Explorations Into the Fertility Transition and Female Labor Force 
Participation in Bangladesh
by
SM SHIHAB SIDDIQUI
A dissertation accepted and approved in partial fulfillment of the
requirements for the degree of
Doctor of Philosophy
in Economics
Dissertation Committee:
Alfredo Burlando, Co-chair
Shankha Chakraborty, Co-chair
Anca D. Cristea, Core Member
Zhibin Yang, Institutional Representative
University of Oregon
Winter 2024
© 2024 SM Shihab Siddiqui
2
DISSERTATION ABSTRACT
SM Shihab Siddiqui
Doctor of Philosophy in Economics
Title: Explorations Into the Fertility Transition and Female Labor Force
Participation in Bangladesh
This dissertation is comprised of three papers. Together, they investigate as-
pects of the fertility and female labor force transition in Bangladesh that started in
the 1960s and continuing today. The dissertation is organized as follows:
Chapter 1 summarizes the context, and provides an overview of the three re-
search chapters. Chapter 2 analyzes the effect of increased industrial work oppor-
tunity on women’s employment, reproductive behavior, and human capital accu-
mulation. Using shift-share instrument, I find that increased industrial work op-
portunity increased women’s employment significantly and reduced human capital
accumulation very modestly among teenage girls. However, there was no effect on
reproductive behavior.
Chapter 3 provides the first comprehensive construction, to the best of my
knowledge, of completed birth estimates of Bangladeshi women who were born all
the way back in 1920. This exercise shows that the rural fertility transition begun
with the 1945-50 cohort, within five years of the fertility transition in urban areas.
I then present suggestive evidence that agrarian economic conditions, mediated by
land availability, was a driver of the transition.
Chapter 4 (co-authored with Shankha Chakraborty) utilizes Oaxaca-Blinder
decompositions to examine why the female labor force participation rate has in-
creased more in Bangladesh compared to contiguous Indian states since 1990s. We
find that while women’s education is positively associated with employment in
Bangladesh, the reverse is true in the selected Indian states. We also find some ev-
idence that women’s relative education compared to men’s has been an important
factor in explaining this. Chapter 5 concludes the dissertation.
This dissertation includes previously unpublished co-authored material.
3
CURRICULUM VITAE
NAME OF AUTHOR: SM Shihab Siddiqui
GRADUATE AND UNDERGRADUATE SCHOOLS ATTENDED:
University of Oregon, Eugene, OR, USA
University of Nebraska-Omaha, Omaha, NE, USA
North South University, Dhaka, Bangladesh
DEGREES AWARDED:
Doctor of Philosophy, Economics, 2024, University of Oregon
Master of Science, Economics, 2018, University of Nebraska-Omaha
Master of Science, Economics, 2015, North South University
Bachelor of Science, Economics, 2013, North South University
AREAS OF SPECIAL INTEREST:
Demography
Development
Economic Growth
Applied Econometrics
PROFESSIONAL EXPERIENCE:
Principal Consultant, Opinion Dynamics Corporation, 2023-present
Graduate Teaching Fellow, University of Oregon, 2018-2023
Graduate Research Assistant, University of Nebraska-Omaha, 2016-2018
Associate, Keystone Business Support Company, 2013-2016
GRANTS, AWARDS, AND HONORS:
Gerlof Homan Graduate Scholarship, Univeristy of Oregon, 2022
Best Field Paper Award, University of Oregon, 2021
Kliensorge Fellowship, University of Oregon, 2018
Barbara Miller Memorial Scholarship, University of Nebraska-Omaha, 2018
Emily Hazel Fellowship, University of Nebraska-Omaha, 2017
4
ACKNOWLEDGMENTS
I wish to extend my deepest appreciation to my committee chairs – Shankha
Chakraborty and Alfredo Burlando, who have consistently offered their unwavering
support and invaluable guidance during my journey through graduate school. I
am also indebted to Anca Cristea, and Zhibin Yang, whose insightful feedback has
significantly contributed to different sections of this dissertation.
I must also express my heartfelt gratitude to my family and friends in and out
of University of Oregon for their limitless support.
5
To my family, who have always been by my side.
6
TABLE OF CONTENTS
Chapter Page
LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1. INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2. GARMENT MANUFACTURING AND WOMEN’S WORK, REPRO-
DUCTION, AND HUMAN CAPITAL ACCUMULATION . . . . . . . . . 15
2.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2. The RMG industry in Bangladesh . . . . . . . . . . . . . . . . . . . . 20
2.3. Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3.1. Female Labor Force Participation . . . . . . . . . . . . . . . . 26
2.3.2. Reproductive Behavior . . . . . . . . . . . . . . . . . . . . . . 26
2.3.3. Human Capital Accumulation . . . . . . . . . . . . . . . . . . 27
2.4. Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4.1. Factory and Export Data . . . . . . . . . . . . . . . . . . . . . 28
2.4.2. Population Census . . . . . . . . . . . . . . . . . . . . . . . . 30
2.5. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.5.1. Factory Locations and Shares . . . . . . . . . . . . . . . . . . 31
2.5.2. Female Labor Force Participation . . . . . . . . . . . . . . . . 32
2.5.3. Reproductive Choices . . . . . . . . . . . . . . . . . . . . . . . 34
2.5.4. Human Capital Accumulation . . . . . . . . . . . . . . . . . . 35
2.6. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.7. Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.8. Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3. RURAL FERTILITY TRANSITIONS: THE ROLE OF LAND CON-
STRAINTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.2. Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.3. Empirical Approach and Data . . . . . . . . . . . . . . . . . . . . . . 57
3.3.1. Empirical Approach . . . . . . . . . . . . . . . . . . . . . . . . 57
7
Descriptive Analysis . . . . . . . . . . . . . . . . . . . . . . . 57
Examining the role of land constrain . . . . . . . . . . . . . . 58
3.3.2. Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.4. Empirical Findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.4.1. Descriptive analysis . . . . . . . . . . . . . . . . . . . . . . . . 61
Correlates of Fertility at the Individual Level . . . . . . . . . . 62
Correlates of Fertility Preference . . . . . . . . . . . . . . . . 62
3.4.2. Role of Migration and Land Constraint . . . . . . . . . . . . . 63
3.5. Discussion of results and conclusion . . . . . . . . . . . . . . . . . . . 64
3.6. Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.7. Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4. FEMALE LABOR FORCE PARTICIPATION IN BANGLADESH AND
NEIGHBORING INDIAN STATES: 1987-2011 . . . . . . . . . . . . . . . 74
4.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.2. Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.3. Empirical Approach and Data . . . . . . . . . . . . . . . . . . . . . . 78
4.3.1. Empirical Approach . . . . . . . . . . . . . . . . . . . . . . . . 78
4.3.2. Data Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.3.3. Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . 80
4.4. Empirical Findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.4.1. Cross-sectional Correlates . . . . . . . . . . . . . . . . . . . . 82
4.4.2. Decomposition results: Human capital and reproductive behavior 82
4.4.3. Decomposition results: Structural Change . . . . . . . . . . . . 83
4.5. Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . 85
4.6. Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.7. Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5. DISSERTATION CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . 101
APPENDICES
A. COMPLETED FERTILITY ESTIMATES . . . . . . . . . . . . . . . . . . 103
B. INDUSTRY CODE HARMONIZATION . . . . . . . . . . . . . . . . . . . 106
REFERENCES CITED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
8
LIST OF FIGURES
Figure Page
2.1. Female Labor Force Participation in Bangladesh and Elsewhere
(1990:2019) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.2. Fertility Transition in Bangladesh and Elsewhere (1970:2019) . . . . . . . 39
2.3. Expansion of the Ready-Made Garment (RMG) Industry in Bangladesh
(1978:2006) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.4. Spread of the RMG industry in Bangladesh (1991-2006) . . . . . . . . . 41
2.5. Importance of knit Versus Woven in the Bangladeshi RMG Industry
(1991:2006) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.6. Age Specific Marriage and Own Children in Household in the Sample
Areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.7. Empirical Cumulative Distribution of RMG factory Size in 1991 and 2001 44
3.1. Fertility Transitions Around the World 1950:2000 . . . . . . . . . . . . . 66
3.2. Rural and urban Fertility Transitions in Bangladesh . . . . . . . . . . . . 67
3.3. GDP, Land and Fertility in Bangladesh . . . . . . . . . . . . . . . . . . . 68
4.1. Female Labor Force Participation in Bangladesh and Elsewhere
(1990:2019) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.2. Female Labor Force Participation in South Asia (1990:2019) . . . . . . . 88
4.3. Potential Sources of Divergence in FLFPR Trajectory of Bangladesh and
Neighboring Indian States (1987-2011): Education . . . . . . . . . . . . . 89
4.4. Potential Sources of Divergence in FLFPR Trajectory of Bangladesh and
Neighboring Indian States (1987-2011): Relative Education of Women . . 90
4.5. Potential Sources of Divergence in FLFPR Trajectory of Bangladesh and
Neighboring Indian States (1987-2011): Education and Reproductive
Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.6. Potential Sources of Divergence in FLFPR Trajectory of Bangladesh
and Neighboring Indian States (1987-2011): Education, Reproductive
Behavior, and Structural Change . . . . . . . . . . . . . . . . . . . . . . 92
4.7. Potential Sources of Divergence in Rural FLFPR Trajectory of
Bangladesh and Neighboring Indian States (1987-2011): Education,
Reproductive Behavior, and Structural Change . . . . . . . . . . . . . . 93
9
4.8. Potential Sources of Divergence in Urban FLFPR Trajectory of
Bangladesh and Neighboring Indian States (1987-2011): Education,
Reproductive Behavior, and Structural Change . . . . . . . . . . . . . . 94
10
LIST OF TABLES
Table Page
2.1. Summary Statistics of RMG Factories in the Sample . . . . . . . . . . . 45
2.2. Differences between Sub-districts With or Without Factory, Census 1991 46
2.3. Correlation Between Intensity of Woven Specialization and Measures of
Infrastructure and Demographic Conditions in 1991 Within Sub-districts
with RMG Factories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.4. LPM Estimating Whether a Sub-district Has Factory by 1991 and 2001 . 48
2.5. 1991 and 2001 αk and αW and Sub-sequent Changes in Sub-district
Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.6. Export Exposure and Female Labor Force Participation . . . . . . . . . . 50
2.7. Stability of Estimated Coefficients . . . . . . . . . . . . . . . . . . . . . . 51
2.8. Effect of Export Exposure on Marriage and Age-specific Fertility Rates . 52
2.9. Export Exposure and Human Capital Accumulation of Bangladeshi
Girls (Ages 5-19) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.1. Correlates of Fertility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.2. Correlates of Fertility (Controlling for Stated Preference) . . . . . . . . . 70
3.3. Correlates of Stated Fertility Preference . . . . . . . . . . . . . . . . . . 71
3.4. Summary Statistics: Population, Density, Migration, and Land Availability 72
3.5. Correlations Between Selected Measures . . . . . . . . . . . . . . . . . . 72
3.6. Land Constraint and Fertility . . . . . . . . . . . . . . . . . . . . . . . . 73
4.1. Labor Force Participation (Overall and Industry-wise) in Bangladesh
and Selected Indian States . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.2. Yearly Changes in Labor Force Participation (Overall and Industry-wise)
Rates in Bangladesh and Selected Indian States in PP terms . . . . . . . 95
4.3. Demographic Conditions in Bangladesh and Selected Indian States . . . 96
4.4. Yearly Changes in Demographic Conditions in Bangladesh and Selected
Indian States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.5. Cross-sectional Correlates of FLFPR in Districts of Bangladesh and
Selected Indian States1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
11
4.6. Influence of Human Capital Endowment and Coefficient Differences
in Explaining the Divergent Trajectory of FLFPR in Bangladesh and
Neighboring Indian States . . . . . . . . . . . . . . . . . . . . . . . . . . 99
4.7. Influence of Reproductive Behavior and Human Capital Endowment and
Coefficient Differences in Explaining the Divergent Trajectory of FLFPR
in Bangladesh and Neighboring Indian States . . . . . . . . . . . . . . . 99
4.8. Influence of Differences in Urbanization and Industrialization in Ex-
plaining the Divergent Trajectory of Key FLFPR Sub-categories in
Bangladesh and Neighboring Indian States . . . . . . . . . . . . . . . . . 100
12
CHAPTER 1
INTRODUCTION
Two of the most important developments of the last 200 years is a sharp
decline in number of children that women have on an average (fertility), and in-
creased participation of women in the formal labor force. One mechanism suggested
by the literature is that urbanization and industrialization initially shifted occu-
pations that made child rearing costly and increased returns to labor and human
capital accumulation. The decline in fertility and increase in women’s employment
outside the household were then consequences of these changes. Similar dynamics
are likely under way in developing countries since the 1960s.
Bangladesh has been undergoing a rapid fertility transition since the 1970s.
The total fertility rate declined from a high of 6.9 to less than two in a span of five
decades,1 marking it as one of the fastest fertility transition. Over the same period,
the GDP per capita grew from 2015 USD equivalent of 385 to 1,593;2 whereas the
share of industry went from six percent to 33 percent. The industrialization went
hand in hand with an increase in urbanization by 32 percentage points, reaching
40 percent in 2020. Additionally, the female labor force participation rate (FLFP)
rate increased by 11 percentage points between 1990-2020. This dissertation, across
three chapters, aims to add to our understanding of the changes in fertility and
women’s employment through the lens of structural change and gendered industrial-
ization, and evolving rural economic conditions.
In chapter two, I begin with an analysis of the effect of female labor demand
in the Ready-Made Garment (RMG) industry on FLFP, fertility, and human capi-
tal accumulation decisions of Bangladeshi women. Using a shift-share instrument,
I find that increased industrial work opportunity increased women’s employment
significantly across all age group and reduced human capital accumulation very
modestly among teenage girls. However, there was no effect on reproductive behav-
ior.
Motivated by the inability of women-employing industrialization and the gen-
eral lack of explanations of most of the fertility transition in Bangladesh, in Chap-
ter three I use descriptive analysis to date the fertility transition in rural and urban
1All data in this paragraph are from The World Bank (2021)
2PPP GDP data is not available for Bangladesh prior to 1991. For reference, the
2020 PPP GDP per capita was 5,905.
13
Bangladesh, and examine typical correlates of fertility. Then I examine the role
that rural economic conditions may have played in initiating rural fertility transi-
tion in Bangladesh. I find that the rural fertility transition begun with the 1945-50
cohort, within five years of the fertility transition in urban areas; and present sug-
gestive evidence that agrarian economic conditions, mediated by land availability,
was a driver of the transition.
Chapter four, co-authored with Prof. Shankha Chakraborty, co-chair of
this dissertation, examines the divergence in women’s employment trajectory in
Bangladesh and contiguous neighboring Indian states that are likely to share com-
mon initial conditions. We find that women’s education is positively associated
with employment in Bangladesh, but the reverse is true in the contiguous Indian
states. We also present some evidence that women’s relative education compared
to men’s has been an important factor in explaining the divergence in FLFP trajec-
tory.
14
CHAPTER 2
GARMENT MANUFACTURING AND WOMEN’S WORK, REPRODUCTION,
AND HUMAN CAPITAL ACCUMULATION
2.1 Introduction
There has been dramatic changes to fertility and women’s educational attain-
ment in developing countries since the 1960s. One potential source of these changes
is industrialization associated with increased employment opportunities for women.
By increasing opportunity costs of child-bearing and child-rearing, and by increas-
ing returns to human capital accumulation in agrarian economies, industrializa-
tion may reduce fertility and increase educational attainment of women through
increased female labor force participation (FLFP) (Morgan and Hagewen (2005);
Aaronson et al. (2014)). Additionally, it can encourage women’s schooling if returns
to education increases (Heath & Mobarak, 2015).
Since its inception in the late 1970s, the Ready Made Garments (RMG) indus-
try has provided millions of Bangladeshi women their first opportunity for employ-
ment outside of agriculture in a high fertility, low FLFP, and low human capital
accumulation context. The gradual expansion of the RMG industry allows me to
examine the salience of low-skilled employment opportunity in the industrial sector
in explaining long-term fertility reduction and increases in educational attainment
of women.
The RMG industry dominates the manufacturing landscape and export earn-
ings of Bangladesh. In Fiscal Year 2016-2017, the RMG industry accounted for 81
percent of Bangladeshi exports – employing about 4 million people, and contribut-
ing 6.6 percent and 33 percent to overall and industrial labor force participation
in Bangladesh respectively (Bangladesh Burea of Statistics, 2020). Matsuura and
Teng (2020) estimates that about 60 percent of the workers in RMG sector are
women.1 It is perhaps not surprising that the growth of the RMG industry has co-
incided with a steady increase in FLFP in Bangladesh from 24.6 percent in 1990 to
36 percent in 2019. This contrasts South Asia’s overall experience (see Figure 2.1)
where FLFP dropped from 29 percent in 1990 to 23.6 percent in 2019 (The World
Bank, 2021).
1Matsuura and Teng (2020) also notes that estimates from different sources
range from 58-80 percent.
15
Between 1990 and 2016, the total fertility rate in Bangladesh decreased from
4.49 to about 2.20.2 This change occurred concurrently with an increase in women’s
youth literacy rates by more than 60 percentage point.3 Using available census sub-
samples, and a novel RMG factory dataset obtained by combining multiple datasets
from Bangladesh Garments Manufacturers and Exporters Association (BGMEA),
I first estimate the extent to which the expansion of the exporting RMG industry
contributed to the increase in FLFP in Bangladesh between 1991-2011. Second, I
estimate the long-term effect of the increased employment opportunities on repro-
ductive and human capital accumulation choices of Bangladeshi women.
The spatiotemporal expansion of the RMG industry is correlated with the spa-
tiotemporal development pattern in Bangladesh. I overcome this endogeneity by
using a Bartik-style instrument. Specifically, the identifying variation in exposure
to employment opportunity is generated by spatial differences in knit- and woven-
based product specialization within areas with RMG factories. My estimates indi-
cate that FLFP increased substantially – in regions exposed to the garment indus-
try, 31 percent of the increase in FLFP between 1991-2011 is due to the garment
industry. I also find that school enrollment of working age teenage girls decreased
due to increased exposure to employment opportunities in the RMG industry. How-
ever, I find no changes in the reproductive choices that I examine (age specific
marriage and fertility rates) and in other aspects of human capital accumulation
(school enrollment at younger ages, and literacy and years of schooling of working
age teenage girls).
The findings in this paper add to three important discussions. First, it informs
us about how important export-oriented manufacturing has been to Bangladesh’s
ongoing development process, adding to the literature discussing the mechanisms of
manufacturing- and export-led development. It shows that industrialization associ-
ated with increased FLFP does not explain the remarkable fertility reduction and
increased education of Bangladeshi women between 1991-2011. This is particularly
relevant given concerns about structural transformation bypassing the manufac-
turing sector in many of the currently developing countries as discussed in Rodrik
2See Figure 2.2 for a comparison of Bangladeshi fertility transition with other
countries.
3The corresponding increase for men was less dramatic, from 52 percent to 91
percent (The World Bank, 2021).
16
(2016). Second, this paper adds to the literature examining the impact of exposure
to international trade on various economic and non-economic aspects of the lives of
workers (Atkin, 2016; Autor et al., 2013, 2019; Li, 2018).
Third, it documents the role of the RMG industry in fostering overall develop-
ment and gender equality in Bangladesh over 1991-2011. In a work with a similar
focus, Heath and Mobarak (2015) investigated how greater access to RMG factory
jobs changed Bangladeshi women’s marriage, childbearing, school enrollment, and
employment decisions. Their outcome and control data comes from a 2009 survey
of 1395 households in sixty villages in four sub-districts of Bangladesh. Their fac-
tory data comes from a 2014 survey of local individuals knowledgeable about the
timing of RMG industry expansion in their own localities. Using 1991, 2001, and
2011 census sub-samples, I am able to investigate the effects of the expansion of the
RMG industry across Bangladesh over a longer period. Moreover, factory level data
on location, size, and product type allows me to use differences in product special-
ization within areas with RMG factories as my identifying variation, ameliorating
concerns related to endogeneity coming from unobserved location characteristics.
Additionally, the prevalence of women in RMG industry in Bangladesh has been de-
clining since 2010 (Matsuura & Teng, 2020), and this fact is in line with what can
be expected as technology improves in a manufacturing sector (Tejani & Kucera,
2021).4 The RMG industry was particularly hit by Covid-19 as well (Kabir et al.,
2020). The findings of this study also informs debate about the immediate future
path of Bangladeshi development.
The FLFP in industrial sector in Bangladesh was quite low at the inception
of the RMG industry. In 1991, only 0.5 percent percent of working aged women
was employed in industrial sectors in Bangladesh whereas the corresponding fig-
ure for men was 2.7 percent.5 Over the next decades, the RMG industry expanded
and became the largest industrial and export sector of Bangladesh. Expansion of
the RMG industry in different regions increased demand for labor in those regions.
The increase in labor demand for women may be more salient as the relative im-
portance of female employment in textile and clothing related industries has been
observed across time and space. Field-Hendrey (1998) and Burnette (2008) docu-
4There has been a reduction in number of RMG factories despite continued
growth in export value since 2013 (Reihan & Bidisha, 2018).
5Based on my own calculations using Census 1991 data.
17
ment that women made up large portions of the labor force in textile and related
industries in USA and in England in 1800s, well before their respective rapid FLFP
transition. Kucera and Tejani (2014) find that textiles and related industries were
the strongest drivers of women’s share of employment among a broad swath of
countries at different levels of development during 1981-2008. Women make up
the majority of labor force in the Bangladeshi RMG industry as well (Matsuura
& Teng, 2020). In addition to the direct employment effects of the RMG industry,
widespread employment in one industrial sector has the potential to increase over-
all FLFP by fostering pro-FLFP cultural norms, or by generating and spreading
knowledge about the lack of impact of FLFP on children, as in Fogli and Veldkamp
(2011) and Fernández (2013). These channels could be especially important in the
historically low FLFP and pro-natal context of Bangladesh.
From a theoretical point of view, FLFP and fertility are endogenous. Employ-
ment opportunities for female workers increase the opportunity cost of having chil-
dren, and decreases fertility and desired fertility (Aaronson et al., 2014). FLFP
could also lead to changes in marriage and divorce by altering the cost and ben-
efit of marriage for both men and women (Autor et al., 2019; Greenwood et al.,
2017). Additionally, industrial employment may change norms regarding women’s
reproductive decisions, a key driver of fertility (S. Amin et al., 1998; Baudin, 2010;
Bhattacharya & Chakraborty, 2012). Interestingly, Bangladesh was undergoing a
rapid fertility transition since the 1970s, decades before the emergence of the RMG
industry. One feature of the Bangladeshi fertility transition is that it started in
rural areas shortly after the transition started in urban areas, where factories are
mostly located (see 3.4.1). Thus the effect of RMG industry expansion on fertility
is an empirical question.6
Opportunities to work in manufacturing sector could increase returns to educa-
tion, as argued in (S. Amin et al., 1998; Heath & Mobarak, 2015). Export demand
from contextually higher skilled sectors has been found to increase schooling in
India (Oster & Steinberg, 2013; Shastry, 2012) and in China (Li, 2018). On the
other hand, Atkin (2016) finds that opportunities in export-oriented manufactur-
6Given the prevalence of younger workers in the RMG industry, it is also pos-
sible that RMG expansion led to delayed child bearing without changing fertility
overall. About 90 percent of the women working in RMG factories are younger
than 40 years old, gaining first experience of paid employment in the RMG sector
(Matsuura & Teng, 2020).
18
ing increases school drop-out rate. Since the returns to skills learned in school in
the RMG industry in Bangladesh is not well established, whether expansion of the
RMG industry increased or decreased human capital accumulation of Bangladeshi
women remains unclear.
The methodology of this inspired by Autor et al. (2013) who derive a measure
of US local labor market exposure per worker to Chinese import competition based
on standard trade models. Their measure starts with industry specific measures of
exposure, obtained by apportioning observed changes in import in an industry j to
each local labor market by that local labor market’s share of total US employment
in industry j. The sum of industry specific exposures in a local labor market is
divided by local employment numbers to obtain per worker measures. Autor et al.
(2013) instrument their share measures with 10 year lags and import measures with
Chinese imports to other countries.
In my case, there are two key challenges in employing a similar measure of
export exposure coming from the RMG industry. First, the placement of RMG fac-
tories are correlated with development patterns in Bangladesh.7 Second, the RMG
industry has an overwhelming importance in Bangladeshi industrial and manufac-
turing employment. Thus, patterns of industrial and manufacturing employment
will also be correlated with development pattern. I get around these challenges
by focusing my analysis only to areas that had export-oriented RMG factories. I
exploit spatial differences in knit- and woven-based product specialization within
areas with RMG factories to obtain identifying variation in exposure to the export-
oriented RMG industry in Bangladesh.
My estimates indicate that the average export exposure during 1991-2011 in
areas with RMG factories increased overall FLFP rate by 3.4 percentage points
and industrial FLFP by 4.5 percentage points.8. For comparison, during 1991-2011,
overall FLFP in Bangladesh increased by 2.45 percentage points and industrial
FLFP increased by 2.04 percentage points. I also find that school enrollment of
working age (14-19) teenage girls decreased by 2.7 percentage points. However, I
find no impact on other aspects of human capital accumulation, and reproductive
choices. My FLFP findings are consistent with previous findings in the literature
that the RMG industry increased FLFP in Bangladesh (Heath & Mobarak, 2015)
7See section 2.2 and section 2.5.1
8854 USD per working age (15-64) person
19
and that increased import competition reduced employment in the United States
(Autor et al., 2013, 2019). Decreased school enrollment of working age teenagers
due to opportunities in export-oriented manufacturing was also observed in the case
of Mexico (Atkin, 2016). However, my finding related to schooling and reproduc-
tive choices differs from the literature focusing on Bangladesh (Heath & Mobarak,
2015).
The rest of the paper is organized as follows - section 2.2 provides a brief in-
troduction to the RMG industry in Bangladesh, section 2.3 describes the empirical
approach, section 2.4 discusses the data, section 2.5 presents and discusses the re-
sults and section 2.6 concludes the paper.
2.2 The RMG industry in Bangladesh
The textile and clothing sector is one of most traded sector in the world. It
accounted for about 4 percent of the total global trade in 2018 (World Integrated
Trade Solution, 2022). As of 2015, Bangladesh was the second largest exporter in
the textile and clothing sector, behind China (World Integrated Trade Solution,
2022). Within the textile and clothing sector in Bangladesh, the export-oriented
RMG industry is overwhelmingly important, accounting for about 93 percent of
Bangladesh’s textile and clothing exports in 2015 (UN Comtrade, 2022). The defin-
ing feature of the RMG industry is that it mass-produces finished textile products
to generalized measurements, as opposed to tailoring textiles to an individual’s
measurement.
Exporting RMG industry started its journey in independent Bangladesh in
1978 with Desh Garments. The factory started with 130 workers trained in South
Korea (BGMEA, 2022a). Since then, the industry expanded rapidly. Figure 2.3
shows the increase in number of export-oriented RMG industry factories between
1978 and 2006. The industry took off in late 1980s, and its expansion accelerated in
the 1990s. About 779 factories were likely established between 1978-1991, whereas
about 2075 factories were likely established between 1992-2001.9 By 1991, the
9The numbers are obtained by matching factory information in BGMEA Di-
rectory 2000-01, and 2009-10 to two other BGMEA datasets. Few factories could
not be matched, and date of establishment had to be estimated for some factories.
Both non-match incidence and error rates in estimating date of establishment are
relatively low. Hence, these numbers are best interpreted as lower bounds on the
20
RMG industry accounted for more than 50 percent of Bangladeshi exports. Eu-
rope and North America emerged as the key importers of RMG products, possibly
buoyed by the Multifiber Agreement, and those areas remain the largest markets
(BGMEA, 2022b).10
Figure 2.4 shows the spatial spread of export-oriented RMG factories in
Bangladesh between 1991 - 2006. In 1991, 37 sub-districts (admin level 3) of 9 dis-
tricts (admin level 2) had RMG factories in them.11 Most of these sub-districts are
located in or near the two economically important districts containing the capital
city (Dhaka) and the port city (Chittagong). These sub-districts contained about
11 million people in 1991, roughly about 10 percent of Bangladesh’s population.
By 2006, RMG factories had spread to 63 sub-districts in 18 districts, containing
roughly 14 percent of Bangladesh’s population. 12 In my analysis below, I find that
relative to availability of workers, quality of infrastructure at the sub-district level
is much more correlated with the existence of RMG factory in a sub-district in
1991 and 2001 (see section 2.5.1 for details). Similarly, Kagy (2014) also find that
the quality of infrastructure and utilities are the key drivers of RMG factory loca-
tion choice, whereas access to educated workforce is not a concern for most factory
owners. In a survey of RMG factory owners, 96 percent of owners considered good
quality roads and access to buildings with gas and electricity to be “very important"
for choosing location of RMG factories, while 70 percent of them considered access
to educated workforce as being “not important" (Kagy, 2014) .
The RMG industry produces two broad categories of products – knit (HS code
61) and woven (HS code 62). Knit fabric is comprised of a single yarn looped re-
peatedly to produce cloth whereas woven fabric is made with multiple strains of
yarn criss-crossed over and under each other. Some of the common knit products
include cotton T-shirts and sweater, while jackets, shirts and pants are examples
of common woven products. Most factories in Bangladesh specialize working with
number of RMG factories established in the respective time periods. See section 2.4
for details.
10Multifiber Agreement ended in 2004, but does not seem to have impacted
factory formation or export growth. See Reihan and Bidisha (2018) and Figure 2.3.
11For comparison, there were 485 sub-districts and 64 districts in 1991.
12There has been administrative boundary restructuring at the sub-district level
within our study period of 1991-2011. All population numbers are estimated based
on the 1991 census boundaries.
21
either knit fabric or woven fabric, although several factories are able to work with
both types of fabrics. Producing woven fabric is more energy and capital intensive,
and woven products command about 10 percent higher per unit price. However,
the value-added in knit export in Bangladesh is higher (Sytsma, 2022) and woven
factories tend to employ more women relative to knit factories (Matsuura & Teng,
2020).
Figure 2.5 shows the changing importance of knit manufacturing relative to
woven manufacturing within the Bangladeshi RMG industry. Most of the RMG
factories established by 1991 engaged in production of woven garments. Over time,
more knit factories have opened and, by 2006, Bangladesh was exporting compara-
ble value of knitted and woven products.
2.3 Empirical Strategy
The goal of this paper is to estimate the long-run effect of increased employ-
ment opportunities stemming from the exporting RMG industry on FLFP, fertility
and marriage rates, and on several aspects of human capital accumulation choices
of Bangladeshi women. Ideally, the value of RMG exports from a sub-district will
be entirely driven by export shocks that are orthogonal to other channels that af-
fect outcome in that sub-district. In that ideal case, I could simply regress sub-
district outcomes on value of RMG exports from a sub-district as follows:
Ys,t = β Orthogonal Exports,t + ηt + Zs,tβz + γs,t (2.1)
where Ys,t are the outcome variables in sub-district s at a period t.
Orthogonal Exports,t measures export from a sub-district s in time t that is or-
thogonal to the other channels affecting outcomes, ηt are time fixed-effects, and Zs,t
is a vector of sub-district specific controls.
However, this simple analysis cannot be performed for two reasons. First is
that the spatiotemporal expansion of the RMG industry is correlated with the spa-
tiotemporal pattern of overall development in Bangladesh. The second is that due
to data limitations, I do not observe exports from a sub-district. The primary con-
cern for identification in this context is that RMG factories are likely to be located
in more developed sub-districts. Differences in these development characteristics
will influence the size of the RMG industry as well as FLFP, reproductive, and hu-
22
man capital accumulation choices of women. This concern is accentuated since the
sample period of 1991-2011 is a period of rapid change in Bangladesh.13
I overcome the challenge of omitted and unobserved development character-
istics in two steps. First, I restrict my analysis to sub-districts that had factories
in them by the end of 2006. The end of year 2006 was chosen as the mid-point
between 2001 and 2011 censuses. This restriction of sample to only sub-districts
with factories avoids comparing outcomes of sub-districts with RMG factories to
sub-districts that do not have RMG factory, ameliorating bias from unobserved
development characteristics. I then estimate a first difference regression equation of
the following form:
∆Ys,t = β∆Exports,t + δt + Zs,t−10βz +Xs,t−10βx + ϵs,t (2.2)
where ∆Ys,t is the change in outcome variables in sub-district s over a period end-
ing at t. The outcome variables in this paper are estimated from 1991, 2001 and
2011 census sub-samples. The decadal changes over 1991-2001 and 2001-2011 in
each sub-district are stacked. ∆Exports,t measures change in export from a sub-
district s in the decade ending at t and δt are period fixed-effects. Zs,t−10 is a vector
of sub-district specific controls common to all regressions. Zs,t−10 includes two prox-
ies of infrastructure quality at a sub-district (beginning of the decade electrification
rate and urbanization rate) and three measures of sub-district demographics (be-
ginning of the decade education levels of men and women between the ages 15-64,
and population density). Xs,t−10 are beginning of the decade controls that vary
depending on the outcome variable. They are discussed in relevant sub-sections
below. This sample restriction and working with first difference of outcomes and
exports ameliorates concerns regarding unobserved location characteristics driving
the outcomes.
The second problem is that I do not observe the value of exports originating
from sub-districts. So, I construct proxy for exports that is orthogonal to local
development using available data: number of knit and woven producing machines in
13For example, the economy grew at an average rate of 5.8 percent per year over
1991-2011. In addition to improvements in FLFP and women’s education, and
reductions in fertility discussed in the introduction; this period saw a doubling of
available workforce, rapid urbanization, and improvements in infrastructure. The
electrification rates increased from 14 percent in 1991 to 60 percent in 2011 (The
World Bank, 2021).
23
a sub-district. Specifically, I construct and use a Bartik-style instrument of decadal
changes in export exposure per potential worker in a sub-district following Autor et
al. (2013).14 The RMG industry produces two broad categories of products – knit
products and woven products.15 I first create a measure of total export exposure
in a sub-district by apportioning values of knit and woven exports originating in
Bangladesh to each sub-district in proportion to that sub-district’s share of knit
and woven machines relative to total knit and total woven machines in Bangladesh
respectively. This measure of total export exposure in a sub-district is divided by
the working age population in that sub-district to derive the export exposure per
potential worker, or simply - Export Exposure. Thus, in a sub-district s ending at
decade t, the Export Exposure is:
∑9 ExportK ∑9 ExportW
Export Exposure K BD,t−i W BD,t−is,t = αs,t−i ∗ + αs,t−i ∗ (2.3)L L
i=0 s,t−i i=1 s,t−i
MachinesKs,t−i Machines
W
Where K W s,t−iαs,t−i = , and α = (2.4)MachinesK s,t−iBD,t−i MachinesWBD,t−i
K and W denotes knit and woven respectively, BD denotes the total for
Bangladesh, Machines is the number of machines, and L is the 15-64 year old pop-
ulation.
Changes in value of knit or woven exports in a decade could be related to both
changes in outcome Y and changes in knit or woven shares in sub-districts. Addi-
tionally, location-time specific shocks could change the outcome, as well as change
the export exposure measures in a sub-district. I account for this by setting the
knit shares, woven shares, and working age population to the beginning of the
decade knit shares, woven share, and working age population respectively. With
this modifications, the decadal change in export exposure per potential worker in
14Autor et al. (2013) use standard trade models to derive a measure of Chinese
import exposure per worker at the different US commuting zones (CZs), their unit
of local labor markets. Industry specific measures of import exposure at the CZs
can be obtained by apportioning observed changes in import in an industry j to
each CZ by that CZ’s share of total US employment in industry j. The sum of in-
dustry specific exposures in a CZ is divided by CZ employment numbers to obtain
per worker measure of import competition in a CZ.
15See Section 2.2 for details.
24
sub-district s and in the decade ending in t is as follows:
∆ExportK ∆ExportW
∆ Export Exposure K BD,t W BD,ts,t =αs,t−10 ∗ + αs,t−10 ∗ (2.5)Lt−10 Lt−10
Replacing ∆Exports,t in equation 2.2 with ∆ Export Exposures,t in equation 2.5,
the first difference regression equation that I estimate takes the following form:
∆Ys,t = β∆ Export Exposures,t + δt + Zs,t−10βz +Xs,t−10βx + ϵs,t (2.6)
Thus, the identifying variation in my regressions stem from spatial variation
in knit- and woven-based product specialization within areas with RMG facto-
ries. The increase in total knit and woven exports from Bangladesh over decades
(see Figure 2.5) were likely driven by changing comparative advantage as Chinese
wages were increasing (BBC, 2012; Zhang et al., 2016) and unlikely to be correlated
with the changes in outcomes in different sub-districts of Bangladesh. Additionally,
all of Bangladesh did experience the phase-out of Multi-Fiber Agreement in 2005.
However, the identification do not rely on the independent or as-if independent
“shifters" assumption highlighted in Borusyak et al. (2022) since I only have the
two sub-sectors - knit and woven, and two periods of change. Rather, the identi-
fication in this paper comes from exploiting the variation in exposure generated
by differences in knit and woven shares across sub-districts exposed to the RMG
industry. These shares are then scaled by temporal differences in changes in knit
and woven exports from Bangladesh.
Following Goldsmith-Pinkham et al. (2020), the identifying assumption in this
paper’s context is that the relative specialization into knit versus woven products
in a sub-district is unrelated to the ϵs,t in equation 2.6. This implies that the dif-
ferences in shares should not influence the outcome through any other confounding
channel. The assumption would be violated in the following scenario. Say, for in-
stance woven factories employ relatively more women. This may relatively increase
the number of potential mothers in a sub-district with relatively higher intensity
of woven factories. In turn, governments or non-governmental organizations may
be encouraged to channel relatively more resources to schooling in areas with rel-
atively more woven production. This will improve schooling for females in those
sub-districts without a corresponding improvement in demand for women’s labor,
violating our identifying assumption.
25
2.3.1 Female Labor Force Participation
I first investigate the impact of increased export exposure on FLFP by estimat-
ing equations of the following form:
∆Ys,t = β∆Export Exposures,t+ δt+Zs,t−10βz +βx1Y
Male
s,t−10+βx2Pop
15−64
s,t−10+ ϵs,t (2.7)
Changes in overall and industrial FLFP rate for working age women (ages 15-64)
are investigated first.16 In two surveys separated by 9 years, Heath and Mobarak
(2015) and Matsuura and Teng (2020) found that roughly 70 percent of the women
working in the RMG industry are 29 years old and younger. In addition, both
surveys found that about 20 percent were teenagers. These age groups are also
more involved in reproductive and schooling choices. Hence, I also investigate the
changes in overall and industrial FLFP for 15-29 year old women and for 15-20 year
old women.17
In addition to common controls mentioned in Section 2.3 equation 2.7 adds
two more controls to equation 2.6. Beginning of the period labor force participation
rate for men in corresponding age groups for all or industrial sectors, as relevant,
is added to each of the regression to capture changes in local industry structures
unrelated to the expansion of the RMG industry. While it is likely that men’s labor
force participation is influenced by the RMG industry and its supporting industries,
it makes up a small portion of men’s employment.18 In addition, start of the period
population of 15-64 year old is included as a control of labor supply conditions.
2.3.2 Reproductive Behavior
Figure 2.6 shows how rates of being ever married and the number of own chil-
dren in household varies for females 10-40 year old in sub-districts with factories
by 2006. Opportunities in the RMG industry changes the economic options of
women. And given the age profile of RMG industry workers discussed above, it also
16None of the census asks whether a respondent worked in garment industry.
2011 census does break occupation down to manufacturing, hence I am using
industrial FLFP instead of manufacturing FLFP.
17Labor force participation data is only available for 15+ ages.
18Even if half the RMG industry employees were men, less than 5 percent of the
working men would be working in the RMG industry.
26
changes costs and benefits of marriage and of bearing and raising children during
ages that are very important for marriage and fertility in Bangladesh. I investigate
how changes in export exposure influences two key components of reproductive
choices of women – marriage rates and fertility rates at different ages. Marriage
rate is investigated by estimating equations of the following form:
∆Ys,t = β∆Export Exposures,t + δt + Zs,t−10βz + ϵs,t (2.8)
where ∆Ys,t includes changes in rates of ever being married among 15-20 and 21-
30 year old women. Influence on women 30 years and above are not investigated
since almost all women of that age have had married at least once by that age
among Bangladeshi women in our sample. No additional controls are included in
regressions investigating marriage rates at different age groups. Next, I investigate
how fertility changes by estimating regressions as follows:
∆Ys,t = β∆Export Exposures,t + δt + Zs,t−10βz + ϵs,t (2.9)
where ∆Ys,t are measures of changes in fertility rates among 15-20, 21-30, and 31-40
year old women. Since I do not observe realized fertility or birth histories, I use
number of own child in household as a proxy measures of fertility.
2.3.3 Human Capital Accumulation
Exposure to exports from RMG industry could encourage increased educa-
tional attainment by increasing returns to basic education. However, it could also
reduce educational attainments by encouraging drop-outs among working age
teenage girls. Additionally, efforts to learn while in school may also increase if re-
turns to human capital increases. These hypotheses are tested using regressions of
the following form:
∆Ys,t = β∆Export Exposure Ms,t + δt + Zs,t−10βz + βxYs,t−10 + ϵs,t (2.10)
where ∆Ys,t include changes in enrollment rates for girls aged 5-9, girls aged 10-13,
14-19; and changes in years of schooling and literacy among working age teenage
girls (ages 14-19). These regressions include beginning of the period outcomes for
males of the corresponding age group as a control for changes in quality of educa-
tional institutions in these regions.
27
2.4 Data
2.4.1 Factory and Export Data
Data on location, product type and number of machines of all factories in 1991
and 2001 is needed to estimate the share αKs and αWs in 1991 and 2001. To that
end, four datasets from the Bangladesh Garments Manufacturers and Exporters
Association (BGMEA) are combined to obtain the factory dataset. Two of the four
datasets provide year of establishment data, needed to identify factories in different
sub-districts that existed in 1991 and 2001. The other two datasets provide data
on number of machines in each factory in 1991 and 2001. BGMEA membership
application and annual fees are inexpensive,19 and provides a way for factories to
establish their legitimacy with foreign buyers. The reported number of machines
can be a good proxy for factory capacity and easily verifiable by prospective buyers.
Hence, the data on factories and their numbers of machines are likely to be of good
quality.
BGMEA Members Directory contains data on name, BGMEA membership
number, addresses associated with the entry (often separated into address of the
factory, and address of the office), number of machines, capacity, and the type of
products produced by the factory. I have directory entries for the years 2000/01
and 2009/10, which means that I observe firm-level characteristics at these two
points in time. For ease of exposition, I call these two datasets directory-based
datasets from hereon. Next, I also obtained a separate 2015 factory list from BG-
MEA and a factory list that I scraped from the current members list section of
BGMEA website on March of 2021. I call the latter two datasets list-based datasets
from hereon. In addition to the data in directory-based datasets, list-based datasets
also include the year of establishment of the firm.20 However, I cannot solely rely
19Currently, the application and annual fees are approximately USD 400 and
USD 100
20Specifically, for each entry, the 2015 list contains data on name, name of cor-
porate owner, email, telephone, cellphone, and fax numbers; contact person and
their designation, primary banker, factory address, product types, whether the unit
specializes in knit or woven, number of machines; and employees. The list scraped
from BGMEA on 2021 contains data on name, BGMEA membership number, office
address, factory address, number of machines, capacity, and year of establishment
of each entry.
28
on the list-based datasets for two reasons. First, the lists do not include factories
that were operational in 1991 and/or 2001, but that have left the industry before
the lists were compiled.21. This can be a serious source of bias. Second, factories
may have changed their number of machines in the 2-4 decades between the obser-
vation dates in the list-based datasets and 1991 and 2001.
These four datasets are combined to crate a panel of factory level data. The
first step in constructing the factory dataset is matching entries in the directory-
based dataset with entries in the list-based datasets by their name or unique BG-
MEA membership number. This yielded 71 percent and 80 percent matches for en-
tries in the 2000-01 and 2009-10 BGMEA directory respectively, providing their fac-
tory address and year of establishment. Sub-districts for these matches are assigned
by geo-coding the factory addresses using Awesome Table. For the remainder of
entries in the BGMEA directory 2000-01, sub-districts are determined manually by
obtaining their factory address and using Google Maps to assign sub-districts. In
many instances, the BGMEA directories do not explicitly classify an address associ-
ated with an entry as a factory address or an office address. In those instances, the
address is classified as a factory address if the phone number of the factory and the
phone number of the office for that entry is the same. When both addresses and
phone numbers of an entry are not classified in the directories, I am unable to as-
sign a sub-district as I cannot infer whether the address associated with the entry is
of the factory, or of the office space. Consequently, locations of 13 percent of entries
in the BGMEA directory 2000-01 remains undetermined and are not included in
my factory dataset.
The years of establishment for BGMEA entries that could not be matched
with the list-based datasets are unavailable. Since its inception in 1983, BGMEA
assigned membership number in a sequence starting with one, and with increments
of one.22 I impute the missing year of establishment using an OLS regression cap-
turing the relationship between year of establishment and BGMEA number within
the BGMEA directory entries that could be matched with the list-based datasets.
21A substantial bout of factory exits occurred in the aftermath of 2013 Rana
Plaza tragedy (Reihan & Bidisha, 2018). Events like these high-light the challenges
of relying only on current list-based datasets
22This fact, and the number of entries in the 2000-01 directory can be used to
infer that at most 19 of the 3026 factories exited the industry between 1983 and
2000-01.
29
Whether a factory existed in 1991 or 2001 is then determined by the year of estab-
lishment, or its estimate if needed. In the final factory dataset used to estimate
1991 and 2001 αKs and αWs , 16 percent of the year of establishment is estimated,
yielding an error rate of 1.14 percent in assigning whether a factory existed in 1991
or 2001.23 Additionally, missing number of machines is assigned the mean value of
machine and constitutes 5 percent of the sample.
Table 2.1 shows the summary statistics of factories in my sample. There are
2853 factories, the average factory had 198 machines with a standard deviation of
259.
Figure 2.7 shows the empirical cumulative distribution of machine size of knit
and woven factories established on or before 1991, and between 1991-2001. No
evidence of in changes in factory size is evident, implying similar factory level tech-
nology at work in the export oriented RMG industry at least till early 2000s.
30 percent of the factories in my sample existed in 1991. 30 percent of them
specialized in knit production, 60 percent on woven production and the remaining
factories produced both.
Bangladesh imports a large portion of the knit and woven textile used in the
RMG industry. Workers in Bangladesh add value to the textile by various pro-
duction activities such as cutting and sewing. As a measure of labor demand, es-
timated values of export originating from Bangladesh is used. For knit (woven)
products, export originating in Bangladesh at year t is estimated by subtracting
value of total knit (woven) textile imports from total value of knit (woven) product
exports. The data of these values are obtained from (UN Comtrade, 2022).24
2.4.2 Population Census
The outcome and control data comes from Bangladesh Census 1991, 2001 and
2011 sub-samples, obtained from Minnesota Population Center (2020). Individual
23When applied to factories whose year of establishment is known, this procedure
yields a 7 percent error rate. The procedure is applied to 16 percent of the factories
in the final dataset.
24Sytsma (2022) reports that Textiles fall under the HS2 heading HS60 (knit),
and HS4 headings (woven): 5007, 5111, 5112, 5113, 5208, 5209, 5210, 5211, 5212,
5309, 5310, 5311, 5407, 5408, 5512, 5513, 5514, 5515, 5516, 5602, 5603, 5801, 5802,
5809, 5903, 5906, and 5907. Knit-based products have an HS code of 61 and woven-
based products have an HS code of 62.
30
data are aggregated to the sub-district level. The spread of RMG industry across
different sub-districts in Bangladesh is correlated with infrastructure quality25. In-
frastructure quality could change the outcome variables differentially on its own.
Hence analysis in the main specifications is restricted to sub-districts that already
had RMG industry factories in them by the end of 2006, the mid point of the last
sample period. Table 2.2 shows the difference in key demographic and infrastruc-
ture measures in 1991 census between sub-districts that are included in the sample,
i.e. had factories by 2006, and sub-districts that are excluded from sample, i.e.
did not have factories by 2006. Whereas Table 2.3 shows the correlation between
specialization on woven production and the same demographic and infrastructure
measures in the sub-district in 1991.26 There are clear differences in density, ur-
banization, electrification, average age, gender distribution and average years of
schooling in areas with RMG industry factories by 2006, and areas without. How-
ever, within factory exposed areas, specialization is uncorrelated with the same
measures.
2.5 Results
2.5.1 Factory Locations and Shares
I first confirm that quality of infrastructure, rather than availability of workers,
is the main determinant of RMG factory location in Bangladesh. Table 2.4 presents
results of linear probability models (LPM) of factory locations in 1991 and 2001.
In each case, the dependent variable is an indicator of whether a sub-district had
RMG factories by 1991 and 2001 respectively. All explanatory variables are in stan-
dardized units. Measures of infrastructure variables – the electrification rate and
the urbanization rate (i.e. share of population in urban areas), are both statistically
significant and meaningful in explaining factory location. One standard deviation
increase in electrification rate is associated with a 13-16 percentage point increase
25see section 2.2 and section 2.5.1
26Measured as ratio of woven-producing machines to total RMG industry ma-
chines in a sub-district. Since the RMG industry is classified into knit and woven in
our analysis, specialization in knit = 1-specialization in woven.
31
in probability of a sub-district having a factory by 1991 or 2001.27 The impact cor-
responding to one standard deviation increase in urbanization rate is 5-7 percentage
point. On the other hand, measures of demographic variables – density of work-
ing age (15-64) population and average years of schooling in a sub-district are less
systematically associated with a a sub-district having a factory by a given year.
Years of schooling do not have a statistically significant impact. Whereas one stan-
dard deviation increase in density of working age population is associated with a
3-6 percentage point increase in probability of having a factory, much weaker than
the combined impact of infrastructure measures. Additionally, dropping density of
working age population barely changes the variation explained by the models, with
electrification rate absorbing the effect.
While the above confirms factory locations are endogenous in the unrestricted
sample, I next provide evidence that the knit and woven shares are not correlated
with evolution of development in the restricted sample. One concern for identifica-
tion in this paper’s context is that knit share (αks,t) and woven share (αWs,t) in one
period may change characteristics of a location which in turn can change the out-
come variables. I regress αks,t and αWs,t in 1991 and in 2001 on changes in observed
sub-district characteristics between 1991-2001, and 2001-2011 respectively. Table
2.5 presents the results of the regressions. I find no evidence of correlation between
the start of the period knit and woven shares in a sub-district, and the subsequent
decade’s changes in electrification and urbanization rate, population (15-64 year
old) and male’s labor force participation in that sub-district. There is a very weak
correlation with reduced schooling of males less than 15 years old, however. The
adjusted R2 is close to zero in all specifications.
2.5.2 Female Labor Force Participation
Table 2.6 presents estimates of regressions of the form of equation 2.7. The
dependent variables are FLFP rates of various age groups expressed in percentage
27Fujii and Shonchoy (2020) argues that spatial differences in electrification
in Bangladesh in the sample period is dependent on infrastructure, number of
households, state of industrial and commercial development and existing social and
community institutions. Hence a more accurate interpretation is that electrifica-
tion and the unobserved development patterns associated with electrification is
associated with a 13-16 percentage point increase in probability of having a factory.
32
and the export exposure measure is in one thousand dollar per working age popula-
tion. Column 1-2 presents estimates for ages 15-64, column 3-4 presents estimates
for ages 15-29 and column 5-6 presents estimates for ages 15-20. Odd numbered
columns correspond to overall FLFP and even numbered columns correspond to
industrial FLFP. The results indicate that exposure to export-oriented RMG indus-
try increases overall and industrial FLFP across all age groups.Table 2.7 reports
that the estimates in specification in column 1 are stable to inclusion of controls.
The coefficient estimates are between 4.06 and 7.56, and are all statistically signif-
icant at the 1 percent level. The increase in industrial FLFP is stronger than the
increase in overall FLFP, as expected when demand for female labor increases in
the industrial sub-sector. Additionally, the effect on FLFP is stronger for younger
age groups. This is consistent with findings that younger women are more prevalent
in the RMG industry of Bangladesh (Heath & Mobarak, 2015; Matsuura & Teng,
2020).
The point estimates imply the effect of export exposure on FLFP is large. To
see why, consider that the average value of export exposure in my sample is 854
dollars over two decades, roughly 6 months - 12 months of income in the time pe-
riod. The average exposure changes overall FLFP by 3.47 percentage points. In
sub-districts included in my sample, overall FLFP increased by 10.8 percentage
point over 1991-2011. 28 Thus, about 31 percent of the change in overall FLFP
over 1991-2011 in my sample areas can be attributed to the expansion of the RMG
industry in Bangladesh. Additionally, the point estimate for increase in industrial-
FLFP is 5.36. Thus, about 55 percent of the 8.13 percentage point increase in
industrial-FLFP can be explained by exposure to the RMG industry. With similar
calculations, the results indicate that for ages 15-29, 32 percent of the increase in
overall FLFP, and 51 percent of the increase in industrial FLFP can be attributed
to the expansion of the RMG industry. The numbers for 15-20 year old are 36 per-
cent for overall FLFP and 49 percent for industrial FLFP. It can be seen that even
though the coefficient is largest for the youngest age group, the economic signif-
icance is actually lower for them, reflecting a broader trend of increasing FLFP
among younger cohorts in Bangladesh.
My findings of positive long-term effect of increases in exposure to the RMG
28For context, the FLFP in Bangladesh increased by 3.59 percentage points over
the same time period.
33
industry on FLFP in Bangladesh is expected and is in-line with previous literature
(S. Amin et al., 1998; Heath & Mobarak, 2015; Kabeer & Mahmud, 2004). How-
ever, the magnitude of effect that I find is smaller. My estimates indicate that at
the average value of exposure, overall and industrial FLFP increases by about 3.47
and 5.36 percentage points respectively. Using a difference-in-difference estimator,
Heath and Mobarak (2015) found that women in villages within commuting zones
of RMG factories were about 15 percentage point more likely to have worked out-
side of home. Moreover, the effect was stronger among women exposed to RMG
factories during critical exposure period (ages 10 - 23) by an additional 12 percent-
age points.29 When examining the effect of exposure to the RMG industry by using
variation in product specialization within garment exposed sub-districts, I find a
smaller effect, possibly pointing to differences between areas close to garments and
further with limited spillovers in female employment between them.
Examining impact of import competition using a similar methodology, Autor
et al. (2013) found that import competition reduced US manufacturing employ-
ment by about a quarter over 1990-2007. Autor et al. (2019) found that one unit
of import exposure, roughly equivalent to the average decade level exposure be-
tween 1990-2014, reduced manufacturing employment as a share of the population
for both sexes by 1.06 pp. My estimates also suggest a large impact of the RMG
industry on Bangladesh FLFP. I also find that the impact is larger among younger
women. However, my magnitudes of impact of exposure to trade are larger than
what Autor et al. (2019) and Autor et al. (2013) finds in the case of US, perhaps
reflective of the relatively more important role of RMG industry in the Bangladeshi
FLFP context.
2.5.3 Reproductive Choices
I assess the impact of expansion of the RMG industry on marriage rates of 15-
20, and 21-30 year old women; and on fertility rates of 15-20 year old women, 21-30
years old women, and 30-40 year old women by estimating regressions of the form
in equation 2.8 and 2.9. Table 2.8 presents the results. The results indicate that
expansion of the export-oriented RMG industry in Bangladesh did not have a mea-
surable impact on marriage rates or on fertility rates in Bangladesh. My findings
29Heath and Mobarak (2015) determined whether villages were in commuting
zones of factories in consultation with officials from BGMEA.
34
are in contrast to both the existing literature in the Bangladeshi RMG industry
context, and in the trade exposure literature context. Using total years of exposure
to capture the overall impact of RMG factories on marriage and child-bearing deci-
sion, Heath and Mobarak (2015) found that 6.4 years of RMG exposure (mean in
their sample) reduced the probability of getting married and having first children
by about 0.3 and 0.23 percentage point. In the trade exposure literature, Autor
et al. (2019) found that negative shocks to female-dominated industries tend to
increase family formation and fertility. Thus, positive shocks to female-dominated
industry can be expected to reduce family formation and fertility.
The high R2 in the regressions, and apparent lack of strong evidence of statisti-
cal significance of any of the controls other than the time trend suggest a very im-
portant role of time trend in analysis of reproductive behavior among Bangladeshi
women over 1991-2011. One interesting feature of the employment in the RMG in-
dustry is the strong encouragement from husbands to partake in the employment
(Matsuura & Teng, 2020), reducing potential trade-offs between marriage and work-
ing outside that can exist in a low FLFP context.
Figure 2.2 shows that between 1970-1990, the TFR in Bangladesh dropped by
about 2.5. The RMG industry emerged and expanded during a period with momen-
tum in favor of declining fertility. This, combined with the estimates of Heath and
Mobarak (2015) suggests that in the backdrop of a rapid fertility transition, the
relative importance of the expansion of the RMG industry to Bangladeshi fertility
transition is minor at best.
2.5.4 Human Capital Accumulation
Table 2.9 presents estimates of regressions of the form of equation 2.10.
Columns 1, 2 and 3 presents results for school enrollment, expressed in percentages,
for ages 5-9, 10-13 and 14-19 year old girls respectively. Column 4 and 5 presents
the results for literacy rate and years of schooling for working age teenage girls.
The results indicate that exposure to the export-oriented RMG industry did not
influence school enrollment of younger girls (ages 5-9 and ages 10-13). However,
school enrollment for working-age girls (ages 14-19) is reduced. Despite this, expo-
sure to the RMG industry did not have a statistically significant impact on literacy
rate and years of schooling of working age girls. At the mean export exposure value,
enrollment for 14-19 year old girls reduces by 2.71 percentage point. The average
35
school enrollment rate for that age group was 40.6 in 2001 and 2011. Thus, expo-
sure to the RMG industry has reduced school enrollment by 6.6 percentage. The
point estimate on years of schooling is negative, and almost statistically signifi-
cant. Using a similar interpretation, mean export exposure led to a decline in 0.11
years of schooling. Taken together, my estimates suggest that the expansion of
the garment industry is associated with a small decline in women’s education. In
contrast, Heath and Mobarak (2015) found positive impact of exposure to RMG in-
dustries on human capital accumulation in Bangladesh. They found that increased
exposure increased educational attainment for women and men by 0.22 and 0.26
years respectively. A positive effect of 0.22 years on women’s education is outside
my 95 percent confidence interval. However, they did not find evidence suggesting
increases in enrollment.
The negative impact of opportunities in export-oriented manufacturing on
school enrollment for working-age students is also observed in case of Mexico.
Atkin (2016) finds that opportunities in low-skilled manufacturing export industry
reduces enrollment between grades 9-12 in Mexico. In the case of China, Li (2018)
finds that high skill export shock increased high school and college enrollment in
between 1990 to 2005. The RMG industry is a low-skilled manufacturing industry.
Additionally, results of my FLFP regressions (2.6) show that years of schooling and
FLFP are negatively related. Taken together, it is likely that working age girls who
drop out due to opportunities in the RMG industry may not have continued in
much further in schools anyway. Thus the opportunities in the RMG industry may
have reduced school enrollment of 14-19 year olds without a statistically significant
impact on years of schooling .
2.6 Conclusion
In this paper, I analyze the effect of the rapid expansion of the RMG indus-
try in Bangladesh on labor force participation, reproductive and human capital
accumulation choices of Bangladeshi women. RMG exports from Bangladesh grew
at a rate of 16 percent a year between 1991-2011, with the industry expanding
from 37 sub-districts in 1991 to 63 sub-districts by 2006. Both historically and in
present, employment in textile and closely related industries are often female dom-
inated even in low female labor force participation contexts. Most workers in the
export oriented Bangladeshi RMG industry are women as well. The expansion of
36
the RMG industry presented opportunities in industrial employment for millions of
Bangladeshi women for the first time. My analysis indicates that the expansion of
the RMG industry has contributed extensively on the overall and industrial FLFP
in Bangladesh, perhaps at the expense of small reductions in schooling for working
age teenage girls. However, I find no impact on marriage rates and fertility rates
among different ages group amid an ongoing fertility transition. Taken together,
the analysis presented here confirms the direct effect of increased engagement of
women in the labor market due to labor demand in the RMG industry. However,
there is no evidence to indicate that the industry contributed to the remarkable
changes in reproductive choices and human capital accumulation of Bangladeshi
women over 1991-2011.
So, what explains the change fertility transition in Bangladesh and what ex-
plains the rural FLFPR in Bangladesh? The next chapter tackles the fertility ques-
tion by first documenting the transition in a finer detail, conducting exploratory
analysis, and examining the role of land constraint in the rural fertility transition of
Bangladesh.
The fourth chapter examines the issue of FLFPR in rural Bangladesh.
37
2.7 Figures
Figure 2.1. Female Labor Force Participation in Bangladesh and Elsewhere
(1990:2019)
Bangladesh had one of the lowest FLFP rate in 1990 among developing re-
gions. While most other developing regions saw its FLFP decline over 1990-2019,
Bangladesh’s FLFP increased. Notably, FLFP of Bangladesh’s neighbors in South
Asia decreased overall. Data source: (The World Bank, 2021) .
38
Figure 2.2. Fertility Transition in Bangladesh and Elsewhere (1970:2019)
In 1970, Bangladesh had one of the highest fertility rate in the world. Since 1970s,
it experienced a rapid fertility transition, bringing its TFR to near replacement rate
towards the end of late 2020s. Data source: (The World Bank, 2021)
39
Figure 2.3. Expansion of the Ready-Made Garment (RMG) Industry in Bangladesh
(1978:2006)
The number of RMG factories have been increasing steadily over 1991-2006. Data
source: My own compilation from various BGMEA sources, See section 2.4.
40
41
Figure 2.4. Spread of the RMG industry in Bangladesh (1991-2006)
The RMG industry emerged in Bangladesh in the 1980s. By 1991, 37 of the 485 sub-districts had RMG factories in
them. By 2001, there were 54 sub-districts with factories and by 2006, 63 sub-districts had factories. Data source: My
own compilation from various BGMEA sources, See section 2.4.
Figure 2.5. Importance of knit Versus Woven in the Bangladeshi RMG Industry
(1991:2006)
At the beginning of 1991, most of the RMG factories specialized in woven-based
products. Since then, more knit factories opened up relative to woven factories.
Exports of knit increased as well. By late 2000s, Knit and woven exports were com-
parable in value. Data source: Export values are from BGMEA (2022b). Factory
data is my own compilation from various BGMEA sources, See section 2.4.
42
Figure 2.6. Age Specific Marriage and Own Children in Household in the Sample
Areas
The left-hand panel shows age-specific rates of ever being married for Bangladeshi
females of ages 10-40. It shows that a large portion of Bangladeshi teenage girls do
in fact get married. Additionally, rates of ever being married flattens from ages 30
and onward. The right hand side shows number of own children in household, my
measure of fertility in this study. It can be seen that the number of own children in
household starts declining for women in their late 30s, perhaps due to family forma-
tion of their own children. These facts motivated my choice of analyzing the effect
of exposure to RMG factories on marriage rates for females of ages 15-20 and 20-30
year old; and on fertility rates for 15-20, 20-30, and 30-40 year old Bangladeshi
females. Data Source: Bangladesh Census 1991, 2001, 2011.
43
Figure 2.7. Empirical Cumulative Distribution of RMG factory Size in 1991 and
2001
The woven and knit shares are measured in 1991 and 2001. I examine whether
technology of factories that were formed between 1991-2001 were similar to tech-
nology of factories formed before 1991. Both knit producing and woven producing
factories have very similar empirical cumulative distribution function for factories
established before 1991 and factories established between 1991-2001. Thus, tech-
nology likely remained the same even though a few new entrants in 1991-2011 were
much larger than existing factories. Data source: Export values are from BGMEA
(2022b). Factory data is my own compilation from various BGMEA sources, See
section 2.4.
44
2.8 Tables
Table 2.1. Summary Statistics of RMG Factories in the Sample
Statistic N Mean St. Dev.
Num. of Machines 2,849 197.9 259.1
Existed in 1991 2,849 0.3 0.4
Knit Factory 2,849 0.3 0.5
Woven Factory 2,849 0.6 0.5
There are 2849 factories in the factory dataset. factories had 198 machines on an
average. 30 percent of the factories existed in 1991. 30 percent of them specialized
in knit production, 60 percent specialized in woven production and the remain
produced both.
45
Table 2.2. Differences between Sub-districts With or Without Factory, Census 1991
Variable N Factory by 20061 No Factory by 20061 p-value2
N = 64 N = 421
Density 485 12,580 (27,765) 865 (965) <0.001
Access to electricity 485 0.42 (0.35) 0.08 (0.11) <0.001
Urban 485 0.52 (0.45) 0.13 (0.16) <0.001
Age 485 22.19 (0.92) 21.65 (0.95) <0.001
Female 485 .465 (0.033) 1.488 (0.012) <0.001
Schooling (Yrs) 485 3.17 (1.32) 2.02 (0.63) <0.001
1 Mean (SD)
2 Wilcoxon rank sum test
Sub-districts that had a RMG factory in them by 2006 were clearly more developed
even by 1991. Factory-exposed sub-districts were much more educated, dense, ur-
banized, and had better access to electricity by 1991. They also contained slightly
older individuals and more males by 1991. This correlation between having RMG
factory and development level informed the decision to restrict analysis only to
regions with factories by 2006.
46
47
Table 2.3. Correlation Between Intensity of Woven Specialization and Measures of Infrastructure and Demographic
Conditions in 1991 Within Sub-districts with RMG Factories
Density Electrification Urbanization Age Sex Yrs. of School
Density
Electrification 0.54***
Urbanization 0.35* 0.88***
Age 0.30 0.35* 0.25
Female -0.42** -0.83*** -0.67*** -0.49**
Yrs. of School 0.44** 0.86*** 0.70*** 0.51** -0.71***
Intensity of Woven1 0.16 0.22 0.13 -0.31 -0.17 0.26
∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
1 Measured as ratio of woven-producing machines to total RMG industry machines in a sub-district
This table shows that intensity of woven and knit production within sub-districts with RMG factories is not correlated
with sub-district level measures of density, electrification, urbanization, average age, sex distribution, and years of school-
ing in 1991.
Table 2.4. LPM Estimating Whether a Sub-district Has Factory by 1991 and 2001
Dependent variable:
Has Factory by 1991 = 1 Has Factory by 2001 = 1
(1) (2) (3) (4)
Electrification Rate 0.13∗∗∗ 0.16∗∗∗ 0.14∗∗∗ 0.15∗∗∗
(0.03) (0.03) (0.02) (0.02)
Urbanization Rate 0.06∗∗∗ 0.05∗∗ 0.05∗∗∗ 0.07∗∗∗
(0.02) (0.02) (0.02) (0.02)
Density (Age 15-64) 0.06∗∗∗ 0.03∗∗∗
(0.02) (0.01)
Yrs. of Schooling 0.01 0.01 −0.01 0.001
(0.02) (0.02) (0.02) (0.02)
Constant 0.13∗∗∗ 0.14∗∗∗ 0.06∗∗∗ 0.06∗∗∗
(0.01) (0.01) (0.01) (0.01)
Observations 485 485 485 485
R2 0.52 0.51 0.43 0.42
Adjusted R2 0.51 0.50 0.43 0.42
∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
The dependent variables are whether a sub-district had RMG factory by 1991
and 2001 respectively. All explanatory variables are standardized. The results
indicate that RMG factory location choice is primarily driven by infrastructure
considerations (as measured by electrification and urbanization rate). Labor market
conditions, as measured by density of working age population and years of school-
ing seem less important.
48
Table 2.5. 1991 and 2001 αk and αW and Sub-sequent Changes in Sub-district
Characteristics
Dependent variable:
αk 91 αW 91 αk 01 αW 01
(1) (2) (3) (4)
∆ Electrification Rate 0.05 −0.004 0.05 −0.002
(0.03) (0.03) (0.03) (0.03)
∆ Urbanization Rate 0.27 0.20 0.09 0.06
(0.27) (0.24) (0.21) (0.18)
∆ Pop (Age 15-64) −0.00 0.00 −0.00 0.00
(0.00) (0.00) (0.00) (0.00)
∆ Male Ind. LFP 0.10 0.09 0.08 0.08
(0.12) (0.11) (0.12) (0.11)
∆ Male Schooling (<15) −0.13 −0.14∗ −0.14 −0.15∗
(0.09) (0.08) (0.09) (0.08)
Constant 0.01 0.02∗ 0.01 0.02∗
(0.01) (0.01) (0.01) (0.01)
Observations 53 53 53 53
R2 0.10 0.13 0.08 0.12
Adjusted R2 0.002 0.03 −0.01 0.02
∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
αk and αW at the beginning of 1991 and 2001 is uncorrelated with changes in elec-
trification rate, urbanization rate, population, industrial labor force participation of
men, and men‘s schooling in subsequent decade.
49
50
Table 2.6. Export Exposure and Female Labor Force Participation
Dependent variable:
Ages 15-64 Ages 15-29 Ages 15-20
∆ FLFPRs,t ∆ FLFPR-Inds,t ∆ FLFPRs,t ∆ FLFPR-Inds,t ∆ FLFPRs,t ∆ FLFPR-Inds,t
(1) (2) (3) (4) (5) (6)
∆ Export Exposure 4.06∗∗∗ 5.36∗∗∗ 5.23∗∗∗ 6.88∗∗∗ 6.34∗∗∗ 7.56∗∗∗s,t
(0.01) (0.01) (0.01) (0.01) (0.01) (0.01)
Electrification Rates,t−1 0.03 0.02 0.04 0.02 0.05 0.04
(0.03) (0.02) (0.04) (0.03) (0.04) (0.04)
Urbanization Rates,t−1 0.01 0.01 −0.01 0.005 0.01 0.005
(0.02) (0.01) (0.02) (0.02) (0.02) (0.02)
Yrs. School (15-64,M)s,t−1 0.04∗∗∗ 0.01 0.04∗∗∗ 0.01 0.04∗ 0.01
(0.01) (0.01) (0.02) (0.01) (0.02) (0.01)
Yrs. School (15-64,F) ∗∗∗ ∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗s,t−1 −0.05 −0.03 −0.05 −0.03 −0.05 −0.04
(0.01) (0.01) (0.01) (0.01) (0.02) (0.01)
Densitys,t−1 −0.00 0.00 −0.00 0.00 −0.00 −0.00
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
Pop (15-64)s,t−1 0.00 0.00 0.00 0.00 0.00 0.00
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
Y Ms,t−1 0.19 −0.001 0.15 0.03 0.06 −0.04
(0.13) (0.07) (0.11) (0.09) (0.11) (0.13)
Period Fixed Effects ✓ ✓ ✓ ✓ ✓ ✓
Observations 126 126 126 126 126 126
R2 0.73 0.73 0.70 0.74 0.65 0.74
Adjusted R2 0.70 0.71 0.67 0.72 0.62 0.71
∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Female Labor Force Participation (FLFP) is measured in percentages. Export exposure is measured in thousands of dollars per
working age (16-64 year old) person. The results show that increases in export exposure led to increases in FLFP. As expected,
the effect is stronger for younger women, and for industrial FLFP.
Table 2.7. Stability of Estimated Coefficients
Dependent variable:
FLFP (Ages 15-64)
(1) (2) (3) (4) (5)
∆ Export Exposure 4.52∗∗∗ 5.00∗∗∗ 4.82∗∗∗ 4.53∗∗∗ 4.06∗∗∗
(0.60) (0.52) (0.62) (0.63) (0.70)
Electrification Rate −1.39 4.23 3.42
(2.45) (2.76) (2.82)
Urbanization Rate 1.49 0.89 0.60
(1.71) (1.60) (1.62)
Yrs. School (15-64, M) 3.30∗∗∗ 3.68∗∗∗
(1.14) (1.17)
Yrs. School (15-64, F) −4.78∗∗∗ −4.64∗∗∗
(1.18) (1.18)
Density 0.00 −0.00
(0.00) (0.00)
Pop (15-64) 0.00
(0.00)
Y Ms,t−1 18.80
(12.90)
Period Fixed Effects No ✓ ✓ ✓ ✓
Observations 126 126 126 126 126
R2 0.32 0.66 0.67 0.72 0.73
Adjusted R2 0.31 0.66 0.65 0.70 0.70
∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
FLFP is measured in percentages. Export exposure is measured in thousands of
dollars per working age (16-64 year old) person. The results provide evidence of
stability of estimated coefficient on export exposure under different nested specifica-
tions.
51
52
Table 2.8. Effect of Export Exposure on Marriage and Age-specific Fertility Rates
Dependent variable:
∆ Marriage Rates,t ∆Fertility Rates,t
Ages 15-20 Ages 21-30 Ages 15-20 Ages 21-30 Ages 30-40
(1) (2) (3) (4) (5)
∆ Export Exposuret −0.22 0.38 0.22 −0.16 3.98
(0.53) (0.43) (0.47) (1.63) (2.44)
Electrification Rates,t−1 5.08∗∗ −1.82 2.37 −7.81 −2.54
(2.31) (1.89) (2.08) (7.12) (10.69)
Urbanization Rates,t−1 −0.29 2.02∗ −0.74 6.00 4.90
(1.34) (1.09) (1.20) (4.13) (6.20)
Yrs. School (15-64,M)s,t−1 −1.70∗ −1.45∗ −1.40 −3.18 −8.47∗
(0.96) (0.78) (0.86) (2.95) (4.42)
Yrs. School (15-64,F)s,t−1 1.75∗ 1.32 1.93∗∗ 5.41∗ 11.80∗∗
(0.99) (0.81) (0.89) (3.04) (4.57)
Densitys,t−1 −0.00∗∗ −0.00 −0.00 0.00 −0.00
(0.00) (0.00) (0.00) (0.00) (0.00)
Period Fixed Effects ✓ ✓ ✓ ✓ ✓
Observations 126 126 126 126 126
R2 0.60 0.74 0.87 0.86 0.89
Adjusted R2 0.58 0.72 0.86 0.85 0.88
∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Marriage rates are in percentages. Number of own children in household is used as estimates of fertility rates. Export
exposure is measured in thousands of dollars per working age (16-64 year old) person. The results indicate that increases
in export exposure did not have any effect on marriage and fertility rates of Bangladeshi women.
53
Table 2.9. Export Exposure and Human Capital Accumulation of Bangladeshi Girls (Ages 5-19)
Dependent variable:
∆ Enrollment Rates,t ∆ Literacy Rates,t Yrs. Schools,t
Ages 05-09 Ages 10-13 Ages 14-19 Ages 14-19 Ages 14-19
(1) (2) (3) (4) (5)
∆ Export Exposures,t 0.56 −1.19 −3.17∗∗∗ −0.96 −0.13
(0.90) (0.88) (1.20) (0.83) (0.08)
Electrification Rates,t−1 −9.15∗∗ −17.84∗∗∗ −26.78∗∗∗ −12.15∗∗∗ −1.36∗∗∗
(3.91) (3.66) (5.04) (3.57) (0.34)
Urbanization Rates,t−1 1.04 −0.12 −0.30 −0.42 0.07
(2.31) (2.21) (3.06) (2.11) (0.20)
Yrs. School (15-64,M) −0.43 5.28∗∗∗s,t−1 7.91∗∗∗ 6.81∗∗∗ 0.71∗∗∗
(1.60) (1.61) (2.55) (2.55) (0.18)
Yrs. School (15-64,F) 1.49 −4.92∗∗∗ −5.29∗∗ −6.96∗∗∗ −0.68∗∗∗s,t−1
(1.67) (1.58) (2.18) (1.53) (0.15)
Densitys,t−1 0.00 0.00 0.00 0.0000 0.00
(0.00) (0.00) (0.00) (0.00) (0.00)
Y M ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗s,t−1 −0.58 −0.54 −0.52 −0.50 −0.45
(7.72) (6.83) (12.87) (8.79) (0.12)
Period Fixed Effects ✓ ✓ ✓ ✓ ✓
Observations 126 126 126 126 126
R2 0.89 0.92 0.67 0.93 0.93
Adjusted R2 0.88 0.91 0.64 0.93 0.93
∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Enrollment rate and literacy rates are in percentages. Export exposure is measured in thousands of dollars per working
age (16-64 year old) person. The results indicate that increases in export exposure led to decreases in enrollment for
working age teenage girls. However, export exposure did not have an effect on other dimensions of human capital accu-
mulation of Bangladeshi girls.
CHAPTER 3
RURAL FERTILITY TRANSITIONS: THE ROLE OF LAND CONSTRAINTS
3.1 Introduction
Fertility rates and socioeconomic development exhibit a strong negative asso-
ciation. Around its independence in 1971, Bangladesh was one of the most densely
populated and poor countries of the world. The country also had a very high fertil-
ity rate. Since then, it underwent an unusually rapid fertility transition. The total
fertility rate (TFR) dropped from 6.92 to 2.05 between 1970 - 2020. This drop is
faster than any experienced by United Nations (UN) country blocks between 1950
- 2020.1 The transition in Bangladesh took place without the use of coercive family
planning strategies, and amid persistent unfavorable conditions of high but falling
mortality, low economic development and contraception hesitance (Caldwell et al.,
1999).2
The national TFR may hide substantial spatiotemporal differences in the on-
set of fertility transitions. Stoeckel and Choudhury (1969) document a 27 percent
decrease in TFR in some rural areas between 1958 and 1967. However, fertility re-
mained positively associated with husband’s education and familial wealth in rural
Bangladesh in the early 1970s (Stoeckel & A. K. M. Alauddin Chowdhury, 1980),
suggesting a preference for larger families despite possible declines in fertility. In
urban settings at around the same time, fertility and contraceptive use were nega-
tively associated with husband’s income and wife’s education as well (Chaudhury,
1978). Taken together, these facts suggest the possibility of spatiotemporal varia-
tion in the onset of the Bangladeshi fertility transition.
Despite the importance of declining fertility for economic development and
for increasing the rights and autonomy of individual women within households
(Aaronson et al., 2014) and of women on the political stage Doepke et al. (2012),
much of the decline in fertility in Bangladesh is not well understood. In first part of
this paper, I document the fertility transition in Bangladesh in finer spatial detail
and establish the timing, magnitudes and correlates of persistent declines in fertility
and fertility preference among different birth cohorts of Bangladesh. In the second
1See Figure 3.1
2In late 1990s, Bangladesh was the poorest country to have a TFR below 5
(Caldwell et al., 1999)
54
part of the paper, I seek to add to our understanding of the Bangladeshi fertility
transition by analyzing the influence of increasing agricultural land constraint on
the fertility transition of Bangladesh.
The rest of the paper is organized as follows - section 3.2 discusses the current
state of the literature, section 3.3 describes the data and the empirical approach,
section 3.4 presents the results, and section 3.5 discusses the findings.
3.2 Literature Review
Three broad categories of models have been used to explain fertility transitions
(Shenk et al., 2013). Risk/mortality based models postulate that lower mortality
reduces fertility. Cultural transmission models emphasize the role of evolution of
preferences and its diffusion, whereas other models emphasize the increasing returns
to human capital, opportunity cost of women and sectoral shifts.
The earliest research to focus on the specific context of Bangladesh stud-
ied 547 couples in government housing in 1963 in what is now the capital city in
Bangladesh. The authors found a positive correlation between education and fam-
ily size, a strong preference for male children, a greater desire for more children
among women than men, and limited use of contraceptives despite knowledge of
it (Roberts et al., 1965). Among elite professionals, support for family planning
started to be positively related with education soon thereafter in 1970s Chaudhury,
1975. In more rural areas, between 1967-1971, large declines in fertility were docu-
mented amid low contraceptive usage (Stoeckel & Choudhury, 1969), and a positive
correlation between fertility and education, and landholding (Stoeckel & A. K. M.
Alauddin Chowdhury, 1980).
Declines in childhood mortality may reduce fertility either by changing family
size preference, or through child replacement motivations. Chowdhury et al. (1976)
finds no evidence of the child replacement motivation at work in Bangladesh at
around 1970s. But in a subsequent work, Hossain et al. (2007) finds evidence of
volitional replacement of deceased child in the 1982-93 period, suggesting that un-
der lower fertility conditions, target family size may become more salient. Both of
these studies examine the influence of mortality on fertility from the perspective of
couples. Widespread changes in family size preference as a result of large declines
in childhood mortality from its high in 1900 remains unexplored.
In addition to a reduction in child mortality, several explanations have been
55
proposed to account for the fertility decline in Bangladesh. The most often studied
aspect of the Bangladeshi fertility transition has been the role of family planning
program. The Matlab family planning program seems to have led to a long-term
fertility decline explaining about 14-17 percent of the decline in fertility in the
treatment villages (Joshi & Schultz, 2013; Sinha, 2005). However, in terms of ex-
penditure per person terms, the Matlab family planning had at least three times
the resources of what was allocated to national family planning efforts (DeGraff,
1991). Using nationally representative surveys between 1969-1983, R. Amin et al.
(1987) find that the national family planning programs increased awareness of con-
traceptive options from 28 percent to 81 percent of the population, but did not
change fertility preference. However, whether the increase in contraceptive aware-
ness is a direct consequence of the family planning programs, or provided the impe-
tus to a diffusion mechanism as discussed in (Bhattacharya & Chakraborty, 2017;
Munshi & Myaux, 2006) is unclear.
Since 1970, and particularly since 1990, female labor force participation and
women’s schooling had increased substantially in Bangladesh. Heath and Mobarak
(2015) find a substantial increase in women’s employment, and a modest increase
in age of marriage and age of child-bearing due to exposure to the ready made
garments industry, Bangladesh’s largest employer of women. Chapter 2 of this dis-
sertation also finds that the impact of the ready made garments industry on the
labor force participation of Bangladeshi women is substantial. However, I find that
the impact on fertility is statistically insignificant. On the other hand, Hahn et al.
(2018) find an 8-12 percent decline in fertility resulting from a plausibly exogenous
increase in women’s education at a time when the TFR was around 4. Fujii and
Shonchoy (2020) find that electrification in rural areas lowered the number of chil-
dren per woman by 0.2 over a five year period. However, it should be noted that
the electrification rate in Bangladesh was only 32 percent in 2000 The World Bank
(2021), by which time TFR had already fallen to 3.3 from near seven in 1970.
Even if we disregard the arguments against exclusion restrictions in the afore-
mentioned studies, most of the change in TFR in Bangladesh remains unexplained.
By my calculations, the aforementioned studies with its emphasis on family plan-
ning program, education, and electrification explain about a quarter of the fall in
fertility between 1970-2020, and about three quarters remain unexplained. Addi-
tionally, women’s economic opportunity in the form of employment in the garment
industry at best explain only a minor amount of the fertility decline.
56
The majority of the Bangladeshi population lives in rural areas, and is en-
gaged in agriculture. This was more true at the dawn of the fertility transition in
Bangladesh (see chapter 1). To my knowledge, the role of rural agrarian economic
conditions in helping initiate the fertility decline in rural Bangladesh, and by exten-
sion in Bangladesh overall, has not been studied. Beginning in 1900s, the childhood
mortality rate in Bangladesh began to fall, and was halved by 1970 (UN DESA,
2022). This led to a fast population growth in what was already one of the most
densely-populated locations in the world. In turn, farm sizes shrinked , and changes
in agricultural technology led to more capital-intensive farming (Caldwell et al.,
1999), reducing labor demand in rural areas. For the second part of this research, I
attempt to partially explain the fertility transition in rural Bangladesh through an
income effect of declining land availability in rural Bangladesh.
3.3 Empirical Approach and Data
3.3.1 Empirical Approach
This research chapter uses two different methods to answer two different re-
search questions. Using descriptive analysis, I determine the timing of the onset
of rural and urban fertility transition in Bangladesh. And using a difference-in-
difference (DiD) approach, I investigate the influence of land constraints on rural
fertility in Bangladesh.
Descriptive Analysis
To determine the timing of the onset of fertility transition in Bangladesh, I
create a retrospective panel data of birth histories from several rounds of represen-
tative national surveys. This retrospective panels from different individuals from
different surveys are then stacked together. Sample weights are adjusted following
Schmidt and Elkasabi (2020).
The birth history panels allow me to estimate realized age-specific fertility
rates for 5 year birth cohorts between 1926-81. Following the same cohort over time
controls for the impact of age compositions on measured fertility outcomes such
as crude birth rates and TFR. The same measure is estimated for rural and urban
population. Using these age-specific fertility rates of different cohorts, I document
the timing of the onset of persistent declines in fertility among Bangladeshi women.
57
I further investigate and establish some of the correlates of completed fertility at
the individual level using simple linear regressions.
The findings of the descriptive analysis, discussed in Section 3.4.1, reveals
i) a five to ten year delay on the onset of fertility transition in rural Bangladesh,
compared to urban Bangladesh; and ii) that land owners had higher fertility rates.
In agrarian economies, land is a key but fixed resource. Population growth due to
falling child mortality will reduce land per labor, which will in turn reduce income
based on land in rural economies. Thus, fertility can start to decline due to the
the negative income effect from reduced land availability. I test this mechanism by
methods described below.
Examining the role of land constrain
I estimate the effect of land constrain on fertility in rural Bangladesh using a
DiD approach. Specifically, I estimate an equation of the following form:
Fertilityi,t = β0 + β1Rurali + β2Xs,t + β3Rural ∗Xs,t + δt + ϵi,t (3.1)
Where, Fertilityi,t is the total number of birth given by a woman till time t. Ru-
ral is an indicator that the person was born in rural areas, and was staying in a
rural area at the time of interview. The first explanatory variable I examine, Xs,t,
is Increased Land Availabilityi,t. It is a measure of district specific plausibly exoge-
nous increase in land availability. δt are year fixed-effects. If land constrains helped
with fertility reduction, increased land availability will increase fertility.
According to the 1981 Bangladeshi census data, 70.7 percent of the rural labor
force was engaged in agriculture in 1981, whereas 65.8 percent of the households
owned land (Bangladesh Burea of Statistics, 1984). This points to an existence of
a large group of land-less agricultural workers. The landless agricultural laborers
will be affected by land constraints through the supply of agricultural labor. In fact,
Stoeckel and A. K. M. Alauddin Chowdhury (1980) provide evidence of an already
emerging reduced fertility for landless agricultural workers. Hence, I also construct
and use a plausibly exogenous Net Migration Rates,t as an explanatory variable.
Areas with high out-migration will lead to higher fertility through the positive
income effect of reducing rural labor supply and increases rural wages. Below, I
discuss construction of these two variables.
58
Construction of Net Migration Rate and Land Availability Owing to
continual religious strife, Bangladesh has been loosing its Hindu population, the
largest minority religious group of Bangladesh, to out-of-country-migration at least
since the early 1900s. One of the two fastest out-migration occurred in the lead-
up, during, and aftermath of the independence struggle of Bangladesh in 1971.3
Consequently, the Hindu population reduced by 5 percentage points between 1961
to 1974. The region also suffered from famine in 1973-74, increasing incentives to
out-migrate.
With neighboring India presenting a less hostile environment for the Hindu
population, most of the out-migration was by members of the Hindu commu-
nity. This resulted in a reduction share of non-Muslim in the country by 6.3 per-
centage points (pp) between 1961-1981. As per (Internaltional Statistical Insiti-
tute, 1976), there was no difference in fertility rates of Muslims and non-Muslims
in Bangladesh at least till 1976. This allows me to construct the measure of
Increased Land Availabilitys,t that could be unrelated to other factors influencing
fertility.
I first estimate the 1981 district level population that would have resulted in
the absence of large scale out-of-country migration as follows:
Muslim
Expected Population Bangladesh,1981s,1981 = ∗ Population (3.2)Muslim s,1961Bangladesh,1961
Since the Muslim population did not out-migrate from Bangladesh at a large scale,
the ratio of Muslim population in 1981 to the Muslim population in 1961 pro-
vides an estimate of population growth that would have happened absent the re-
ligious component to Bangladesh’s war of independence and related non-Muslim
out-migration.4 Note that the ratio captures effects of war, natural fertility and
mortality, and out-migrations independent of the religious strife angle.
By applying the ratio to district level population in 1961, I estimate the pop-
ulation that would have existed if religious strife was not part of the 1961-1981
3Pro-Muslim majority identity was used by the then Pakistani government as a
tool to counteract pro-independence sentiment in current Bangladesh.
4The first wave of large scale out-migration of non-Muslim population happened
in connection with the 1947 partition of India. Lack of available sub-division,
which later was converted to districts, level Hindu and Muslim population data of
1941 census prevents me from using the first wave of out-migration to construct
measures of land availability and net out-migration.
59
demographic changes in Bangladesh. The difference between the expected popu-
lation and actual observed population is the net migration, and the net migration
rate is the difference normalized by district level 1961 population:
Expected Populations,1981 − PopulationNet Migration Rate s,1981s,1961−81 = (3.3)Populations,1961
The estimate of Increased Land Availabilitys,1981 as follows:
Non-river Area Non-river Area
Increased Land Availability s ss,t = − (3.4)Populations,1981 Expected Populations,1981
Identification The identification depends on three key assumptions. First, the
identification assumes that the population pressure affects rural areas but not ur-
ban areas. Differences in land availability through out-migration of non-Muslims
creates differences in land availability for farming, and agricultural labor supply.
Thus the familial income effect acts on rural women, but not urban women. This is
because more than 90 percent of the population was engaged in agriculture during
1961-1981. And almost of rural Bangladesh was engaged in land-based agriculture.
But agriculture is not practiced at scale in urban regions. Consequently, differences
in land availability affects rural women more.
Second, the identification assumes that women who were of fertile age in 1961-
1981 were not affected by these to the same degree. This is restrictive, and plausi-
bly not true given the continuous nature of out-migration. Not accounting for the
treatment contamination will likely bias the estimates towards zero.
Third, it is assumed that out-migration is unrelated to local labor market con-
ditions. And this assumption is likely to hold since the out-migration is primarily
due to religious strife (Mithun (2019); Datta (2013)).
3.3.2 Data
The key data sources for this project are household surveys conducted through
the World Fertility Survey (WFS) in 1975-76 (Internaltional Statistical Insititute,
1976), eight waves of standard versions of the Demographic Health Survey’s (DHS)
Bangladesh survey between 1993 and 2018, and 2001 special issue DHS survey
(ICF International, 2021). Combined, these surveys provide data on 217,878 adult
women and their households. This data was collected from numerous different clus-
ters in Bangladesh. Administratively, Bangladesh is divided into 8 divisions, 64
60
districts, and 495 sub-districts at present. However, administrative boundaries un-
derwent substantial changes over the time.
The WFS survey is representative at the national and rural-urban level, and
the DHS surveys are representative at the national, rural-urban, and divisional
levels at the time of the survey. These surveys provide a detailed look into the
realized and desired fertility behavior of Bangladeshi women and couples. A rich
set of additional socioeconomic and demographic information about the surveyed
women and their households is available as well.
3.4 Empirical Findings
3.4.1 Descriptive analysis
I first estimate age specific fertility rates (ASFR) of 5-year birth cohorts
of Bangladeshi women born between 1926-1981. Estimated ASFR table for
Bangladesh as a whole as well as ASFRs in rural and urban areas are presented
in Appendix A. As shown in Figure 3.2 the fertility transition begins with the 1946-
1950 birth cohort. Particularly, the women in this cohort have fewer children by 25
years of age. Relative to previous generations, they also showed strongest decline
when they were 40-45 years of age.
Underneath the overall numbers, there is rural-urban differences. The com-
pleted fertility in rural Bangladesh has been higher than in urban Bangladesh
since 1926-30 birth cohorts. The difference in completed fertility rates started at
about 0.06 children, increased to about 0.3 for the 1925-40 birth cohort and then
increased to 1 child per woman by the 1941-45 cohort.
Based on persistent declines in cohort fertility, the transition in urban areas
starts with the 1941-45 cohort, whereas in the case of rural areas, the transition
begins with the 1946-50 cohort. In urban areas, large declines in age-specific fertil-
ity relative to the previous cohorts start with ages 25 onward. Thus, the transition
may have started by 1970. In the rural areas, the transition happens most strongly
after the age of 35. Thus, the rural transition might have started by 1980. One
interesting feature of the transition in Bangladesh is the relatively small gap in
timing of onset of transition between rural and urban areas of the country.
61
Correlates of Fertility at the Individual Level
In this sub-section, I establish some of the key correlates of fertility by running
regressions of the following form:
Fertilityi = β0 +Xiβx + Ziβz + δt + δbirth + ϵi (3.5)
The correlates checked include education of husband and wife, working in agricul-
tural sector and land ownership. I also include birth year fixed effects and survey
period fixed effects to account for time-trends. The regression results are presented
below. In the baseline specification, I do not include a measure of fertility prefer-
ence – stated ideal number of children, as the measure is likely to be endogenous to
fertility itself. In an additional specification, I use fertility preference as a covariate
as well. Since I am using data from multiple periods with changing population, I
use adjusted sample weights in the regression.
The results of the regression is in table 3.1. Both wife‘s, and husband’s educa-
tion are negatively correlated with fertility. However, the strength of relationship
between fertility and wife’s education seem to be much greater. Among occupa-
tions, working in agriculture is strongly associated with higher fertility. In addition,
the limited data does not provide any evidence of relationship between the amount
of land and fertility. The qualitative and quantitative strength of correlates dis-
cussed above remain similar in regressions with fertility preference as a control (see
table 3.2). However, stated fertility preference is strongly correlated with fertility.
Since fertility preferences are such an important correlate of realized fertility,
next I check the correlates of fertility preference.
Correlates of Fertility Preference
Correlates of fertility preference are established by running regressions of the
following form:
Fertility Prefi = β0 +Xiβx + Ziβz + δt + δbirth + ϵi (3.6)
The correlates checked include measures of education of husband and wife, working
in agricultural sector and land ownership. Demographic controls include birth year
fixed effects and I include survey period fixed effects to account for trends. The
regression results are presented below. Since I am using data from multiple periods
with changing population, I use adjust sample weights in the regression.
62
Regression results in table 3.3 indicate that the education level of couples,
particularly of wife’s, have the strongest impact on fertility preference. In addition,
fertility preference is higher in rural areas. However, fertility preference do not
seem to depend on working in agriculture or owning land.
3.4.2 Role of Migration and Land Constraint
I investigate the effect of land availability and net migrations on fertility by
estimating equations of the form of 3.1. The Xs,t are constructed measures of
Net Migration Rates,1961−81 and Increased Land Availabilitys,1981. Table 3.4 presents
summary statistics on population, density, out-migration, and land availability. The
population and thereby the population density almost doubled between 1961 to
1981. As expected, the mean of expected district-level population in 1981 is higher
than the mean of actual population in 1981, and net migrations positive (i.e. on av-
erage, places experienced out-migration), and the net migration is about 7 percent
of the 1961 population. Table 3.5 presents correlations between selected measures.
As expected, areas with larger population in 1961 had greater levels of population
and expected population in 1961. Moreover, those areas also saw greater outward
migration and greater extra land per person.
The sample is restricted to years before and including 2001, to minimize pick-
ing up effects of land availability occurring through children of those exposed to the
land availability shock by 1981. Moreover, the rural sample consists only of women
who were interviewed in a rural area, and were raised in a rural area. The urban
sample consists of women interviewed in urban area who were raised in urban area.
This controls for the effect of selective rural-urban migration.
Table 3.6 present the estimates of regression equations 3.1. The results show
that rural areas had higher levels of fertility, and conform to findings in the descrip-
tive analysis, and the literature more broadly. The results also show a statistically
significant and positive effect of land availability. However, the coefficient on land
availability and rural interaction is statistically insignificant.
In the specification with net migration rate, similar level differences in total
births are observed between rural and urban regions. Net out-migration is nega-
tively relatively with total births. Here, however, the coefficient on net migration
rate interacted with rural is positive and statistically significant: implying that
outward migration had a positive impact on realized fertility in rural areas.
63
3.5 Discussion of results and conclusion
This research seeks to add to the understanding of the dramatic fertility tran-
sition in Bangladesh. To do so, I first document that the fertility transition begins
with the 1941-45 cohort in urban areas and the rural transition starts almost imme-
diately with the 1946-1950 birth cohort. The transition happens within the context
of falling mortality rates. Looking into the correlates of fertility, husband and wife’s
education, and changes in fertility preference are strongly associated with lower
fertility. Increased education is also associated with preference for fewer children.
This is not surprising given the negative education and fertility relationship around
the world.
The transition in Bangladesh started in between 1960-70, when GDP per
capita was below 500 USD in 2015 dollars (The World Bank, 2021). At the indi-
vidual level, participation in agriculture, especially while owning land, is associated
with increased fertility. Mortality rates in Bangladesh have been falling since early
1900s. The improved mortality conditions have reduced land availability.5 Thus, I
investigate the possibility that reduced returns to agricultural activities have led to
smaller families by examining the role of increased land availability and less supply
of labor in rural areas on rural fertility.
My findings suggest that while land availability directly may not have in-
duced fertility reductions, overall effect of rural economic competition through
outward migration may have been a contributor to the rural fertility transition in
Bangladesh. Given Bangladesh is not unique, and perhaps typical among the broad
swath of developing countries undergoing fertility transition in rural areas in 1960s,
this presents an additional mechanism of initiation of rural fertility transitions.
There are three key limitations of the empirical exercise linking fertility and
rural economic conditions. First, the current set-up makes it hard to speak to the
importance of the effect size. This might be addressed by using long differences
or other panel data methods, and by redefining our Xs,t to make commenting on
effect size more straight-forward. Another potential solution is to create district
level panel of fertility measures using re-weighting techniques. Second, the measure
of Increased Land Availabilitys,1981 likely contains large error due to data quality.
This may explain the null effect of increased land availability on fertility in rural
5See figure 3.3
64
areas while out-migration differentially increased fertility in rural areas. Obtain-
ing better measures of arable land could be possible using GIS techniques. Third,
a formal model of the argument linking mortality declines to increased rural eco-
nomic competition and subsequent fertility decisions of households could help with
interpretation of the findings.
In the next chapter, I investigate the changes in women’s employment in
Bangladesh with a focus on rural areas, by comparing to the experience of neigh-
boring Indian states.
65
3.6 Figures
Figure 3.1. Fertility Transitions Around the World 1950:2000
66
Figure 3.2. Rural and urban Fertility Transitions in Bangladesh
67
Figure 3.3. GDP, Land and Fertility in Bangladesh
68
3.7 Tables
Table 3.1. Correlates of Fertility
Dependent variable:
Fertility
(1) (2) (3) (4)
Female Educ −0.290∗∗∗ −0.317∗∗∗ −0.275∗∗∗ −0.300∗∗∗
(0.007) (0.015) (0.018) (0.013)
Male Educ −0.120∗∗∗ −0.084∗∗∗ −0.070∗∗∗ −0.118∗∗∗
(0.006) (0.013) (0.014) (0.011)
Rural 0.279∗∗∗ 0.397∗∗∗
(0.011) (0.029)
Engaged in Agri. 0.033∗∗∗ 0.001 −0.048 0.029
(0.012) (0.021) (0.040) (0.023)
Land Owner 0.056∗∗ 0.033
(0.023) (0.033)
Land Amount −0.00000
(0.00002)
Engaged in Agri X Land Owner 0.095∗∗
(0.048)
Engaged in Agri X Land Amount 0.00001
(0.00003)
Observations 106,338 30,058 23,719 27,480
Adjusted R2 0.510 0.552 0.558 0.524
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
69
Table 3.2. Correlates of Fertility (Controlling for Stated Preference)
Dependent variable:
Fertility
(1) (2) (3) (4)
Female Educ −0.251∗∗∗ −0.277∗∗∗ −0.232∗∗∗ −0.260∗∗∗
(0.007) (0.015) (0.018) (0.013)
Male Educ −0.107∗∗∗ −0.070∗∗∗ −0.058∗∗∗ −0.103∗∗∗
(0.006) (0.012) (0.014) (0.010)
Rural 0.223∗∗∗ 0.320∗∗∗
(0.011) (0.028)
Fertility Preference 0.451∗∗∗ 0.401∗∗∗ 0.408∗∗∗ 0.472∗∗∗
(0.006) (0.011) (0.012) (0.011)
Engaged in Agri. 0.028∗∗ 0.007 −0.066∗ 0.042∗
(0.012) (0.021) (0.040) (0.022)
Land owner 0.060∗∗∗ 0.024
(0.023) (0.033)
Land Amount −0.00000
(0.00002)
Engaged in Agri.XLand Owner 0.132∗∗∗
(0.048)
Engaged in AgriXLand Amount −0.00002
(0.00003)
Observations 103,154 28,217 22,075 26,894
Adjusted R2 0.534 0.572 0.578 0.555
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
70
Table 3.3. Correlates of Stated Fertility Preference
Dependent variable:
Fertility Preference
(1) (2) (3) (4)
Female Educ −0.084∗∗∗ −0.103∗∗∗ −0.103∗∗∗ −0.088∗∗∗
(0.004) (0.008) (0.010) (0.007)
Male Educ −0.032∗∗∗ −0.042∗∗∗ −0.041∗∗∗ −0.034∗∗∗
(0.003) (0.007) (0.008) (0.006)
Rural 0.109∗∗∗ 0.158∗∗∗
(0.006) (0.016)
Engaged in Agri 0.004 −0.026∗∗ −0.006 −0.015
(0.006) (0.012) (0.022) (0.012)
Land Owner 0.002 0.015
(0.013) (0.018)
Land Amount 0.00001
(0.00001)
Engaged in Agri x Land owner −0.026
(0.026)
Engaged in Agri x Land Amount 0.00001
(0.00002)
Observations 103,154 28,217 22,075 26,894
Adjusted R2 0.084 0.062 0.056 0.086
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
71
72
Table 3.4. Summary Statistics: Population, Density, Migration, and Land Availability
Statistic N Mean St. Dev. Min Max
Population 61 54 987,955 620,288 86,422 3,663,965
population 81 54 1,684,688 1,136,591 141,045 6,268,521
Expected Population 81 54 1,826,901 1,147,022 159,809 6,775,307
Density61 54 447 277 27 1,631
Density81 54 719 347 66 1,909
Net Migration 54 142,212 701,546 −3,589,919 2,194,943
Net Migration Rate 54 0.07 0.61 −2.74 1.22
Extra Land Per Person (in Acres) 54 −0.002 0.202 −1.198 0.356
Table 3.5. Correlations Between Selected Measures
Pop 61 Pop 81 Expected Pop 81 Extra Land Per Person Net Migration
Pop 61 1.00 0.81 1 0.22 0.32
Pop 81 0.81 1.00 0.81 0.002 −0.29
Expected Pop 81 1.00 0.81 1.00 0.22 0.32
Extra Land Per Person 0.22 0.002 0.22 1.00 0.36
Net Migration 0.32 −0.29 0.32 0.36 1.00
Table 3.6. Land Constraint and Fertility
Dependent variable:
Total births
(1) (2)
Rural 0.678∗∗∗ 0.665∗∗∗
(0.007) (0.007)
Increased Land Availability 115.841∗∗∗
(23.636)
Increased Land Availability x Rural −34.447
(24.033)
Net Migration Rate −0.059∗∗∗
(0.017)
Net Migration Rate x Rural 0.147∗∗∗
(0.018)
Observations 1,372,174 1,372,174
R2 0.077 0.077
Adjusted R2 0.077 0.077
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
73
CHAPTER 4
FEMALE LABOR FORCE PARTICIPATION IN BANGLADESH AND
NEIGHBORING INDIAN STATES: 1987-2011
This chapter is co-authored with Shankha Chakraborty. I had an essential
role in developing the initial idea that led to this project. I also wrote a significant
amount of code that generates the results in this paper. I also wrote and edited
many sections of the paper.
4.1 Introduction
South Asia has one of the lowest Female Labor Force Participation Rate
(FLFPR) in the world, and the region as a whole has been experiencing a decline
in FLFPR in recent decades (see Figure 4.1). Within South Asia, FLFPR has been
declining for countries with relatively higher levels of FLFPR in 1990, including In-
dia; and increasing in countries with relatively lower levels of FLFPR in 1990 such
as Bangladesh (see Figure 4.2).
There is considerable variation within the geography of South Asian countries.1
Consequently, there are differences in the relative importance of agriculture, spe-
cialization in the type of agriculture, religion, and culture among these countries.
Additionally, the recent policy and economic environment have varied substantially.
All these factors affect both the levels and dynamics of FLFPR (see section 4.2).
In this paper, we examine the diverging experience of increasing FLFPR in
some parts of South Asia and decreasing FLFPR in other parts. We focus on the
case of Bangladesh between 1991-2011 and Indian states that border Bangladesh
– i.e. Mizoram, Tripura, Meghalaya, Assam, and West Bengal between 1987-2009.
Bangladesh and these Indian states, except Mizoram,2, are part of the Gangetic
plains and thereby likely share similar history of agriculture, and have been part of
multiple unitary polities before.3 Section 4.2 discusses the importance of this fact.
1For instance, Sri Lanka is an island and Bhutan is a mountainous land-locked
country. Bangladesh is mostly the delta of Gangetic plains whereas Afghanistan is
mostly a dry desert.
2Mizoram was part of Assam till 1972, and shares ethnic and economic similar-
ity to the Eastern Divisions of Sylhet and Chittagong of Bangladesh
3The eastern part of the Gangetic plains were often unified under one polity.
74
The time period choices were dictated by data and the geographic choices are mo-
tivated by the size-able increase in FLFPR that Bangladesh experienced in recent
decades. In contrast, FLFPR declined in India. Moreover, the focus on Bangladesh,
and its neighboring Indian states reduces time-invariant regional differences that
pose challenges to drawing conclusions since all states included.
We use Oaxaca-Blinder decomposition to investigate the diverging FLFPR tra-
jectory in Bangladesh and selected Indian states. Results of Oaxaca-Blinder decom-
positions suggests that differences in coefficients are more influential in explaining
the divergent FLFPR dynamics in Bangladesh and its neighboring Indian states
regarding women’s education, reproductive behavior, and measures of structural
change. This implies that the association between changes in education, reproduc-
tive behavior, and economic conditions are qualitatively different in Bangladesh
and its neighboring Indian states, and the underlying mechanisms of changes in
women’s employment were different too.
Particularly, we find that women’s education reduced FLFPR in selected In-
dian state but increased the same in Bangladesh. We also find evidence suggesting
that this is likely due to a stronger increase in women’s education in Bangladesh
relative to men’s when compared to the neighboring Indian states. Our analysis
also finds very weak evidence for increased urban and industrial employment of
men creating labor market opportunities for rural women.
The rest of the chapter is organized as follows - section 4.2 discusses the liter-
ature, section 4.3 describes the data and empirical approach, section 4.4 presents
and discusses the results, and section 4.5 concludes the paper.
4.2 Literature Review
The existing literature attempts to explain both the levels and dynamics of
women’s employment. Many of the factors that explain the differences in FLFPR
In ancient history of the Indian sub-continent, most of these regions were under
Mauryan Empire, Gupta Empire, Pala empire, and to a more limited extent, the
Greater Sena Empire. In medieval and early modern India, the Mughal Empire, the
Bengal Subha, and the British East India company controlled these regions. Note
that many of these polities including the Pala empire, the Greater Sena Empire,
the Bengal Subha, and early incarnations of the British East India company were
“regional" powers, indicating natural regional demarcations that includes these
regions.
75
across different regions are rooted in deep historical and geographical factors of
those regions. For instance, Hansen et al. (2015) finds that in a sample of Euro-
pean county, longer experience with agriculture is associated with reduced levels of
contemporary women’s employment. Their suggested mechanism is that increased
agriculture increased fertility through income effect, which in turn led to women’s
specialization in home production.
Alesina et al. (2011, 2013) provides evidence that historical plough use helped
shape gender norms surrounding fertility and women’s employment. Particu-
larly, their 2013 study suggests that historical plough use reduces women’s rela-
tive productivity in agriculture, leading to lower FLFPR norms in countries with
longer history of plough use. With similar mechanisms in mind, Fredriksson and
Gupta (2023) suggest that historical irrigation is negatively associated with current
FLFPR. Bolstering the same line of argument, Carranza (2014) finds that within
India, loamy soil structure that allows for deeper tillage suppressed female labor
demand historically and is associated with lower current FLFPR. Despite the po-
tentially historically determined norms regarding women’s employment, such norms
can and do change (see Fogli and Veldkamp (2011) and Fernández (2013)).
The results of these historical agricultural pattern apply well to the case of
Bangladesh and its neighboring Indian states. Considering that Bangladesh, and
Tripura, Meghalaya, Assam, and West Bengal (all Indian states under study except
Mizoram, previously under Assam) are part of the Gangetic plains, lower levels of
FLFPR in these areas relative to rest of the world are in line with the predictions
of the literature on FLFP.
Most of the studies examining the FLFPR dynamics explains changes as a re-
sult of three broad and often intertwined factors. First, is the role of fertility. Using
historical and contemporary data from different sources, Aaronson et al. (2021) find
that at lower levels of development, fertility and women’s labor supply are unre-
lated. However, they also find a negative relationship between fertility and FLFPR
at further stages of development. Using infertility shocks from 1990s data in Latin
America, Aguero and Marks (2008) finds no fertility-FLFPR relationship. Using
GDP per capita as a proxy for levels of development, the aforementioned studies
suggest a null fertility-FLFPR relationship in our data of 1991-2011 Bangladesh
and 1987-2009 neighboring Indian states.
A second set of literature focuses on the dynamics of development and FLFPR.
Empirical work such as Goldin (1995) and theoretical models such as Rees and
76
Reizman (2012) posits that at initial stages of development, movement away from
agriculture reduces FLFPR, and further along the path of development, increases
FLFPR though reduction in fertility, increase in women’s education, and structural
change. However, Gaddis and Klasen (2014) finds no evidence that this a general
pattern. Instead, they argue that the exact dynamics depends on the initial condi-
tions of labor market. While some research finds that time use trends in developing
regions, such as modern Africa, are similar to historical US experience (Dinkelman
& Ngai, 2022), the evidence for a general relationship between development and
FLFPR is limited.
The third string of literature suggests the specifics history and development
process matter. For instance, Goldin and Olivetti (2013) and Rose (2018) find that
increased employment opportunity for women through US men’s mobilization in
World War II, and increased factory work opportunities for war-time industrial
production, led to a size able short-term increase in FLFPR. However, both of
them find very limited long-term impact. In contrast, permanent shortage of men
due to World War II casualties were associated with persistent increases in FLFPR
in France (Boehnke & Gay, 2022). Additionally, studies such as Autor et al. (2013),
Autor et al. (2019) and the first chapter of my dissertation highlights that changes
in women-specific long-term employment opportunity due to trade changes FLFPR.
Focusing on the case of Bangladesh, the second chapter of this dissertation,
as well as Heath and Mobarak (2015) find that the expansion of garment indus-
try in Bangladesh have been a very important contributor to increases in FLFPR
in Bangladesh. Despite a similar impetus and timing to liberalize in 1990s (Dev,
2000), the Indian states neighboring Bangladesh did not see emergence of a simi-
larly large and female-biased industry.
Investigating the particular case of India, Afridi et al. (2018) find that the
lower levels of FLFPR among married women in rural India is attributable to in-
creased education of rural women that increased their relative returns to home
production. Additionally, Tumbe (2015) provides evidence of increased FLFPR in
Indian agriculture due to rural-urban migration of men for better paid work.
Taken together, the current literature does not allow us to hypothesize whether
differences in observed covariates of fertility or different development processes ex-
plain the differences in evolution of FLFPR in Bangladesh and neighboring Indian
states.
77
4.3 Empirical Approach and Data
4.3.1 Empirical Approach
We use the Oaxaca-Blinder decomposition method ((Oaxaca, 1973) and (Blin-
der, 1973)) to examine the diverging FLFP dynamics in Bangladesh and Indian
states that border Bangladesh – i.e. Mizoram, Tripura, Meghalaya, Assam, and
West Bengal.4 The Oaxaca-Blinder decomposition first estimates the same regres-
sions model for the two groups separately by using data from two groups separately.
The estimated regression can be differenced to examine the contribution of a) differ-
ences in mean value of covariates and ii) differences in coefficients, in explaining the
differences in outcome. We used the three-fold decomposition to ask how relevant
are i) differences in the evolution of observed explanatory variables (endowments),
and ii) differences in correlation between endowment and outcomes (coefficients) in
explaining the diverging Bangladeshi and neighboring Indian States’ experience.5
The decomposition estimated takes the following form:
∆̄YB − ∆̄YI = (∆̄X ′ ′ ′B − ∆̄XI) β̂I + ∆̄XI(β̂B − β̂I) + (∆̄XB − ∆̄XI) (β̂B − β̂I) (4.1)
Where I denotes selected Indian states and B denotes Bangladesh. The left-
hand side, ∆̄YB − ∆̄YI , measures the differences in yearly-∆FLFP between selected
Indian states and Bangladesh. This can be interpreted as the differences in trajec-
tory of FLFPR in the two regions. (∆̄XB−∆̄XI) is the differences in the trajectories
of covariates. Hence, the term (∆̄XB − ∆̄XI)′β̂I estimates how much of the differ-
ence can be attributed to the changes in covariates if both countries had the Indian
β̂I coefficients. The term ∆̄X ′I(β̂B − β̂I) measures the effect of having different
coefficients if both regions had the Indian covariate trajectories. The remaining
term provides us the interaction effect and does not have a clear interpretation in
this analysis. We incrementally add covariates of interest in the same specification
above.
Following the literature and descriptive analysis in Section 4.3.3, the first de-
composition analysis (Decomposition 1) examines the influence of differences in
human capital accumulation trajectories of both men and women (measured with
4The R package ‘oaxaca’ (Hlavac, 2022) was used for computation.
5Two-fold decomposition assumes that there exists a set of natural correlations
between outcome and explanatory variables, and hence not used in this analysis
78
Yrs. of Schooling) on the divergence in FLFPR in Bangladesh and neighboring
Indian states. Influence of both men’s and women’s education is tested. In both
countries, younger women are likelier to be employed. We account for this by us-
ing share of 15-39 year old in the district population as a covariate. In addition,
we also test the influence of changes in education levels of women relative to men
(∆ Y rsofEducF ).
Y rsofEducM
The second decomposition analysis (Decomposition 2) examines the in-
fluence of differences in the trajectory of observed reproductive behavior mea-
sured with marriage rate (∆ % Married), and rate of having under five children
(∆% With U-5 Child) at home, after controlling for human capital accumulation
and age structure.6
In the third decomposition analysis (Decomposition 3), we examine whether
structural changes were an important driver of the divergent FLFPR trajectory
between Bangladesh and the selected Indian states. This is done in two steps. First,
the we examine the effects of changes in two proxy measures of structural change –
(i) district level urbanization, and (ii) share of men working in the industrial sector,
on the divergence in FLFPR between Bangladesh and selected Indian states. This
analysis is considered as the baseline decomposition specification as it contains all
available covariates typically used in decompositions.
Second, we then examine whether industrialization affected FLFPR differently
in rural versus urban areas of Bangladesh and neighboring Indian states. In all
specifications, educational attainment and reproductive behavior variables are con-
trolled for. In the specifications investigating the rural and urban FLFPR changes,
we use ratio of F/M of working age population as a measure of competition from
men in the rural labor market; and use changes in share of men working in indus-
try in the urban area of the district as proxy for industrialization and men’s “pull
factor”.
Standard errors in decomposition are estimated non-parametric bootstrap
methods from Efron (1979).
6The Bangladeshi data does not allow us to construct reliable completed fertility
rates.
79
4.3.2 Data Sources
Data for the outcome and control variables are obtained by aggregating micro-
data to district levels.7 The Bangladeshi micro-data comes from the Census 1991,
2001, and 2011 sub-samples from Minnesota Population Center (2020). Micro-data
for Indian states neighboring Bangladesh comes from the Socio-Economic Survey,
Household Schedule 10: Employment and Unemployment conducted by the Na-
tional Sample Survey Organization, Government of India from Minnesota Popula-
tion Center (2023); referred to as NSS surveys hereafter. The 1987, 1991, 2004, and
2009 waves of the NSS surveys are used since they have district identifiers.
In the case of Bangladesh, estimates of outcome and control variables are ob-
tained for all 60 districts with harmonized boundaries for 1991, 2001, and 2011. As
its Indian counterpart, Indian states of Mizoram, Tripura, Meghalaya, Assam, and
West Bengal are chosen as they share geographic boundaries, geographic similar-
ities, cultural and historical similarities with Bangladesh. Estimates of outcome
and control variables are obtained for 42 districts from the five Indian states for
1987, 1991, 2004, and 2009. The NSS surveys have a much richer occupational clas-
sification. The analysis is limited to broader categories of Agriculture, Industry,
and Service as these are the only occupational classifications available in the 2011
Bangladeshi census. Appendix B describes the construction of key FLFPR and
education variables.
Given the variation in lengths of periods between the Bangladeshi census and
the Indian NSS surveys, and the variation in lengths of periods within different
waves of Indian NSS surveys, differences in outcome and control variables are ex-
pressed in yearly changes throughout the analysis unless stated otherwise.
4.3.3 Descriptive Statistics
Table 4.1 and Table 4.2 present the gender-specific overall, industrial, and
agricultural labor force participation rate.8 Towards the beginning of the study pe-
7Aggregation to a finer geographic unit is possible for Bangladesh, but not for
the Indian states neighboring Bangladesh.
8The overall FLFPR measure differs significantly from ILO-based estimates
presented in 4.1. For example, the 1991 Bangladeshi FLFPR was 6.61 percent in
the census-based estimate, whereas ILO estimate obtained from The World Bank
80
riod, the male labor force participation rate (MLFPR) were similar in Bangladesh
(88.05% in 1991) and neighboring Indian States (86.38% in 1987). That similarity
has persisted towards the end as well (83.26% in 2011 in Bangladesh, 85.98% in
2009 in neighboring Indian states). Looking at finer details, the industrial MLFPR
was lower in Bangladesh, and has been increasing. The industrial MLFPR in neigh-
boring Indian states also increased over similar periods. Both regions experienced
a decline in MLFPR in agriculture. This suggests a shift of male workers from the
agricultural sector to the industrial sector.
The FLFPR in Bangladesh was one of the lowest in the world in 1991 (6.61%),
and much lower than neighboring Indian states in 1987 (17.19%). Roughly over the
next two decades, Bangladesh gained 2.91 percentage point (pp) in FLFP, while
neighboring Indian states experienced a decline of 0.65 pp. Similarly, the industrial
FLFPR increased in Bangladesh whereas it reduced in neighboring Indian states.
Since both FLFPR and agricultural-FLFPR have the same base in my
estimates, the importance of agriculture can be measured by (∆FLFPR :
Agr)/(∆FLFPR). Using this metric, roughly 60% of the increase in FLFPR in
Indian states between 1987-2004 were increases in FLFPR in agriculture. In later
years, 55% of the decline afterwards are accounted for by the decline in agricultural
FLFPR. In Bangladesh, almost 100% of the increase in FLFPR between 1991-2001
is due to increase in agricultural FLFPR. However, between 2001-2011 agriculture
becomes much less important, losing about 4 percentage point (pp) while FLFPR
declined by about 2 percentage point.
Table 4.3 and Table 4.4 shows the levels and changes in schooling, children
at home, share of population aged 15-39, and marriage rates of women. Women
in India appear to have higher education, and remain more educated in our sam-
ple period. The fertility decline is perhaps stronger in Bangladesh. The share of
population 15-39 year old was lower in Bangladesh but rapidly catching up. Both
geographic units have similar F/M ratio and the changes in F/M are negligible
considering the regions as a whole. Moreover, Bangladesh has urbanized slightly
more as well.
(2021) puts it closer to 25 percent. There are two key sources of this difference.
First, given the data availability, we do not make adjustments for factors such as in-
stitutionalized persons. Second, ILO estimates statistically re-adjusts the numbers
to include proportion of women who are looking for work at much longer horizons
than what the census asks.
81
Taken together, the descriptive statistics suggest that periods of increases in
FLFPR and agricultural-FLFPR seem to coincide with declining male participation
in agriculture in both the regions. In addition to examining the influence of avail-
able covariates discussed in 4.3.1, we also investigate whether male-out migration
pattern of India explains the concentrated relationship between F/M and FLFPR
in India.
4.4 Empirical Findings
4.4.1 Cross-sectional Correlates
Table 4.5 presents results of regressions of available covariates on FLFPR
in Bangladesh in 1991 and 2011, and in the selected Indian states in 1987 and
2009. Overall, the regressions do not reveal any strong and systematic correlation
between FLFPR and the covariates in Bangladesh or Indian states neighboring
Bangladesh. It is interesting to note that the correlations do not change systemati-
cally despite liberalization of both economies in the early 1990s (Dev, 2000).
In the Bangladeshi data, district level increases in urbanization rate appear to
become a statistically and economically relevant correlate in 2011. On the other
hand, years of schooling (F), share of 15-39 year old in population, share of women
with young children appear to be at home, and local labor market conditions as
measured by MLFPR, appears to be uncorrelated with FLFPR. However, F/M is
negatively correlated with FLFPR in Bangladeshi districts in both 1991 and 2011.
In the data for selected Indian states, urbanization rate, share of 15-39 year
olds in population and F/M is correlated with the FLFPR in both 1987 and 2009.
However, years of schooling and share of women with under 5 children at home are
positively correlated in the 2009 data, and MLFPR is weakly negatively correlated
in the 2009 data.
Taken together, the simple correlations point to the possibility that differences
in coefficients are likely important, suggesting different underlying process related
to evolution of FLFPR in Bangladesh and neighboring Indian states.
4.4.2 Decomposition results: Human capital and reproductive behavior
Table 4.6 presents results of Decomposition 1 – decompositions using only edu-
cation and age structure variables . The left-hand side is ∆FLFPRB −∆FPFPRI ,
82
i.e. the differences in trajectories of FLFPR between Bangladesh and neighboring
Indian states. Decomposition analysis 1 shows that differences in trajectories of fe-
male and male years of schooling are an important source of divergence in women’s
employment in the two regions after controlling for changing age structure.
In specification 1, both men’s and women’s years of education is used as rele-
vant human capital variables. While differences in covariates are statistically impor-
tant, the effect is dominated by differences in coefficients (see Figure 4.3). Individu-
ally, none of the differences in age structure, and gender specific schooling changes
has independent explanatory power. However, at the mean levels of changes in
India, the coefficients as a whole and individually are statistically significant. Par-
ticularly, increases in women’s education reduces the gap between the countries
whereas increases in share of 15-39 year old and men’s education opened up the gap
in FLFPR rate trajectories.
In specification 2, we include changes in the ratio of women’s to men’s edu-
cation. The relative importance of endowment differences increase substantially.
Moreover, figure 4.4 shows that differences increased women’s education relative
to men has a statistically significant explanatory power. When coefficients are
considered, the estimates are nosier, likely due to correlation between changes in
men’s and women’s education and their relative education. However, the sign of the
effect remains the same compared to specification 1. Together, results, of specifica-
tion 1 and 2 suggests that changes in women’s in education had different effects on
FLFPR in Bangladesh and in its neighboring Indian states, and that these differ-
ences are moderated by differences in women’s education relative to men.
Specification 3 includes share of women that are married, and share of women
with under five children in specification 1. Taken together, the results in Table 4.5
and Figure 4.5 does not change any of the qualitative findings from specification
1. Differences in trajectories of marriage rates, and rates of having under 5 chil-
dren does not add to the explanation of the diverging trajectory of FLFPRs in
Bangladesh and selected Indian states.
4.4.3 Decomposition results: Structural Change
Decomposition 3 is aimed investigating the role of structural change in explain-
ing the divergence of FLFPR trajectories in Bangladesh and selected Indian states.
Specification 4 focuses on both rural and urban data combined, and uses changes in
83
urbanization rate, and changes in share of men working in industry in the district
as proxy for industrialization. Since the estimating regression covers the entirety of
both regions, and includes covariates of all sources of divergence we are considering,
this specification is our baseline.
Table 4.8 and Figure 4.6 presets results for specification 4. First, the effect
of differences in coefficients is statistically significant, whereas effect of covariate
differences is not. Moreover, neither of the structural change variable appear to be
statistically important. Both the coefficient and endowment term of the share of
men in industrial activity seem to be positive, but not statistically significant.
In both regions, industrial work accounts for very little of employment and
agriculture is the dominant source of employment. Additionally, the descriptive
statistics point towards a faster rate of industrialization in Bangladesh in this pe-
riod of time. Hence, in specification 5, we test a mechanism where more men leave
the rural areas to gain employment in urban industrial sector of the same district,
thereby making it easier for women to enter the rural agricultural workforce.9 We
do so by including changes in share of men in industry in the urban region of the
district as a control, and using F/M ratio as a measure of relative abundance of
working age women relative to men in a district.
Table 4.8 and Figure 4.7 presents results of specification 5. The results indicate
a similar picture where coefficient effects are dominant, particularly for education
related variables. However, increases in F/M appear to be a positive, but statisti-
cally insignificant contributor to FLFPR in rural areas of Bangladesh.
Specification 6 results are presented in 4.8 and 4.8, and shows that the models
do not have a strong explanatory power in explaining the relatively minor diver-
gence in urban FLFPR trajectories between Bangladesh and selected Indian states.
The differences in trajectory of FLFPR in Bangladesh and its neighboring In-
dian states in the decades of 1990s and 2000s appear to be driven by how changes
in education relates to FLFPR in those regions. Increases in women’s education
affected Bangladeshi FLFPR positively, but it affects FLFPR negatively in the se-
lected Indian states. Results from specification 2 suggests that the greater increase
in relative education of women may have made it easier for Bangladeshi FLFPR to
9This is a limitation since industrial employment is typically concentrated in
Dhaka-Chittagong corridor of Bangladesh. Industrialization rate of a few districts
may be more relevant in the Bangladeshi case.
84
start catching up to the selected Indian states. Additionally, there is very weak evi-
dence that industrialization helped increase FLFPR relatively more in Bangladesh,
particularly in farm-employment by nicreasing relative abundance of women in
rural areas of Bangladesh.
The results for Indian states, despite being surprising from a theoretical point
of view, are in line with the FLFPR literature of rural India. Afridi et al. (2018)
used decomposition analysis to show that between 1987-1999, almost all of the
reduction in FLFPR of rural married women in India is attributable to changes
in household and demographic characteristics. Similarly, most of the reduction in
FLFPR in 1999-2011 is also attributable to the same changes. Afridi et al. (2018)
posits that these changes resulted from an increased education of rural women that
increased their relative returns to home production. Unfortunately, the previous
literature for Bangladesh is limited.
In a limited capacity, the findings are also consistent with Tumbe (2015)
who provide evidence that men’s rural-urban out-migration for employment lead
to women’s entry in the labor force in India. The Bangladeshi findings are also
broadly consistent with the descriptive statistics showing an waning importance of
agriculture in Bangladeshi women’s employment, and with related literature docu-
menting the emergence of a large and critical “women employing" garment sector
(see chapter 2 and Heath and Mobarak (2015)).
4.5 Discussion and Conclusion
Bangladesh, and Indian states neighboring Bangladesh – Mizoram, Tripura,
Meghalaya, Assam, and West Bengal; experienced rapid increases in FLFPR in
1990s. Between 1991-2011, Bangladeshi FLFPR increased by about 4.56 percentage
point. Between 1987-1999, the number for the selected Indian sates is 5.04 (see
Table 4.1. Most of these increases were from participation into agriculture. Since
then, Bangladeshi women has increased their industrial FLFPR despite a drop in
overall and agricultural FLFPR. In the selected neighboring states of India, the
drop was not accompanied by increased in FLFPR.
This paper attempts to identify the drivers of this divergence by using Oaxaca-
Blinder decompositions. The results paint a picture of two similar regions with
different underlying process of changes in FLFPR. When influence of schooling,
reproductive behavior, and structural change is considered, the results indicate
85
that differences in women’s education relative to men’s is associated with increased
FLFPR in Bangladesh relative to the selected Indian states. Reductions in pro-
marriage and pro-natal behavior does not appear to be important.
The decomposition analysis also focused on one particular channel. Tumbe
(2015) shows the importance of out-migration of men in engendering FLFPR agri-
culture in India. We included changes in F/M , the only available proxy to capture
effect of male out-migration, in the analysis. We find that differences in trajectories
of ∆(F/M) and associated coefficients are weakly associated with the diverging
FLFPR between Bangladesh and its neighboring Indian states in the cases of over-
all FLFPR, agricultural-FLFPR in rural areas, but not in the case of industrial
FLFPR.
When placing this findings in context of the overall literature the key conclu-
sion is that the underlying processes of changes in FLFPR in Bangladesh between
1991-2011 and the selected Indian states between 1987-2009 are different. Increases
in FLFPR is positively associated with women’s education in Bangladesh, but not
in the case of Indian states. This is likely due to an increase in the relative level of
education of women compared to men in Bangladesh. On the other hand, FLFPR
is unrelated with reproductive choices. These findings are in-line with agriculture
being the primary employer of women in both regions. It appears that although
the initial increase in women’s employment is due to agriculture, faster changes in
non-farm employment of men in urban areas of Bangladesh may have played a role
as some men exited the rural labor market.
86
4.6 Figures
Figure 4.1. Female Labor Force Participation in Bangladesh and Elsewhere
(1990:2019)
Note: Bangladesh had one of the lowest FLFPR rate in 1990 among developing
regions. While most other developing regions saw its FLFPR decline over 1990-
2019, Bangladesh’s FLFPR increased. Notably, FLFPR of Bangladesh’s neighbors
in South Asia decreased overall. Data source: (The World Bank, 2021) .
87
Figure 4.2. Female Labor Force Participation in South Asia (1990:2019)
Note: Within South Asia, countries with a high level of FLFPR in 1990 experi-
enced a decline, and the reverse was true for countries with relatively lower FLFPR
in 1990, including Bangladesh. It is also interesting to note that Bangladeshi
FLFPR was the highest in Muslim-majority South Asian countries in 1990, and in
keeping in line with other Muslim-majority countries in South Asia, experienced an
increase in FLFPR since 1991. Data source: (The World Bank, 2021) .
88
Figure 4.3. Potential Sources of Divergence in FLFPR Trajectory of Bangladesh
and Neighboring Indian States (1987-2011): Education
89
Figure 4.4. Potential Sources of Divergence in FLFPR Trajectory of Bangladesh
and Neighboring Indian States (1987-2011): Relative Education of Women
90
Figure 4.5. Potential Sources of Divergence in FLFPR Trajectory of Bangladesh
and Neighboring Indian States (1987-2011): Education and Reproductive Behavior
91
Figure 4.6. Potential Sources of Divergence in FLFPR Trajectory of Bangladesh
and Neighboring Indian States (1987-2011): Education, Reproductive Behavior,
and Structural Change
92
Figure 4.7. Potential Sources of Divergence in Rural FLFPR Trajectory of
Bangladesh and Neighboring Indian States (1987-2011): Education, Reproduc-
tive Behavior, and Structural Change
93
Figure 4.8. Potential Sources of Divergence in Urban FLFPR Trajectory of
Bangladesh and Neighboring Indian States (1987-2011): Education, Reproduc-
tive Behavior, and Structural Change
94
95
4.7 Tables
Table 4.1. Labor Force Participation (Overall and Industry-wise) in Bangladesh and Selected Indian States
Geographic Unit Year FLFPR FLFPR-Ind FLFPR-Agr MFLPR MFLPR-Ind MFLPR-Agr
Bangladesh 1991 6.61 0.59 1.16 88.05 4.13 49.09
Bangladesh 2001 11.17 0.75 5.74 77.98 4.35 39.25
Bangladesh 2011 9.02 2.54 1.3 83.26 10.59 41.02
Indian States1 1987 17.19 3.57 7.65 86.38 13.4 45.38
Indian States 1999 22.23 4.64 10.28 86.22 12.49 41.32
Indian States 2004 24.79 5.69 12.08 86.61 13.97 40.93
Indian States 2009 16.54 3.41 7.52 85.98 16.49 40.68
1Mizoram, Tripura, Meghalaya, Assam, and West Bengal
Table 4.2. Yearly Changes in Labor Force Participation (Overall and Industry-wise) Rates in Bangladesh and Selected
Indian States in PP terms
Geographic Unit Period ∆FLFPR ∆FLFPR-Agr ∆FLFPR-Ind ∆MFLPR ∆MLFPR-Ind ∆MLFPR-Agr
Bangladesh 1991-2001 0.44 0.46 0.01 -1 0.01 -0.90
Bangladesh 2001-2011 -0.24 -0.44 0.17 0.53 0.60 0.26
Indian States1 1987-1999 0.42 0.22 0.08 -0.03 -0.02 -0.45
Indian States 1999-2004 0.66 0.32 0.25 0.09 0.28 -0.16
Indian States 2004-2009 -1.69 -1.01 -0.39 -0.11 0.57 -0.13
1Mizoram, Tripura, Meghalaya, Assam, and West Bengal
96
Table 4.3. Demographic Conditions in Bangladesh and Selected Indian States
Geographic Unit Year Yrs of School (F) F/M Ratio % With U-5 child %15-39 Years Old % Married (F) % Urban
Bangladesh 1991 1.83 0.94 41.86 37.02 80.06 19.46
Bangladesh 2001 3.25 0.94 32.86 40.79 76.86 23.51
Bangladesh 2011 4.38 1 26.9 41.46 79.46 23.32
Indian States1 1987 2.87 0.92 31.89 41.48 71.28 20.6
Indian States 1999 4.12 0.94 26.08 42.42 73.41 19.23
Indian States 2004 4.63 0.94 23.63 42.89 74.33 21.17
Indian States 2009 5.5 0.91 19.81 44.21 74.14 20.9
1Mizoram, Tripura, Meghalaya, Assam, and West Bengal
97
Table 4.4. Yearly Changes in Demographic Conditions in Bangladesh and Selected Indian States
Geographic Year Years of F/M % with % 15-39 % %
Unit School (F) Ratio U-5 child (pp) Years Old Married (F) (pp) Urban (pp)
Bangladesh 1991-2001 0.1391 0.0001 -0.8875 0.0036 -0.3109 0.2838
Bangladesh 2001-2011 0.1101 0.0061 -0.5847 4e-04 0.2825 -0.1749
Indian States1 1987-1999 0.1135 0.0017 -0.5041 8e-04 0.1645 0.0431
Indian States 1999-2004 0.0885 0.001 -0.3644 7e-04 0.2336 0.2726
Indian States 2004-2009 0.17 -0.0064 -0.7482 0.0026 0.0124 0.0586
1Mizoram, Tripura, Meghalaya, Assam, and West Bengal
98
Table 4.5. Cross-sectional Correlates of FLFPR in Districts of Bangladesh and Selected Indian States1
Dependent variable:
FLFP
Bangladesh 1991 Bangladesh 2011 India 1987 India 2009
(1) (2) (3) (4)
% of Population in Urban Areas 0.022 0.244∗∗∗ −0.262 −0.086
(0.037) (0.063) (0.262) (0.137)
Avg. Years of Schooling (F) −0.002 −0.0003 0.043 0.061∗∗∗
(0.009) (0.009) (0.040) (0.014)
% 15-39 years Old −0.119 −0.112 −0.877 0.333
(0.213) (0.289) (1.771) (0.693)
% With U-5 Child 0.010 0.216 −0.097 1.337∗∗∗
(0.118) (0.193) (0.796) (0.477)
MLFPR −0.051 −0.137 0.372 −1.695∗∗
(0.174) (0.211) (1.372) (0.674)
F/M Ratio −0.461∗∗∗ −0.503∗∗∗ 0.489 −0.158
(0.101) (0.148) (0.534) (0.231)
Observations 60 60 37 37
R2 0.548 0.668 0.090 0.603
Adjusted R2 0.497 0.631 −0.092 0.524
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Notes: No strong correlation of the observed covariates and FLFP. The one exception is the F/M ratio.
99
Table 4.6. Influence of Human Capital Endowment and Coefficient Differences in Explaining the Divergent Trajectory of
FLFPR in Bangladesh and Neighboring Indian States
Specification 1 Specification 2
Yearly Endowment 0.228*** 0.725***
∆(∆FLFPR) SE (Endowment) (0.114) (0.174)
= 0.370 Coefficient 0.708*** 0.408∗∗∗
in pp terms SE (Coefficient) (0.305) (0.280)
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Specification 1 uses changes in male and female years of schooling, and share of population 15-39 as the explanatory
variables.
Specification 2 adds changes in female years of schooling over male years of schooling as an additional variable.
Table 4.7. Influence of Reproductive Behavior and Human Capital Endowment and Coefficient Differences in Explaining
the Divergent Trajectory of FLFPR in Bangladesh and Neighboring Indian States
Specification 3
Yearly Endowment 0.153
∆(∆FLFPR) SE (Endowment) (0.134)
= 0.370 Coefficient 0.772***
in pp terms SE (Coefficient) (0.297)
Note:
∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Specification 3 includes share of women married, and share of women with under five children in specification 1
100
Table 4.8. Influence of Differences in Urbanization and Industrialization in Explaining the Divergent Trajectory of Key
FLFPR Sub-categories in Bangladesh and Neighboring Indian States
Specification 4 Specification 5 Specification 6
Yearly ∆(∆FLFP) Yearly ∆(∆FLFP:Agr) Yearly ∆(∆FLFP:Ind)
0.370 pp 0.364 pp 0.076 pp
Endowment 0.147 0.172 -0.009
SE (Endowment) (0.148) (0.174) (0.045)
Coefficient 0.776∗∗∗ 0.939∗∗∗ 0.073
SE (Coefficient) (0.277) (0.308) (0.066)
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Specification 4 includes changes in urbanization rate, and changes in share of
men working in industry in the district as proxy for industrialization.
Specification 5 focuses on agricultural FLFP in rural areas. In addition to vari-
ables in specification 3, it also includes changes in share of men working in industry
in the urban area of the district as proxy for industrialization and men’s “pull
factor”, and changes in ratio of F/M as a measure of competition from men in rural
labor market.
Specification 6 focuses on industrial FLFP in urban areas. In addition to vari-
ables in specification 3, it also includes changes in share of men working in industry
in the urban area of the district as proxy for industrialization and men’s “pull
factor”, as a measure of competition from men in the labor market.
CHAPTER 5
DISSERTATION CONCLUSION
Bangladesh has been undergoing a rapid fertility and FLFP transition since
the 1970s and 1990s respectively. The three chapters in this dissertation aims to
add to our understanding of the changes in fertility and women’s employment
through the lens of structural changes through industrialization and evolving ru-
ral economic conditions.
In chapter two, I examine the effect of women-favoring industrialization on
women’s employment, reproductive, and human capital accumulation behavior.
I ameliorate some of the concerns related to unobserved location characteristics
that may influence the outcome variables by exploiting a combination of spatial
variation in RMG product specialization across regions, and temporal variation in
product exports to estimate the FLFP, fertility and education response to female
labor demand shocks. I find evidence of increases in FLFP, and a small reduction
in fertility. However, I find no effect on school enrollment of girls.
In chapter three, I analyze the spatiotemporal variation in the fertility transi-
tion in Bangladesh. I begin with a systematic documentation of the fertility transi-
tion in different regions of Bangladesh since 1950s using birth histories. I document
that the fertility transition begun in urban areas with the 1941-45 birth cohort,
and soon in rural areas in the 1945-51 birth cohort. Descriptive analysis suggests
that women’s education and agricultural land constraint could have been important
drivers of the transition. Using a difference in difference approach, I find that rural
land constraints have not directly affected fertility transitions, but may have played
an important role through income effect of increased labor supply in rural labor
market.
The analysis in Chapter four suggests that in India, increased women’s edu-
cation reduces women’s economic activity but the reverse is true in Bangladesh. I
present evidence weakly suggesting that women’s education relative to men’s have
increased more in Bangladesh, contributing to the difference in women’s employ-
ment trajectory.
Taken together, this dissertation finds evidence in favor of parts of mechanism
of industrialization leading to fertility and FLFP transition. Keeping in line with
the literature, differences in FLFPR trajectory appear to be very context depended.
In addition, it also finds limited evidence suggesting the role of rural economic
101
conditions through land availability as an originator of rural transition.
102
103
APPENDIX A
COMPLETED FERTILITY ESTIMATES
104
105
APPENDIX B
INDUSTRY CODE HARMONIZATION
For Indian data, occupational code 10 (Agriculture, fishing, and forestry) is
mapped to Agriculture. Codes 20 (Mining and extraction), 30 (Manufacturing),
40 (Electricity, gas, water and waste management), 50 (Construction), 130 (Other
industry, n.e.c.) are classified as Industry. Codes 60 (Wholesale and retail trade),
70 (Hotels and restaurants), 80 (Transportation, storage, and communications), 90
(Financial services and insurance), 100 (Public administration and defense), 110
(Services, not specified), 111 (Business services and real estate), 112 (Education),
113 (Health and Social Work), 114 (Other Services), 120 (Private household ser-
vices) are classified as services.
For Bangladeshi data, codes 04 (Agriculture) in 1991 and 2011 are mapped to
Agriculture. Codes 05 (Industry), 06 (Water/electricity/gas), and 07 (Construction)
in 1991 and 2011 are mapped to Industry. Codes 08 (Transport/communication),
09 (Hotel/restaurant), 10 (Business), and 11 (Service) are classified as Service in
1991 and 2011.
106
REFERENCES CITED
Aaronson, D., Dehejia, R., Jordan, A., Pop-Eleches, C., Samii, C., & Schulze, K.,
The Effect of Fertility on Mothers’ Labor Supply over the Last Two Cen-
turies (en), The Economic Journal 131 (2021), no. 633, 1–32, https://doi.
org/10.1093/ej/ueaa100
Aaronson, D., Lange, F., & Mazumder, B., Fertility Transitions Along the Exten-
sive and Intensive Margins, American Economic Review 104 (2014), no. 11.
Afridi, F., Dinkelman, T., & Mahajan, K., Why Are Fewer Married Women Join-
ing the Work Force in Rural India? A Decomposition Analysis over Two
Decades, Journal of Population Economics 31 (2018), no. 3, https://doi.org/
10.1007/s00148-017-0671-y
Aguero, J. M., & Marks, M. S., Motherhood and Female Labor Force Participation:
Evidence from Infertility Shocks, American Economic Review 98 (2008),
no. 2, 500–504, https://doi.org/10.1257/aer.98.2.500
Alesina, A., Giuliano, P., & Nunn, N., Fertility and the Plough. American Economic
Review 101 (2011), no. 3, 499–503.
, On the Origins of Gender Roles: Women and the Plough, Quarterly Jour-
nal of Economics 128 (2013), no. 2, 469–530.
Amin, R., Choudhuri, S. R., Mariam, A. G., & McCarthy, J., Family Planning
in Bangladesh, 1969-1983, International Family Planning Perspectives
13 (1987), no. 1, 16–20, https : / /doi . org / 10 . 2307 /2948103 Publisher:
Guttmacher Institute.
Amin, S., Diamond, I., Naved, R. T., & Newby, M., Transition to Adulthood of Fe-
male Garment-Factory Workers in Bangladesh, Studies in Family Planning
29 (1998), no. 2, 185–200.
Atkin, D., Endogenous Skill Acquisition and Export Manufacturing in Mexico, The
American Economic Review 106 (2016), no. 8, 2046–2085, Publisher: Ameri-
can Economic Association.
Autor, D., Dorn, D., & Hanson, G., When Work Disappears: Manufacturing De-
cline and the Falling Marriage Market Value of Young Men, AER:Insights 1
(2019), no. 2, 161–178.
107
Autor, D., Dorn, D., & Hanson, G. H., The China Syndrome: Local Labor Market
Effects of Import Competition in the United States, American Economic
Review 103 (2013), no. 6, 2121–68, https://doi.org/10.1257/aer.103.6.2121
Bangladesh Burea of Statistics, (1984), Analytical findings and national tables 1981
(tech. rep.), Bangladesh Burea of Statistics.
, (2020), Statistical pocketbook 2019 (tech. rep.), Bangladesh Burea of Statis-
tics.
Baudin, T., A role for cultural Transmission in fertility Transitions, Macroeco-
nomic Dynamics 14 (2010), no. 4.
BBC, Chinese factories turn to Bangladesh as labour costs rise (en-GB), BBC
News (2012).
BGMEA, About Garment Industry of Bangladesh, retrieved October 16, 2022, from
https://www.bgmea.com.bd/page/AboutGarmentsIndustry 2022.
, Export Performance, retrieved October 16, 2022, from https://www.bgmea.
com.bd/page/export-performance-list 2022.
Bhattacharya, J., & Chakraborty, S., Fertility choice under child mortality and
social norms, Econom. Lett. 115 (2012), no. 3, 338–341.
, Contraception and the Demographic Transition. Economic Journal 127
(2017), no. 606, 2263–2301.
Blinder, A. S., Wage Discrimination: Reduced Form and Structural Estimates, Jour-
nal of Human Resources 8 (1973), no. 4, 436–455.
Boehnke, J., & Gay, V., The Missing Men, Journal of Human Resources 57 (2022),
no. 4, 1209–1241, https : / /doi . org / 10 . 3368 / jhr . 57 . 4 . 0419 - 10151R1
https://jhr.uwpress.org/content/57/4/1209.full.pdf.
Borusyak, K., Hull, P., & Jaravel, X., Quasi-Experimental Shift-Share Research
Designs, The Review of Economic Studies 89 (2022), no. 1, 181–213, https:
//doi.org/10.1093/restud/rdab030
Burnette, J., Women Workers in the British Industrial Revolution, 2008.
Caldwell, J. C., Barkat-e-Khuda, Caldwell, B., Pieris, I., & Caldwell, P., The
Bangladesh Fertility Decline: An Interpretation (en), Population and De-
108
velopment Review 25 (1999), no. 1, 67–84, https://doi.org/10.1111/j.1728-
4457.1999.00067.x
Carranza, E., Soil Endowments, Female Labor Force Participation, and the Demo-
graphic Deficit of Women in India, https://doi.org/10.1257/app.6.4.197
2014.
Chaudhury, R. H., Attitude of Some Elites Towards Introduction of Abortion as a
Method of Family Planning in Bangladesh (en), The Bangladesh Develop-
ment Studies 3 (1975), no. 3, 479–494.
, Female Status and Fertility Behaviour in a Metropolitan Urban Area of
Bangladesh. Population Studies 32 (1978), no. 2, 261–273.
Chowdhury, A. K. M. A., Khan, A. R., & Chen, L. C., The Effect of Child Mortal-
ity Experience on Subsequent Fertility: in Pakistan and Bangladesh, Popula-
tion Studies 30 (1976), no. 2, 249–261, https://doi.org/10.2307/2173608
Publisher: [Population Investigation Committee, Taylor & Francis, Ltd.]
Datta, A., Pakistan–Bangladesh Partition 1971 and forced migration (en), In
The Encyclopedia of Global Human Migration, John Wiley & Sons, Ltd,
2013, https : / / doi . org / 10 . 1002 / 9781444351071 .wbeghm586 _eprint:
https://onlinelibrary.wiley.com/doi/pdf/10.1002/9781444351071.wbeghm586.
DeGraff, D. S., Increasing Contraceptive use in Bangladesh: The Role of Demand
and Supply Factors (en), Demography 28 (1991), no. 1, 65–81.
Dev, S. M., Economic Liberalisation and Employment in South Asia: II, Economic
and Political Weekly 35 (2000), no. 3, 135–146.
Dinkelman, T., & Ngai, L. R., Time Use and Gender in Africa in Times of Struc-
tural Transformation, Journal of Economic Perspectives 36 (2022), no. 1,
57–80, https://doi.org/10.1257/jep.36.1.57
Doepke, M., Tertilt, M., & Voena, A., The Economics and Politics of Women’s
Rights. Annual Review of Economics 4 (2012), no. 1, 339–372.
Efron, B., Bootstrap Methods: Another Look at the Jackknife, Ann. Statist. 7
(1979), no. 1, 1–26, https://doi.org/10.1214/aos/1176344552
Fernández, R., Cultural Change as Learning: The Evolution of Female Labor Force
Participation over a Century, American Economic Review 103 (2013), no. 1,
472–500, https://doi.org/10.1257/aer.103.1.472
109
Field-Hendrey, E., The Role of Gender in Biased Technical Change: U.S. Manu-
facturing, 1850- 1919, The Journal of Economic History 58 (1998), no. 4,
1090–1109, Publisher: [Economic History Association, Cambridge University
Press].
Fogli, A., & Veldkamp, L., Nature or Nurture? Learning and the Geography of
Female Labor Force Participation, Econometrica 79 (2011), no. 4, 1103–
1138, https : / / doi . org / https : / / doi . org / 10 . 3982 / ECTA7767
https://onlinelibrary.wiley.com/doi/pdf/10.3982/ECTA7767.
Fredriksson, P. G., & Gupta, S. K., Irrigation and gender roles, Journal of Develop-
ment Economics 163 (2023), 103076, https://doi.org/https://doi.org/10.
1016/j.jdeveco.2023.103076
Fujii, T., & Shonchoy, A. S., Fertility and rural electrification in Bangladesh (en),
Journal of Development Economics 143 (2020), 102430, https://doi.org/10.
1016/j.jdeveco.2019.102430
Gaddis, I., & Klasen, S., Economic Development, Structural Change, and Women’s
Labor Force Participation: A Reexamination of the Feminization U Hypoth-
esis, Journal of Population Economics 27 (2014), no. 3, 639–681, https :
//doi.org/10.1007/s00148-013-0488-2
Goldin, C., The U-Shaped Female Labor Force Function in Economic Development
and Economic History, Investment in women’s human capital and economic
development (T. P. Schultz, [Ed.]; 61–90), University of Chicago Press, 1995,
61–90.
Goldin, C., & Olivetti, C., Shocking Labor Supply: A Reassessment of the Role of
World War II on Women’s Labor Supply, American Economic Review 103
(2013), no. 3, 257–62, https://doi.org/10.1257/aer.103.3.257
Goldsmith-Pinkham, P., Sorkin, I., & Swift, H., Bartik Instruments: What, When,
Why, and How (en), American Economic Review 110 (2020), no. 8, 2586–
2624, https://doi.org/10.1257/aer.20181047
Greenwood, J., Guner, N., & Vandenbroucke, G., Family Economics Writ Large,
Journal of Economic Literature 55 (2017), no. 4, 1346–1434.
Hahn, Y., Islam, A., Nuzhat, K., Smyth, R., & Yang, H.-S., Education, Mar-
riage, and Fertility: Long-Term Evidence from a Female Stipend Program
in Bangladesh (en), Economic Development and Cultural Change 66 (2018),
no. 2, 383–415, https://doi.org/10.1086/694930
110
Hansen, C. W., Jensen, P. S., & Skovsgaard, C. V., Modern Gender Roles and
Agricultural History: The Neolithic Inheritance, Journal of Economic Growth
20 (2015), no. 4, 365–404, https://doi.org/10.1007/s10887-015-9119-y
Heath, R., & Mobarak, M. A., Manufacturing Growth and the Lives of Bangladeshi
Women, Journal of Development Economics 115 (2015), 1–15.
Hlavac, M., (2022), Oaxaca: Blinder-oaxaca decomposition in r (tech. rep.), R pack-
age version 0.1.5.
Hossain, M. B., Phillips, J. F., & Legrand, T. K., THE IMPACT OF CHILD-
HOOD MORTALITY ON FERTILITY IN SIX RURAL THANAS OF
BANGLADESH (en), Demography 44 (2007), no. 4, 15.
ICF International, (2021), Demographic and health surveys (various) [bangladesh:
Standard dhs 1990-00, 2004, 2007, 2011, 2014, 2017-18]. ICF International.
Internaltional Statistical Insititute, (1976), World fertility survey: Bangladesh, 1975-
76 (wfs).
Joshi, S., & Schultz, T. P., Family Planning and Women’s and Children’s Health:
Long-Term Consequences of an Outreach Program in Matlab, Bangladesh
(en), Demography 50 (2013), no. 1, 149–180, https ://doi .org/10.1007/
s13524-012-0172-2
Kabeer, N., & Mahmud, S., Globalization, gender and poverty: Bangladeshi women
workers in export and local markets, Journal of International Development
16 (2004), no. 1, 93–109, https://doi.org/https://doi.org/10.1002/jid.1065
https://onlinelibrary.wiley.com/doi/pdf/10.1002/jid.1065.
Kabir, H., Maple, M., & Usher, K., The impact of COVID-19 on Bangladeshi ready-
made garment (RMG) workers, Journal of Public Health (Oxford, England)
(2020), fdaa126, https://doi.org/10.1093/pubmed/fdaa126
Kagy, G., (2014), Female labor market opportunities, household decisionmaking
power, and domestic violence: Evidence from the bangladesh garment indus-
try (tech. rep.) (Working Paper).
Kucera, D., & Tejani, S., Feminization, Defeminization, and Structural Change in
Manufacturing, World Development 64 (2014), 569–582.
111
Li, B., Export expansion, skill acquisition and industry specialization: evidence from
china, Journal of International Economics 114 (2018), 346–361, https://doi.
org/https://doi.org/10.1016/j.jinteco.2018.07.009
Matsuura, A., & Teng, C., (2020), Issue brief: Understanding the gender compo-
sition and experience of ready-made garment (rmg) workers in bangladesh
(tech. rep.).
Minnesota Population Center, (2020), Integrated public use microdata series, inter-
national: Version 7.3 [bangladesh 1991, 2001, 2011 sub-samples].
, (2023), Socio-economic survey, household schedule 10: Employment and
unemployment [india 1987, 1999, 2004, 2009].
Mithun, M. B., Reasons Behind the Forced Migration of Bangladeshi Hindu Reli-
gious Minorities to India, International Journal on Minority & Group Rights
26 (2019), no. 3, 461–483, https://doi.org/10.1163/15718115-02603002
Publisher: Brill Academic Publishers.
Morgan, S. P., & Hagewen, K. J., Fertility (en), Handbook of Population (D. L.
Poston & M. Micklin, [Eds.]; 229–249), Handbooks of Sociology and Social
Research, Springer US, Boston, MA, 2005, 229–249, https://doi.org/10.1007/
0-387-23106-4_9
Munshi, K., & Myaux, J., Social norms and the fertility transition, Journal of De-
velopment Economics 80 (2006), no. 1, 1–38, https://doi.org/https://doi.
org/10.1016/j.jdeveco.2005.01.002
Oaxaca, R., Male-Female Wage Differentials in Urban Labor Markets, Internat.
Econom. Rev. 14 (1973), no. 3, 693–709.
Oster, E., & Steinberg, B. M., Do IT service centers promote school enrollment?
Evidence from India (en), Journal of Development Economics 104 (2013),
123–135, https://doi.org/10.1016/j.jdeveco.2013.05.006
Rees, R., & Reizman, R., GLOBALIZATION, GENDER, AND GROWTH, Re-
view of Income and Wealth 58 (2012), no. 1, 107–117, https : / / doi .
org / https : / / doi . org / 10 . 1111 / j . 1475 - 4991 . 2011 . 00483 . x
https://onlinelibrary.wiley.com/doi/pdf/10.1111/j.1475-4991.2011.00483.x.
Reihan, S., & Bidisha, S. H., (2018), Female employment stagnation in bangladesh.
report on economic dialogue on inclusive growth in bangladesh (tech. rep.),
Asia Foundation.
112
Roberts, B. J., Yaukey, D., Griffiths, W., Clark, E. W., Shafiullah, A. B. M., &
Huq, R., Family Planning Survey in Dacca, East Pakistan, Demography 2
(1965), 74–96, https://doi.org/10.2307/2060107 Publisher: Springer.
Rodrik, D., Premature deindustrialization (en), Journal of Economic Growth 21
(2016), no. 1, 1–33, https://doi.org/10.1007/s10887-015-9122-3
Rose, E. K., The Rise and Fall of Female Labor Force Participation During World
War II in the United States, The Journal of Economic History 78 (2018),
no. 3, 673–711, https://doi.org/10.1017/S0022050718000323
Schmidt, L., & Elkasabi, M., (2020), Accumulating birth histories across surveys for
improved estimates of child mortality (tech. rep.), ICF, Rockville, Maryland,
USA.
Shastry, G. K., Human Capital Response to Globalization: Education and Infor-
mation Technology in India, Journal of Human Resources 47 (2012), no. 2,
287–330.
Shenk, M. K., Towner, M. C., Kress, H. C., & Alam, N., A model comparison ap-
proach shows stronger support for economic models of fertility decline, Pro-
ceedings of the National Academy of Sciences 110 (2013), no. 20, 8045–8050,
https://doi.org/10.1073/pnas.1217029110 Publisher: Proceedings of the
National Academy of Sciences.
Sinha, N., Fertility, Child Work, and Schooling Consequences of Family Planning
Programs: Evidence from an Experiment in Rural Bangladesh (en), Eco-
nomic Development and Cultural Change 54 (2005), no. 1, 97–128, https:
//doi.org/10.1086/431259
Stoeckel, J., & A. K. M. Alauddin Chowdhury, Fertility and Socio-Economic Sta-
tus in Rural Bangladesh: Differentials and Linkages, Population Studies 34
(1980), no. 3, 519–524, https://doi.org/10.2307/2174808 Publisher: [Popula-
tion Investigation Committee, Taylor & Francis, Ltd.]
Stoeckel, J., & Choudhury, M. A., East Pakistan: Fertility and Family Planning
in Comilla (en), Studies in Family Planning 1 (1969), no. 39, 14–16, https:
//doi.org/10.2307/1964902
Sytsma, T., Improving Preferential Market Access through Rules of Origin: Firm-
Level Evidence from Bangladesh. American Economic Journal: Economic
Policy 14 (2022), no. 1, 440–472.
113
Tejani, S., & Kucera, D., Defeminization, Structural Transformation and Technolog-
ical Upgrading in Manufacturing, Development and Change 52 (2021), no. 3,
533–573.
The World Bank, (2021), World development indicators, The World Bank.
Tumbe, C., Missing men, migration and labour markets: evidence from India, The
Indian Journal of Labour Economics 58 (2015), 245–265, https://doi.org/
https://doi.org/10.1007/s41027-016-0017-4
UN Comtrade, Download trade data: UN Comtrade: International Trade Statistics,
2022.
UN DESA, World Population Prospects 2022, 2022.
World Integrated Trade Solution, Products by Sector exports and imports by coun-
try 2018, 2022.
Zhang, M., Kong, X. X., & Ramu, S. C., The Transformation of the Clothing Indus-
try in China, Asia Pacific Business Review 22 (2016), no. 1, 86–109.
114