability, Figure 10 also reports enrolment share of females in Quomi madrasas.
Interestingly enough, even in the state unregulated part of the religious education sector, a
rise in female enrolment is evident. This is particularly so in the post-1999 years. In the
absence of any national level enrolment figures on un-regulated madrasas, however, it is not
possible to generalize on the basis of this pattern that like their counterpart in the state-
regulated sector, Quomi madrasas in Bangladesh are also becoming increasingly feminised.
However, our sample data certainly highlights yet another previously unknown phenomenon,
namely the role played by state un-regulated madrasas in educating females in rural
Bangladesh. In other words, while share of these madrasas in total female enrolment still
remains miniscule (i.e. 1.4%) in our study area, it is nonetheless much higher than what it
was a decade ago.


17
By feminization we only mean that madrasas have gone from all-male institutions to co-ed institutions.
28
5.2 Structure: A selective institutional profile
71. How do madrasas financially sustain themselves in rural Bangladesh? Who monitors
them? What is specifically included in the curriculum? How are classrooms organized in
these schools? Do these institutions reinforce or influence gender norms? Do they provide
other services besides schooling? These are all important issues in their own right and we will
pay detailed attention to them in future research, however, for now we briefly highlight some
salient features to give readers an idea of the complexity of these institutions.
72. Fiscal aid: Figure7 reports the distribution of our sample secondary schools by
management type also highlighting whether the school receives fiscal aid from the
government. Sixty-Four percent of all sample secondary recognized schools and madrasas are
privately managed but aided by the government where another 1.24% is fully government
schools. This implies that the remaining 34% secondary educational institutions are outside
the state sector which includes all Quomi madrasas. This confirms that majority of the state
recognized secondary religious schools operate with state-funding. While 24% of our sample
institutions are Aliyah madrasas, only 29% of this Aliyah sub-sample (or 7% of the total
sample) survive entirely on self-finance. At the same time, it highlights that a fraction of
Aliyah madrasas in rural Bangladesh operate with limited state oversight as do their Quomi
counterparts.
Figure 7: Distribution of sample secondary schools (N=403)
by management type
1%
9%
41%
24%
7%
18%
Govt school
Private school
Aided school
Aided Aliyah
Private Aliyah
Private quomi

73. Funding: The above observation that Quomi madrasas receive no aid from the state
implies that their sources of funding will differ significantly from schools and madrasas in the
state sector. Indeed from our analysis of data on annual earnings of Quomi madrasas, we find
that 57% of the total income comes from private donations. This confirms the popular belief
that these types of institutions survive under community patronage and with private donations
from Muslim households and individuals (e.g. through a process of alms-giving or zakat). In
addition, 30% of their total income does not fall under any category
18
. In comparison,
unaccounted income comprises only 4% of total income in Aliyah madrasas. At the same

18
The issues of foreign funding and the growing madrasa system outside of government regulation is a common
concern in the popular press. While the large fraction of unaccounted income reported could be attributed to
foreign funding, we are unable to verify this.
29
time 11% of total income comes from student fees, pointing out that not all of these
traditional madrasas provide free education. Nonetheless, revenue earned from student fees is
insignificant as a fraction of total revenue in Quomi madrasas if compared to the private un-
aided schools where 65% of total income comes from student fees.
74. External monitoring: In 54.1% cases, our sample Quomi madrasas were inspected by
someone in the last one year and in 62% cases, the visitor was someone from the respective
education Board. In contrast, in 44% cases, our sample unaided Aliyah madrasas were
inspected by someone in the last one year and in only 15% cases, the visitor was someone
from an education Board. Therefore, compared to Aliyahs, Quomis seem to be part of a more
active network where they are subject to some degree of internal monitoring.
75. This point becomes clearer when we look at data on affiliation to a central education
authority - 85% of Quomis in our sample are affiliated to an education Board. However, the
number of Boards providing affiliation was found to be 14 and all of these are managed by
private bodies outside the state sector (i.e. they are not approved by Governments madrasa
board). This is not surprising as these madrasas have no connection with the government, as
was the case during the British Colonial period, when they were called Khariji i.e. outside
the government jurisdiction. One of these private Boards -- the Befaqul Madaris of
Bangladesh (i.e. Quomi Madrasa Education Board) is the umbrella under which the
majority of our sample unaided Quomis have been found to be organized.
76. Gender orientation: Seventy-Four percent of our samples Quomis are all male-
institutions while only a mere 9.5% are coeducational
19
. In contrast, 80% of our sample
Aliyah madrasas (aided and unaided together) are coeducational. Similarly, 83% of our
sample schools (government, aided and private unaided together) are coeducational.
77. A hallmark of madrasas in South Asia is their visible focus on religious symbolism.
They are known to implement strict dress codes along an orthodox Islamic style, particularly
for girls. Our survey design explicitly looked into this issue and we had a specific interview
with the principals about the prevalence and nature of dress code by gender. The issue of
dress code is also important given the increasing trend amongst female secondary students to
attend madrasas. If even aided and recognized madrasas officially imposes gender norms in
dressing, this may have implications for governments broader social implication of
subsidizing female education in madrasas.
Table 6: Presence and nature of policy regarding uniform for female students

No
official
uniform

If yes, what type of uniform?
Burka Shalowar-Kamiz Nikab Head-scarf/cover N
School 0.11 0.01 0.87 0.03 0.38 199
Aliyah Madrasa 0.15 0.69 0.30 0.10 0.35 123
Quomi Madrasa 0.00 1.00 0.32 0.21 0.32 19
Note: (a) Calculation is based on co-educational or girls-only institutions (i.e. boys-only institutions are
excluded from the calculations). (b) The table above is based on multiple responses. Therefore, a madrasa can
report having all 4 types of uniforms burka (body is covered but face is visible), salwar-kamiz (tunic-pant set),
nikab (burka plus only eyes are visible), and head scarf. (c) All number refers to fractions. (d) Source: Authors
calculation based on data from QSSMEB.


19
In all of our coeducational Quomi madrasas, however, boys and girls had separate classroom.
30
78. Table 6 above reports data on official policies related to female dress-codes in school.
First note that there is not a single Quomi madrasas (out of 19 coeducational or girls-only) in
our study area that does not impose a dress code on female students. On the other hand, their
counterpart in the state-regulated sector appears to practice a slightly more liberal regime -
15% of the Aliyah madrasas in our sample do not have any official dress code for their
female students. The same pattern prevails when we look at the nature of the required attire
female students are officially expected to wear. In all Quomi madrasas, girls are required to
wear burka (compared to 69% of Aliyah madrasas). Similarly, the incidence of Nikab (full
cover of body, head and face) is most common in traditional madrasas: 21% of our Quomi
madrasas implement this as a dress code compared to only 10% Aliyah madrasas.
79. We explicitly asked principals to comment on any official policies relating to the
practice of purdah inside the madrasa/school compound. Purdah is a nebulous concept and
can range from strict segregation of the sexes to wearing a loose scarf around ones head
when out in public. Once again, Quomi madrasas appear to strictly implement the policy of
purdah irrespective of whether it is outside or inside the classroom. But so do state-
regulated madrasas: 85% of Aliyah madrasas require female students to practice purdah
inside the classroom compared to 95% of Quomi students (see Table 7 below).
Table 7: Presence of policy regarding purdah for female students
Yes: inside the classroom Yes: outside the classroom
School 0.18 0.18
Aliyah Madrasa 0.85 0.84
Quomi Madrasa 0.95 1.00
Note: (a) Based on head-teacher response. (b) Excludes all institutions where there is no girl student. (c) All
number refers to fractions. (d) Source: Authors calculation based on data from QSSMEB.

80. Now interpreting the information presented above is complicated. It could very well be
that madrasas are simply catering to the preferences of some conservative rural households
and imposing dress codes that are culturally more appealing to parents in rural Bangladesh.
Even 3% of our secular aided private schools had policies mandating Nikab. Are these
schools signalling to parents who prefer this drastic (and up till recently, a rare) form of
appearance for their daughters that they are more accommodative of their preferences
compared to other institutions, or are they imposing their own conservative social norms on
poor parents? Is this an education issue or a social issue? These are deeply complex
questions which ultimately require policymakers to make uncomfortable normative political
decisions.
81. Student accommodation: A common perception is that traditional madrasas are
attractive because they provide services outside of education such as food and lodging.
Indeed 33% of our Quomi madrasas are fully residential (compared to only 4% of Aliyah
madrasas). As a matter of fact, only 87% of our Quomi madrasas offer at least some
residential facilities (compared to only 19% of Aliyah madrasas). This finding is therefore
consistent with the popular belief that madrasas offer a free education, and board to their
students, and thus they appeal to poor households.
82. Class size: One of the most widely studied institutional aspect of schools is number of
students enrolled in a given grade or class-size. Bigger class size is associated with lower
learning outcomes. Looking across institutions type, we find significant difference in this
proxy for education quality. The distribution of class size is more favourable in the Quomi
31
sector: the mean class size (in grade 8 or its equivalent) in Quomi is 10 compared to 48 and
32 in school and Aliyah, respectively. However, care is needed in comparing class size data
between Quomi and registered schools/madrasas. First, as already pointed out, Quomi
madrasas are predominantly residential where learning takes place throughout the day and is
not necessarily limited to the classroom. In many cases, the same room may be used for
residential and teaching purposes. Second, in a majority cases, Quomi classrooms do not have
desks and chairs - rather, students are seated on ground. Such differences in classroom
organization once again make it less straightforward to compare class size (defined as the
number of students in a classroom) between Quomis and educational institutions in the state
financed sector.
83. Difference in subjects offered: We will not go into this issue in detail in this report,
however, we just highlight the fact that there are major differences between School/Aliyah
(aided and unaided) and Quomi madrasas. It is interesting to note that besides not offering
any Math or English, many Quomis do not have Bengali (the national language) as a subject
in grade 8 or its equivalent (see Table 8 below). This goes back to the Urdu pedigree of the
Deoband line. However, this does not mean that secular subjects are not taught in Quomi
madrasas at all. When assessed in terms of curriculum used for grade 5 (or its equivalent),
only around 5% Quomis were found not to teach Bengali, Mathematics or English at all.
Therefore, majority (over 90%) of Quomis do have at least one teacher responsible for
teaching primary level Mathematics and/or English, albeit at a lower grade equivalence
(primary level).
Table 8: Subjects not offered in the institution (%)
Type No Bengali No English No Math No Science

Aided-Aliyah 0.0 0.0 0.0 0.0
Unaided Aliyah 0.0 0.0 0.0 0.0
Quomi 40.6 26.6 56.3 29.7
Source: QSSMEB. Data corresponds to grade 8 (or its equivalent).

84. Professional background of Teachers: Compared to educational institutions in the
state or state-aided sector, Quomi madrasas are predominantly staffed by untrained teachers.
A staggering 82% of all Math teachers in Quomi madrasa are untrained compared to their
counterparts in schools and Aliyah madrasas (16% and 42% respectively). Note that the
problem of unqualified teachers is not limited to Mathematics. Similarly, 80% of all English
teachers in Quomi madrasa are untrained compared to their counterparts in schools and
Aliyah madrasas (19% and 44% respectively). The poor qualifications of Quomi teachers are
also evident when assessed in terms of years of education acquired in mainstream educational
institutions, the only place where they could in the past learn Math and English. On average a
Math and English teacher in our sample Quomi only have SSC-level education from the
mainstream, state-recognized education system. Compare this to aided school and Aliyah
madrasas where Maths and English teachers are on average are educated in secular schools
beyond the HSC-level. Overall, therefore, not only do Quomi madrasas often not offer
subjects such as Maths and English, they do poorly in terms of teacher recruitment for
teaching these subjects.
85. We have only provided a partial characterization of Quomi madrasas in this section. A
host of other institutional factors have not been commented on and instead left for future
work. In this context, we are planning to revisit two unique aspects of the Bangladeshi
32
madrasa system, namely, feminization of registered madrasas and extensive reform of the
curriculum followed in these schools. Furthermore, QSSMEB also provides a unique
description of a traditional madrasa in Bangladesh where we focus on physical conditions of
madrasas, nature of curriculum (e.g. modern subjects taught; religious content of the
curriculum) and textbooks used, gender norms (e.g. dress codes and rules regarding how male
teachers interact with female students), educational, demographic and socio-economic profile
of teachers, issues related to school governance (e.g. accountability of students/teachers);
learning environment (e.g. classroom organization; seating arrangements, teacher-student
positioning) and student background (e.g. parental education; past school type attended by the
child).
5.3 Quality: Levels of student learning by school type and gender
86. We administered four distinct cognitive tests to 8
th
grade students
20
during the
institutional survey. First, a maths test instrument was constructed by using items previously
used in the Trends in International Mathematics and Science Study (TIMSS). The TIMSS has
been administered three times by the International Association for the Evaluation of
Educational Achievement (IEA), the latest in 2003. The study has released several secondary-
standard (grade 8) mathematics items indicating what each item measures and the results of
every participating country. The TIMSS instrument assessed competency in data,
measurement, number, algebra and geometry using 125 multiple-choice format mathematics
items. On the basis of pilot test experience and given the time constraint in the field, a total of
25 original items were retained in our test instrument
21
. That is what we mean when we refer
to the TIMSS Math test throughout the report not the actual TIMSS administered by IEA
given that Bangladesh does not participate in this (nor any other) internationally comparable
standardized test. On the basis of the national curriculum we devised a test to assess
proficiency in English (20 items), General Knowledge (6 items) and Islamic studies (10
items). The test on Islamic studies, however, was restricted to only Muslim students. In
Bangladesh Islamic studies is a compulsory subject in government approved schools for all
Muslim students. The test was given to all students of Grade 8 who were present on the day
of the survey
22
. If a school/Aliyah madrasa had multiple-sections of the same Grade, one
section was randomly selected. In case of Quomi madrasas identification of Grade 8 was a
not so obvious. This is because they often use a subject-based system rather than a grade-
based system
23
. In such cases, we requested the madrasa authorities to select the equivalent
grade on our behalf and carried out the test in the selected cohort of pupils assuming
equivalence. However, because of potential ambiguity over equivalent grade selection in
Quomi madrasas, combined with the variation in organization structure within the Quomi, we
have restricted our analysis of learning outcomes to schools and Aliyah madrasas only. In
total, we tested 9,436 pupils in Grade 8 where 68%, and 24% of the students belong to
general school and Aliyah, respectively. Therefore, excluding the Quomi students (N=663 or
7% of the original sample), we were left with 8733 observations. In this sub-section, we first

20
We also administered basic numeracy and literacy tests to children during the household survey. Analysis of
household based assessment is planned for future studies.
21
For a similar exercise to assess student learning in rural secondary schools, see Asadullah et al (2006).
22
Apart from English, all tests were based on multiple-choice questions (MCQ) format.
23
The non-standard grade structure of Quomi madrasas in Bangladesh has been also noted by other researchers
in previous work. For instance, Sattar in his detailed monograph on madrasa education system points out that
some Quomis organize education in a total of 9 grades while others use a 15-grade system (see Sattar, p. 378,
2004). Such divergence in institutional structure can be partly explained by the fact that Quomi madrasas in
Bangladesh are organized under different regional associations (as explained in section 5.2) which pursue
independent pedagogical models and standards.
33
discuss how raw student test scores in our dataset vary by school type and gender on one
hand, and within and between schools on the other. In order to highlight the heterogeneity in
quality, however, we also discuss the entire distribution of the test score statistics. In the next
sub-section, we then discuss the correlates of individual test scores with a focus on
institutional factors.
87. Table 9 reports average math and English scores (i.e. fraction of correct answers) by
institution-type, gender and tests. For presentational ease, we use the following groupings:
schools (government schools, private schools, private aided-schools); Aliyah madrasa
(aided and unaided). On average, only 35% of the math test questions could be correctly
answered by students in our sample. This number is even lower (i.e. 33%) for English.
Interestingly, the performance is far from satisfactory even if we consider performance in the
general knowledge test: only 39% of the questions were correctly answered on average.
However, when assessed in matters related to religion (i.e. Islam), the performance was very
satisfactory: students on average correctly answered 75% of all the test items. Our sample
average Math score is strikingly similar to mean student performance in similar tests for
secondary school students elsewhere in South Asia. For instance, in a recent study on two
Indian states Orissa and Rajasthan, Das and Zajonc (2008) report mean scores (i.e. % correct
answer) of 34 and 37 respectively. Their findings are based on a survey of 9
th
grade students
in Math carried out in the year 2005 using a 36-item TIMSS test.
Table 9: Average Math and English Test Scores by Management-type
Math Math-M Math-F ENG ENG-M ENG-F

Gov-school 0.33 0.41 0.28 0.40 0.49 0.33
(107) (47) (60) (107) (47) (60)
Private-school 0.35 0.36 0.35 0.37 0.37 0.37
(519) (197) (322) (519) (197) (322)
Aided-School 0.36 0.39 0.34 0.37 0.38 0.36
(5,805) (2,286) (3,519) (5,805) (2,286) (3,519)
Aided-Aliyah 0.33 0.37 0.31 0.26 0.28 0.25
(1,983) (742) (1,241) (1,983) (742) (1,241)
Private Aliyah 0.31 0.34 0.30 0.28 0.30 0.27
(359) (86) (273) (359) (86) (273)
Total 0.35 0.37 0.33 0.33 0.33 0.33
(8,773) (3,358) (5,415) (8,773) (3,358) (5,415)
Note: (a) Numbers in parenthesis indicate sample size (b) each cell number refers to fraction of total questions
being answered correctly.
Source: QSSMEB.
88. Table 9 above only reports the fraction of correct answers in different subjects. As such,
it is not obvious whether the questions have been set at a reasonable degree of difficulty. As
a matter of fact, the state of poor-quality becomes even more evident if we consider few
examples of student performance in individual questions where a majority of students were
unable to answer. For instance, let us focus on two simple questions related to division and
fractions. Only 62% (54%) of school (Aliyah) students could correctly answer the following
question: Divide:
15
4
35
8

. By 8
th
grade, only 11% of school children and 8% of Aliyah
children can correctly divide 24.56 by 0.004.
89. Turning to questions relating to other domains of learning, a similar picture can be
painted. For instance, the question " The number of 750 mL bottles that can be filled from
600 L of water is_________? " could be correctly answered by only 13% sample students on
34
average (where 12% and 17% of school and Aliyah students provided correct answers,
respectively). A simple measurement test Which of these is the longest time? {(a) 15000
seconds (b) 1500 minutes (c) 10 hours (d) 1 day} is too difficult for 46% of grade 8 children
in our sample. When disaggregated by institution type, only 56% and 51% of the school and
Aliyah students, respectively, could correctly answer the question. In a simple problem
related to probability -- In a bag of cards 1/6 are green, 1/12 are yellow, 1/2 are white and
1/4 are blue. If someone takes a card from the bag without looking, which colour is it most
likely to be? -- only 34% of our sample students could correctly answer this question.
90. The English test assessed students in 3 domains: correct use of words in sentence,
(knowing) word meaning, and sentence construction. The dismal state of English language
learning is evident when we take into account the basic nature of the individual questions.
When asked to fill in the blank using the correct form of the word highlighted in the
parenthesis, 80% of our sample students failed to do so for the following sentence: He (go)
____ to school yesterday. When disaggregated by institution type, only 23% and 15% of
school and Aliyah students, respectively, could correctly answer the question. Similarly,
another 76% of our sample students failed to complete the following sentence: My father
currently (live) _____ in Dhaka.
91. In case of general knowledge test, 46% students could correctly name the current
president of Bangladesh at the time of the survey i.e. J anuary 2008. Only 18% students could
correctly write down the date (day, month and year) when Bangladesh became independent.
Performance in the last question is shocking considering the fact that students are taught these
facts at a much earlier stage (e.g. primary grades) in Bangladesh.
Figure 8: Distribution of test score by school type and gender
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
M
a
t
h
s
M
a
t
h
s

(
M
a
l
e
)
M
a
t
h
s

(
F
e
m
a
l
e
)
E
N
G
E
N
G

(
M
a
l
e
)
E
N
G

(
F
e
m
a
l
e
)
G
K
G
K

(
M
a
l
e
)
G
K

(
F
e
m
a
l
e
)
I
s
l
a
m
I
s
l
a
m

(
M
a
l
e
)
I
s
l
a
m

(
F
e
m
a
l
e
)
Test score by type & gender
F
r
a
c
t
i
o
n

o
f

a
n
s
w
e
r
s

c
o
r
r
e
c
t
General school
Aliyah Madrasa

Source: Authors calculation based on data from the "Quality of Secondary School Madrasa Education in Bangladesh". Data
corresponds to grade 8 students fromsample secondary schools and madrasas.
92. Figure 8 reports % of question items correctly answered by our sample students across
school types, gender and tests. A number of interesting patterns follow from the figures
presented above. Something that we can conclusively say is that girls fare worse than boys in
both Math and English in almost all school types we will come back to this point later on.
Looking at the % of Math questions answered correctly, school students (36%) have a slight
advantage over Aliyah (33%). A slightly bigger advantage is noticeable in case of the English
test: school students correctly answered 37% of the test items whereas Aliyah students could
35
answer only 27% of the answers correctly. When judged in terms of performance in general
knowledge, there was no major difference across students by school type. However, a
performance advantage prevails for Aliyah madrasa students in Islamic studies while 70%
of school students answered all Islamic studies questions correctly, 80% of Aliyah students
answered all Islamic studies questions correctly.
93. These patterns highlighted above are based on mean student test scores. As such, it hides
important distributional issues. For instance, within the state sector, there can be some very
good schools and madrasas. If so, the under-performance of a school/madrasa cannot be
exclusively explained by the underlying institutional regime - i.e. state regulation or the lack
thereof. Simple one-way analysis of variance (ANOVA) indicates that the within variation in
test scores (e.g., variation between pupils studying in schools) is greater than the between
variation in test scores (e.g., variation between schools and Aliyah)
24
. Figure 9 below plots
individual-level test score by school type across the entire distribution of the test score. For
the sake of brevity, our discussion will focus on two extreme points of the test score
distribution, the bottom and top deciles. Of the sample pupils 73% (27%) of the students
belong to general school (Aliyah madrasa) shown by the unbroken (broken) horizontal line in
Figures 9a and 9b. If there are not major differences in the distribution of test scores between
institution-type, then the fraction of a given institution-type in a particular decile should be
similar to its sample fraction. Hence the height of the decile bar should be equal to that of the
respective overall sample share horizontal line.

Source: (a) Authors calculation based on data fromQSSMEB. (b) Data corresponds to grade 8 students from327 sample
secondary schools and Aliyah madrasas. (c) Numbers in parenthesis indicate mean overall test score (rounded up to
the nearest integer) in a given decile.

24
Results not presented, however, available from authors upon request.
Figure 9b: English score (decile distribution by institution type)
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
1 (7) 2 (15) 3 (20) 4 (25) 5 (30) 6 (38) 7 (45) 8 (50) 9 (57) 10 (73)
Test scor e deciles
D
e
c
i
l
e

s
a
m
p
l
e

f
r
a
c
t
i
o
n
School
Aliyah madrasa
Overall sample
share: school
Overall sample
share: Madrasa
Fi gure 9a: Math score (deci l e di stri buti on by i nsti tuti on type)
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1 (12.1) 2 (20) 3 (24) 4 (28) 5 (32) 6 (36) 7 (40) 8 (46) 9 (54) 10 (66)
Test scor e deciles
D
e
c
i
l
e

s
a
m
p
l
e

f
r
a
c
t
i
o
n
School
Aliyah madrasa
Overall sample
share: school
Overall sample
share: Madrasa
36
94. On average, the bottom 10% could correctly answer only 12% of the Maths question
while even the top 10% could only correctly answer 66% of the questions. Now, the relative
share of school and Aliyah in the bottom 10% population is 60% and 40%, respectively. This
suggests that students of Aliyah madrasas are somewhat excessively represented in the lowest
performing declie. On the other hand, the relative share of school and Aliya students in the
top 10% population is 73% and 27%, respectively - this suggests that both types are
proportionally represented in the top decile (relative to their sample fraction).
95. The above analysis suggests that the distribution of individual Math test score is slightly
skewed to the left for Aliyahs and even for general schools. This is also evident if we look at
the school-specific distribution more formally using non-parametric Kernel density functions.
Appendix Figure 1a plots Kernel density estimates of individual students Math score. Here
we can see that the Aliyah distribution has a slight lower mean and a more left-skewed
distribution compared to the school distribution which closely follows a bell-shaped normal
distribution.
96. The non-normal nature of the test score distribution of Aliyah students is more evident if
we look at the English test score data. The relative share of school and Aliyah students in the
top 10% student population is 90% and 10%, respectively. This suggests that almost all of
the top scorers in English come from schools, while Aliyah students are grossly under-
represented in the top decile (relative to their sample fraction). The distribution of student
performance in English test is therefore skewed to the right for schools and to the left for
Aliyah. Once again, we produce kernel density plots of individual-level test scores by school
type (see Appendix Figure 2a). The bottom heavy nature of the test score distribution of
madrasas becomes more evident when we use plot aggregated data (see Appendix Figure 2b).
97. Lastly, we carry out a similar distributional analysis of student performance at the
institutional level. This is achieved by aggregating test score data at school/madrasa level.
Therefore, observed differences in aggregated test score highlight the between (instead of
within) school difference in quality. Figure 10 below plots aggregate data on test score by
school type across various deciles of the aggregate test score distribution.
37

Source: (a) Authors calculation based on data from QSSMEB. (b) Data corresponds to 327 sample secondary
schools and Aliyah madrasas. (c) Numbers in parenthesis indicate mean overall test score (rounded up to the
nearest integer) in a given decile.

98. Once again, for the sake of brevity, our discussion will focus on two extreme points of
the test score distribution. Note that 62% of our sample educational institutions (excluding
Quomis) belong to general school. This is shown by the bold solid line in Figure 10.
Therefore, we can see that visually at the aggregate level (i.e. between institutions),
differences in test score appears to be much larger. In the bottom (top) 3 deciles, the share of
schools is less much (more) than its population share in English test (see Figure 10b). For
Aliyah madrasas, it is the opposite. For instance, in the bottom (top) 3 deciles, the share of
madrasas is much more (less) than its population share as shown by the dashed line with
crosses (see Figure 10b). A somewhat similar albeit less striking pattern can be seen in the
aggregate distribution of Math score. For example, schools (madrasas) are under (over)
represented in the bottom decile compared to their sample share. However, in the top decile,
madrasas have a slightly higher share (relative to population fraction) whilst schools are
under-represented (relative to population fraction).
99. The very low level of achievement in Math and English is a matter of serious concern.
Nonetheless, this evidence is not surprising. It is fully consistent with findings from an earlier
attempt to assess student learning in rural secondary schools. Asadullah et al. (2007)
employed a similar Mathematics test to Grade 8 students and found that on average they
could correctly answer only 38% of the sample questions. This is very similar to the % of
correctly answered Math questions (i.e. 34%) in our QSSMEB sample. Findings of low
achievement documented in this and the previous studies therefore call for greater insights
Fi gure 10b: Engl i sh score (deci l e di stri buti on by i nsti tuti on type)
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1 (7) 2 (14) 3 (20) 4 (24 5 (28) 6 (33) 7 (36) 8 (40) 9 (45) 10 (59)
Test scor e deciles
D
e
c
i
l
e

s
a
m
p
l
e

f
r
a
c
t
i
o
n
School
Aliyah madrasa
Overall sample
share: school
Overall sample
share: Madrasa
Figure 10a: Math score (decile distribution by inst itution t ype)
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
1 (14) 2 (21) 3 (25) 4 (28) 5 (31) 6 (33) 7 (36) 8 (39) 9 (44) 10 (58)
Test scor e deciles
D
e
c
i
l
e

s
a
m
p
l
e

f
r
a
c
t
i
o
n
School
Aliyah madrasa
Overall sample
share: school
Overall sample
share: Madrasa
38
into the determinants of student achievement. In the next sub-section, therefore, we focus on
explaining differences in these outcomes in a regression framework using individual/student
level data.
5.4 Quality: Correlates of test scores
100. The regression analysis excludes data on Quomi madrasas and is restricted to the sample
with no missing cases for any correlate. This lead to exclusion of another 125 cases yielding a
regression sample of 8,648 observations. For the sake of brevity, we only discuss regression
estimates of correlates of student learning achievement using scores from Math and English
(regression results for the remaining tests are presented in the Appendix Tables 3a and 3b).
The dependent variable has been scaled by the number of questions items in the respective
test so that they now correspond to % of questions being correctly answered (on a scale of 0-
100). Table 10 reports regression estimates for Math score (summary statistics of variables
used in the regression are presented in the Appendix Table 2). Column 1 reports a
parsimonious specification only controlling for a dummy indicator for Aliyah madrasa
attendance (with schools being the base-category) while column 2 changes the base category
to aided schools. Therefore, in column 1 the regression constant corresponds to the
unconditional mean test score of school students while in column 2, the constant corresponds
to mean score of aided school students. The latter 3 models employ detailed specifications
and therefore highlight the role of various other factors in explaining learning outcomes
beyond that of institutional type. Estimates reported in specifications 3-5 therefore relate to
conditional gaps in test score by institutional type (with "school" being the omitted category).
Specification 3 adds control for (self-reported) family background (e.g. maternal and paternal
education, household assets and newspapers/magazines at home), while in columns 4 and 5
we introduce controls for school and teacher characteristics, respectively. Column 6 estimates
the model fully controlling for institution-specific (fixed) effects following least square
dummy variable approach.


39
Table 10: OLS regression of correlates of Math score
(1) (2) (3) (4) (5) (6)
Madrasa: Aliyah -2.653 -2.727 -1.955 -2.778 -1.675
(7.39)** (7.49)** (5.71)** (6.77)** (3.05)**
Govt. school -2.444
(1.69)
Private school -0.397
(0.58)
Female -5.058 -4.986 -5.155 -5.541
(16.35)** (16.17)** (16.40)** (22.23)**
Mother_edu: primary 1.026 1.313 1.267 0.046
(2.39)* (3.09)** (2.94)** (0.14)
Mother_edu: some secondary 1.249 1.150 1.112 0.237
(2.75)** (2.55)* (2.45)* (0.69)
Mother_edu: secondary complete 1.281 1.378 1.307 0.604
(2.46)* (2.67)** (2.52)* (1.55)
Mother_edu: post-secondary 2.464 2.327 2.212 1.155
(3.61)** (3.45)** (3.26)** (2.27)*
Father_edu: primary 0.382 0.716 0.741 0.612
(0.75) (1.43) (1.46) (1.62)
Father_edu: some secondary 0.410 0.442 0.604 0.680
(0.91) (0.99) (1.33) (2.02)*
Father_edu: secondary complete 0.647 0.902 1.050 0.759
(1.37) (1.93)+ (2.22)* (2.16)*
Father_edu: post-secondary 2.199 2.312 2.401 1.413
(4.45)** (4.72)** (4.88)** (3.82)**
Newspaper at home -0.329 -0.172 -0.060 0.484
(0.98) (0.52) (0.18) (1.82)+
Household asset index 0.797 0.554 0.598 0.344
(4.86)** (3.38)** (3.62)** (2.70)**
log(school land) -0.562 -0.508
(4.52)** (3.98)**
School inspected -5.198 -5.403
(13.32)** (13.55)**
School has mobile -0.465 -0.066
(1.26) (0.17)
log(# of rooms in sch) 1.582 0.367
(3.55)** (0.74)
Ln(fees in tot) 0.566 0.512
(2.72)** (2.39)*
Log(class size) 0.218 -0.045
(0.64) (0.13)
Fraction of female teacher 1.386
(1.00)
Fraction of trained teachers 2.840
(3.90)**
Fraction of temporary teacher -6.383
(3.64)**
log(mean teacher salary) -0.050
(0.21)
Constant 35.753 35.827 46.617 33.876 38.537 43.698
(191.46)** (182.10)** (3.85)** (2.79)** (3.15)** (3.51)**
N 8648 8648 8483 8393 8107 8483
R
2
0.01 0.01 0.15 0.17 0.17 0.57
Note: (a) Regression constant suppressed; (b) t-statistics in parenthesis; (c) base category for parental education
variable is never schooled or less than primary (i.e. grade 5) education; (d) Columns 3-5 include district
dummies; (e) Column 6 includes a full set of institution dummies. (f) All regressions control for child age and
age-squared. (g) Asset index is computed by aggregating binary indicators of household ownership of various
assets using the principal component method.

40
101. In all specifications, irrespective of the type of control variables used, there is a clear
statistically significant learning disadvantage associated with madrasa schooling. Starting
with the most parsimonious specification, column 1 tells us that the mean score by school
students is 35.75% correct answer while for madrasa students, this is only 2.6% less. In
model 2, the base category is changed to aided school so that the regression constant now
captures mean score of aided schools students. The fact that the regression constant does not
differ across specifications 1 and 2 implies that within the school sector, there is not a large
difference in test score. Indeed, this is also confirmed by the coefficients on dummies for
government and private unaided schools in column 2. Columns 3-5 report alternative
estimates of learning gaps associated with madrasa attendance by additionally controlling for
family background, school characteristics and teacher attributes in a sequential manner. In all
these four specifications, once again madrasa attendance variable has a large negatively
signed coefficient which is statistically significant. However, given the low overall mean
achievement in the data, this negative penalty is not large in magnitude. In other words, the
observed test score gap between Aliyah and schools is rather minor.
102. What is noteworthy is the relatively low R
2
obtained from our model. As discussed later
in this section, test score variance decomposition analysis indicates substantial association
between institutional factors and learning outcomes. In case of Math (English) test, 42%
(34%) of the total variation could be explained by institutions net of observed child and
family characteristics. The smaller R
2
simply indicates that while institutions do matter,
available indicators of institutional effectiveness as captured by observed characteristics
(such as class size, teacher characteristics and most importantly institution type) do a poor
job in explaining the institution effect. This is yet another reason why the observed negative
test score gap between school and Aliyahs should not take centre stage in the current
discourse over education quality in rural Bangladesh.
103. Apart from test score gaps by institution type, there is a significant learning
disadvantage experienced by girls in our sample. The coefficient on the female dummy is
large, negative and statistically significant. As a matter of fact, average gender gap in Math
score is much larger than the average negative gap associated with madrasa schooling. The
substantive significance of this gap is also evident if compared to the gap observed between
school and Aliyah students. Comparing the regression coefficients on female and Aliyah
dummy, we find that the gender penalty in our data is several times the penalty associated
with Aliyah madrasa attendance (see column 5).
104. Almost identical results are obtained from the regression analysis of English test score
(see Table 11 below). However, compared to Math, the penalty associated with being female
is less severe - but not so for madrasa attendance. The coefficient on madrasa dummy is now
even a bigger negative
41
Table 11: OLS regression of correlates of English score
(1) (2) (3) (4) (5) (6)
Madrasa: Aliyah -10.412 -10.400 -9.311 -10.809 -8.132
(23.10)** (22.74)** (20.74)** (19.88)** (11.18)**
Govt. school 2.794
(1.54)
Private school -0.436
(0.51)
Female -2.063 -1.686 -2.097 -1.108
(5.09)** (4.13)** (5.04)** (3.12)**
Mother_edu: primary 2.167 2.200 2.355 1.145
(3.85)** (3.90)** (4.13)** (2.47)*
Mother_edu: some secondary 2.442 2.276 2.433 1.544
(4.11)** (3.81)** (4.05)** (3.17)**
Mother_edu: secondary complete 2.323 2.296 2.519 1.825
(3.40)** (3.36)** (3.67)** (3.28)**
Mother_edu: post-secondary 3.726 3.485 3.596 1.599
(4.16)** (3.89)** (4.00)** (2.21)*
Father_edu: primary 0.349 0.551 0.596 0.418
(0.52) (0.83) (0.89) (0.78)
Father_edu: some secondary 0.766 0.737 0.823 1.264
(1.30) (1.25) (1.37) (2.64)**
Father_edu: secondary complete 1.102 1.244 1.158 1.759
(1.78)+ (2.01)* (1.85)+ (3.51)**
Father_edu: post-secondary 3.972 3.756 3.619 3.354
(6.13)** (5.78)** (5.55)** (6.37)**
Newspaper at home 0.057 -0.089 0.043 0.132
(0.13) (0.20) (0.10) (0.35)
Household asset index 0.821 0.675 0.742 0.352
(3.82)** (3.11)** (3.39)** (1.94)+
log(school land) -0.653 -0.617
(3.96)** (3.65)**
School inspected -3.334 -3.646
(6.45)** (6.90)**
School has mobile 0.169 0.438
(0.34) (0.87)
log(# of rooms in sch) 4.258 3.330
(7.20)** (5.08)**
Ln(fees in tot) 0.577 0.722
(2.10)* (2.54)*
Log(class size) -2.446 -2.753
(5.44)** (5.80)**
Fraction of female teacher 7.905
(4.29)**
Fraction of trained teachers 4.930
(5.12)**
Fraction of temporary teacher -15.332
(6.59)**
log(mean teacher salary) 0.330
(1.03)
Constant 37.124 37.113 88.234 78.176 79.355 -15.675
(158.26)** (150.16)** (5.56)** (4.86)** (4.90)** (0.88)
N 8648 8648 8483 8393 8107 8483
R
2
0.06 0.06 0.13 0.14 0.15 0.48
Note: (a) Regression constant suppressed; (b) t-statistics in parenthesis; (c) base category for parental education
variable is never schooled or less than primary (i.e. grade 5) education; (d) Columns 3-5 include district
dummies; (e) Column 6 includes a full set of institution dummies. (f) All regressions control for child age and
age-squared. (g) Asset index is computed by aggregating binary indicators of household ownership of various
assets using the principal component method.
42
105. We also carried out similar regression analysis using test score data on general
knowledge and Islamic studies (see Appendix Tables 3a and 3b). We found no madrasa-
school learning gap in general knowledge. This finding is not surprising as the regression
models control for family and parental background of the child and hence we do not expect
much residual learning that is shaped by the institution. However, significant institution-type
effect is observed in the regression model explaining student score in Islamic studies test,
with the effect being in favour of madrasas.
106. A large gender gap in test score is also evident from Appendix Tables 3a and 3b. In
general knowledge test, for instance, while there is no penalty for enrolment in Aliyah, there
still remains a very large gender gap. These findings together with results reported in Tables
10 and 11 imply that in rural secondary education sector in Bangladesh, the learning
disadvantage associated with a childs gender is much larger and persistent than that of
religious schooling.
107. The gender difference in test score is important for another reason. Compared to Aliyah
madrasas, secondary schools in rural Bangladesh (as well as those in our sample) have more
girls who on average tend to do worse compared to boys. In other words, madrasas in our
sample on average have lower score despite the fact that they have more boys who tend to
perform better.
108. To further explore the origin of gender gap in learning outcomes, Appendix Table 4
reports estimates of gender gaps in Math and English test scores separately for the school and
madrasa sample, net of institution fixed effects. The regressions also hold observed
differences in family backgrounds constant. Yet for both samples, the gender disadvantage
remains sizable across all outcomes
25
. However, the disadvantage is more pronounced in the
Aliyah sector. In Math test, for instance, girls on average have 5.2 percentage points deficit
(when compared to boys who on average correctly answer 38.9%). The female disadvantage
is more evident in the madrasa sample. Whilst madrasa boys could answer 37% questions
correctly, this is 6.3% less for girls of same madrasas. Compared to Math, the gender deficit
is much smaller in case of English test. However, it is still significant for the madrasa sample
and sizable (i.e. 1.5 percentage point). There is no significant difference between English test
score of boys and girls enrolled in the same school, however.
109. Therefore, while girls perform poorly in Math and English, their performance is worse
in the madrasa sample. This finding is also consistent with patterns observed in the raw data
(see Figure 8) and therefore highlights a new challenge for policy makers in Bangladesh
increasing share of female enrolment in Aliyah madrasas may do little in terms of closing the
gender gap in achievement.
110. There are other interesting findings from the regression analysis that we do not address
given our primary focus is on the madrasa coefficient. For instance, there is a systematically
positive correlation between household wealth (measured by an aggregate asset index) and
test score. Amongst institutional factors, there is a significant positive relationship between
learning outcomes and trained teachers, while there is a significant negative relationship
between temporary teachers and learning outcomes particularly for English. Similarly, our

25
Similar analysis is also presented for General Knowledge and Islamic studies tests (see Appendix Table
5). In both tests, girls once again significantly under-perform when compared to boys enrolled in the same
institution. However, the gap is particularly large in case of General Knowledge score -- in case of school
(Aliyah madrasa) sample, it is 5.35 (6.81) percentage points.
43
analysis suggests a negative and statistically significant association between larger class size
and English test score. However, in case of Math, the effect is not significant. Although
popularly perceived and rigorously documented as an important influence on learning
outcomes, there is no systematic evidence of this effect from Bangladesh so far. An earlier
attempt to identify the class size effect in secondary schools found no systematic evidence in
support of the hypothesis that smaller class sizes aid student learning (see Asadullah 2005).
Given the institutional differences across schools and madrasas in our sample, the issue needs
a careful treatment and hence is left for future research.
111. Finally, it should be borne in mind that the regressions presented above do not show
causation
26
. They only highlight a negative correlation between religious school attendance
and learning outcomes. First, as found in a previous study on Bangladesh, religious school
students do have lower test scores in mathematics when compared to their peers in secular
schools. However, conditional on selection into a given school type, there is no significant
learning differences across religious and secular schools (Asadullah et al. 2007). That finding
was similar to Newhouse and Beegle (2006) who find little difference in learning outcomes
between madrasa and non-madrasa students in Indonesia. It suggests that students who would
otherwise perform worse even in schools are disproportionately likely to enrol in madrasas.
In this report we do not elaborate upon the selection question an issue we hope to revisit in
our future research (which will also bring in detailed household-level controls). Re-analysis
of this issue taking into account the problem of selection will also make a general
contribution to the broader international literature on state-aided faith schools. Nonetheless,
preliminary econometric analysis of the issue yields results consistent with the previous study
on the issue for Bangladesh -- When we instrument for selection into institution-type (i.e., net
out factors that influence decision to enrol a child in an Aliyah madrasa from factors that are
associated with learning outcomes), the Aliyah madrasa coefficient becomes insignificant for
both mathematics and English specifications
27
. This further underscores our point that quality
is low across the board regardless of institution-type.
112. The other issue left for future research is the analysis of educational production in
Quomi madradsas. As mentioned earlier, many of the Quomis do not use a grade based
system. Therefore, enumerators in the QSSMEB survey had to rely on principals or teachers
for identification of the equivalent grade and hence, selection of students to be tested. This
practice meant that an analysis of test score data collected on Quomis could be subject to
statistical bias. Quomi authorities may have exploited the ambiguity over grade equivalency
and deliberately identified an older, selective cohort of students for testing purposes. In the
absence of a grade-based class registrar, it was not possible for the enumerator to physically
verify these issues. Indeed, analysis of student characteristics in sample Quomi madrasas
indicate that they are much older compared to their counterparts in registered schools and
madrasas (aged 16 years compared to the sample average of 13; see Appendix Table 1).
113. In sum, our detailed analysis of various correlates of learning in the state regulated
segment of secondary education sector in rural Bangladesh highlights the following. First,
while statistically significant, test score difference between schools and madrasas is not

26
For example the negative correlation between school inspections and test scores could be due to the fact that
poorly performing schools are more likely to be visited.
27
We do not present the IV-regressions in this report, but are available from the authors upon request. We use
total number of schools and Quomi madrasas in the union as excluded instruments to predict selection into
Aliyah madrasas. In the first stage regression, both instruments were highly significant and negatively
influenced the probability of enrolment in Aliyah madrasa. Most importantly, they both comfortably passed the
over-identification tests.
44
substantively: overall level of learning is so low that the apparent learning disadvantage
associated with (Aliyah) madrasa attendance is not large. Second, even after detailed control
for a battery of factors all of which are widely believed to promote learning, we can only
explain 17% (15%) of the total variation in Math (English) test score. Note that the final
regression model (i.e. column 5) in Tables 10 and 11 control for conventional educational
inputs such as class-size, institution type, total land area of the institution, whether the
institutions is being inspected by an external authority, total number of rooms in the school,
total amount of fees charged in a year and a wide selection of teacher attributes (such as
fraction teachers being female, fraction of trained teachers, fraction of teachers being on a
temporary contract, and mean teacher salary). Inclusion of such an extensive set of controls
does partly explain the learning gap between schools and madrasas: the penalty associated
with madrasa attendance is reduced by nearly 40% (26%) in case of Math (English) as we
move from column 1 to column 5 in Table 10 (Table 11). However, the overall gain in the
explanatory power of the model remains modest. In other words, while the conventional
correlates of learning do have some explanatory power in the context of rural schools in
Bangladesh, most of the process of test score production still remains unknown. It certainly
does not mean that no learning takes place in school. In both Tables 10 and 11, the
explanatory power of the regression reported in column 6 is substantial: Holding school
specific fixed effects constant, we can explain 57% (48%) of the variation in Math (English)
test score. However, the commonly perceived observed institutional features of schools and
madrasas do very little in explaining this otherwise large association between institutions and
learning outcomes.
114. To further probe into this issue, Appendix Table 6 reports results from (regression
based) variance decomposition analysis. Total variance in test score is decomposed in 3 parts:
portion explained by (a) district of residence, (b) child and family characteristics and (c)
institution effects. For this purpose, we started with a simple model of student score only
controlling for district fixed effects. We then sequentially added further controls for child
(and family) characteristics and institution fixed effects. On the basis of resultant changes in
R
2
value across the three estimated models, it was possible to work out what fraction of
achievement variation was explained by observable characteristics specific to the child and
school above and beyond the district effect. After accounting districts, child and family
specific observables, and institutions, the remaining variation is idiosyncratic and is reported
as the fourth element in the table. This analysis is carried out separately for student scores in
each of the four tests. We also report the result separately for the school and Aliyah sample.
The usual caveat applies, as before. Given that our regression models are not causal, the
decomposition analysis carried out does not imply causal relations either.
115. In the full sample, institutions account for 42% (34%) of the total variation in Math
(English) test score. When compared to test score on other subjects, institutions have
strongest effect in Math. Similar patterns are obtained when sub-sample specific results are
considered. In the school (madrasa) sub-sample, institutions explain an additional 38 %
(51%) of Math test score variation above districts and observable child/family characteristics.
This is almost three times the amount of variation explained by districts and household
characteristics in school as well as in the madrasa sample. In sum, the institution effect is
very large and all of this cannot be dismissed on the basis of selection on unobservables:
institutions do matter! Once again, looking across subjects, we find that the institution
effect is relatively small in general knowledge and Islamic studies. However, in case of the
latter, the effect is larger for madrasa sample.
45
116. In sum, the lack of sizable between-sector (school vs. Aliyah) difference in learning
outcomes in our study does not mean that no large between-institution difference exists.
Rather, the institutions role appears even more significant when compared to the importance
of family influence and place of residence in explaining test score variance. But more
research is needed to unpack the black-box policy of the institution effect (e.g., detailed
information on teacher time-on-task). We neither know what observed institutional metric
and teacher attribute can explain the effect best and/or to what extent these are conflated with
the effect of unobserved family-specific correlates or learning.

46
Chapter 6: Conclusion
6.1 Summary
117. The evidence presented herein constitutes by far the most detailed account of secondary
educational institutions in Bangladesh, and unregistered traditional madrasas in general in
any Muslim majority country. The evidence presented on the incidence of religious education
in this report comes from a variety of sources, namely: (a) multiple rounds of nationally
representative household survey data; (b) nation-wide census records on registered secondary
educational institutions; (c) complete census of all educational institutions in 48 unions across
12 districts; (d) a near census assessment of secondary students enrolled in grade 8 (or its
equivalent) in the sample unions; and (e) complete population census in 96 villages in our
sample unions. While the analysis in this report only represents an initial foray into the vast
data collected in this endeavour (please see appendix for examples of specific studies to be
conducted with the dataset), a number of conclusions can be drawn on the basis of the
findings reported in this study.
118. We do find some evidence that madrasas are more likely to be located in poor rural
areas, and there is a relationship between household poverty and madrasa attendance. The
income effect is, however, small in magnitude indicating that there are other major factors
besides household wealth that influences madrasa enrolment. Regardless of what factors
ultimately shape household decision to send their children to madrasas, the incidence in terms
of enrolment share of unregistered traditional madrasas in primary and secondary education
is very small. This finding is similar to the only other serious study on the incidence of
madrasas which finds that enrolment share of madrasas in three districts of Punjab Pakistan is
also extremely small (Andrabi et al, 2005). It is also consistent with the available data on
enrolment share of madrasas in India where most of the Islamic seminaries operate outside
the state sector: A high level government commissioned study on the educational status of the
Muslim community in India report only 3% of Muslim children among the school going age
going to Madarsas (Sachar, 2006). However, unlike India and Pakistan, Bangladesh does
have a large registered reformed madrasa sector. For example, the combined enrolment share
of madrasas (Aliyah and traditional combined) is larger than the NGO share in primary.
While there is a lot of emphasis given to the role of NGOs in the primary sector of
Bangladesh, there is very little discussion on the role played by madrasas in service provision
in rural Bangladesh. This finding is extremely important for the exact size of the madrasa
education sector in Bangladesh has been subject to intense media speculation. Again, going
back to our earlier citation of Ahmed (2005) on behalf of the International Crisis Group
(ICG): Bangladesh's madrasa sector has mushroomed, reaching an estimated 64,000
madrasas from roughly 4,100 in 1986, with little if any government oversight. While there
indeed has been an expansion in the share of students studying in the madrasa sector, it has
not been without any government oversight to a large extent it has rather been due to a
conscious government policy to bring secondary school madrasas into the mainstream.
119. The first recommendation from our report is that there needs to be a more nuanced
policy discussion regarding madrasas reform we must distinguish between Aliyah and
Quomi madrasas not only for syntactic reasons but for policy relevance. Private Aliyah
madrasas funded by the government and regulated by the state sanctioned Madrasa Board
now have a similar curriculum to that of aided private secular schools. Private unaided
traditional Quomi madrasas still have a major difference in curriculum content and in
pedagogic orientation. Theoretically at least, further Aliyah reform is relatively easier than
47
trying to reform Quomis. The transaction costs of Aliyah reform should be lower given that
the state has to negotiate with one central Board, not to mention that Aliyahs themselves are
dependent upon state financing. On the other hand, the overwhelming number of Quomis are
not registered, operating without any government oversight, some under a plethora of
informal undocumented boards, and most of all financially independent of the state.
120. For the sake of argument, let us assume away the small traditional madrasa sector. Is
there an issue with the rising share of enrolments in registered government-aided Aliyah
madrasas? This study tries to explore this from the dimension of learning outcomes.
28
It is
our hope that the policy dialogue can be anchored along the metrics of learning assessments
in subjects such as Mathematics, Science, English, and Bengali. To that end our report draws
upon the near census assessment of secondary students enrolled in Grade 8 in the sample
unions to focus on achievement levels in Math and English. We do find that Aliyah madrasas
fare worse than their school counterparts, particularly in English.
121. The second recommendation from this report is that we need to reduce the quality gap,
particularly in English, between public-aided secondary schools and Aliyahs Aliyahs must
be held responsible for improving learning outcomes. Efforts to improvement learning
outcomes should also have a gender focus. The gender gap in quality of learning outcomes is
most pronounced in madrasas. Thus, while Aliyah madrasas have played an instrument role
in reducing gender inequality in access, the Government should ensure that they are also not
inadvertently increasing gender inequality in learning. This finding also highlights the
shortcomings of only relying upon SSC pass rates as an indicator of quality. The SSC pass
rate is higher for aided-Aliyah madrasas compared to aided-schools
29
. However, we find that
in terms of assessment of actual numerical and literacy skills, Aliyahs fair worse than
schools. Currently the SSC examination system is under different regional and institutional
boards making it difficult to compare quality across providers.
122. Having underscored the learning deficiency in Aliyahs, we should immediate stress the
point that it should not be about Aliyahs raising their standards to match schools both need
to be held accountable for raising learning outcomes. The punch line of this report remains
that quality of schooling in rural Bangladesh is poor regardless of institution type. Thus, the
third recommendation is that the government should find more innovative ways to link
substantial public resources that it pours into these aided private institutions, religious or
otherwise, with concurrent improvements in numeracy and literacy skills. This requires
holding each and every type of provider accountable to improving quality standards. In that
regard the government has to take learning assessments seriously. It would be ideal if
Bangladesh participated in international testing such as TIMSS or institutionalize a national
assessment of numeracy and literacy skills. Furthermore widespread dissemination of
information on learning outcomes should be made available to parents, communities, schools,
media, and larger civil society to increase ownership of the quality discourse. More
information flows and parental/community involvement should also help to improve the
governance and accountability of these institutions. Ultimately, it is up to the state to link its
financial support to providers with minimum performance standards based on student

28
There is the dimension of social outcomes that this paper has not explored in detail. Issues such as these
require careful exploration of a host of individual, household, institutional, and community factors that
ultimately shape social attitudes. For example, Asadullah and Chaudhury (2009, Forthcoming) have shed some
light on this issue, and do find that female Aliyah madrasa graduates have a more traditional view of female
roles (schooling, work, family life) compared to their counterparts in schools.
29
http://www.banbeis.gov.bd
48
learning outcomes. The Government has done an excellent job tying provision of schooling to
teacher pay it needs to be more creative in beginning to seriously tie performance to teacher
payment as well, particularly in the non-state sector where it is less encumbered by civil-
service bureaucratic constraints.
123. Besides fiscal incentives tied to performance, the government should also facilitate
identification and placement of qualified teachers. Recently the government has carried out
several policy reforms and project interventions to improve the quality of aided institutions.
For example, under a World Bank supported education sector reform credit, the government
has implemented a National Teacher Registration and Certification Authority (NTRCA) body
to ensure that only qualified teachers could be hired in these aided schools and Aliyahs.
Initiatives such as NTRCA are a major step in the right direction towards improving
transparency and accountability. The government has also recently launched an innovative
project with World Bank support (Secondary Education Quality and Access Enhancement
Project - SEQAEP) to improve learning outcomes. Under SEQAEP, a series of fiscal
incentives tied to improvement in performance are given to institutions, teachers, and pupils.
Furthermore, acknowledging the difficulty these private institutions have in recruiting
qualified mathematics and English teachers to work in rural areas, the government is helping
to identify and post a pool of qualified mathematics and English teachers to work with
teachers and pupils in project schools on a temporary basis. Programs and projects such as
these are essential in improving the quality of education in rural Bangladesh.
124. Given the difficulty in coordinating improvements in quality and governance in
thousands of aided private schools, some have called for increasing the provision of public
secondary schooling. The government voted into office in December 2008, has already
expressed the need for setting up some more government high schools in rural areas. It is not
automatic that these new government high schools will do better than aided private schools.
Currently government high schools (on average) do better (in terms of 10
th
grade secondary
school completion exams) than their private aided counterparts - given that the government
focuses on a handful of high schools located predominantly in urban areas/district
headquarters with tight admission standards and staffed with highly qualified teachers.
However, it is not known to what extent this apparent government school advantage is owing
to selection of students of higher socio-economic backgrounds and screening out of lower-
ability students through entrance exams. Besides, it will be non-trivial to replicate the model
high school experience in remote rural areas. Note that instead of increasing the supply of
public high school staffed with civil-servant teachers, these new schools could be instead still
managed by the government, however, staffed by non-civil servant teachers on a contract
basis with payment/contract tied to performance. While the government contemplates
various options, at least in the short to medium term, it should continue to focus on improving
quality and accountability in these private aided institutions.
125. The first National Education Commission in Bangladesh 1974, headed by Dr. Kudrat-e-
Khuda, recommended uniform, free and compulsory primary education of 8-years' duration.
Anyone wishing to pursue Madrasa education could do so after the 8th grade according to the
Commission. Interestingly, a new commission formed in 2003 had the following
recommendations: introduction of the same textbooks in Bangla, English and Mathematics in
Ebtidai Madrassas (primary Aliyah equivalent) as in public (and government-aided) primary
schools. The current Government has announced that the Kudrat-e-Khuda 1974 Commission
findings will form the basis of its overall education reform agenda. We hope that key
49
insights from that Commission, as well as more recent serious empirical studies will help to
inform the madrasa reform dialogue.
126. Now coming back to the issue of unregistered Quomi madrasas what is the role that
the government can play? Even though a tiny fraction of rural children attend these
institutions, the state should have an obligation to ensure quality education for all. The issue
of school choice, particularly in rural Bangladesh is quite complicated though. Many of these
households who send their children to Quomis might actually have had alternatives they
could have sent their child to a school or an Aliyah madrasa, but still picked a Quomi
madrasa. Is the customer always right? If the parents only care about whether their child
excels in religious studies, then as we point out in the report Quomis are actually doing a
good job. Unlike traditional madrasas in Pakistan, traditional madrasas in Bangladesh seem
to have undergone some structural changes even in the absence of any state intervention.
Some have started to admit girls in recent years in addition to undertaking some
modernization of the curriculum. Still, the pace of this change is unsystematic and slow. The
fourth recommendation from this report is that we cannot ignore Quomis given that the
overwhelming majority of their graduates have very limited exposure to subjects that are
universally recognized as critical to a modern economy. The government should engage in a
dialogue with Quomi madrasas to explore options such as introducing modern subjects,
paired with fiscal incentives (a la Aliyahs), and oversight under a unified regulatory body.
127. A final recommendation from the authors is that the policy discourse on madrasas in
Bangladesh should have a sound empirical basis. Unfortunately, that is currently not the
situation. For example, in a recent edition of the Harvard International Review, Evans
(2008) states that madrasas in Bangladesh have risen as competition against private English
medium schools which could not be further from reality.
128. While the focus of this report has been on the secondary sector, it is important to note
that there has not been any serious attempt to reform madrasas in the primary sector. Even
though a small fraction of children graduate from primary madrasas, there is serious concern
about quality of learning had persistence of underperformance through secondary cycle
(Asadullah et al. 2007). Compared to MOE, there has not been any major reform initiative
from MOPME. Similar to the secondary sector, this reform agenda must be embodied in a
larger reform agenda to improve quality across the board for all primary institutions given the
low quality of learning.
50
References
Abdalla, Raisuddin and Hussein (2004) Bangladesh Educational Assessment: Pre-primary
and Primary Madrasah Education Bangladesh. Dhaka: The United States Agency for
International Development (USAID).
Ahmad, Mumtaz (2004) "Madrassa Education in Pakistan and Bangladesh". In Satu Limaye,
Robert Wirsing and Mohan Malik (eds.) Religious Radicalism and Security in South
Asia. Asia-Pacific Center for Security Studies: Honolulu.
Ahmed, Samina (2005) Testimony prepared for the US Senate Foreign Relations Committee
Hearing on 'Combating Terrorism through Education - the Near East and South Asian
Experience'. Washington D.C.
Asadullah, M. Niaz (2005) The Effect of Class Size on Student Achievement: Evidence from
Bangladesh, Applied Economics Letters, Vol. 12(4), 15, pp. 217-221.
Asadullah, M. Niaz (2006) "Returns to education in Bangladesh", Education Economics, 14,
457-712.
Asadullah, M. Niaz (2009), "Returns to Private and Public Education in Bangladesh and
Pakistan: A Comparative Analysis", J ournal of Asian Economics, Vol. 20(1), pp. 77-86.
Asadullah, M. Niaz and Chaudhury, N. (2006) "Religious schools, social values and
economic attitudes: Evidence from Bangladesh", QEH Working paper 139, University
of Oxford (Forthcoming in World Development).
Asadullah, M. Niaz and Chaudhury, N. (2008a) "Holy Alliances: Islamic high schools, public
subsidies, and female schooling in Bangladesh". In Mercy Tembon and Lucia Fort
(eds.) Girl's Education in the 21st Century: Gender equality, Empowerment and
Economic Growth. World Bank: Washington, DC. (Forthcoming in Education
Economics).
Asadullah, M. Niaz and Chaudhury, N. (2008b), "Madrasas and NGOs: Complements or
Substitutes? Non-State Providers and Growth in Female Education in Bangladesh",
World Bank Policy Research Working Paper no 4511.
Asadullah, M. Niaz and Chaudhury, N. (2009) "Reverse Gender Gap in Schooling in
Bangladesh: Insights from Urban and Rural Households", J ournal of Development
Studies, Vol. 45, issue 8, pp 1360-1380.
Asadullah, M. Niaz and Chaudhury, N. and Dar, A. (2007) "Student Achievement
Conditioned Upon School Selection: Religious and Secular Secondary School Quality
in Bangladesh", Economics of Education Review, Vol. 26(6), pp 648-659.
BANBEIS (2006). Secondary School Key Educational Indicators prepared by the Bangladesh
Bureau of Educational Information and Statistics. Accessed on 15/10/2006 from
website
http://www.banbeis.gov.bd/db_bb/secondary_education_1.htm
CAMPE (2005) "The State of Secondary Education Progress and Challenges".
http://www.campebd.org/content/EW_2005.htm
Das, J ishnu and Tristan Zajonc (2008) India Shining and Bharat Drowning: Comparing Two
Indian States to the Worldwide Distribution in Mathematics Achievement, Policy
Research Working Paper 4644.
Evans, A. (2008) "Madrasah Education: Necessity or Rational Choice?" Harvard
International Review, Vol. 30 (3).
Fahimuddin (2004) Modernization of Muslim education in India. Delhi, Adhyayan.
Golding, Claudia and Lawrence F. Katz (2008). The Race between Education and
Technology; Harvard University Press; Cambridge, USA.
Griswold, Eliza (2005) "The Next Islamist Revolution?", J anuary 23, New York Times;
Magazine Desk.
51
Ladbury, Sarah (2004) "Madrassahs and Islamic Education in South Asia: What the donor
community needs to know?", Report prepared for the World Bank, South Asia
Department.
Lintner, Bertil (2003) "Religious Extremism and Nationalism in Bangladesh", Faultlines, 14
(The Institute of Conflict Management, New Delhi).
Khan, Amirullah, Mohammad Saqib and Zafar Anjum (2003) "To kill the mockingbird"
(mimeo).
McEwan, Patrick J . (2001) "The Effectiveness of Public, Catholic, and Non-Religious Private
Schools in Chile's Voucher System", Education economics.
Mercer, M., Murshid, T., Kabir, M. and Khan, A (2006) "Madrasas in Bangladesh and
Pakistan: an analysis" (mimeo).
Metcalf, Barbara (1978) The Madrasa at Deoband: A Model for Religious Education in
Modern India, Modern Asian Studies, Vol. 12(1), pp. 111-134
Newhouse, D., and Beegle, K. (2006). The effect of school type on academic achievement:
Evidence from Indonesia. J ournal of Human Resources, 41 (3): 529-557.
Sachar, Rajindar (2006) Social, Economic and Educational Status of the Muslim
Community of India: A Report, Prime Ministers High Level Committee Cabinet
Secretariat Government of India.
Sattar, Abdus (2004) Bangladesh madrasa shikhkha o samaj jibone tar provab (Madrasa
education in Bangladesh and indications in social life). Dhaka: Islamic Foundation.
Stern, J essica (2001) "Preparing for the War on Terrorism", Current History, Volume 100,
Numbers 642-650.




52
Appendix Table A: Aliyah madrasa by grade, equivalence, duration, type of study and
core subjects
Grade
Mainstream
equivalent
Duration
(years)
Type of study Core subject Optional subjects
Ebtidai Primary 5 General and
religious
Quran, Bangla, English,
grammar, mathematics,
history, geography, science,
jurisprudence, sports,
handicrafts
Urdu and Persian
Dakhil Secondary 5 General,
science,
Mujabbid,
Hifzul Quran
Quran, Arabic, Bangla,
English, jurisprudence,
mathematics, geography,
Islamic history, science
Urdu, Persian, Hadith
Alim Higher
secondary
2 General,
science,
Mujabbid
Religious: Quran, Hadith,
Bangla jurisprudence.
Science: Arabic,
mathematics, science.
Humanities: Arabic, Islamic
history, economics
Either English, Urdu or
Persian
Fazil Bachelors 2 General, science Tafsir, Hadith , Arabic,
Bangla.
General: J urisprudence,
Islamic history, Aqaid and
Balagat.
Science: English, physics,
chemistry, mathematics,
biology
Either English, Urdu or
Persian
Kamil Masters 2 Theology Hadith, Tafsir, Fiqh, Adab
(i.e. traditions,
interpretation, jurisprudence,
etiquette)

Source: Mercer et al. (2006)
53
Appendix Table B: Quomi madrasa by grade, equivalence, duration, type and core
subjects
Grade Mainstream equivalent
Duration
(years)
Ebtidai Primary 5
Mutwasita Secondary 5
Sambia Ulya Higher Secondary 2
Fazeelat Bachelors 2
Taqmeel Masters 2
Iimul Qiraat Wat Taujid Post Masters elective 1
Hifzul Quran Post Masters elective 1
Source: Mercer et al. (2006).
Appendix Table 1.a: Madrasha enrolment (%) of girls by education level and year
Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5
Primary NGO 2000 12.2 10.6 0.9 1.5 0.8
Primary Madrasha 2000 2.2 2.4 2.6 0.8 1.6
Secondary Madrasha 2000 3.6 2.9 3.4 6.0 3.0
Primary NGO 2005 6.0 4.0 1.1 1.7 1.6
Primary Madrasha 2005 4.6 4.7 4.5 1.7 0.5
Secondary Madrasha 2005 9.1 9.8 5.3 4.9 3.8
Source: Authors calculation based on HIES 2000 and 2005 data.

Appendix Table 1.b: Madrasha enrolment (%) of boys by education level and year
Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5

Primary NGO 2000 5.9 3.3 2.5 - -
Primary Madrasha 2000 5.0 - 1.7 4.8 3.4
Secondary Madrasha 2000 9.3 3.8 7.7 4.7 7.7
Primary NGO 2005 5.9 1.6 4.2 2.9 0.9
Primary Madrasha 2005 5.9 4.8 5.8 4.7 1.8
Secondary Madrasha 2005 14.3 10.9 5.9 5.9 5.3
Source: Authors calculation based on HIES 2000 and 2005 data.

Appendix Table 1.c: Urban Madrasa Enrolment (%)
Quitile1 Quitile2 Quitile3 Quitile4 Quitile5
Primary Madrasha 2000 - - 1.1 2.0 1.2
Secondary Madrasha 2000 3.2 1.7 4.4 1.6 1.9
Primary Madrasha 2005 4.96 4.65 3.40 4.35 1.30
Secondary Madrasha 2005 7.46 7.96 3.01 4.03 3.05
Source: Authors calculation based on HIES 2000 and 2005 data.

Appendix Table 1.d: Rural Madrasa Enrolment (%)

Quitile1 Quitile2 Quitile3 Quitile4 Quitile5
Primary Madrasha 2000 5.8 2.4 2.8 3.2 3.2
Secondary Madrasha 2000 7.9 4.6 6.2 7.3 6.9
Primary Madrasha 2005 5.38 4.85 6.36 2.16 1.12
Secondary Madrasha 2005 13.39 11.27 6.94 6.27 6.10
Source: Authors calculation based on HIES 2000 and 2005 data.
54
Appendix Table 2: Summary statistics of regression variables
Full School Madrasa: Aliyah Madrasa: Quomi
Variable Mean sd Mean sd Mean sd Mean sd
TIMSS score 34.51 15.06 35.57 14.46 33.14 15.84 29.34 16.40
English score 33.33 19.33 36.88 19.31 26.69 16.70 22.54 17.90
Islam score 74.72 21.40 69.94 20.91 81.54 19.84 92.92 13.04
General knowledge score 38.67 20.49 38.35 20.50 38.52 20.27 42.43 20.77
Female 0.59 0.49 0.61 0.49 0.65 0.48 0.24 0.42
Age 13.32 1.39 13.04 0.96 13.34 1.16 15.95 2.50
Mother_edu: primary 0.19 0.39 0.18 0.39 0.20 0.40 0.18 0.38
Mother_edu: some secondary 0.17 0.37 0.17 0.37 0.17 0.37 0.17 0.37
Mother_edu: secondary completed 0.13 0.33 0.13 0.34 0.12 0.32 0.09 0.29
Mother_edu: post secondary 0.06 0.24 0.08 0.26 0.04 0.19 0.04 0.18
Father_edu: primary 0.12 0.32 0.12 0.32 0.13 0.33 0.09 0.29
Father_edu: some secondary 0.17 0.37 0.17 0.37 0.17 0.37 0.18 0.38
Father_edu: secondary completed 0.16 0.36 0.16 0.37 0.16 0.37 0.14 0.35
Father_edu: post secondary 0.18 0.38 0.19 0.39 0.14 0.35 0.21 0.41
Newspaper at home 0.38 0.49 0.39 0.49 0.33 0.47 0.54 0.50
# of hh assets 0.38 0.19 0.39 0.19 0.35 0.18 0.32 0.17
# of tot. farm animals 2.21 2.57 2.27 2.58 2.38 2.55 0.98 2.05
Govt. school 0.01 0.11 0.02 0.13
Private school 0.06 0.23 0.08 0.27
Aided school 0.62 0.49 0.90 0.30
Aided Aliyah 0.21 0.41 0.85 0.36
Private Aliyah 0.04 0.19 0.15 0.36
Quomi 0.07 0.26 1.00 0.00
log(school land) 4.89 1.28 4.91 1.26 4.77 1.26 5.09 1.49
School inspected 0.77 0.42 0.82 0.38 0.69 0.46 0.53 0.50
School has mobile 0.33 0.47 0.32 0.46 0.25 0.44 0.71 0.46
log(# of rooms in sch) 2.67 0.53 2.61 0.41 2.70 0.30 3.19 1.32
Ln(fees in tot) 4.77 0.89 4.95 0.77 4.20 0.91 5.00 1.04
Fraction of female teacher 0.15 0.14 0.18 0.12 0.09 0.12 0.10 0.23
Fraction of trained teachers 0.55 0.32 0.69 0.25 0.25 0.18 0.21 0.32
Fraction of temporary teachers 0.05 0.12 0.06 0.12 0.03 0.11 0.05 0.17
log(mean teacher salary) 8.49 0.72 8.63 0.36 8.29 1.20 7.92 0.46
District: Barisal 0.08 0.27 0.08 0.26 0.08 0.28 0.14 0.34
District: Bogra 0.07 0.26 0.07 0.26 0.07 0.25 0.08 0.27
District: Chandpur 0.10 0.30 0.11 0.31 0.12 0.32 0.01 0.10
District: Chittagong 0.08 0.27 0.06 0.24 0.07 0.26 0.26 0.44
District: Comilla 0.14 0.35 0.16 0.36 0.13 0.33 0.08 0.27
District: Faridpur 0.08 0.27 0.08 0.27 0.07 0.25 0.08 0.27
District: Jessore 0.11 0.32 0.12 0.32 0.11 0.32 0.05 0.21
District: Kurigram 0.07 0.25 0.07 0.25 0.08 0.27 0.02 0.14
District: Laksham 0.10 0.30 0.10 0.31 0.11 0.31 0.05 0.23
District: Mymensing 0.06 0.24 0.06 0.24 0.06 0.25 0.07 0.26
N (sample size) 9407 6431 2322 654

55
Appendix Table 3a: OLS regression of determinants of General Knowledge test
(1) (2) (3) (4)
Madrasa: Aliyah -0.043 0.619 -0.548 0.767
(0.09) (1.29) (0.94) (0.99)
Female -4.943 -5.220 -5.454
(11.38)** (11.91)** (12.24)**
Mother_edu: primary 1.043 1.341 1.641
(1.73)+ (2.22)* (2.69)**
Mother_edu: some secondary 1.947 2.149 2.384
(3.05)** (3.36)** (3.70)**
Mother_edu: secondary complete 1.532 1.788 1.917
(2.09)* (2.44)* (2.61)**
Mother_edu: post-secondary 1.442 1.883 2.066
(1.50) (1.96)+ (2.15)*
Father_edu: primary 0.521 0.503 0.854
(0.73) (0.70) (1.19)
Father_edu: some secondary 0.691 0.382 0.573
(1.09) (0.60) (0.89)
Father_edu: secondary complete 0.081 0.058 -0.014
(0.12) (0.09) (0.02)
Father_edu: post-secondary 2.048 2.126 2.057
(2.95)** (3.05)** (2.95)**
Newspaper at home 0.542 0.517 0.477
(1.15) (1.09) (1.00)
Household asset index 0.880 0.877 0.868
(3.82)** (3.76)** (3.70)**
log(school land) 0.114 0.276
(0.64) (1.53)
School inspected -1.599 -1.900
(2.88)** (3.36)**
School has mobile 1.050 1.490
(2.00)* (2.75)**
log(# of rooms in sch) -2.457 -0.810
(3.87)** (1.15)
Ln(fees in tot) -0.353 -0.386
(1.20) (1.27)
Log(class size) -2.241 -2.118
(4.65)** (4.17)**
Fraction of female teacher 0.090
(0.05)
Fraction of trained teachers 4.018
(3.89)**
Fraction of temporary teacher -1.227
(0.49)
log(mean teacher salary) -1.397
(4.07)**
Constant 38.593 69.494 87.666 91.194
(149.89)** (4.09)** (5.08)** (5.26)**
N 8648 8483 8393 8107
R
2
0.00 0.12 0.12 0.13
Note: (a) Regression constant suppressed; (b) Columns 2-4 include district dummies; (c) t-statistics in
parenthesis. (e) All regressions control for child age and age-squared. (g) Asset index is computed by
aggregating binary indicators of household ownership of various assets using the principal component method.
56
Appendix Table 3b: OLS regression of determinants of Islamic studies test score
(1) (2) (3) (4)
Madrasa: Aliyah 11.370 1.201 1.140 1.267
(22.40)** (24.15)** (18.84)** (15.38)**
Female -0.749 -0.726 -0.720
(16.30)** (15.67)** (15.18)**
Mother_edu: primary 0.110 0.105 0.094
(1.75)+ (1.67)+ (1.47)
Mother_edu: some secondary 0.158 0.146 0.160
(2.34)* (2.15)* (2.32)*
Mother_edu: secondary complete 0.152 0.153 0.158
(1.97)* (1.98)* (2.02)*
Mother_edu: post-secondary 0.197 0.191 0.202
(1.92)+ (1.86)+ (1.95)+
Father_edu: primary -0.072 -0.055 -0.056
(0.96) (0.73) (0.74)
Father_edu: some secondary 0.151 0.150 0.142
(2.27)* (2.24)* (2.08)*
Father_edu: secondary complete 0.170 0.186 0.165
(2.42)* (2.65)** (2.30)*
Father_edu: post-secondary 0.353 0.364 0.350
(4.82)** (4.95)** (4.71)**
Newspaper at home 0.076 0.062 0.078
(1.54) (1.26) (1.55)
Household asset index 0.048 0.044 0.040
(1.97)* (1.81)+ (1.60)
log(school land) -0.115 -0.101
(6.12)** (5.19)**
School inspected -0.344 -0.364
(5.96)** (6.12)**
School has mobile -0.017 -0.000
(0.30) (0.00)
log(# of rooms in sch) 0.129 0.127
(1.93)+ (1.70)+
Ln(fees in tot) 0.101 0.110
(3.26)** (3.42)**
Log(class size) -0.202 -0.258
(3.99)** (4.79)**
Fraction of female teacher -0.315
(1.50)
Fraction of trained teachers 0.439
(3.97)**
Fraction of temporary teacher -0.705
(2.69)**
log(mean teacher salary) -0.056
(1.56)
Constant 70.172 10.348 10.278 10.761
(256.79)** (5.87)** (5.74)** (5.93)**
N 8011 7862 7776 7511
R
2
0.06 0.15 0.16 0.17
Note: (a) Regression constant suppressed; (b) Columns 2-4 include district dummies; (c) t-statistics in
parenthesis. (d) Sample restricted to Muslim children only. (e) All regressions control for child age and age-
squared. (f) Asset index is computed by aggregating binary indicators of household ownership of various assets
using the principal component method.
57
Appendix Table 4: Within-institution estimates of gender gap in Math and English
scores
Math English
School Aliyah School Aliyah
(1) (2) (1) (2) (1) (2) (1) (2)
Female -5.215 -5.281 -6.353 -6.374 -0.620 -0.816 -1.542 -2.107
(17.73)** (17.83)** (14.25)** (13.94)** (1.42) (1.88)+(2.72)** (3.66)**
Age -1.231 0.357 -5.149 0.227
(0.64) (0.16) (1.83)+ (0.08)
age, sq 0.021 -0.036 0.124 -0.052
(0.30) (0.47) (1.19) (0.54)
Mother_edu: primary 0.229 -0.431 1.084 1.193
(0.58) (0.76) (1.87)+ (1.68)+
Mother_edu: some secondary 0.529 -0.506 1.708 1.002
(1.28) (0.85) (2.81)** (1.33)
Mother_edu: secondary complete 1.065 -0.686 1.883 1.501
(2.27)* (0.99) (2.73)** (1.72)+
Mother_edu: post-secondary 1.790 -1.552 1.973 -0.811
(3.07)** (1.43) (2.30)* (0.59)
Father_edu: primary 0.743 0.278 0.458 0.446
(1.62) (0.42) (0.68) (0.54)
Father_edu: some secondary 0.474 1.281 1.236 1.294
(1.17) (2.14)* (2.09)* (1.72)+
Father_edu: secondary complete 1.086 -0.160 1.960 1.178
(2.54)* (0.26) (3.13)** (1.54)
Father_edu: post-secondary 1.478 0.999 3.598 2.395
(3.36)** (1.47) (5.57)** (2.80)**
Newspaper at home 0.407 0.684 -0.021 0.496
(1.29) (1.39) (0.05) (0.80)
Asset index 0.287 0.469 0.593 -0.469
(1.90)+ (1.97)* (2.68)** (1.57)
Constant 38.917 50.215 37.207 38.829 37.500 81.327 27.709 32.765
(176.71)** (3.88)** (108.14)** (2.49)* (114.98)**(4.28)** (63.33)** (1.67)+
N 6306 6196 2342 2287 6306 6196 2342 2287
Number of institution dummies 202 201 125 123 202 201 125 123
R-squared 0.05 0.06 0.08 0.09 0.00 0.03 0.00 0.03
Note: Asset index is computed by aggregating binary indicators of household ownership of various assets using
the principal component method.


58
Appendix Table 5: Within-institution estimates of gender gap in General Knowledge
and Islamic Studies test scores
GK Islam
School Aliyah School Aliyah
(1) (2) (1) (2) (1) (2) (1) (2)
Female -5.352 -5.590 -6.814 -6.620 -0.620 -0.638 -0.771 -0.775
(11.26)** (11.73)** (9.56)** (9.08)** (11.34)** (11.70)** (9.51)** (9.31)**
Age 0.338 -6.602 -0.150 -0.695
(0.11) (1.88)+ (0.43) (1.74)+
age, sq -0.047 0.224 0.002 0.022
(0.41) (1.82)+ (0.13) (1.56)
Mother_edu: primary 0.809 1.372 -0.025 0.149
(1.27) (1.52) (0.34) (1.46)
Mother_edu: some secondary 2.104 0.227 0.093 0.095
(3.16)** (0.24) (1.21) (0.88)
Mother_edu: secondary complete 1.945 0.866 0.068 0.209
(2.57)* (0.78) (0.78) (1.66)+
Mother_edu: post-secondary 1.109 0.318 0.056 0.275
(1.18) (0.18) (0.52) (1.38)
Father_edu: primary 0.648 -0.272 -0.002 -0.087
(0.88) (0.26) (0.02) (0.73)
Father_edu: some secondary 1.004 2.333 0.188 0.120
(1.54) (2.45)* (2.52)* (1.10)
Father_edu: secondary complete 0.588 0.935 0.190 -0.065
(0.86) (0.97) (2.40)* (0.59)
Father_edu: post-secondary 2.736 1.948 0.310 0.131
(3.86)** (1.80)+ (3.83)** (1.06)
Newspaper at home 1.096 -1.686 0.100 -0.004
(2.15)* (2.15)* (1.73)+ (0.04)
Asset index 0.534 0.348 0.043 0.003
(2.20)* (0.92) (1.55) (0.06)
Constant 41.841 43.484 42.954 90.194 7.399 8.914 8.655 13.898
(117.51)** (2.08)* (78.09)** (3.62)** (179.38)** (3.81)** (138.13)** (4.89)**
N 6306 6196 2342 2287 5689 5593 2322 2269
Number of institution dummies 202 201 125 123 201 200 125 123
R-squared 0.02 0.03 0.04 0.05 0.02 0.03 0.04 0.05
Note: Asset index is computed by aggregating binary indicators of household ownership of various assets using
the principal component method.
59
Appendix Table 6: Variance decomposition: Districts, child and family backgrounds
and schools

% of variance explained by:
Math English General
Knowledge
Islamic
studies
Full sample Same district 10 5 9 5
Same household characteristics 5 9 3 10
Same institution 42 34 30 20
Residuals 43 52 58 65

School sample Same district 9 4 11 7
Same household characteristics 5 4 2 4
Same institution 38 34 25 19
Residuals 48 58 62 70

Madrasa sample Same district 14 11 7 5
Same household characteristics 5 2 5 1
Same institution 51 47 41 24
Residuals 30 40 47 61
Notes: (i) Cells contain the % of variance explained by (a) a district fixed effect; (b) a district fixed effect and
child age, age-squared, gender, mothers education, fathers education, number of household assets, number of
farm animals at home, whether household keeps newspapers/magazines; (c) a school fixed effect and child age,
age-squared, gender, mothers education, fathers education, number of household assets, number of farm
animals at home, whether household keeps newspapers/magazines. (ii) All estimations exclude Quomi
madrasas.
60
Appendix Figure 1a: Kernel density plot of student-level score (on a scale of 0-100) in
Maths test
0
.
0
1
.
0
2
.
0
3
D
e
n
s
i
t
y
0 20 40 60 80
timss_score
School
Normal distribution
Aliyah madrasa
kernel =epanechnikov, bandwidth =2.2593
Kernel density estimate


Appendix Figure 1b: Kernel density plot of aggregate score (on a scale of 0-100) in
Maths test

0
.
0
1
.
0
2
.
0
3
.
0
4
D
e
n
s
i
t
y
0 20 40 60 80
(mean) timss_score
School
Normal distribution
Aliyah madrasa
kernel =epanechnikov, bandwidth =3.1060
Kernel density estimate

61
Appendix Figure 2a: Kernel density plot of student-level score (on a scale of 0-100) in
English test
0
.
0
0
5
.
0
1
.
0
1
5
.
0
2
.
0
2
5
D
e
n
s
i
t
y
0 20 40 60 80 100
eng_score
School
Normal distribution
Aliyah madrasa
kernel =epanechnikov, bandwidth =2.8988
Kernel density estimate


Appendix Figure 2b: Kernel density plot of aggregate score (on a scale of 0-100) in
English test
0
.
0
1
.
0
2
.
0
3
D
e
n
s
i
t
y
0 20 40 60 80 100
(mean) eng_score
School
Normal distribution
Aliyah madrasa
kernel =epanechnikov, bandwidth =4.2678
Kernel density estimate

62
Appendix

An agenda for potential future research
1. The need for further detailed work on Bangladesh is because (a) this is a country with a large
Muslim population where a traditional madrasa sector co-exists with a large number of modernized
religious schools, (b) madrasas operating in and outside the state sector are both exposed to and
compete with a diverse set of non-religious state and non-state schools, and (c) religious and non-
religious schools in the state-regulated sectors are exposed to a variety of incentive schemes. This
provides a fertile ground for policy orientated research on madrasas which can inform the wider
debate on madrasa education in South Asia and elsewhere.
2. In this report we have only highlighted issues concerning incidence and quality of learning of
madrasas in a descriptive manner. Even then we have primarily touched upon the institutional portion
of the survey the comprehensive household portion of the survey has yet to be fully analyzed.
Given the richness of the data collected at both the institution and household level as part of the
World Bank QSSMEB survey, we are planning to do more detailed analysis of issues covered in the
report (e.g., quality of learning) as well as additional issues of interest (e.g., madrasa education and
social attitudes). Furthermore, it is our hope that QSSMEB can serve as a baseline for spanning a
dialogue on issues that will require longitudinal data, particularly if we want to address post-
graduation outcomes. Following a cohort of school and madrasa graduates from the QSSMEB sample
over time will allow policymakers to engage in an empirically-based discussion on issues related to
post-secondary schooling decisions, labour market participation, occupational choice, earnings, civic
and political participation. Regardless of any additional follow-up survey rounds in the future, here is
an example of questions we plan to revisit or address with the existing cross-sectional data from
QSSMEB:
3. Supply decisions of madrasas: In section 3 of this report, we highlighted the positive correlation
between incidence of madrasas and sub-district level poverty. However, that analysis remained
limited to madrasas under the state sector. As a result, we could not comment whether this pattern
holds for traditional madrasas that operate without any state regulation. The data collected in this
study therefore provides a rare opportunity to formally investigate this issue. In future research, for
instance, we intend to examine whether traditional madrasas arise in villages where there is a scarcity
of public schools and where there is greater income poverty. Such analysis will be informative for
other countries as well where the rise of traditional madrasas is popularly explained away by state
failure to provide public schools in poor rural areas. In addition, GPS coordinates on traditional
madrasas including all other secondary educational institutions will provide a rare description of the
geography of religious schools. This way, it would be possible to physically verify whether these
institutions indeed prevail in poorer neighbourhoods within a given locality, in addition to being
located in a poor region.
4. More needs to be done to understand the relative supply of traditional Quomi madrasas. Our
sample indicates that the largest share of these traditional madrasas is in Sylhet division. Interestingly,
this is also a division where the number of registered secondary schools and Aliyah madrasas are
lowest in the country. This points towards a negative correlation between availability of state/state
aided schools and traditional madrasas. However, looking at other regions such as Chittagong and
Bogra, support for this hypothesis is not evident.
63

Source: Authors calculation based on data from the QSSMEB. Data corresponds to 403 sample secondary schools and
madrasas.
5. Similarly, more research is needed on the geography of religious schools. Do they form clusters
across various layers of the education? Or is there no systematic link between the supply of madrasas
across different levels of education within a locality? This issue is important for it has important
implication for the current debate on madrasa reform. As pointed out in Asadullah and Chaudhury
(2008b), a large number of children are educated in NGO-run primary. Presence of these non-formal
primary schools in a locality significantly raises the female enrolment in registered secondary
madrasas. This finding needs to be revisited using data on traditional madrasas which has also seen a
modest rise in female enrolment in recent years.
6. Demand for religious education: Using household data from HIES, in this report we have
reported descriptive regressions explaining the probability of enrolment in madrasas controlling for
socio-economic background of the student, parental educational profile and household income. While
we highlighted the correlation between madrasa attendance and poverty, this issue needs to be
revisited. Currently, we do not know whether children from poor families are more likely to attend
madrasas because their parents cannot afford private expenses associated with enrolment in
mainstream schools or the explanation is more due to household preferences. In the absence of
information on the religiosity of the household and religious preference in HIES dataset, for
example, we could not formally test the hypothesis of that religious school choice in rural Bangladesh
determined by the religious proclivity of households. Such research however is possible using the
household part of QSSMEB data.
7. Gender disparity in learning outcomes: The education debate in Bangladesh is still stuck around
the issue of access. Bangladesh does indeed rightly deserve credit for already achieving gender-parity
in primary and education, something which still remains a challenge for several South Asian
countries. However, it is high-time to shift the focus of the debate towards quality and the most to
gain from this will be girls. We have presented evidence that the learning gap for girls is significant
which is also consistent with earlier studies (e.g. Asadullah et al 2007) on the state of student learning
in secondary sector in rural Bangladesh. Besides doing more detailed analysis with institutional data
64
(e.g., brining in gender-specific class-room practices), we plan to thoroughly examine this issue with
the detailed household data. Even though the incidence of sibling-pairs (i.e., households with both a
boy and girl in school around the same grade level) is small, we will draw upon that sub-sample given
that we can factor out household unobservable. Additionally we can explore in depths as to whether
this is caused by differences in intra-household allocations towards education input and time use.
8. Quality of primary madrasas relative to other schools (Government, Government-Aided and
NGOs): While documentation of quality in the primary sector is important in its own right, it also has
important implications on how it effects learning achievement in secondary schooling. Asadullah et
al. 2008 find that graduates of primary madrasa do particularly poorly in secondary schools and
registered Aliyah madrasas. As highlighted in our report, there is a significant number of both
recognized and unrecognized madrasas in the primary sector. However, very little evidence is
available on their relative quality when assessed in terms of student learning. While we did not
conduct any testing in primary schools, we did conduct numeracy and literacy tests of all school-
going aged children in our detailed household survey. Furthermore, in a companion survey to
QSSMEB we have also collected similar household-level data from a large survey spread over 4
divisions and 15 districts which was purposefully designed to cover poor areas (survey was conducted
by Gana Shahajjo Kendra, a national NGO). For example, the remote costal area of Charfassion
which falls in that sample is unfortunately known for being a particularly vulnerable part of
Bangladesh. The institution survey included government primary schools, registered government-
aided primary schools, NGO schools, and madrasas. Detailed socio-demographic data was collected
from 43,199 households from 65 villages. There was also a specifically matched institution-household
survey for 1,600 5
th
grade primary school pupils. Furthermore these pupils were tested in
Mathematics, English, General Knowledge, and Science. Therefore, analysis based on this
complementary survey dataset will not only provide new evidence on the relative quality of madrasas
education in the primary sector, it will also shed light on the state of education in remote poverty-
stricken areas. There the dynamics of poverty, adverse shocks, vulnerability, and schooling can be
explored in greater details.
9. Non-cognitive outcomes: In addition to data on cognitive performance, we have gathered
valuable information on other aspects of child development such as: extra-curricular activities, socio-
economic attitudes, educational and occupational aspiration, and subjective well-being.