Post on 01-Feb-2022
transcript
The Pennsylvania State University
The Graduate School
College of Agricultural Sciences
POVERTY DYNAMICS AND HOUSEHOLD RESPONSE: DISASTER SHOCKS IN RURAL BANGLADESH
A Thesis in
Agricultural, Environmental and Regional Economics and Demography
by
Anuja Jayaraman
© 2006 Anuja Jayaraman
Submitted in Partial Fulfillment of the Requirements
for the Degree of
Doctor of Philosophy
August 2006
The thesis of Anuja Jayaraman was received and approved* by the following Jill. L. Findeis Professor of Agricultural, Environmental and Regional Economics and Demography Thesis Advisor Chair of Committee Carolyn E. Sachs Professor of Rural Sociology and Women’s Studies Gretchen T. Cornwell Research Associate and Assistant Professor of Rural Sociology and Demography Bee-Yan Roberts Professor of Economics Stephen M. Smith Professor of Agricultural and Regional Economics Committee Member and Head of Department of Agricultural Economics and Rural Sociology * Signatures are on file in the Graduate School.
Abstract
South Asia has the largest concentration of the world’s poor, with over half a
billion people surviving on less than a dollar a day. One of the Millennium Development
Goals (MDG) aims to halve the proportion of the world’s people whose income is less
than one dollar a day and the proportion of people who suffer from hunger by the year
2015. The success of poverty alleviation programs in South Asia is critical if this MDG
is to be met. Within South Asia, Bangladesh has the highest incidence of poverty and
only India and China have larger numbers of poor people. It is estimated that nearly half
of Bangladesh’s population of 135 million people live below the poverty line. The
Human Poverty Index reported by the Human Development Report places Bangladesh at
the 86th position among 103 developing countries. Apart from high poverty levels and
low gender empowerment rates, the country also faces yearly natural disasters in the form
of floods. In this dissertation, we first analyze issues relating to chronic and transient
poverty following a major catastrophic event using a short panel of household data from
Bangladesh. Bangladesh experienced the largest floods of the century in 1998. Increase
in private borrowing was one of the medium-term impacts of the floods. The
International Food Policy Research Institute’s Food Management and Research Support
Project (IFPRI-FMRSP) household survey of rural Bangladesh for the years 1998-99 is
used for the analysis. The dissertation attempts to identify the characteristics that
distinguish between those who are able to eventually escape poverty following the flood
(the transient poor) versus those unable to leave poverty (chronic poor). The study uses
cost-of-basic-needs (CBN) poverty lines calculated by the World Bank for Bangladesh
for the year 2000. We use multinomial logit models to asses the determinants of chronic
iii
and transient poverty, comparing them to Bangladeshi households that were never poor.
We also use Censored Quantile Regression models to identify the correlates of each kind
of poverty.
We find that household size, dependency ratio, number of working members, land
ownership, location, social assistance and education characterize the chronically poor.
Ownership of physical and human capital make households less likely to be chronically
poor. Larger household size and dependents in the household push families towards
chronic poverty. Increase in number of working members in the family bring in more
income and reduce the chances of households being chronically poor. Given that
Bangladesh is an agrarian society and faces yearly floods, it is not surprising that
households with heads employed in the trade and self-employment sectors are less likely
to be chronically poor compared to those in the agricultural sector. Long term
investments in human and physical assets clearly help households out of chronic poverty.
Apart from household size, dependency ratio, number of working members, and land
ownership, the transient poor are characterized by their access to credit. Credit access
and remittances explain transient poverty better. Our models are not able to characterize
the transient poor as well as the chronically poor.
After having studied the poor and their characteristics, we seek to study how
individuals interact and operate within a family or household. We asses intra-household
dynamics (e.g., variations in household bargaining behaviors) with a focus on the
household’s expenditure patterns. Receipt of credit is taken as the measure of bargaining
between the head and the spouse. Food and non-food share equations were individually
estimated using random effect OLS and Tobit models to test if participation in credit
iv
markets influences food and non-food expenditure shares. Endogeneity corrections were
incorporated whenever tests indicated an endogenous relationship between total
household consumption and a particular expenditure share. 2SLS models and
simultaneous Tobit models were used to correct for endogeneity between the expenditure
shares and total expenditure. Our results indicate that more men compared to women
participate in the credit market. As is typical of any rural-developing economy,
household expenditure share is highest for food. Amount borrowed by the household
head has effects on food expenditure, adult goods and education expenditure. Amount of
credit taken by the household head negatively affects food expenditure and positively
affects share spent on adult goods. The negative effect on food expenditure has policy
implications related to nutritional intake of children in the household. Women and girls
in the household may also suffer from resultant nutritional deficiencies. Women’s use of
credit has a positive impact on expenditure on children’s goods, durable goods, education
and housing. Results show that resources in the hands of women have implications for
improvement in child outcomes, especially educational outcomes. The positive and
significant impact of spouse’s credit on housing share indicates that resources in the
hands of women also go towards improvement in household and related outcomes. We
also find that households are more likely to spend in round 2 and round 3 than in round 1
on food, education and personal care and more likely to spend on adult goods, children’s
goods, durable goods, fuel, health and housing in round 1 than in round 2 or 3.
v
Table of Contents List of Tables viii List of Figures ix Acknowledgements x 1 Chapter 1: Research Problem 11.1 Introduction 11.2 Poverty and Vulnerability to Shocks 31.3 Shocks in Bangladesh 71.4 Poverty in Bangladesh 81.5 Consumption Smoothing 111.6 Gender and Poverty 131.7 Study Objectives 141.8 Organization of the Dissertation 16 2 Chapter 2: Literature Review and Theoretical Framework 172.1 Introduction 172.2 Poverty Dynamics 182.2.1 Definition of Chronic and Transient Poverty 192.2.2 Measuring Chronic and Transient Poverty 202.2.3 Empirical Studies 222.2.4 Characteristics of the Chronically Poor 252.2.5 Characteristics of the Transient Poor 262.3 Intrahousehold Resource Allocation Models 272.3.1 Unitary Household Model 272.3.2 Agricultural Household Models 302.3.3 Collective Models 332.3.4 Cooperative Model 332.3.5 Cooperative Bargaining Models 362.3.6 Non-Cooperative Model 42 3 Chapter 3: Data and Descriptions 443.1 Country of Study: Bangladesh 443.2 Data Characteristics 503.2.1 Sampling Procedure 533.3 Research Trip to Bangladesh, February 2005 543.4 Data Description 573.4.1 Regional Variations 61 4 Chapter 4: Methods 664.1 Poverty Dynamics 664.2 Measurement of Transient and Chronic Poverty 674.2.1 Approach I 674.2.1.1 Estimation Technique 684.2.2 Approach II 69
vi
4.2.2.1 Estimation Technique 704.2.3 Distinguishing Between Approaches 734.3 Poverty Lines 744.4 Household Expenditure and Credit 754.4.1 Fixed Versus Random Effects 764.4.2 Empirical Framework 774.4.3 Econometric Issues 79 5 Chapter 5: Poverty Dynamics Results 835.1 Introduction 835.2 Descriptive Analysis 845.2.1 Incidence of Poverty 845.2.2. Time-Specific Profile of Poverty 865.3 Stochastic Dominance (First-order) Test 905.4 Receiver Operating Characteristic (ROC) Analysis 915.5 Econometric Analysis: Poverty Dynamics 955.5.1 Methodology I 955.5.1.2 Results from Multivariate Analysis 1025.5.2 Methodology II: Censored Quantile Regression 1075.6 Conclusion 110 6 Chapter 6: Household Expenditure and Credit as
Bargaining Measure 1116.1 Introduction 1116.2 Credit Availability 1126.3 Results 1146.3.1 Descriptive Statistics 1156.3.2 Credit and Household Expenditure 1226.3.3 Amount of Credit and Household Expenditure 1236.3.4 Credit Participation and Household Expenditure 130 7 Chapter 7: Conclusions 1357.1 Introduction 1357.2 Policy Implications 1417.3 Future Research 142 Reference 144
vii
List of Tables
Table 1.1 Types of risks faced by the households 4Table 1.2 Policy instruments for risk reduction 5Table 1.3 Trends in poverty and inequality in the 1990s in Bangladesh 10Table 2.1 Five-tier classification of the poor 21Table 2.2 Studies decomposing the poor into relevant categories 24Table 3.1 Country profile 44Table 3.2 Division profile 48Table 3.3 Timing of the rounds 50Table 3.4 Summary of the content of the household and community-level
questionnaire 52
Table 3.5 Selected Thanas 53Table 3.6 Demographic characteristics of sample of households in rural
Bangladesh, 1998-99 59
Table 3.7 Financial Asset ownership of sample households in rural Bangladesh, 1998-99
60
Table 3.8 Amount of credit taken by gender and region 65Table 4.1 CBN region poverty lines 75Table 5.1 Consumption expenditure and poverty 85Table 5.2 Number of periods poor 86Table 5.3a Occupation of the household head and poverty measures 87Table 5.3b Educational attainment of the household head and poverty measures 88Table 5.3c Age and gender of the household head and poverty measures 89Table 5.4 Area under the ROC curve for individual poverty indicators and over
all model using upper poverty line 94
Table 5.5 Number of poor in Bangladesh by poverty categories 96Table 5.6 Characteristics of sample households in rural Bangladesh 100Table 5.7 Estimates and marginal effects from multinomial logistic regression:
persistent and sometimes poor 105
Table 5.8 Total, chronic and transient poverty by region 107Table 5.9 Censored quantile regression results (85th quantile) 109Table 6.1 Number of households in which men, women or both take loans 116Table 6.2 Formal and informal loans amount by head and spouse 117Table 6.3 Use of informal credit (%) 120Table 6.4 Use of formal credit (%) 121Table 6.5 Mean and standard deviations 122Table 6.6 Effect of credit amount on household expenditure shares, OLS and
Tobit estimates 126
Table 6.7 Effect of credit (dichotomous) on household expenditure shares, OLS and Tobit estimates
131
viii
List of Figures
Figure 3.1 Political map of Bangladesh 45Figure 3.2 Flood affected area in Bangladesh 47Figure 3.3 Administrative divisions of Bangladesh 49Figure 3.4 Mean Consumption levels by lower poverty line 61Figure 3.5 Head count rate calculated using the lower poverty line 63Figure 3.6 Poverty gap ratio calculated using the lower poverty line 64Figure 3.7 Squared poverty gap ratio calculated using the lower poverty line 65Figure 5.1 Stochastic dominance curve 91Figure 5.2 ROC curve for poverty models 93Figure 6.1 Credit receipt of household head and spouse 117Figure 6.2 Food expenditure share by region 119Figure 6.3 Non-food expenditure shares by region 119
ix
Acknowledgements
I express my sincere gratitude to my advisor Prof. Jill L. Findeis for making this
dissertation possible. She has trained me to be a good researcher and also helped me
become a better person. Her deep commitment and optimism to research has been a
constant source of inspiration. I am grateful to each of my committee members; Profs.
Gretchen Cornwell, Bee-Yan Roberts, Carolyn Sachs, and Steve Smith for their support
and help. I would especially like to thank Prof. Cornwell for encouraging me to
undertake a qualitative study in Bangladesh. This study was funded by grants from the
Office of International Programs, College of Agricultural Sciences; Population Research
Institute; Women in Science and Engineering Institute; and the Department of
Agricultural Economics and Rural Sociology. The focus group discussions would not
have been possible without the help of Bangladesh Institute of Development Studies and
in particular Mr. Mohammad H.R. Bhuyan. I have also benefited from numerous
discussions with Hema and Tesfayi. I am extremely grateful to Prof. Francis Dodoo for
providing me with funding during the final stages of my dissertation.
I would like to thank my parents, Gita and Jayaraman for encouraging me to come
to Penn State and pursue my dream. They have prayed for my success and happiness. I
am grateful to my husband Chandrasekhar for all the help and support. Life in State
College would not have been fun without my good friends: Abhiroop, Priya, Smita, Ram,
Viji and Julia. I would also like to mention Meenakshi, Rajesh, Vijay and Anuja for
helping me at every stage. Finally, a word of thanks to Durga Athai for constant supply
of goodies!
x
1
Chapter 1
Research Problem
1.1 Introduction
South Asia has the largest concentration of the worlds’ poor1, with over half a
billion people surviving on less than a dollar a day. One of the Millennium Development
Goals (MDG) is halving the proportion of the world’s people whose income is less than
one dollar a day and the proportion of people who suffer from hunger by the year 2015
(OECD 2001). The success of poverty alleviation programs in South Asia is critical if
this MDG is to be met. Within South Asia, Bangladesh has the highest incidence of
poverty and only India and China have larger numbers of poor people. It is estimated
that nearly half of Bangladesh’s population of 135 million people live below the poverty
line (World Bank 2003a).
The United Nation’s General Assembly declared 1996 as the International Year
for the Eradication of Poverty. This was done "recognizing that poverty is a complex and
multi-dimensional problem with origins in both the national and international domains,
and that its eradication in all countries, in particular in developing countries, has become
one of the priority development objectives for the 1990s in order to promote sustainable
development.” (United Nations’ General Assembly Resolution 48/183 1993: p. 1). The
World Bank Group defines poverty as a multidimensional phenomenon where to be poor
not only means to be hungry and to lack access to shelter and resources but also means to
be illiterate, have poor health, not receive adequate nutrition and be vulnerable to shocks,
violence and crime. Not being in poverty entails individuals leading a life free from 1 Poverty is still a global problem in the 21st century, with 2.8 billion people living on less than $2 a day and 1.2 billion living on less than $1 a day (World Bank 2001b).
2
anxiety (World Bank 2001a, OECD 2001). Poverty has also been defined as the
combination of two interacting deprivations, namely physiological and social (Hazell and
Haddad 2001). Physiological deprivation includes deprivation resulting from lack of
income, food, education, shelter, sanitation and health. It can be quantified, and
household-level data capturing physiological deprivation are frequently collected and
readily available. Social deprivation is difficult to quantify because it includes elements
such as autonomy, time information, dignity and self-esteem (Hazell and Haddad 2001).
The traditional method of measuring poverty is to use the consumption or income
concept (defined as income poverty). An individual is deemed poor if his/her
consumption or income falls below the set minimum. The poverty line sets this
minimum standard specific to each society (Lipton and Ravallion 1995). Poverty has
different implications for individuals, families and societies. The International Labor
Organization defines different levels of poverty: individual, family and society-level
poverty (ILO 2003). Poverty results in poor health, lower working capacity, lower
productivity and a shorter life expectancy among individuals, and for families it leads to
inadequate schooling and income, early parenthood, poor health, and often early death.
At the societal level, poverty is an impediment to growth, stability and sustainable
development.
According to Hulme and Shepherd (2003), policy makers often define the poor as
those individuals who have not been integrated into the market economy and policy goals
often tend to view the poor as belonging to a single homogeneous category. Further,
policy makers tend to focus on only those poor whom the market can help (Hulme and
Shepherd 2003). Given that poverty alleviation is one of the most important challenges
3
faced by the international community today (ILO 2003), an understanding of the
dynamics of poverty alleviation is critical to the formulation of appropriate policy.
Poverty measures such as the head count ratio2 are static measures that are useful for
gauging the prevalence of poverty but do not indicate the severity of poverty or
fluctuations in economic welfare indicators over time including (but not limited to)
income and consumption.
1.2 Poverty and Vulnerability to Shocks
It is clear that one of the most important aspects of poverty is vulnerability. The
poor are the most vulnerable to health hazards, economic downturns, natural
catastrophes, and even man-made violence (World Bank 2001b). The World Bank
defines vulnerability as the likelihood of being affected by shocks, which have negative
impacts on the income and consumption of poor households. Further, households in most
developing countries face a high level of income variability due to factors beyond their
control, and their poverty makes them particularly vulnerable to shocks.
Shocks can be common or idiosyncratic. Common (or aggregate) shocks are
experienced by everyone in a particular group, community or geographical region while
an idiosyncratic shock affects only a particular individual or household (Dercon 2001a).
2 The most basic measure of poverty is the head count, which is the count of the poor below the poverty line. The head count index is the head count of those in poverty as the fraction of the total population. Other measures are the poverty gap index and the squared poverty gap index (Ray 1998). The depth of poverty is measured by the poverty gap index that calculates the average income shortfall from the poverty line. The squared poverty gap index measures the severity of poverty taking into account both distance separating the poor from the poverty line and income inequality (Ray 1998). These measures can be represented using the following equation where z is the poverty line, y is per capita expenditure and N is population size (World Bank 2002b): Pα= Σ [(z-y)/z]α/N with α = 0, 1 or 2, where α = 0 gives the head count index, α = 1 gives the poverty gap index and α = 2 gives the squared poverty gap index.
4
Table 1.1 categorizes the types of risks faced by individuals, households, communities
and geographical regions. Risks can be classified based on the level at which they occur
(micro, meso and macro) and can also be classified on the basis of type: natural, health,
social, economic, political and environmental (World Bank 2000). At the micro level,
individuals or specific households face a particular set of micro-level shocks focused at
the individual or household level, e.g., old age or domestic violence. Meso-level shocks
include shocks to groups of households or villages (for example, excessive rainfall,
landslides and epidemics), and national and international shocks are best classified as
macro shocks (World Bank 2000).
Table 1.1: Types of Risks Faced by Households Type of risk Risk affecting
individual or households (micro)
Risk affecting groups of households or
communities (meso)
Risk affecting regions or nations (macro)
Natural Rainfall, landslide, volcanic eruption
Earthquake, flood, drought, high winds
Health Illness, injury, death, disability, old age,
Epidemic
Social Crime, domestic violence
Terrorism, gang activity Civil strife, war, social upheaval
Economic Unemployment, resettlement, harvest failure
Change in food prices, growth collapse, hyperinflation, balance of payments, financial or currency crisis, technology shock, terms of trade shock, transition costs of economic costs
Political Riots Political default on social programs, Coup d’etat
Environmental Pollution, deforestation, nuclear disaster
Source: World Bank 2000.
5
While common shocks such as drought and floods increase the hardships
experienced by the poorer households, idiosyncratic shocks like illness, accidents and
death also have a deleterious impact on the economic well-being of households more
generally. There is a growing literature examining the impacts of shocks on poor
households (Fafchamps 1999; Dercon 2001a; Heitzmann et al. 2001). The World Bank
(2002a) cites a study of South Indian villages which showed that “the higher risk repre-
sented by a shift in the onset of the monsoon could cut the farm profits of households in
the lowest wealth quartile by 35%, compared to a 15% reduction for median households
and no effect on the wealthiest” (World Bank 2002a: p. vii).
Table 1.2: Policy Instruments for Risk Reduction Source of risk Spreading
awareness Public health, safe water & sanitation
Labor standards, workplace safety
Education & training
Natural disasters √ Epidemics √ √ Illness, disability √ √ Old age √ Economic crises Labor market risk √ √ Harvest failure, food price
Crime & violence √ √ Environmental resettlement
√ √ √
Source of risk Access to agro-
technology Sound macro-economic policy
Transparency in decision making
Redistribution, e.g. land reform
Natural disasters Epidemics Illness, disability Old age Economic crises √ √ Labor market risk √ √ Harvest failure, food price
√ √ √
Crime & violence Environmental resettlement
√
Source: World Bank 2002a: p. 15.
6
Table 1.2 suggests potential policy instruments to reduce risks. These instruments
promote growth, increase the quality of human capital and reduce poverty. Public
awareness could play an important role in helping individuals and households to cope
with natural disasters, epidemics, crime and violence, and environmental resettlement.
The table also indicates that having a sound health, education and macro-economic
infrastructure provides a cushion against various shocks and risks such as economic
crisis, crop failure, labor market risks and epidemics (World Bank 2002a).
Kochar (1995), while analyzing idiosyncratic crop shocks faced by agrarian
households in India, found that well-functioning labor markets and not credit markets
helped farmers to smooth consumption. These households were able to increase the
number of hours of work and thereby avoid depleting their savings and the selling off of
assets. Dercon and Krishnan (2000) observed high seasonal variability in consumption
and poverty in short-panel data from Ethiopia. They attribute the variability to the
presence of shocks. Households were found to be prone to both common shocks
(rainfall) and idiosyncratic shocks (household-specific crop and livestock shocks).
Dercon and Krishnan (2000) found that poverty transitions during the period under study
could be explained by people’s vulnerability to shocks and seasonal incentives in terms
of prices and employment opportunities that affected household consumption.
Households employ various coping strategies once affected by a shock, as they
attempt to return to their original level of consumption. Disinvestment is a popular
means of coping. Liquid assets such as jewelry are initially disposed, followed by more
productive assets, making it increasingly difficult for households to return to their pre-
crisis state. Intensity, frequency and length of the shock have different impacts and
7
require different consumption-smoothing strategies. Consumption smoothing becomes
increasingly difficult for households with successive shocks (Alderman 1996). Poor
agrarian households are most susceptible to climatic shocks that directly affect crop
production and indirectly affect labor income. Household savings, asset accumulation,
income diversification, reallocation of labor, temporary migration and group-based risk
sharing are among the coping strategies used by households (Fafchamps 1999).
1.3 Shocks in Bangladesh
In 1998, Bangladesh experienced one of the largest floods of the century, which
covered more than two-thirds of the country and caused a loss of 2.04 million metric tons
of rice crop (del Ninno et al. 2001). While the overall economic impact of this flood was
less severe than previous flood occurrences and caused less damage than anticipated
(Benson and Clay 2002), the floods significantly damaged the crops and other productive
assets and further contributed to underemployment. Fortunately, trade liberalization in
the early 1990s made large-scale private food imports possible, and government food
transfers and non-governmental organization activities averted a major food crisis in
Bangladesh. However, there were short-term and medium-term negative impacts
attributable to the major flooding that occurred. In the short-term, consumption declined
and there were observed increases in the incidence of illness, especially among children.
The medium-term impact was an increase in private borrowing and negative impacts on
nutritional intake (del Ninno et al. 2003).
In Bangladesh, the agricultural sector and the labor market were the most
negatively affected after the 1998 flood. Households coped by growing alternative crops
8
and feeding alternative feed to livestock, and by finding alternative employment
opportunities. Private borrowing, used mainly for buying food, was the most widely used
coping mechanism relied on by Bangladeshi households (del Ninno et al. 2001).
Subsequent food insecurity resulted in households buying food on credit, reducing food
consumption, and borrowing money to buy food. The resulting changes in food
consumption had health implications, especially among children: an increase in stunting
and wasting among Bangladeshi preschoolers was observed (del Ninno et al. 2003).
Admittedly, shocks are an integral part of a developing economy and policies
should be geared towards equipping the poor to cope. The standard policy prescriptions
aimed at poverty reduction typically focus on ways to improve the mean utility of a poor
household. In contrast, policies that reduce the variance of households’ well-being over
time are becoming increasingly popular. Here a distinction is being made between
policies aimed at reducing chronic poverty and those to reduce transient poverty. The
World Bank (2002a) recommends a social protection strategy that extends “beyond the
traditional poverty reduction measures, to focus on creating opportunities for households
to manage risk better, primarily through a variety of instruments that perform the role of
safety nets” (World Bank 2002a: p. vi).
1.4 Poverty in Bangladesh
Bangladesh has received considerable international focus because of its high
density of population and low income. It is recognized as one of the most disaster-prone
countries in the world (Benson and Clay 2002). These factors make Bangladesh one of
the most vulnerable societies in the world. From the time of its independence in 1971,
9
Bangladesh has made considerable progress on all fronts (World Bank 2003b). The
country has achieved commendable reductions in population growth rates, child mortality
and child malnutrition (World Bank 2002b). There has also been successful disaster
management, increasing emancipation of women and growth of grass-root activism
through Non-government Organizations (NGOs) and Community-based Organizations
(CBOs) (World Bank 2003b).
There has been a reduction in both ‘income poverty’ and ‘human poverty’ in
Bangladesh since independence3. Human poverty in Bangladesh has declined at a faster
rate than income poverty in the past two decades but income poverty reduction has been
faster in the 1990s compared to the 1980s. Overall income poverty has declined at the
rate of 1 percent per annum (World Bank 2003b). Despite this progress, there has also
been an increase in inequality, both income and gender, during this period.
The difference in poverty between the poor and the poorest group is stark, where
45 percent of the poor live in extreme poverty4. Further, extreme poverty is higher
among female-headed or female-managed households. Table 1.3 shows that while all
poverty measures declined from 1991-92 to 2000, the Gini index of inequality indicates
an increase in income inequality in Bangladesh over this time period (World Bank
2003b).
3 Taking into account the multidimensionality of poverty, it is possible to categorize it into income poverty and human poverty. Income poverty measures poverty in quantitative terms. Over the 1990s, there has been increase in consumption expenditure, that is, a decline in income poverty (World Bank 2003b). The concept of human poverty was first defined in UNDP’s Human Development Report in the year 1997. It focuses on what people can or cannot do. The human poverty index captures health, literacy and economic provision deprivation (UNDP 2000). 4 Extreme poverty is defined as taking less than 1800 Kcal as per the direct calorie measure (World Bank 2003b).
10
Table 1.3: Trends in Poverty and Inequality in the 1990s in Bangladesh 1991/92 2000 Change per year (%)
Headcount Rate National Rural Urban
58.8 44.9 61.2
49.8 36.6 53.0
-1.8 -2.2 -1.6
Poverty Gap National Rural Urban
17.2 12.0 18.1
12.9 9.5
13.8
-2.9 -2.5 -2.8
Squared Poverty Gap National Rural Urban
6.8 4.4 7.2
4.6 3.4 4.9
-3.8 -2.7 -3.8
Gini Index of Inequality National Rural Urban
0.259 0.307 0.243
0.306 0.368 0.271
2.1 2.3 1.4
Source: BBS, Preliminary Report of Household Income and Expenditure Survey 2000, Dhaka, 2001 and World Bank, op.cit.(World Bank 2003b).
Roughly half of Bangladesh’s population still lives in extreme poverty (World
Bank 2002b). Poverty declined by 9 percent over 1990-2000 but the absolute number of
poor remained stable because of population growth. It is the eighth most populous
country in the world, with a total population of 135.7 million and a population growth
rate of 1.7 percent per annum in 2002 (World Bank 2004).
There have also been changes in the structural composition of the Bangladeshi
economy. During the 1990s, the share of agriculture in the Gross Domestic Product
(GDP) declined and that of the service and manufacturing industry sectors increased.
Structural adjustments in terms of trade liberalization since the 1980s brought about
macroeconomic stability, improved fiscal and monetary management and encouraged
private sector investment in the economy (Benson and Clay 2002). The question
becomes, given these important changes in Bangladesh, how can the vulnerability of
these households be reduced so as to better adjust to exogenous shocks. Among
Bangladesh’s poor, this is a critical question.
11
1.5 Consumption Smoothing
Provision of a safety net should include availability of credit to the rural
households. Ray (1998) divides credit requirements into fixed capital, working capital
and consumption credit. He defines fixed capital as the credit required for investment in
new production units which could be used to cover the fixed cost of the units. Working
capital includes costs of day-to-day running of the unit and finally, consumption credit
includes credit needed by the poor to smooth consumption due to shocks. This type of
credit need arises in agrarian societies where seasonality dictates earnings of the
households. Among the consumption smoothing mechanisms, consumption credit is
most prominent in the developing world and most of it is informal (Fafchamps 1999).
Credit could come from formal or informal sources. Formal sources include government
banks, commercial banks and NGOs and informal sources include borrowing from a local
money lender, landlord, friends and neighbors. Informal credit markets are found to
charge high, exploitative interest rates but play an important role in rural credit markets.
Other income-smoothing mechanisms used by households are remittances and transfers
from family, friends and other institutions such as NGOs. These could be viewed as
returns to social capital (Dercon 2001a).
Policy makers in the developing world have found it difficult to provide access to
credit to small rural borrowers and microfinance is an attempt to reach the poor deprived
of financial services (UNCDF 2005). In Bangladesh, microcredit programs play an
important role and these programs focus on the poor. An estimated 13 million poor
people benefit from microfinance and credit availability in Bangladesh (UNCDF 2005).
NGOs have become actively involved in the microfinance sector. The four large
12
institutions that play a crucial role in this sector are BRAC (Bangladesh Rural
Advancement Committee), Grameen, Association for Social Advancement (ASA) and
Proshika (Zaman 2004).
Zaman (2004), in his exposition of the evolution of the microfinance sector in
Bangladesh, observes that the earliest microcredit models came about in Bangladesh in
the 1970s in an attempt to rehabilitate people post-independence. The 1980s witnessed
growth in NGO in the provision of credit and which subsequently in the 1990s developed
into the ‘Grameen-model’ of credit delivery. The majority of borrowers are women who
are targeted by these programs. Individuals receive loans as members of a group. Group
liability ensures repayment and extending credit to women also has an element of
empowerment. Kabeer (2001), in her evaluation of empowerment potential of credit
programs in rural Bangladesh, finds improvement in household outcomes when women
are given loans but cautions that patriarchal society in Bangladesh is still a constraint for
women.
Focused leadership and supportive legal systems have contributed to the success
of the microcredit movement (Morduch 1999). It is argued that these formal financial
services do not reach the poorest of the poor. In Bangladesh, microfinance prominence is
lowest among the poorest group and highest among the second lowest quintile group
(Hashemi and Rosenburg 2006; Morduch 1999). These programs have improved the
lives of poor households but it is important to note that they are costly to implement and
need to be heavily subsidized (Morduch 1999).
13
1.6 Gender and Poverty
Individuals interact and operate within a family or household. To improve the
well-being of individuals, development policies not only have to take into account how
resources are allocated within the family or household but also the impact of this resource
allocation on individuals. This process of resource allocation and the resulting outcomes
are referred to as “intrahousehold resource allocation” (Haddad et al. 1997; Quisumbing
2003). Differential intrahousehold resource allocation has implications for inequality and
poverty among household members (Strauss and Beegle 1996).
Numerous studies have shown that equality of men and women in a society has
positive effect on growth of the economy, poverty alleviation and individual well-being
of household members. Productive assets in the hands of the women have led to
reduction in poverty levels and empowerment of women (World Bank 2002c).
Empowerment indicators include both participation in the political process and control
over household resources. Access to credit and financial resources are limited for women
especially in rural areas (Bamberger et al. 2002). Moreover, the intrahousehold resource
allocation literature supports the fact that transfer of resources improves the bargaining
position of women in the household and thereby improves resource allocation and child
outcomes.
As part of this research focus group discussions were held in Bangladesh. The
rationale for the focus groups was to explore not only the coping strategies employed by
the households during and after floods but also to see if there was evidence of any form
of bargaining within the household. These discussions raised a number of questions:
What decision making power did the wife have in the household? Did that affect
14
intrahousehold resource allocation? Did receipt of credit or any other form of resource
give her more power to make decisions in the household? Could the decisions taken in
the household be gender differentiated? The qualitative study was not only important in
understanding the social and cultural context in Bangladesh and therefore helped in better
interpretation of the empirical results but also indicated that receipt of financial help
could possibly increase women’s say in family matters. They also pointed that that
borrowing money especially from informal sources was one of the major coping
mechanisms employed by Bangladeshi households.
1.7 Study Objectives
Given this perspective, this research analyzes issues related to chronic and
transient poverty following a major catastrophic (flood) event using a short panel of
household data from Bangladesh. This analysis is at the household level where the
research examines how the poverty level changes and how poor and non-poor families
are different from each other. Households have been found to adjust to shocks such as
floods by borrowing, selling assets and altering expenditures (del Ninno et al. 2001).
Among these coping mechanisms, we focus on credit receipt by the household head and
the spouse and the resultant differential household-level outcomes. Here the focus is on
individuals in the household and an attempt is made to study how their interaction affects
household resource allocation. The data used in this study were primarily collected to
identify policy prescriptions for sustainable improvement in household food security in
the period following the 1998 Bangladesh flood. The specific research objectives are:
15
1 To identify the determinants of poverty in Bangladesh, with a specific focus on
differentiating those who experienced poverty following the 1998 flood in
Bangladesh and those who did not.
2 To determine the differences between those among the poor (identified in
objective 1) who are able to eventually escape poverty following the flood (the
transient poor) versus those unable to leave poverty (chronic poor). It is
hypothesized that the poor are heterogeneous and within the poverty group those
who endure poverty for a sustained period of time are characteristically different
from those who move into and out of poverty.
3 To understand how intrahousehold dynamics (e.g., variations in household
bargaining behaviors) lead to differences in outcomes of specific interest (e.g.,
food expenditure). The household bargaining model will be used to analyze the
effects of receipt of credit on consumption choices within these poor households
following the flood event. The focus is on the household’s expenditure patterns.
The study hypothesizes that receiving credit from formal and informal sources
will affect decision-making capacity of women in Bangladeshi households. This
in turn is expected to have implications for child health and nutritional outcomes.
It is important to point out that the study is not looking at how the credit amount
is spent by the receiver but instead assesses how household expenditure patterns
are associated with credit receipt by the husband and wife. Availability of data on
credit receipt by individual members of the household enables us to look at
gender differentials resulting from transfer of resources in the hands of women.
16
We also restrict our analysis to male-headed households since there are too few
female-headed households in our data.
1.8 Organization of the Dissertation
Following this introduction, Chapter 2 of the dissertation provides a review of
literature relating to poverty and the ability of households to cope with exogenous shocks.
It also addresses the bargaining literature and the theoretical model relevant for the
agricultural household. Chapter 3 presents country and data descriptions. Chapter 4
outlines the methods and the estimation strategy used in the dissertation. Chapter 5 and
Chapter 6 present the results of the analysis. Finally, Chapter 7 describes the research
conclusions and policy implications.
17
Chapter 2
Literature Review and Theoretical Framework
2.1 Introduction
The World Bank Group defines poverty as a multidimensional phenomenon
where to be poor not only means to be hungry and to lack access to shelter and resources
but also means to be illiterate, have poor health, fail to receive adequate nutrition and be
vulnerable to shocks, violence and crime. The poor can be divided into those who remain
poor continuously over time and those who enter and exit poverty from time to time. A
large proportion of the poor include people moving into and out of poverty (Baulch and
Hoddinott 2000). As noted in Section 1, the poor are the most vulnerable to health
hazards, economic downturns, natural catastrophes, and even man-made violence (World
Bank 2001a). The World Bank defines vulnerability as the likelihood of being affected
by shocks, which have negative impacts on the income and consumption levels of poor
households. Further, households in most developing countries face high income
variability due to factors beyond their control, and their poverty makes them particularly
vulnerable to shocks. Depending on how well households are able to cope, they could
remain in poverty or may be able to move out of poverty following the shock (Baulch and
Hoddinott 2000).
The coping strategies that result in consumption smoothing in response to a shock
reflect poverty dynamics5. Recent advances in computation coupled with the more
frequent collection of panel data at the household level have contributed to the study of
both the dynamics of poverty and the coping strategies that households use over time as
5 Poverty dynamics is defined as the movement into and out of poverty (Baulch and Hoddinott 2000).
18
they attempt to escape poverty. Recent studies of poverty, a topic widely covered in the
literature, have focused on the dynamics of poverty, including what it means to be in
poverty for the long term versus in the transient poverty state. At the same time, new
theories of household interactions have emerged in the economics literature, allowing
examination of intrahousehold effects. This section examines both of these topics – i.e.,
the recent literature that examines poverty dynamics, and in specific, chronic versus
transient poverty, and the literature on the new household models that focus on
intrahousehold behaviors.
2.2 Poverty Dynamics
The percentage of people living below the poverty line is an aggregate measure
that is useful in some respects but limited in others. A limitation of the aggregate
measures is their failure to provide any indication of how poor households are faring or
whether their economic status is changing over time. It is important to disaggregate the
poor to understand their circumstances and dynamics. It is critical to understand the
differences among the different types of households and individuals within households
who are classified as ‘poor’. One salient difference is the difference between those
households and individuals who move into and out of poverty versus those who fail to
move out of poverty over time. This calls for incorporating a dynamic perspective into
poverty analysis, with differentiation between the chronic and transient poor.
Households who experience poverty and deprivation for prolonged periods are
defined as chronically poor and those who move into and out of poverty (temporary) are
the transient poor (Hulme and Shepherd 2003). These two types of poverty require
19
different policy measures. Chronic poverty eradication measures include long-term
investments such as increasing human and physical capital and the returns to assets. On
the other hand, policies to help the poor cope with idiosyncratic shocks are appropriate to
tackle the problem of transient poverty. Understanding why households move into and
out of poverty can help to target the poor more effectively than assessing only static
welfare indicators; static measures may (wrongly) include those experiencing short-term
misfortunes but otherwise are not poor based on their permanent incomes or levels of
consumption. Similarly, excluding those experiencing only temporary spells out of
poverty but in fact who are in poverty the majority of the time is also a weakness (Baulch
and Hoddinott 2000). Baulch and Hoddinott (2000) emphasize that understanding factors
affecting poverty dynamics help in designing safety net policies and, most importantly,
help to target the vulnerable.
Finally, while income and consumption measures of poverty are often used to
study chronic and transient poverty, there is an increasing emphasis on adopting a
multidimensional approach in poverty studies by considering other measures such as
educational attainment, nutritional intake and ownership of assets in the analysis (McKay
and Lawson 2002).
2.2.1 Definition of Chronic and Transient Poverty
It is widely accepted that the poor are a heterogeneous group. Studies of poverty
dynamics generally treat a household as a single economic unit6. Jalan and Ravallion
(2000) define transient poverty as the poverty that is caused by variability in
6 Assumes a unitary model framework.
20
consumption. A household with mean consumption below the poverty line across all
periods is defined to be experiencing chronic poverty. Hulme and Shepherd (2003)
define the chronically poor as those who experience poverty for a period of five years or
more and transient poor as those who move into and out of poverty. They argue that the
five-year period is a significant length of time and studies show that individuals who are
poor for five years or more have a high probability of remaining poor for the rest of their
lives. The chronic poor suffer from persistent deprivation. The chronically poor are also
those who need external help to get out of the poverty trap and they remain poor despite
implementation of policies to tackle poverty (Aliber 2003). Chronic poverty also is
transmitted from one generation to another, and children within chronically poor
households are more likely to be caught in the poverty trap and likely to remain poor the
rest of their lives (Aliber 2003).
2.2.2 Measuring Chronic and Transient Poverty
Income and consumption7 measures at the household level are common measures
of chronic and transient deprivation and longitudinal or panel data sets are ideally suited
to study poverty movements or transitions8. However, the multidimensional definition of
poverty requires that other welfare indicators – for example, asset ownership, nutritional
intake, educational enrollment, the human deprivation index -- could be included to
provide a more inclusive measure of poverty.
Even households that are sometimes poor are heterogeneous. The question
becomes how to rigorously define the transient poor (Baulch and Hoddinott 2000). For 7 Also called metric welfare measure (Baulch and Hoddinott 2000). 8 Atleast a three-year panel data set is needed (Baulch and Hoddinott 2000).
21
example, using the definition that the chronically poor are poor continuously over time
and the transient poor experience at least one spell out of poverty, in five-year panel data,
a household classified as poor for four years would be categorized as transient poor but
then so would be a household that is poor just one year. Treating both households alike is
probably not reasonable. Hulme and Shepherd (2003) categorize poor as ‘always poor’,
‘usually poor’, ‘churning poor’, ‘occasional poor’ and ‘never poor’ (see Table 2.1). Both
income and non-income indicators are used in their categorization.
Table 2.1: Five-tier Classification of the Poor Aggregate category Specific category Definition Chronic poor
Always poor Usually poor
Those whose poverty score is below the poverty line for each period. Those whose mean poverty score over all periods is below the poverty line but not poor in every period.
Transient poor Churning poor Occasionally poor
Those with a mean poverty score around the poverty line and who are poor in some periods. Those with mean a poverty score above the poverty line and that have experienced at least one period of poverty.
Non-poor Never poor Those whose mean score is always above the poverty line.
Source: Hulme and Shepherd (2003).
The poverty literature delineates two approaches to study transitory and chronic
poverty using income or consumption data. First, the spells approach is used, where the
number or length of poverty spells experienced classifies the poor as chronic or transient
(McKay and Lawson 2002). A transition matrix can be used to provide information on
the proportion or number of people moving into and out of poverty using deciles,
quintiles or poverty lines, and also gives information about the poverty experiences of
22
households in the intervening period (Oduro 2002). Using this approach, Oduro (2002)
found that consumption measures are generally less variable compared to income
measure of poverty. This is because households have the ability to smooth consumption
over time. Therefore, the number of poor in the transient category increase when the
income measure of poverty is used and different welfare measure could yield different
estimates of transient and chronic poverty. The spells approach is plagued by
measurement error occurring in the process of collecting income and consumption data
(Hulme and Shepherd 2003). Difficulty in measuring the values of own production (as
well as problems recalling and imputing values) and in determining the revenue and cost
of farm and non-farm enterprises result in errors (Baulch and Hoddinott 2000).
A second approach, the components approach, decomposes the permanent component of
household income from its transitory variations. Households that have permanent
components below the poverty line are then defined as the chronically poor (McKay and
Lawson 2002).
It is recommended that to have a complete picture, the spells and component
approach based on income or consumption have to be complemented by the used of
qualitative studies. That is, there is a need to move beyond defining poverty in monetary
terms.
2.2.3 Empirical Studies
An important question is why some households are not able to move out of
poverty while others appear to move into and out of poverty over time – they escape
being poor but often fall back into poverty at least temporarily. Some studies have been
23
able to predict chronic poverty better than transitory poverty (Haddad and Ahmed 2003;
Baulch and Hoddinott 2000; Jalan and Ravallion 2000). Analysis of chronic poverty
across 25 countries shows that chronic poverty is spatially concentrated, affected by
demographic composition of the household and determined by human and physical
capital and labor markets (Yaqub 2002).
Jalan and Ravallion (2000) define total poverty as the sum of chronic and
transient poverty and decompose households into total, chronic and transient poor using
the squared poverty gap index. The censored quantile regression technique is used to
identify the determinants of both chronic and transient poverty. Their model is able to
predict chronic poverty better than transient poverty, with the determinants of chronic
and total poverty being similar. Haddad and Ahmed (2003), using a two-round panel
data set for Egypt, measure changes in per capita consumption among households. They
calculate squared poverty gap measures for each household and divide the sample
population into total, transient and chronically poor groups. The censored quantile
regression method is again applied to identify the determinants of chronic and transient
poverty. As was also true for Jalan and Ravallion (2000), their model predicts chronic
poverty better than transient poverty, with the determinants of total and chronic poverty
again being quite similar.
Kedir and McKay (2003) analyze three waves of household survey to study
chronic poverty in Ethiopia using total household expenditure per month as the welfare
indicator and the population is classified into ‘always poor’, ‘two-period poor’, ‘one-
period poor’ and ‘never poor’. Multinomial regression analysis indicates that household
24
composition, unemployment, lack of asset ownership, lack of education, ethnicity, and
the age and gender of the household head are important determinants of chronic poverty.
Table 2.2: Studies Decomposing the Poor into Relevant Categories Country Number
of waves
Welfare measure Percentage of poor
Always Sometimes Never South Africa (Carter 1999)*
2 Expenditure per capita 22.7 31.5 45.8
Ethiopia (Dercon and Krishnan 1999)*
2 Expenditure per capita 24.8 30.1 45.1
India (Gaiha 1998)* 3 Income per capita 33.3 36.7 30.0 India (Gaiha and Deolalikar 1993)*
9 Income per capita 21.8 65.8 12.4
Cote d’Ivoire (Grootaert and Kanbur 1995)*
2 Expenditure per capita 14.5 20.2 65.3
Cote d’Ivoire (Grooteart and Kanbur 1995)*
2 Expenditure per capita 13.0 22.9 64.1
Cote d’Ivoire (Grootaert and Kanbur 1995)*
2 Expenditure per capita 25.0 22.0 53.0
Zimbabwe (Hoddinott, Owens and Kinsey 1998)*
4 Income per capita 10.6 59.6 29.8
China (Jalan and Ravallion 1999)*
6 Expenditure per capita 6.2 47.8 46.0
Pakistan (McCulloch and Baulch 1999)*
5 Income per adult equivalent 3.0 55.3 41.7
Russia (Mroz and Popkin 1999)* 2 Income per capita 12.6 30.2 57.2 Chile (Scott 1999)* 2 Income per capita 54.1 31.5 14.4 Indonesia (Skoufias, Suryahadi and Sumarto 2000)*
2 Expenditure per capita 8.6 19.8 71.6
Egypt (Haddad and Ahmed 2003)
2 Average per capita consumption
19.02 20.46 60.52
Ethiopia (Kedir and McKay 2003)
3 Median consumption expenditure
21.5 36.2 51.1
Pakistan (McCulloch and Baulch 1999)
5 Annual income 15.31 43 41.69
*Source: Baulch and Hoddinott (2000): p.7.
McCulloch and Baulch (1999) studied the implications of decomposing the poor
into the chronic and transient poor on the basis of household characteristics for targeting
the poor. They estimated multinomial logit and ordered logit models to identify the
determinants of the chronically and transient poor. They conclude that income smoothing
25
policies are more appropriate for tackling the problem of transitory poverty while growth
policies designed to increase the mean income level help the poor to escape being
chronically poor. Table 2.2 provides a summary of selected studies of chronic and
transient poverty.
2.2.4 Characteristics of the Chronically Poor
Education is a powerful and important predictor of chronic poverty. Studies have
found that an increase in number of years of education decreases the probability of being
chronically poor (McCulloch and Baulch 1999; Jalan and Ravallion 2000; Aliber 2003;
McCulloch and Calandrino 2003). Human capital accumulation in Bangladesh is an
important form of asset holding for the poor, which equips them to participate in the
growth process (World Bank 2002b).
Larger households are more likely to experience chronic poverty. This is true
among households that have limited access to resources and assets. McCulloch and
Baulch (2000), Jalan and Ravallion (2000), Haddad and Ahmed (2003) and Aliber (2003)
in their study of Pakistan, China Egypt and South Africa, respectively, found this to be
true. Older household heads and female-headed households are also more likely to be
chronically poor (Aliber 2003). All things equal, the same is true for households with a
greater number of children, more members above the age of 60 and for households with
more disabled members.
Place of residence determines the opportunities and facilities available to the
households (McKay and Lawson 2002). Remote geographical locations are
disadvantaged in terms of access to resources. The likelihood of being persistently or
26
chronically poor in such locations is higher. McKay and Lawson (2002) also find that
chronic poverty is a major problem in rural areas because of lack of employment
opportunities and resources.
Lack of physical assets is associated with chronic poverty (McCulloch and Baulch
2000; Aliber 2003). Assets such as livestock and land help poor households not only
generate income but are also a form of investment. Poorer households commonly hold a
greater share of their assets in the form of liquid assets such as livestock and financial
assets (World Bank 2002a).
The sector of occupation of the household head is shown to be very important in
most studies. Haddad and Ahmed (2003), in their study of chronic and transient poverty,
report that being employed in the manufacturing, recreation or non-farm sectors
decreases the likelihood of being chronically poor as compared to being engaged in the
agricultural sector. Seasonal, casual and retrenched farm workers are also vulnerable
(Aliber 2003).
2.2.5 Characteristics of the Transient Poor
Some factors affect both chronic and transitory poverty but there are others that
are associated with transient poor alone. Poverty levels in general decline rapidly with
increases in education of the household head (World Bank 2002a). There is also a strong
negative association between transient poverty and educational attainment (Haddad and
Ahmed 2003; Jalan and Ravallion 2000). Jalan and Ravallion (2000) find higher
transitory poverty among smaller Chinese households. Adoption of new technology and
adverse price fluctuations can result in temporary poverty (McKay and Lawson 2002).
27
The adoption of new agricultural techniques involves risk taking on the part of farmers,
which, in turn, causes variability in their income. The study of Argentinean households
by Cruces and Wodon (2003) found that the risk of running a business made employers
vulnerable to transient poverty, and the provision of social security by the public sector
made households engaged in this sector more resistant to transient poverty.
2.3 Intrahousehold Resource Allocation Models
Individuals interact and operate within a family or household. To improve the
well-being of individuals, development policies not only have to take into account how
resources are allocated within the family or household but also the impact of this resource
allocation on individuals. This process of resource allocation and the resulting outcomes
are referred to as “intrahousehold resource allocation” (Haddad et al. 1997; Quisumbing
2003). Differential intrahousehold resource allocation has implications for inequality and
poverty among household members (Strauss and Beegle 1996).
2.3.1 Unitary Household Model
The traditional approach to intrahousehold resource allocation is the unitary
approach where the household is viewed as a single economic unit. Becker (1965)
proposed that the household, sharing a single set of preferences, maximized utility by
combining time, goods purchased in the market and goods produced at home. Becker’s
theory of time allocation assumes that households are both consumers and producers. As
producers, household combine nonlabor inputs and time using the cost minimization
approach and, as consumers, maximize their utility subject to prices and resource
28
constraints. Unitary models assume that all members of a household share the same
preference function and pool their resources. In Becker’s time allocation model,
households maximize the following utility function
U = U (Z1,…., Zm) ≡ U (f1,…., fm) = U (x1,….,xm; T1,…., Tm) [2.1]
where Zi are goods produced within the household, x1 is the vector of market goods, and
Ti is a vector of time inputs used in producing the ith commodity. The utility of the
household is maximized subject to:
1. Household production function which combines time and market goods to
produce goods:
Zi= fi (xi, Ti) [2.2]
The production function is characterized by fixed coefficients as:
T1 ≡ ti Zi [2.3]
xi ≡ bi Zi [2.4]
where ti is input of time required to produce one unit of household good (Z) and bi
is the input of market good required to produce a unit of household good (Z).
2. Goods constraint
∑m
tt xp1
= I = V + Tw ẅ [2.5]
where I is total income, V is other income, Tw denotes market work and ẅ is the
vector of wage rates.
3. Time constraint
Tc = T – Tw [2.6]
29
where T is the total time available during the day and Tc is the time spent on
consumption (leisure).
Equations 2.3, 2.4, 2.5 and 2.6 can be combined to yield the full-income constraint:
Σ (pibi+ t ẅ) Zi = V + T ẅ [2.7]
Households thereby make consumption and production decisions based on exogenous
factors, namely market prices, wages and non-earned income (Schultz 2001).
Apart from being applied to standard demand analysis, unitary models were
extended to include determinants of education, health, fertility, migration and labor
supply (Haddad et al. 1997). This approach is popular because it is straightforward and
household-level data are readily available. Attempts have been made to assess questions
of intrahousehold resource allocation within the unitary framework, despite its treatment
of households as ‘black box’ (Pitt 1997; Alderman and Gertler 1997). For example, Pitt
(1997) looks at household resource allocation using intrahousehold conditional demand
equations. These equations determine how allocations such as time and food to one
member affect allocations to others. This model requires prices of person-specific goods
for estimating demand equations. Absence of prices lead to issues of identification and
Pitt suggests solutions to overcome this problem. Further, Alderman and Gertler (1997)
use the unitary framework to show how gender plays a role in human capital investment
within households with different levels of resources. They find that the demand for
daughters’ human capital is more income and price elastic in cases where there is a son
preference. This was empirically tested in their study of Pakistan and expected
relationships were observed.
30
2.3.2 Agricultural Household Models
The agricultural household model describes household behavior using the unitary
model. The vast majority of households in rural areas in developing countries are
engaged in agricultural activities. The agricultural household production model is a
model of both production and consumption; in this model, the household is both a
consumer and producer and hired labor is assumed perfectly substitutable with family
labor. The model assumes the presence of perfect labor markets where excess labor can
be employed in the non-agricultural sector and households can also hire labor under
situations of excess demand for labor (Schultz 2001). Within the agricultural household
models, production and consumption decisions can be analyzed either sequentially
(separable) or simultaneously (nonseparable model).
The basic agricultural household model posits a farm household that is assumed to
maximize a household utility function (Singh et al.1986a):
U = U (Xa, Xm, Xl) [2.8]
where utility is a function of agricultural staples (Xa), market goods (Xm) and leisure (Xl).
The household production function is represented as Q (L, A), where L is labor and A is a
fixed quantity of land. Household utility is maximized subject to a budget constraint,
time constraint and the production technology:
pm Xm = pa (Q - Xa) – w (F - L) [2.9]
Xl + F = T [2.10]
Q = Q (L, A) [2.11]
where w is the wage rate, pm and pa are market prices, F is family (household) labor input
and T is total stock of time within the household. The Q (L, A) - Xa is marketed surplus
31
and F - L yields net sales of labor. All prices (w, pm, pa) are exogenous and the
household is a price-taker in all three markets. The three constraints are collapsed into
one full-income budget constraint:
pmXm + paXa + wXl = wT + paQ(L, A) – wL [2.12]
The paQ (L, A) – wL represents farm profits and the left-hand side of equation 2.12
represents total household expenditures including purchase of market goods, the
household’s purchase of its own output and its own purchase of time in the form of
leisure.
Optimal levels of consumption of each of the commodities (Xa, Xm, Xl) and the
total labor input utilized in agricultural production are determined. Optimal value of
labor, output and full income is derived solving the first-order conditions. In the case of
agricultural households, production activities determine income and factors affecting
production influence the household’s full income, which in turn affects household
consumption. Therefore, household production and consumption are separable or
recursive. Separability of the decision implies that production decisions are not
influenced by consumption and labor supply decisions but consumption and labor supply
decisions are dependent on production decisions. That is, production decisions do not
depend on consumption preferences but consumption decisions are influenced by
production through the full income. Households follow a two-step optimality procedure:
first, farm profits are maximized using the optimal combination of inputs and the
household utility function is maximized (Singh et al. 1986a).
Separable models are not applicable under all circumstances and nonseparable
models may be more appropriate in the case of presence of imperfect markets, when sale
32
and purchase prices for goods differ or when markets fail (Singh et al. 1986b). Market
failures are characteristic of developing agrarian economies and nonseparable models are
more applicable (Sadoulet and de Janvry 1995).
However, unitary household models are very restrictive in nature. They do not
indicate how household decisions are made and how resources are allocated among
members (Schultz 2001). Individuals constitute the household and taking individuals as
the unit of analysis theoretically makes more sense. Given that individuals within a
household could have access to different kinds of resources, the assumption of income
pooling may not hold empirically (Mendoza 1997). Also, household production
functions are difficult to estimate as the output produced in the household (for example,
children’s education) is not sold in the market. Individual members of the household
may have different tastes and preferences which could be distinct from that of the
household.
Unitary models also do not take into account intrahousehold allocation of
consumption and the implications of this allocation for welfare. Different allocations to
different members have different welfare outcomes (Mendoza 1997). For example, in
many developing countries human capital investments in men and women have different
implications for the household and child outcomes (Behrman 1997). Development
policies only sometimes target individuals. Unitary models which do not take into
account individual preferences could yield misleading policy directives (Quisumbing
2003).
33
2.3.3 Collective Models
During the 1980s alternatives to the unitary approach to household resource
allocation emerged. Under the collective approach, the household utility function is
disaggregated and the model takes into account the different preferences of each member
of the household (Chiappori 1988, 1992; Browning and Chiappori (1998); Haddad et al.
1997; Quisumbing 2003; Mendoza 1997). Here the focus is on the individuals within the
household rather than on the entire household as one unit, and resources are no longer
pooled. Individuals decide the amount of their income to be transferred to others and the
amount allocated to purchasing common household goods (Doss 1996).
Unitary models can be shown as a special case of the collective models.
Collective household models can be divided into cooperative and non-cooperative models
(Mendoza 1997). All cooperative models assume that households make Pareto efficient
allocations, i.e., no one can be made better off without making someone else worse off
(Chiappori 1988, 1992) whereas non-cooperative models may or may not yield Pareto
optimal outcomes.
2.3.4 Cooperative Model
Cooperative models postulate that individuals form households only if there is a
net gain in doing so. There are two approaches within the cooperative framework. The
first approach assumes that all households have a sharing rule to allocate income among
members (Chiappori 1988, 1992; Browning and Chiappori 1998; Apps and Rees 1997).
The income-sharing rule is a function of the incomes of the husband and the wife and
total household income (Mendoza 1997). The household uses the sharing rule to allocate
34
resources among its members. Doss (1996) outlines four assumptions that are needed to
recover the sharing rule from household expenditure data: 1) requires that some goods be
private, 2) the utility of other members is included as one of the arguments in one’s own
utility function, 3) a separable utility function exists with respect to private and public
goods, and 4) at least one private good is assignable in order to observe who consumes
that good.
Browning et al. (1994) develop a model showing how income affects household
outcomes within the framework of family expenditure data. They assume that
households make Pareto efficient decisions and find that resource allocation decisions
among Canadian couples depend on their current income, age and lifetime wealth.
Browning et al. (1994) consider a two-member household (a, b) where households
maximize the weighted sum of household members’ utility subject to a budget constraint
(Strauss and Beegle 1996):
Max μ UA (xA, xB) + (1-μ) UB (xA, xB) [2.13]
subject to p (xA + xB) = Y [2.14]
where Ui is the utility function of the household members, xi is the private consumption
good, Y is total household income and p is the price vector for the market good. The μ is
the welfare weight for household member A which lies between 0 and 1 and is a function
of prices, household income and other factors such as the distribution of income (Strauss
and Beegle 1996). This model collapses into the unitary model if individuals A and B are
identical or if μ equals 0 or 1. This would imply that everybody has identical preferences
or there is a dictator in the household. Demand for the market good x is a function of
prices, income and μ (xA = xA (p, Y, μ (p, Y))).
35
Strauss and Beegle (1996) derive tests for the collective approach using a two-
stage decision process with an income-sharing rule. The household first pools resources
and allocates income to each individual and then individuals maximize their sub-utility
subject to the income they have been allotted. Suppose θ is the income allotted to one
member out of the total income Y, then in a two-member household the other member
would have (Y – θ) left for him/her. Therefore, in the second-stage, members maximize:
max UA (xA) [2.15]
subject to pxA = θ [2.16]
The sharing rule (θ) is a function of prices, income and other distributional factors and
the model is a unitary model when θ is fixed. The conditional demand curve is xA =
xA (p, θ). It is also postulated that the ratio of the marginal propensity to consume a good
with respect to changes in income of the two individuals is to be the same across all
goods. That is,
Bj
Aj
Bk
Ak
YXYX
YXYX
∂∂∂∂
=∂∂∂∂
//
// [2.17]
In a unitary model this ratio is equal to one and in a collective model this ratio represents
the sharing weights that determine the control of the individual over the resources.
Chiappori (1992) develops a collective model of household labor supply where the
economic agents first share nonlabor income based on the sharing rule and then in the
second-stage make labor supply and consumption decisions.
36
2.3.5 Cooperative Bargaining Models
The other approach was developed by McElroy and Horney (1981) and Manser
and Brown (1980) which explicitly assumes a bargaining rule among members of the
household. Manser and Brown (1980) provide a cooperative bargaining solution to the
issue of marriage and household decision-making where benefits derived from marriage
are distributed between the husband and wife. In the Manser and Brown model, a rule is
derived to resolve household allocative and distributional issues using a Nash-bargaining
model. Individuals have a guaranteed utility level that they enjoy when they do not
cooperate, and they marry or form a family only if the utility they derive from
cooperating is greater than being single. This minimum reservation utility is defined as
the threat point9 and gains from cooperation are a function of the bargaining strength of
the individual family members (Mendoza 1997). In the bargaining approach, control
over the income plays an important role in household decision-making unlike in Becker’s
unitary model where the household collectively controls the total income.
McElroy and Horney’s (1981) Nash-bargaining household decision model
assumes a two-individual household, m and f. Market goods of interest to m are
xm = (x0, x1, x3) at prices pm = (p0, p1, p3) and those of interest to f are xf = (x0, x2, x4) at
prices pf = (p0, p2, p4) where x1 and x2 are the market goods consumed by the husband and
the wife, respectively, and x3 and x4 are the leisure time10 of the husband and wife,
respectively. The x0 is the household good that is consumed which has a public good
characteristic. If individuals are not married then they maximize their individual utility
functions subject to their individual budgets, to derive their indirect utility functions 9 Defined as maximum level of utility outside of the household (McElroy 1990). 10 Leisure time is defined as time not spent in market work (McElroy and Horney 1981).
37
),(0 mmm
om IpVV = and ),(0 ff
fo
f IpVV = where Ik (k = m, f) is non-wage income.
Further, the McElroy and Horney model assumes gains from marriage, with the married
couple maximizing the Nash utility function
)];,()()][;,()([ 00 fffff
mmmmm IpVxUIpVxUN αα −−= [2.18]
subject to a full-income constraint
fm IITppxpxpxpxpxp +++=++++ )( 434433221100 [2.19]
where T is the time endowment of both individuals and the Vi are the threat points of the
individuals m and f (threat of becoming divorced). Threat points represent the utility
individuals would receive if they remain single (reservation utility) and the αi are the
extrahousehold environmental parameters (EEPs) (McElroy 1990). Solution to the
maximization problem yields Marshallian demand equations which are functions of
prices, nonlabor income, and the EEPs:
4,3,2,1,0),,;,,( == iIIphx fmfmii αα [2.20]
The EEPs shift the threat points and have no effect on nonwage income and prices
(McElroy 1990). They include social, legal and institutional parameters that have welfare
impacts on households, and thus enabling the policy component to be explicitly included
in the model (Swaminathan 2003). Examples of EEPs are divorce laws, welfare policies
of single mothers, extended family support networks and local ratios of marriageable men
to women (Schultz 2001). These factors may affect the reservation utilities and family
outcomes within the bargaining process.
Lundberg and Pollak (1993) introduce a ‘separate spheres’ bargaining model
within marriage. Divorce as a threat point is replaced by a non-cooperative equilibrium
38
that reflects traditional gender roles. Within this framework, Lundberg and Pollak (1993)
study the distributional implications of the child allowance schemes to the mother in
United Kingdom. This would make outcomes of cooperative bargaining favor the
woman. Ermisch (2003) suggests that an increase in mother’s income has an income
effect which gets translated into increases in expenditures on children and herself. This
also increases her bargaining power within the household. Assuming that women give
greater weight to children’s needs, improvement in her position could further increase
expenditure on children. Ermisch (2003) argues that any improvement in her position
within the household could have improved child outcomes.
Sahn and Stifel (2002) in their paper show that educational attainment increases
women’s power in the household by improving her employment and income earning
capacity. Examining data from 25 Demographic Health Surveys collected during the
1990s covering 14 African countries, they find parental education to be important in their
children’s anthropometric outcomes and mothers’ education, enabling resources to be
channelized to their children. However, they also find that father’s education has a
greater impact on boys and vice versa for girls especially with respect to height-for-age
outcomes.
Increased participation of women in the Ecuadorian flower industry has had
bargaining effect on men’s contribution in household work (Newman 2002). The study
finds a wage substitution effect which determines how much men contribute in household
work and the higher the wages the woman earns the more is her bargaining power to
redistribute household tasks. Martinelli and Parker (2003) use a Nash bargaining model
to analyze allocation of child’s time between labor and education. Bargaining between
39
parents is also looked at for making consumption and bequest decisions. In particular,
they study conditional and unconditional government transfers11. Their results show that
conditional transfers improve the welfare of the child as well of the mother in bequest-
constrained households. However, in a household with positive bequests, conditional
transfers lead to over accumulation of human capital making unconditional transfers
more appropriate.
Various measures of bargaining power have been used in the literature. Schultz
(1990) uses unearned income as a proxy for the bargaining power of the individual who
controls the income. Schultz (1990) found that in the case of Thai women, increases in
bargaining power increased consumption of leisure and time spent in non-market
activities and women also preferred to have more children. Unearned income also
increased the leisure and time spent in non-market activities of Thai men but the effect of
wife’s non-earned income had a weaker effect on his labor participation decision.
Thomas (1990), studying Brazilian women, found that unearned income in the hands of
women has a larger impact on health and child survival probabilities. Doss (1997) used
ownership of current assets held by Ghanaian women as a proxy for bargaining power;
Doss (1997) found that the woman’s ownership of assets increased her threat point and
improved her bargaining position within the Ghanaian household.
Lundberg et al. (1997) found that there was an increase in spending on women’s
and child clothing in the U.K. when there was a policy change in national child benefit
plans. The new policy transferred resources to the women rather than to men which had
significant positive effects on household expenditure patterns (Lundberg et al. 1997). 11 Conditional transfers are transfers that are conditional on human capital investment (Martinelli and Parker 2003).
40
Quisumbing and de la Briere (2000) examined the differences in the bargaining power of
Bangladeshi husbands and wives using current assets and the value of assets brought into
marriage. Their finding corroborates the findings that increases in the control of
resources by women have beneficial impacts on child outcomes through increases in
expenditures on child clothing and schooling. In a study of rural Malawi, Swaminathan
(2003) used access to credit and land ownership as measures of bargaining power.
However, the results do not support the hypothesis that women’s spending is necessarily
more oriented towards children and the household. However, this result may reflect from
the extremely low incomes of Malawian households, such that only the most essential
expenditures are even possible (Swaminathan and Findeis 2003). Suen et al. (2003)
theoretically present a Nash bargaining analysis of parental transfers to daughters. The
study shows how inter-generational transfer affects intrahousehold allocation of the
daughter by improving her bargaining position in her family. Their analysis yields that
parents have greater incentive to allocate money to married and income-earning
daughters because marriage improves efficiency in both consumption and production of
public goods and that increased dowry actually would reduce probability of divorce.
None of these measures are perfect and the choice of indicators should be guided
not only by its exogeniety to bargaining within the marriage but should also take into
account the cultural relevance of these indicators (Quisumbing and de la Briere 2000).
The cooperative bargaining approach which is applied to study the interaction
between spouses can also be used to study bargaining process taking place between other
members of the family which could include more than one generation. Lundberg and
Pollak (2004) suggest that the bargaining approach can be applied to other family
41
relationships. For example, bargaining between child and parent is a function of outside
options such as the child threatening to leave the house. The same game-theoretic
models of strategic interaction can used to understand the allocation between other
members of the family. Chang, Chen and Somerville (2003) find that compared to
common preference models, the Nash bargaining approach works better when they study
the mobility decisions of households in Taipei, Taiwan. In particular, they examine the
bargaining between older and younger generations in an extended family. Consistent
with the mobility literature where likelihood of mobility lowers with age, their study
shows that as income of the elderly members increase, their bargaining position in
mobility decision increases. That is households with older earning members are less
likely to move.
Extended family structure is common in developing economies. Taking into
account the bargaining power of other members of the family may be important in
understanding intrahousehold allocation outcomes (Quisumbing 2003). Fafchamps and
Quisumbing (1999) explore the affect of human capital, learning by doing and one’s
status in the family on division of labor within household in rural Pakistan where
presence of extended family within the household is common. They find that daughter-
in-laws have little bargaining power in the household and are exploited by mother in-
laws. The result is that daughter-in-laws are involved in household work and not market
work. The bargaining dynamics between the two generations results in her working
harder.
42
2.3.6 Non-Cooperative Model
Under the non-cooperative model, members of the household are involved in
individual optimization and their actions depend on the actions of other members
(Mendoza 1997). Woolley (1988), Ulph (1988) and Carter and Katz (1997) use this
approach which requires that no assumption about income pooling is made. In a two-
person economy, individuals maximize their individual utility functions subject to the
individual’s income constraint. Individual utility is a function of consumption of his or
her private goods and common goods. Utility of the other household member is not
incorporated into one’s own utility function (Haddad et al. 1997).
Woolley (1988) argues that it is difficult to apply cooperative game theory to
marriage, since under cooperative game theory players must negotiate binding contracts
regarding allocation of resources. This may not be possible between family members.
He demonstrates using the Cournot-Nash equilibrium solution that the income differential
between the spouses affects the type of equilibrium and the expenditure pattern of the
household. Bloch and Rao (2002) apply a non-cooperative bargaining and signaling
model of dowry. They use domestic violence is as an instrument of bargaining between
husband and wife, resulting in redistributed resources. They find that the more satisfied
the husband is with the marriage, the less violent he is and more dowry reduces
probability of being violent. What is surprising in their findings is that women from
wealthier households are more likely to be beaten by their husbands to extract greater
dowry.
The difference between the cooperative and non-cooperative models lies in the
fact that under the cooperative approach, individuals are bound by costless enforceable
43
contracts that facilitate distribution of the benefits of cooperation among household
members. This is not the case when taking a non-cooperative approach (Mendoza 1997).
Jose (2003) in his discussion of gender bias in resource allocation in India
supports collective models to unitary models. However, he cautions that bargaining
models are not entirely satisfactory for South Asian countries where factors including
ideology, lineage system, kinship system, religious codes and family structure play an
important role within households. He puts forward ‘process- or situation-specific’
explanations for household resource allocation which acknowledges dynamic and
complex processes that exist in these societies. While examining the participation of
women in home-based production in the garment sector in urban India, Kantor (2003)
finds that access to resources alone does not translate into improvement in position of the
women. This may be true because of stringent social norms that women have to face and
a small contribution to the total household income may have only marginal impacts on
their decision-making ability.
44
Chapter 3
Data and Descriptions
3.1 Country of Study: Bangladesh
Bangladesh is the eighth most populous country in the world with a total
population of 135.7 million persons and a population growth rate of 1.7 percent per
annum in 2002 (World Bank 2004). It is a small country covering 144 thousand square
kilometers (as of 2001) with a population density of 1024 per square kilometer (as of
2002)12. Table 3.1 shows that Bangladesh has a higher population density and total
fertility rate than India13.
Table 3.1: Country Profile Country Total Population Population Density (annual %) Fertility (births per woman) Bangladesh 135.7 million 1.7 3 India 1 billion 1.6 2.9 Pakistan 144.9 million 2.4 4.5
Source: World Development Indicators database, April 2004.
Bangladesh gained independence in 1971 before which it was part of Pakistan.
Dhaka is the capital city and Bengali is the official language of Bangladesh. Islam and
Hinduism are the two popular religions followed by many. Geographically, Bangladesh
is very flat and very prone to flooding. The three main rivers are Brahmaputra, Ganges
and Meghna, which occupy about 7% of the total land area (Benson and Clay 2002).
Figure 3.1 presents the map of Bangladesh which is surrounded by Myanmar on the
southeast and India on three sides. 12 Information in this section is primarily taken from the World Bank country statistics website available at http://www.worldbank.org.bd/WBSITE/EXTERNAL/COUNTRIES/SOUTHASIAEXT/BANGLADESHEXTN/0,,menuPK:295785~pagePK:141132~piPK:141109~theSitePK:295760,00.html and from Encyclopedia Britannica, available at http://www.britannica.com/ebc/article?eu=381825&query=bangladesh&ct= 13 India is the second-most populous country in the world.
46
Bangladesh is also prone to various idiosyncratic shocks such as violence,
economic shocks and illnesses. Flash floods are common. Figure 3.2 indicates the flood
affected regions. Features in blue indicate normal flooding and features in red indicate
flash flooding. Compared to other regions, Chittagong is less affected by normal
flooding and Sylhet is seems to be most affected by flash floods.
Despite making significant progress in poverty reduction, roughly half of
Bangladesh’s population still lives in extreme poverty. Poverty rates declined by 9
percent over 1990-2000 but the absolute number of poor remained stable, due to
population growth. The poor in Bangladesh are characterized by low levels of education,
limited access to human and physical capital, and employment in low-paying, physically-
demanding jobs. Eighty-five percent of the poor in Bangladesh live in the rural areas and
agriculture is the main activity in the region. Female-headed households are found to
have a higher incidence of poverty (World Bank 2002b).
At the same time, there has also been successful disaster management, increasing
emancipation of women and growth of grass-root activism through Non-government
Organizations (NGOs) and Community-based Organizations (CBOs) (World Bank
2003b). There have also been changes in the structural composition of the Bangladeshi
economy. During the 1990s, the share of agriculture in the Gross Domestic Product
(GDP) declined and that of the service and manufacturing sectors increased. Structural
adjustments in terms of trade liberalization since the 1980s brought about
macroeconomic stability, improved fiscal and monetary management and encouraged
private sector investment in the economy (Benson and Clay 2002).
48
The six administrative divisions of Bangladesh are Barisal, Chittagong, Dhaka,
Khulna, Rajshahi and Sylhet14. Figure 3.3 shows the location of these divisions. These
divisions are further divided into 64 districts and each district is further subdivided into
thanas15. Table 3.2 presents the division profiles. Dhaka, Rajshahi and Chittagong
divisions have higher population and Barisal is found to have the highest average
educational levels. Sylhet has the lowest average literacy and also the lowest literacy
levels among women. Agriculture (own farm) and agriculture labor (wage) are uniformly
the most important occupations across all divisions.
Table 3.2: Division Profile Barisal Chittagong Dhaka Khulna Rajshahi Sylhet Location South Southeast Central Southwest Northwest Northeast Population 7462644 23999345 38678000 1446819 29992955 77899816 Literacy (%)
Average Male Female
35.25 39.67 30.76
32.08 39.7 25.3
33.05 39.8 26.5
33.1 40.0 25.8
32.4 38.9 25.5
27.9 33.7 21.8
Main Occupation Agriculture Agriculture labor Commerce Service
Agriculture Agriculture labor Commerce Service
Agriculture Agriculture labor Fishing Service
Agriculture Agriculture labor Commerce Service
Agriculture Agriculture labor Commerce Service
Agriculture Agriculture labor Wage labor Commerce
Source: http://banglapedia.search.com.bd/HT/R_0080.HTM
14 Districts included in the six divisions are as follows: Barisal Division: Barguna , Barisal, Bhola, Jhalkathi, Patuakhali and Pirojpur. Chittagong Division: Bandarban, Brahmanbaria, Chandpur, Chittagong, Comilla, Cox's Bazar, Feni, Khagrachari, Lakshmipur, Noakhali and Rangamati. Dhaka Division: Dhaka, Faridpur, Gazipur, Gopalganj, Jamalpur, Kishoreganj, Madaripur, Manikganj, Munshiganj, Mymensingh, Narayanganj, Narsingdi, Netrakona, Rajbari, Shariatpur, Sherpur and Tangail. Khulna Division: Bagerhat, Chuadanga, Jessore, Jhenaidah, Khulna, Kushtia, Magura, Meherpur, Narail and Satkhira. Rajshahi Division: Bogra, Dinajpur, Gaibandha, Jaipurhat, Kurigram, Lalmonirhat, Naogaon, Natore, Nawabganj, Nilphamari, Pabna, Panchagarh, Rajshahi, Rangpur, Sirajganj and Thakurgaon. Sylhet Division: Habiganj, Moulvibazar, Sunamganj and Sylhet. 15 Information in this section is primarily taken from Wikipedia: The Free Encyclopedia website available at http://en.wikipedia.org/wiki/Main_Page.
49
Figure 3.3: Administrative Divisions of Bangladesh
Source: http://en.wikipedia.org/wiki/Image:Bangladesh_divisions_english.png
50
3.2 Data Characteristics
This dissertation uses the International Food Policy Research Institute’s Food
Management and Research Support Project (IFPRI-FMRSP) household survey of
Bangladesh for the years 1998-99. The households were interviewed in three waves
including approximately 750 households in seven flood-affected thanas (administrative
units). The purpose of collecting the household data was to study the impact of the 1998
floods in rural Bangladesh on food security, employment and household coping strategies
(del Ninno 2001). The first round of the survey was administered between the 3rd week
of November to the 3rd week of December 1998, and the second round between April and
May 1999. Finally, the third round of the survey was conducted exactly a year after the
first round (November-December 1999).
Table 3.3: Timing of the Rounds Time Seasons16 Round 1 November-December 1998 Aman harvest Round 2 April-May 1999 Boro harvest Round 3 November-December 1999 Aman harvest
The fact that these data were collected immediately after the 1998 floods makes
the data set unique for analyzing the effects of the flood as a shock event. Instruments
used for collecting the data were household-level and community-level questionnaires.
Thana (townships), union and village-level information was collected as subcategories
16 The three crops of rice annually produced in Bangladesh are aman (transplanted in June-July and harvested in November-December), boro (transplanted in December-January and harvested in May-June) and aus (transplanted in March-April and harvested in July-August). Aman is the major monsoon season rice crop (Ninno and Dorosh 2003).
51
within the community-level questionnaire17. Detailed household-level data were
collected on the pattern of household expenditures, the pattern of land use at the plot level,
the household’s participation in the rural labor market, the ownership and loss of assets,
borrowing strategies used by the household and anthropometric measures at the
individual level.
In addition, retrospective questions on situations before and during the flood were
asked. The community-level questionnaire focused on agricultural production, local
labor market conditions and other economic conditions at the union level and at village
level, during and after the flood (del Ninno 2001). Table 3.4 gives a brief description of
the information collected using the household and community-level questionnaires.
17 Among thana, union and villages, village is the smallest administrative unit. Several villages come together to form a union and unions form thanas. http://www.mofabd.org/glimpse_of_bangladesh.htm
52
Table 3.4: Summary of the Content of the Household and Community-level Questionnaire Household Level Information Collected 1. Household information Age, gender, civil status, time of absence from the household, individual
sending or receiving money for support. 2. Education Education level for all individuals age 6 and older, dropout, and development
programs running with the school. 3. Status and history of employment, job search, training and public works
Limited to all household members age 10 and over. Labor participation, the main type of work and the reason for not participating, job search strategy and the attitude towards accepting a job (willingness to relocate and minimum wage), the history of employment held before the current employment, training and public works and questions related to the number of weeks spent in public works and job training for each year since 1995.
4. Dependent job18, permanent and daily labor
Primary and secondary dependent job: type of job, industry, time allocated, type of contract, salary and benefits three different time frames.
5. Casual jobs, daily labor Time spent, tasks, wage rates etc. of causal jobs for three time periods. 6. Non-ag self-employment, business activities
Cottage activities, non-agricultural self-employment information.
7. Agricultural activity, access to agricultural land, production and allocation of production
Agricultural production, availability of agricultural land, agricultural assets and livestock, number of weeks worked during the past year and the hours worked last week, access (for each of the past four years) and type and acquisitions of agricultural land (orchard, pastures and cropland).
8. Fishing activity and livestock
Management of ponds and fishing activities and type and number of livestock available and the production of animal products derived from them.
9. Family labor allocation Allocation of family labor among the alternative agricultural activities 10. Social assistance, availability of benefits
Level and the number of months several benefits received, currently and in the last three years.
11. Household furniture and durables
The number of items, the current value and the year of acquisition as well the time and reason for disposal.
12. Credit Amount of credit received, the interest rate and the repayment. 13. Housing and sanitation Quality of the dwelling and the rent paid and monthly expenses. 14. Regular and occasional non-food spending
Non-food expenditures include regular non-food spending for the past month and occasional non-food spending that occurred in the past 12 months.
15. Food expenditure and consumption
Consumption of food at home and away from home, all the items that have been consumed during the last month; quantities consumed from purchases, own production and received from other sources are listed along with the purchase value, if quantities are not known, and current price.
16. Health status
Health status includes type of disability and treatment for chronic illness cost and type of consultation for acute illness that occurred in the past 4 weeks.
17. Anthropometry
Height and weight have been collected for all children below 10 years of age and all females between the ages of 13 and 45.
Community Level Round 1, Nov-Dec 1998 Thana Union Village Round 2, Apr- May 1999 Union Round 3, Nov-Dec 1999 Thana Union Village
Agricultural production 1995 to 1998 Information about the 1998 flood, prices and other characteristics Mostly labor data Labor, NGO programs, prices, rainfall, program intervention, daily wages Intervention programs at thana level Data on program intervention Labor, prices, cost of farming, time of crops, start and receding time of flood water per year (1997-1999), economic activity, law and order, food intervention programs and NGO programs
Source: del Ninno (2001).
18 Dependent job is defined as a job performed on a regular basis for somebody
53
3.2.1 Sampling Procedure
Regions selected for administering the surveys were a fair representation of flood-
affected areas in Bangladesh. Three main criteria were used to select the seven thanas
(Derai, Madaripur, Mohammadpur, Muladi, Saturia, Shahrasti and Shibpur). First, the
depth of the water determined the severity of flooding. The Bangladesh Water
Development Board classified thanas as “not affected,” “moderately affected,” and
“severely affected,”. Second, the level of poverty was used, with thanas with more than
70 percent of the population being classified as poor (del Ninno et al. 2001). Finally,
from the thanas selected based on the first two criteria, selection was made of those
thanas that were included in other studies and that provided a good regional and
geographical balance across the six administrative divisions of Bangladesh (del Ninno
2001). For a list of selected thanas, see Table 3.5.
Table 3.5: Selected Thanas Non-poor Thanas Poor Thanas Total Severely Affected Muladi, Barisal District (Barisal)
Shibpur, Narsingdi District (Dhaka) Mohammadpur, Magura District (Khulna), Saturia, Manikganj District (Dhaka)
4
Moderately Affected Shahrasti, Chandpur District (Chittagong)
Madaripur, Madaripur District (Dhaka) Derai, Sunamganj District (Sylhet)
3
Total 3 4 7 Source: del Ninno et al. (2001)
A multiple-stage probability sampling technique was used to randomly choose the
households to be included in the survey. Three unions were taken from each thana and
six villages were selected from each union, with a probability proportional to the
population in each village. Two clusters in each village were selected using preassigned
random numbers and, finally, three households were chosen from each cluster using a
54
systematic random selection process. Information at the union level was collected using
the community questionnaire, and a separate village-level questionnaire was used to
collect information about rural labor markets during November and December 1998 in 64
villages (del Ninno et al. 2001).
A total of 757 households were interviewed in the first round, 7 households either
refused to be interviewed or were absent at the time of the second round, and 23
households were missing in the third round. Separate male and female questionnaires
were administered, where men were asked about labor and agriculture and women were
questioned about food purchase and allocation and intake of food (del Ninno 2001).
3.3 Research Trip to Bangladesh, February 2005
A qualitative study was also undertaken in Bangladesh in February 2005 as part
of this research. The study was undertaken to gain a better understanding of the social
and cultural context of the research questions (objectives) being addressed in this
dissertation. Attempts were also made to acquire first-hand experience and perspectives
on the traditional ways of life of the people in Bangladesh. The quantitative data from
IFPRI fail to capture the qualitative context and perspectives. The experience gained
from this trip provided a more complete and better interpretation of the quantitative
results. The research conducted involved focus group interviews with women in rural
Bangladesh held over a period of two weeks. Given the time and resources constraints,
the focus groups were comprised only of women. Rural women were mainly engaged in
household work and looking after the livestock within the homestead which made it
easier to access them. A set of broad questions about their experiences during and after
55
the 1998 Bangladesh floods were given to a group to discuss. Some examples of
questions asked in each group included: (1) Do they remember the 1998 Bangladesh
floods? (2) How were they affected by the floods? (3) Were they displaced and when
did they return to their community or village? (4) Did the floods change their
relationship within the household and with their spouse?19
Each focus group consisted of 7 to 9 participants. Care was taken to recruit
participants. In particular, those recruited were women in the 18 to 45 age group, were
married and living with their spouse, and had experienced the 1998 floods. Only one
woman per household was asked to participate in a focus group. Monetary and time
constraints forced us to choose villages in districts close to Dhaka. Districts that were
chosen were Manikgonj, Norshindi and Madripur. They were, respectively, 62
kilometers, 58 kilometers and 52 kilometers from Dhaka. A total of 6 focus group
interviews were conducted, two in each of the three districts.
All participants agreed that floods are a major problem – they were severely
affected by this calamity. Villages closer to the river-bed faced the additional problem of
losing their land to the river. This forced them to leave their homesteads and migrate to
19 Some of the specific questions progressed as follows: How are you? How was your day? Do you have children? How many? Where are they now? Do they go to school? Have you gone to school? Does the village have a school? Do you think floods are a problem? Do you remember the floods last year? Do you remember the floods of 1998? What was your experience? Do you have any local name for the flood? How were you affected by the floods? Did you have to move out of your home or village? When did you get back? How did you get back? Did you get any help from the government? Did you get any help from the NGOs? What is particularly important for you during the floods: children, food or sickness? Do you make any decisions in the household? Do women in your household make any decisions? What kind of decisions do you make? What kind of work do you do in your house? Does your husband help you in household activities? Do you work with your husband? Is there a change during the floods? Do you work more in the house or outside? Do children help you? Are NGOs active in your area? As time passes does their activity decrease or increase? Does the situation get worse as the time passes after the floods?
56
other villages. After losing their land and house, they had additional expenditures related
to renting a house and setting up a new household.
Flooding often brings 3 to 4 feet of water inside the house. The focus group
discussions revealed that during the time of the floods, households went to the high road
for shelter for around two months. They used plastic and bamboo for shelter and leaves
of the banana tree were used as the means of transportation in the water. Alternatively,
households made dais/platforms inside their house and lived there for a couple months
until the water receded. These households did not move out of their house or village.
Some of the households had the option of leaving and living with their relatives in nearby
villages.
Participants generally responded that they take few precautions to prepare for the
flood. Some of the precautionary measures included saving some money, food and
firewood. Many sold their hens, goats and cows during the floods in order to survive.
Many households took monetary loans from wealthier households at high interest rates.
Another common means of coping was to take just one meal per day. Flood water
damaged their houses and sometimes their crops, livestock and fish pond were also
affected.
Some of the women participating in the focus groups were members of the local
NGO, BRAC (Bangladesh Rural Advancement Committee). Some of these borrowing
women reported investing in their husbands’ businesses and a portion of it was spent on
the children and family as well. They also reported not getting any help from NGOs
although some received government assistance in the form of medicines, bread, wheat
and biscuits. However, assistance was not consistently available to all at all times.
57
Another factor that made relief work difficult was the water-hyacinth. The water-
hyacinth made it difficult for boats to move in water and relief workers were not able to
reach some of the households. However, it was also reported that over time NGO
activities are increasing in the local villages.
Among the many problems faced by flood-affected families, health and security
of children is of primary concern. Food security issues are important as well. The
women participants also found resuming normal life to be very difficult. They reported
that getting back to the life they had before the floods takes at least three or four months,
on average. Participants were asked about the role of their spouse in decision-making
within the households. Many of the women admitted that decisions are made jointly --
that their husbands always consulted them.
It became clear that rural households are very poor to begin with and floods push
them further into poverty. They are forced to develop their own coping mechanisms
which include taking loans, selling livestock and eating just one meal a day. Floods of
various magnitudes occur yearly in Bangladesh and households are faced with difficult
conditions on a yearly basis. Long-term solutions are required to alleviate the poverty
and related problems in rural Bangladesh. Within the household, bargaining between
husband and wife was not apparent.
3.4 Data Description
Table 3.5 presents selected demographic characteristics of the household in our
sample, based on round 1 survey data. The average household size is 5.43 persons per
household with a dependency ratio of 1.12. Given that the data cover rural households,
58
agriculture is the dominant occupation where 49 percent of the household heads are
employed. More than 55 percent of the sample households are employed in agriculture
and related activities. The trade (11 percent), industrial (11 percent) and service sectors
(7 percent) are some of the other important sectors, although nowhere close to the
farming sector. Nearly 6 percent of the household heads are out of the labor force or
unemployed. Consistent with the low literacy level in rural Bangladesh, 57 percent of the
household heads are not educated and only 8 percent of the households have heads who
are educated beyond 9th grade. These figures correspond to the national averages
published in country reports.
Table 3.6 also shows the land and buildings owned by the household. As pointed
out in Chapter 2, asset accumulation is an important indicator of well-being and
vulnerability level of a family. The average household in our data owns 1.33 acres of
land and 1.87 buildings. From the data we find that more than 95 percent of the
households owned some land (not in the table). There is not much variation over the
three rounds. This is because land ownership is the most important type of asset holding
in rural economies. Availability of amenities including sanitary latrine, clean water,
material used in household construction and method of garbage disposal are good
indicators of a household’s socio-economic well-being. Sanitary latrines are sealed
toilets which ensure waste is not spread in the surrounding area. Availability of clean
water for household usage has important consequences for the spread of diseases. This is
doubly important in flood situations where the population is already vulnerable to
infections. Only 23.92 percent of the households have access to a sanitary latrine, 14
percent have electricity at home and 21 percent use tin or concrete for house construction.
59
In total, 59 and 41 percent of the families have a fixed place to dispose garbage and use
tube wells for washing purposes, respectively. These amenities are indicators of lack of
development and limited resources available to households in rural Bangladesh.
Table 3.6: Demographic Characteristics of Sample Households in Rural Bangladesh, 1998-99 Characteristicsa Mean/Percentage Household Characteristics
Household size 5.43 Dependency ratio 1.12 Working members 1.55 Location (percentage)
Derai 13.48 Madaripur 14.31 Mohammadpur 14.58 Muladi 14.44 Saturia 14.17 Shibpur 14.86 Shahrasti 14.17
Physical Assets
Land owned (acres) 1.33
Household Head Characteristics Age 45.11 Education
No education 56.83 Less than 5th grade 13.09 5th grade 9.21 6th to 9th grade 13.09 Higher than 9th grade 7.77
Primary occupation
On-farm agricultural work 49.03 Off-farm agricultural work 5.52 Industrial sector 11.46 Trade sector 11.19 Transportation sector 5.52 Construction sector 3.31 Self-employed and service sector 1.38 Miscellaneous services 7.04 Out of the labor force 5.52
Socio-economic characteristics Number of buildings owned 1.87 Access to sanitary latrine 23.92 Access to fixed garbage disposal 59.44 Access to electricity 14.86 Tin and concrete wall material used 21.25 Use of tube well water for washing 41.16
Note: Based on own calculations using round 1 from the IFPRI-FMRSP Bangladesh data 1998-99.
60
We see in Table 3.7 that, including both formal and informal credit, nearly 63
percent of the households have received credit and around 40 percent of the households
received social assistance from Test Relief, Gratuitous Relief, Vulnerable Group
Development, Vulnerable Group Feeding and Food For Work (FFW) schemes. This
social assistance included efforts that began immediately after the floods. Initially,
households were given food and income as part of the relief program and later resources
were channeled towards rebuilding infrastructure and providing agricultural credit (del
Ninno 2001). Table 3.7 also presents these assets at a disaggregated round level. It
should be noted that credit receipts and social assistance are highest in the first round
immediately after the floods. Among other financial fall-backs, around 12 percent of the
households received remittances (Table 3.7). Remittances are lowest in the first round
and steadily increase by round 3. This could be because help from family and friends
may not be immediate as in case of government and NGO relief. Table 3.7 establishes
that the majority of households engaged in borrowing activities and depended on
transfers.
Table 3.7: Financial Asset Ownership of Sample Households in Rural Bangladesh, 1998-99 Percentage Round 1 Round 2 Round 3 Average Credit availability 75.9 60.5 52.1 62.8 Social assistance 51.3 39.8 29.3 40.1 Remittances 9.1 10.4 17.7 12.4
Note: Based on own calculations using round 1 from the IFPRI-FMRSP Bangladesh data 1998-99.
3.4.1 Regional Variations
It is accepted that space matters and policy targeting at the national level has to
take into consideration the regional differences within Bangladesh. Figures 3.4-3.7
61
present consumption levels and poverty measures by region in our sample by each round.
These figures show how poverty levels change over the three rounds of the survey.
Figure 3.4 presents total household consumption levels and Figure 3.5 to 3.7 correspond
to head count, poverty gap ratio and squared poverty gap ratio, respectively.
Figure 3.4: Mean Consumption Levels by Lower Poverty Line
Figure 3.4 shows that, except for in Muladi, consumption levels for flood-affected
households are lower in round 3 than in round 1. Even in Muladi the increase over the
three rounds is very small and compared to other regions consumption expenditure is
low. Muladi is one of the regions severely affected by the flood and not too close to
Dhaka in terms of distance. In Shibpur consumption expenditures are highest in round 2
611.0 643.1
739.8
650.5
682.5
807.7
654.5
629.7
787.2
664.3 663.4 659.2
696.9
629.3
787.1
729.7
812.9
831.4
704.5 834.9
812.3
0 200 400 600 800 0 200 400 600 800
0 200 400 600 800
3
2
1
3
2
1
3
2
1
3
2
1
3
2
1
3
2
1
3
2
1
Derai Madaripur Mohammadpur
Muladi Saturia Shahrasti
Shibpur
Mean Consumption (Taka)Graphs by Region
62
and it has one of highest consumption expenditure levels. Its located in the Dhaka
division could have played a role in keeping the expenditure level steady despite being
severely affected by flooding.
In Derai, Madaripur and Shahrasti, total household consumption falls gradually;
that is, consumption in round 2 is lower than in round 1 and consumption in round 3 is
lower than round 2 (see Figure 3.4). We know from Chapter 2, that Derai and Madaripur
are poor thanas and were moderately affected by the floods. Also, the fact that Derai is
located in the Sylhet division is furthest away from the capital city does not help. With
the relief effort centering in the capital city, this regions’ access to resources and
opportunities including getting attention of the government could be limited.
Mohammadpur and Saturia which are poor thanas have households faring better in round
3 than in round 2.
In terms of number of people below the poverty line, Figure 3.5 indicates that
Derai and Muladi start with large numbers of poor and continue to have high populations
of poor over the survey period. In round 1, Derai has a head count poverty rate of 44
percent which increases to 57 percent by the end of the survey period. The fact that
households were worse-off in the third round is reiterated even in this figure.
63
Figure 3.5: Head Count calculated using the Lower Poverty Line
Depth and severity of poverty is shown in figure 3.6 and 3.7. Among all the
regions, Shahrasti has the lowest depth and severity of poverty and Derai is worst off.
Good performance of Shahrasti could be attributed to being a non-poor thana and not
being severely affected by the flood. This indicates that initial conditions of the region
matter. Shahrasti is located is Chittagong region that is not as affected to the same extent
by yearly floods (see Figure 3.2). As observed earlier, the situation in Mohammadpur
worsens in the second round only to improve in round 3. As will be seen in subsequent
analysis, these regional variations have an important effect on the calculation of poverty
outcomes and in our analysis of the impacts of credit receipt on household expenditure.
57.14 53.06
43.88
43.27
40.38
33.65
42.45 45.28
31.13
51.43 44.76 45.71
45.79
46.73
30.84
34.26 22.22 23.15
40.78 36.89 37.86
0 20 40 60 0 20 40 60
0 20 40 60
3
2
1
3
2
1
3
2
1
3
2
1
3
2
1
3
2
1
3
2
1
Derai Madaripur Mohammadpur
Muladi Saturia Shahrasti
Shibpur
Head Count Rate Graphs by Region
64
Figure 3.6: Poverty Gap Ratio Calculated using the Lower Poverty Line
In the empirical models, we not only study the characteristics of poor but also
assess the impacts of amount of credit received by the head and the spouse on household
expenditures. Table 3.8 shows that the average value of the amount borrowed by the
head is higher than that for the spouse. There are regional differences as well. Women in
the Derai region have the lowest average amount of credit available. This region shows
high levels of poverty during the time of the surveys. Saturia and Shibpur are the two
high credit regions with respect to women. The poverty figures above show that these
regions show lower levels of poverty as well. This is especially true for Shibpur. This
18.58 13.63
12.81
13.23
8.91
6.73
11.65 13.25
7.01
13.57 11.53
12.72
11.75
11.27
5.97
8.35 6.32
4.53
12.66 9.60
11.44
0 5 10 15 20 0 5 10 15 20
0 5 10 15 20
3
2
1
3
2
1
3
2
1
3
2
1
3
2
1
3
2
1
3
2
1
Derai Madaripur Mohammadpur
Muladi Saturia Shahrasti
Shibpur
Poverty Gap IndexGraphs by Region
65
could reflect an association between poverty and credit being made available to women
in rural households.
Figure 3.7: Squared Poverty Gap Ratio calculated using the Lower Poverty Line
Table 3.8: Amount of Credit Taken by Gender and Region Head Spouse N All 5971.90 1056.59 673 Regions
Derai 4887.89 291.11 90 Madaripur 3259.39 1090.69 102 Mohammadpur 6275.15 919.90 102 Muladi 10239.74 531.52 103 Saturia 5000.77 3351.91 84 Shibpur 4286.60 1190.72 97 Shahrasti 7537.90 294.74 95
Note: Based on own calculations from the IFPRI-FMRSP Bangladesh data 1998-99.
7.62 4.75
5.32
5.26
2.80
1.93
4.62 5.80
2.57
4.66 4.00
5.28
4.34
4.20
1.64
2.84 2.15
1.38
5.27 4.11
4.76
0 2 4 6 8 0 2 4 6 8
0 2 4 6 8
3
2
1
3
2
1
3
2
1
3
2
1
3
2
1
3
2
1
3
2
1
Derai Madaripur Mohammadpur
Muladi Saturia Shahrasti
Shibpur
Squared Poverty Gap IndexGraphs by Region
66
Chapter 4
Methods
4.1 Poverty Dynamics
The poverty spell approach and the components approach are two approaches
used in modeling poverty dynamics (McKay and Lawson 2002). The first approach also
called the duration model takes into account the duration of completed poverty spells
(Bane and Ellwood 1983). The second approach focuses on the poverty status of the poor
and distinguishes the chronic from the transient poor (McKay and Lawson 2002).
Parametric and nonparametric techniques have been widely used to study poverty
dynamics.
The descriptive analysis in this study includes testing for changes in the
distribution of poverty over the rounds of the survey using the cumulative distribution of
consumption. This study performs the first-order stochastic dominance test. The
performance of the independent variables selected for the poverty analysis in predicting
the probability of being poor is tested. This is done using a useful but not commonly
applied tool in poverty analysis called Relative Operating Characteristics (ROC) analysis.
The rural poor of Bangladesh are then categorized into chronic and transient poor using a
short panel of household data. Households have been found to adjust to shocks such as
floods by borrowing, selling assets and altering expenditures (del Ninno et al. 2001).
Attempts are made to identify the determinants of poverty in Bangladesh, with a specific
focus on differentiating those who experienced poverty following the 1998 flood in
Bangladesh and those who did not, and then to determine differences between those
67
among the poor who are able to eventually escape poverty following the flood (the
transient poor) versus those unable to leave poverty (the chronic poor).
Negative income shocks can pull the non-poor into poverty. Vulnerable
populations can experience declines in their living standards and suddenly fall into
extreme poverty (Jalan and Ravallion 2000). We define chronic and transitory poverty as
defined by Jalan and Ravallion (2000). They define a household to be transient poor if
(and only if) the household is observed to be poor for at least one date and if its standard
of living varies over time. They focus on inter-temporal variability in living standards of
poverty. Given resource constraints and yearly floods, the question becomes if there are
factors distinguishing between these two types of poverty in Bangladesh.
4.2 Measurement of Transient and Chronic Poverty
4.2.1 Approach I
Following McCulloch and Baulch (1999), households are categorized into three
mutually exclusive groups: never poor, chronically poor and transitory poor.
Never poor if yit > z for all time periods
Chronically poor if Et[yit]< z
Transitory poor if Et[yit]> z but yit < z for some time periods
where Et[yit] is mean income over the time period and z is the poverty line.
A household is chronically poor if its mean expenditure is below the poverty line across
all periods and transitory poor if its mean expenditure is above the poverty line but total
68
per capita household expenditure is not above the poverty line for all periods. Household
expenditure of a chronically poor household may be above the poverty line in a particular
round but not high enough to pull mean expenditure above the poverty line. Households
whose total household expenditure is above the poverty line in all rounds are never poor.
4.2.1.1 Estimation Technique
Given the nominal categorization of the poor, a multinomial logit model will first
be estimated to study the determinants of chronic and transient poverty and to assess how
these two types of poverty differ from each other. Although the dependent variable is
defined using per capita household expenditure across all three rounds, we will use the
values of the independent variables in the initial time period (round 1) for the analysis.
This is done because most of the independent variables used in the analysis are time
invariant. However, the financial-asset variables are used in the analysis vary over time.
The households have three alternative outcomes (not in poverty, chronic poverty
and transient poverty), where the states are numbered 1 to 3. Pr (y = m|x) is the
probability of observing m given x (Long 1997)20.
3,2,1,)exp(
)exp()|Pr( 3
1
===
∑=
jx
xxmy
jji
miii
β
β [4.1]
20 It is assumed that Pr (y = m|x) is a function of linear combination of xβm and the vector βm includes the intercept term and the coefficients estimated for the effect of the independent variables. Probabilities are non-negatives and sum to one. This is ensured by taking exponential of xβm and the )exp( mix β is
divided by sum ∑=
3
1)exp(
jjix β .
69
where yi is the dependent variable, x is the vector of individual, household and
community variables, and β is a vector of coefficients. The maximum likelihood estimate
(Long 1997) for a multinomial logit model is:
∏∑
∏=
==
=my
J
j ji
miJ
mj
i x
xXyL
11
1)exp(
)exp(),|,.....,(
β
βββ [4.2]
where the product over all outcomes is taken and the log of equation 4.2 gives the log
likelihood equation of a multinomial logit model. The maximization of the log of
equation 4.2 would give the required estimates of β. Independence of Irrelevant
Alternatives (IIA) is an important assumption where relative probabilities of any two
outcomes are not dependent on any other factor or alternatives other than their own
attributes (Wooldridge 2002). It is key that multinomial models be used only when the
alternatives are tested and found to be independent (Long 1997). Before estimating the
model, Hausman test of IIA is applied21.
4.2.2 Approach II
Following the Jalan and Ravallion (2000) approach, the ith household’s
consumption stream over the D time periods (three in this case) is defined as (yi1,
yi2,...,yiD). The yit is the welfare measure such as consumption expenditure of the
household which is corrected for differences in demographics and prices. The aggregate
inter-temporal poverty measure for household i is P (yi1, yi2,...,yiD). The Foster-Greer-
Thorbecke class of poverty measure could be used for the calculation of aggregate inter-
temporal poverty. Chronic poverty (Ci) is:
Ci= P (Eyi, Eyi, ..., Eyi) [4.3] 21 See Long (1997) for detailed discussion on Hausman test of IIA.
70
where Eyi is the expected value of consumption over time of household i or the time-
mean consumption for all dates. Transient poverty (Ti) is a component of total inter-
temporal poverty measure (P (.)), which takes into account the inter-temporal variability
in consumption. That is,
Ti = P (yi1, yi2,...,yiD) - P (Eyi, Eyi, ..., Eyi) [4.4]
Jalan and Ravallion (2000) require the poverty measure to be additive over time
and across households. Individual poverty function p(yit) is assumed to be the same for
all households and dates22. It is convex and decreasing up to the poverty line. At the
poverty line, it is zero and vanishes smoothly as the poverty line is approached from
below. This measure penalizes inequality among the poor. The SPG index satisfies these
conditions. It takes the value one if the household is at the poverty line and zero if the
consumption/income is above the poverty line. The squared poverty gap index is defined
as:
p(yit) = (1 - yit)2 if yit < 1
= 0 otherwise [4.5]
This index is evaluated at the household level and averaged across households to
derive the aggregate (total) SPG index. The poverty measure is weighted by household
size and yit is normalized by the household-specific poverty line. Household total poverty
is calculated as the expectation over time of the poverty measure at each point of time
p(yit) McCulloch and Baulch (1999). That is,
∑=
=T
tii p
TP
1
1
[4.6]
where pit is
22 To make this assumption reasonable, appropriate deflators for consumption could be used.
71
0
2
zyif
zyifzyz
p
it
itit
it
≥=
<⎟⎠⎞
⎜⎝⎛ −
= [4.7]
Household chronic poverty does not change over time and can be written as the
expectation (mean) over time of the household’s chronic poverty at each point in time.
That is,
][ 0
][ ][ 2
zyEif
zyEifz
yEzc
itt
ittitt
it
≥=
<⎟⎠⎞
⎜⎝⎛ −
= [4.8]
The households in the IFPRI data set for Bangladesh are divided into three
mutually exclusive groups. The first group consists of households that are persistently
poor. They are poor on every date and include households whose mean consumption is
always below the poverty line. The second group consists of households with mean
consumption below the poverty line but not poor in all periods. The last category
consists of those households whose mean consumption is above the poverty line but are
poor only sometimes.
4.2.2.1 Estimation Technique
We follow Jalan and Ravallion (2000) and Haddad and Ahmed (2003) and use
semi-parametric methods of estimation. They specify their transient poverty models as
follows:
Ti = Ti* if Ti
*>0 where Ti* = xi βT + ui
T [4.9]
= 0 otherwise
72
where Ti* is the latent variable, Ti is the observed transient poverty, βT is the vector of
unknown parameters, xi is the vector of explanatory variables and uTi is the error term.
The chronic poverty model to be estimated is
Ci = Ci* if Ci*>0 where Ci
* = xi βC + uiC [4.10]
= 0 otherwise
where Ci* is the latent variable, Ci is the observed transient poverty, βC is the vector of
unknown parameters, xi is the vector of explanatory variables and ui is the error term.
As in calculation of any poverty measure if a household is not poor then the
measure is taken to be zero. The households which are non-poor generate ‘bunches’ of
zero values in the dependent variables in poverty equations 4.9 and 4.10 and thus the
need to use a censored model arises. These two equations are estimated separately.
Tobit models are a commonly applied censored technique. However, Jalan and
Ravallion (2000) argue that Tobit estimates are not robust to mis-specification of the
error distribution and recommend the use of semi-parametric methods such as the
Censored Quantile Regression (CQR) model. The error term in the poverty equations
may not normally distributed and Probit and Tobit models would give inconsistent
estimates. Also heterogeneity of the households because of their characteristics makes
the error term heteroscedastic. In this dissertation, after the estimation of multinomial
logit models, we will also use Censored Quantile Regression techniques to study the
determinants of total, chronic and transient poverty. CQR models are robust to
heteroscedastic and non-normality assumptions (Muller 1997). The same set of
independent variables (geographic and household characteristics) are used to estimate
poverty types.
73
The censored quantile function for transient poverty is (Muller 2002):
)],0max([1);( '
1
ti
N
in xT
NQ βρθβ θ −= ∑
=
[4.11]
This function is minimized over all β, and ρθ is the weighting function which centers the
data depending on the quantile (θ). N is the sample size. Quantile is defined as the
solution to the minimization of equation 4.11. The censored quantile function for chronic
poverty is:
)],0max([1);( '
1
ti
N
in xC
NQ βρθβ θ −= ∑
=
[4.12]
where
||)]0).(1()01(.[)( λλθθλρθ <−+≥= I [4.13]
In the weighting function, I(.) is the indicator function and a similar model is also
estimated for the chronic poverty equation. Quantile regression methods are also robust
to outliers which are common in poverty analysis. Outliers can be outcomes of
measurement errors common in consumption surveys. Choice of quantile (θ) helps by
focusing on the poor in the data in order to identify their characteristics and can be thus
chosen to reduce the role of those above poverty. Muller (2002) recommends
simultaneous estimation of the two poverty equations (4.11 and 4.12) but simultaneous
regression techniques for censored quantile regression is not available at this point.
4.2.3 Distinguishing Between Approaches
The Jalan and Ravallion (2000) approach identifies the correlates of each kind of
poverty and helps in the identification of policies that are likely to reduce chronic and
transient poverty. On the other hand, the McCulloch and Baulch (1999) approach does
74
not differentiate between the chronically and transient poor but shows how these poverty-
ridden households are different from households that have never been poor (McCulloch
and Baulch 1999).
4.3 Poverty Lines
The present study will also use cost-of-basic-needs (CBN) poverty lines
calculated by the World Bank for Bangladesh for the year 200023. Lower and upper
poverty lines were calculated by dividing the country into 14 geographical regions (nine
urban and five rural) 24. The lower poverty line allows for only minimum allowance for
non-food goods as opposed to the upper poverty line where greater allowance is made for
non-food goods in the calculation of the poverty line (World Bank 2002b).
Real total per capita household expenditure per month is the welfare indicator
used in this study. The data used for the analysis were collected from the following
seven regions: Derai, Madaripur, Mohammadpur, Muladi, Saturia, Shahrasti and Shibpur.
Each of these areas belonged to one of the regions for which the poverty line for 2000
was available. CBN poverty lines are available for the year 2000 and the data being used
in this research were collected in 1998. Therefore, the poverty lines need to be corrected
for price changes over the period 1998 to 2000 using the consumer price index. Table 3.4
gives the region-specific poverty line corrected for changes in prices for the year 1998
23 The poverty lines were calculated by analyzing selected Household Expenditure Surveys (HES) conducted by the Bangladesh Bureau of Statistics (BBS) for the 1990s (World Bank 2002b). 24 The 14 regions used comprised: 1. Dhaka SMA, 2. Other urban areas of Dhaka division, 3. Rural areas of Dhaka and Mymensingh, 4. Rural areas of Faridpur, Tangail, and Jamalpur, 5. Chittagong SMA, 6. Other urban areas of Chittagong division, 7. Rural areas of Sylhet and Comilla, 8. Rural areas of Noakhali and Chittagong, 9. Urban areas of Khulna division, 10. Rural areas of Barishal and Pathuakali, 11. Rural areas of Khulna, Jessore, and Kushtia, 12. Urban areas of Rajshahi, 13. Rural areas of Rajshahi and Pabna, and 14. Rural areas of Bogra, Rangpur, and Dinajpur greater districts.
75
using both upper and lower poverty lines. Once the poverty line is determined, poverty
measures such as head count ratio, poverty gap and squared poverty gap will be
calculated as measures of the extent of deprivation.
Table 4.1: CBN Region Poverty Lines Region Lower
Poverty Line (2000)
Corrected for Price Change
Upper Poverty Line
(2000)
Corrected for Price Change
Rural Dhaka 548 503.73 659 605.76 Rural Sylhet Comilla 572 525.79 738 678.38 Rural Noakhali Chitagong 582 534.98 719 660.91 Rural Barishal Pathuakali 546 501.89 616 566.23 Rural Khulna Jessore Kushtia 527 484.42 624 573.59
Source: BBS and World Bank staff estimates. Amounts are in Tk.(taka) per person per month.
4.4 Household Expenditure and Credit
Provision of a safety net should include availability of credit to the rural
households. Despite the presence of microcredit institutions in Bangladesh, informal
credit is one of the most important consumption-smoothing mechanisms. The goal in the
previous section is to identify the poor and assess if they are indeed heterogeneous. The
poverty literature finds women to be one of the most vulnerable groups and their
empowerment and emancipation is a vital goal in the poverty reduction strategy. Given
that the IFPRI-FMRSP survey on Bangladesh included too few female-headed
households, we were not able to conduct our analysis on the basis of gender of household
headship. Since informal credit was found to as important as the formal, we do not
distinguish between the two in our analysis. We hypothesize that whatever the channel
of resource transfer to women, there will be welfare implications for the household. This,
76
however, does not discredit the efforts of microcredit institutions that target women in
particular.
In particular, the question becomes does credit receipt by men and women in a
family have any effect on the kind of goods purchased and consumed in a household and
any implication for poverty reduction by altering consumption expenditure. As discussed
in Chapter 2, we want to test if resource allocation in favor of women improves
nutritional, child or other household outcomes.
4.4.1 Fixed Versus Random Effects
Panel data techniques can be utilized because the IFPRI-FMRSP survey were of
the same set of 750 households over the three rounds. It is a short panel data comprising
of three round data collected over the period of one year. Use of panel data increases the
degrees of freedom and reduces collinearity among independent variables (Hsiao 2003).
We get efficient estimates when panel data are used. It also lends towards modeling
differences in behavior across individuals (Greene 2000). The classic regression equation
of panel data is as follows (Greene 2000):
ititiit exy ++= 'βα [4.14]
Excluding the constant, there are K regressors in xit. The αi is the individual effect which
is constant over time t and corresponds to individual cross-section unit i. Fixed and
random effects models are the two variations of the panel data estimation approach. The
difference between the two variations is in the treatment of αi. The fixed effect models
take αi to be a group-specific constant term. These models take the individual effect to be
fixed and different across individuals. Random effect models treat αi like an error term,
77
i.e., as a group-specific disturbance and specific effects are treated as random. The
constant term in a fixed effect model captures the cross-sectional differences as follows
and is appropriate to use when unit differences are taken to be parametric shifts of the
regression function:
iiii eXiy ++= βα [4.15]
The assumption made in a fixed effect model is:
TtcxuE iiit ,....,1,0)|( == [4.16]
On the other hand, the random effect model is written as follows:
itiitit euxy +++= 'βα [4.17]
The assumptions made in a random effect model are (Wooldridge 2002):
TtcxuE iiit ,....,1,0)|( == [4.18]
0)()|( == iii cExcE [4.19]
where xi= (xi1, xi2,…., xiT). In this study OLS and Tobit models are estimated using
random effect techniques. As in Swaminathan (2003) and Quisumbing and de la Briere
(2000), the short nature of the panel and use of time-invariant independent variables in
the model suggests the use of random effect models.
4.4.2 Empirical Framework
The data used for the analysis contains detailed household-level data collected on
the pattern of household expenditures, the pattern of land use at the plot level, the
household’s participation in the rural labor market, the ownership and loss of assets,
borrowing strategies used by the household and anthropometric measures at the
individual level (del Ninno 2001). In particular, data were collected on non-food
78
expenditures including regular non-food spending for the past month and occasional non-
food spending that occurred in the past 12 months. Detailed information is available on
consumption of food at home and away from home, quantity purchased, produced and
received from sources in the last one month along with the purchase value, if quantities
are not known, and current price is also known25.
Following Quisumbing and de la Briere (2000) and Swaminathan (2003), this
research considers agricultural households with two members (head and spouse) in the
household and uses the Nash-bargaining model to study the role of credit receipt in
determining bargaining power. The following expenditure function is estimated:
∑ =+++
++
++=
K
1k 16
54
321
cdemographi credit) sspouse'log(
credit) shead'log(area) land household total(
size) household totallog( e)expenditur household totallog(
εδα
αα
ααα
kk
jb
[4.20]
where bj is the expenditure share of the jth good, independent variables include household
demographic variables, location variables and ε is the error term.
Non-food categories used in the analysis are cigarettes/beetel, adult goods,
children's goods, durable goods, education, fuel, health, personal care, housing, travel and
social activities26. For each loan taken by the household, respondents were asked about
the type of loan, amount in both cash and in kind, collateral, interest rate charged, and
repayment details in all three rounds. Member ID was recorded with each loan which
25 Households were asked about roughly 200 foods items they consumed over the three rounds. 26 Adult and children’s goods include clothing and footwear; education category includes school fees, house tutor, boarding fees, books, stationary, education purpose transportation, battery, other educational expenses, electricity and pocket allowance; durable goods include dishes, silverware, pots, lamps, basket/bags and toys; fuel costs include firewood, dried leaves, cow dung, jute leaves, rice bran, straw, matches, kerosene and gas; health expenses include fees for medical care, drugs/medicine, dental fees, lab tests and other treatments; travel expenditure includes rickshaw/van, bus/microbus/minibus, travel to other districts and repairs of bi-cycle/rickshaw; personal care includes bathing soap/shampoo, shaving, tooth powder/brush, hair oils and cosmetics; social events include weddings, funerals, birthdays/ anniversaries, circumcision and cash gifts given and housing includes rent and repairs in the house.
79
made it possible to identify the member in the household who actually took the loan.
Two models will be estimated in the analysis. The first model uses actual loan amount
received by household members and the second model uses a credit dichotomous variable
where if the individual receives credit the variable is coded as 1 and 0 otherwise. We do
not distinguish formal credit from informal credit, as the number of women participating
in credit programs is relatively small in the data precluding disaggregation by type of
credit. Separate equations for food and each of the non-food categories are estimated
controlling for household size, location of household and the round in which data were
collected. Households which are monogamous and whose family structure did not
changed across all three rounds of data collection were identified and included in the
analysis. Given that only 4.38 percent of households were female-headed, we study only
male-headed households. There are 655 such male-headed households.
4.4.3 Econometric Issues
OLS is used to estimate the food share and personal care share equations and
Tobit models are used to estimate the 10 other non-food budget shares. The Tobit model
is used when the dependent variable is zero for a non-trivial fraction of the population
and is otherwise continuously distributed (Wooldridge 2006). Not all households are
found to be spending on all non-food expenditure categories27. Using OLS in such cases
27 In the data, 87.32 percent of the households do not spend on social activities, 10.25 percent don’t spend on cigarettes and beetel (tobacco), 17.34 percent do not spend on adult clothing, 31.05 percent do not spend on children’s clothing, 67.51 percent don’t spend on durable goods, 21.69 percent do not spend on education, 1.29 percent do not have fuel expenditures, 6.49 percent do not have health expenditures, 0.30 percent do not have personal care expenditures, 64.44 percent do not have repair expenses and 45.27 percent do not have travel expenses.
80
would result in inconsistent estimates. The standard Tobit model is written as
(Wooldridge 2002):
),0max(),0(~|,
*
2*
yyNxuuxy
=
+= σβ [4.21]
The conditional log likelihood for the censored Tobit model is:
}2/)log(]/)[]{log0[1)]/(1log[]0[1)( 2σσβφσβθ −−>+Φ−== iiiiii xyyxyl [4.22]
We take into consideration the potential endogeneity of total expenditure with budget
shares. We first test for endogeneity using the regression-based Hausman test as follows
(Wooldridge 2002):
121111 uyzy ++= αδ [4.23]
In equation 4.23 the dependent variable y1, is potentially endogenous with y2. The z1 are
the exogenous variables (includes the constant) and u1 is the error term. Together, z1 and
u1 satisfy the following condition:
0)( 1" =uzE [4.24]
The null hypothesis tested is if y2 is exogenous. First, y2 is regressed linearly on z2 as:
222 vzy += π [4.25]
0)( 2' =vzE [4.26]
errorvyzy +++= '2121111 ραδ [4.27]
Adding the residual (v2) from equation 4.25 to equation 4.27 and obtaining the t-statistic
would indicate the exogeneity of y2. If the coefficient on the residual is significant then
we reject the null hypothesis and y2 is taken to be endogenous. As in Swaminathan
(2003), if the shares are endogenous then we use multiple instruments in two stage least
square estimator (2SLS) and simultaneous-Tobit models. Access to sanitized toilets,
81
material used in building walls of a house, source of water for washing, electricity
availability and availability of fixed garbage disposal techniques are the instruments used
in the 2SLS models. The first stage involves running the following regression to obtain
fitted values for total household expenditure (Wooldridge 2006):
variablescdemographiother y electricit water washingof source material roof garbage toileteexpenditur household Total
654
3210
ααααααα
++++++=
[4.28]
The second stage involves running the share equations on the fitted value and other
independent variables. Endogeneity correction is carried out even in Tobit models
following the Smith-Blundell procedure (Wooldridge 2002). The 2SLS Tobit models
estimated are as follows:
2654
3210
v variablescdemographiother y electricit water washingof source material roof garbage toileteexpenditur household Total
+++++++=
ααααααα
[4.29]
1*232
1
st variableindependenother
eexpenditur houshold total,0max(shares eExpenditur
ev +++
=
αα
α [4.30]
Equation 4.29 is estimated by OLS and the residuals (v2*) are obtained. Next, the Tobit
model is estimated as shown in equation 4.30 substituting v2* in the expenditure share
equation.
After testing for endogeneity, we use the 2SLS models and simultaneous Tobit
models to correct for endogeneity between the expenditure shares and total expenditure.
Where the tests support the null hypothesis of exogeneity, we use the simple random
effect OLS and Tobit models for estimating the share equations. Simultaneous
estimation of the expenditure shares as a system of equations would be most ideal.
However, at this point we are not aware of the regression technique to estimate the
82
system of equations with continuous and censored dependent variable with endogeneity
corrections within a panel data framework.
83
Chapter 5
Poverty Dynamics Results
5.1 Introduction
Analysis of poverty and vulnerability are crucial for understanding the current
well-being of households and for making a good prediction of the future based on present
conditions. Such information helps in formulation of effective poverty reduction policies
and for better targeting of the vulnerable sections of the population. Welfare indicators,
poverty lines and poverty measures are the tools needed for poverty analysis. This study
uses a monetary indicator of well-being. Specifically, per capita total household
consumption expenditure is used. Poverty lines provide the threshold, which distinguish
the poor from the non-poor. The present study uses cost-of-basic-needs (CBN) poverty
lines calculated by the World Bank for Bangladesh for the year 2000.
The standard FGT (Foster, Greer and Thorbecke 1984) poverty indices are used
as poverty measures. Datt et al. (1998) define the poverty headcount as the proportion of
households with consumption below the poverty line. The poverty gap index is the mean
distance between consumption of the population and the poverty line. The squared
poverty gap index takes into account the square of the distance that separates the poor
from the poverty line and places higher weights on households further away from the
poverty line. Headcount index measures the incidence of poverty; poverty gap index
measures the depth as well as incidence of poverty and squared poverty gap index
measures severity of poverty (Datt et al. 1998). The poverty gap and squared poverty
gap measures provide greater insights into poverty.
84
5.2 Descriptive Analysis
5.2.1 Incidence of Poverty
Table 5.1 presents the mean per capita household consumption expenditures and
poverty measures calculated using both lower and upper poverty lines. Table 5.1
indicates statistics from all three rounds of the IFPRI-FMRSP Bangladesh 1998-99
survey. Mean consumption expenditures generally decline between round one and round
three. In round one, 34.83 percent of households were below the poverty line for the
lower poverty line and 48.31 percent were below the upper poverty line. Between round
1 and round 3 there was a 28 percent increase in the number of households below the
lower poverty line and a 24 percent increase in the number of households below the
upper poverty line. By round three, headcount poverty rates increased to 44.66 percent
and 59.97 percent for the lower and upper poverty lines, respectively.
Similar downward trends are observed for the poverty gap and squared poverty
gap indexes. In the second round, there was a 20 percent increase in the poverty gap
index and a 19 percent increase in the squared poverty gap ratio for calculations based on
both lower and upper poverty lines. By round three, the increases in the poverty gap
index and squared poverty gap index are approximately 47 and 50 percent, respectively
(lower poverty line). This suggests that welfare levels of households declined (and
declined significantly) in each subsequent round. Based on these income measures,
households were better off immediately after the floods when their consumption levels
were boosted by government and NGO transfers and help. Since no data were collected
before the flood, we can only hypothesize that consumption levels were artificially
increased in the first round only to fall to normal levels a year after the 1998 floods.
85
Table 5.1: Consumption Expenditure and Poverty Round 1 Round 2 Round 3 Real per capita household expenditure Mean Change over the previous period (%)
774.76
700.90 -9.53
674.26 -12.97
Poverty Lower Poverty Line
Headcount index (%) Change (%) Poverty gap index (%) Change (%)
Squared poverty gap index (%) Change (%)
Upper Poverty Line
Headcount index (%) Change (%) Poverty gap index (%) Change (%)
Squared poverty gap index (%) Change (%)
34.83
8.6
3.2
48.31
14.2
5.8
40.45 16.14
10.3 19.77
3.8
18.75
57.72 19.48
17.0 19.72
6.9
18.97
44.66 28.22
12.6 46.51
4.8
50.00
59.97 24.14
19.5 37.32
8.4
44.83 Note: Based on own calculations from the IFPRI-FMRSP Bangladesh data 1998-99.
Movements into and out of poverty are observed in Table 5.2. The head count
poverty index shows that 18.9 percent of households in Bangladesh (over the three
rounds) are found to be always below the poverty line using the lower poverty line and 25
percent using the upper poverty line. In addition, 41.4 percent of the households move in
and out of poverty. There is not only a high level of chronic poverty but also a high level
of transitory poverty. One out of five households are chronically poor following the
flood event using the lower poverty line and one in four are using the upper poverty line.
This is the case even with NGO and government assistance following the flooding. In
addition, about 40 percent are only temporarily out of poverty, perhaps because of the aid
itself (see Table 5.2 results)
86
Table 5.2: Number of Periods Poor
Number of rounds in which poor Never 1 2 Always Total
% of Households (number)
Lower Poverty Line
Upper Poverty Line
39.6 (282)
33.0 (235)
19.8 (141)
25.0 (178)
21.6
(154)
16.9 (121)
18.9
(135)
25.0 (178)
100 (712)
100 (712)
Note: Number of households in parentheses.
5.2.2. Time-Specific Profile of Poverty
Table 5.3a presents poverty measures for each round by occupation using the
upper poverty line. Overall, household welfare conditions deteriorated from the first to
the third round with all measures showing similar patterns. Households engaged in the
industrial sector are the worst off across all three rounds, followed by unemployed
household heads and those engaged in off-farm agricultural work and the transport sector.
In comparison, the self-employed and those involved in the trade sector and
miscellaneous services are better off. Individuals who are self-employed and employed
in the service sector (employed in NGO) are less likely to be poor than those who are in
the manufacturing, transport and construction sectors (World Bank 2002b). However,
construction activities increased after the floods during the reconstruction phase. Greater
resilience was shown by the agricultural sector after the flood. The industrial sector was
severely affected and did not get the government support to recover at a fast pace (Beck
2005).
87
Table 5.3a : Occupation of the Household Head and Poverty Measures Headcount Poverty Gap Squared Poverty Gap Occupational Status
Agricultural worker (on-farm) Round 1 0.474 0.137 0.056 Round 2 0.594 0.167 0.066 Round 3 0.592 0.175 0.072
Agricultural worker (Off-farm) Round 1 0.595 0.171 0.063 Round 2 0.630 0.276 0.132 Round 3 0.743 0.266 0.114
Industrial enterprise Round 1 0.644 0.223 0.100 Round 2 0.718 0.226 0.090 Round 3 0.795 0.291 0.128
Trade Round 1 0.341 0.082 0.027 Round 2 0.451 0.113 0.042 Round 3 0.495 0.164 0.074
Transport Round 1 0.561 0.133 0.040 Round 2 0.705 0.212 0.081 Round 3 0.703 0.219 0.091
Construction work Round 1 0.500 0.153 0.062 Round 2 0.630 0.229 0.103 Round 3 0.667 0.243 0.096
Self-employed professional Round 1 0.364 0.071 0.020 Round 2 0.333 0.132 0.061 Round 3 0.500 0.177 0.078
Miscellaneous services Round 1 0.333 0.098 0.041 Round 2 0.304 0.080 0.029 Round 3 0.415 0.161 0.076
Unemployed Round 1 0.628 0.219 0.108 Round 2 0.667 0.235 0.120 Round 3 0.692 0.262 0.131
Total Round 1 0.489 0.144 0.059 Round 2 0.582 0.174 0.071 Round 3 0.604 0.197 0.085
Note: Calculations based on upper poverty lines.
88
Table 5.3b highlights that lower educational levels of the household head are
associated with higher poverty rates. Educational attainment improves the employment
and income-earning opportunities of the individual. It is interesting to note that in round
1, heads with less than grade (class) 5 education have lower headcount poverty rates
compared to those with a grade 5 level of education or better. This is reflected even in
the squared poverty gap index.
Table 5.3b: Educational Attainment of the Household Head and Poverty Measures Headcount Poverty Gap Squared Poverty Gap Educational Attainment
No education Round 1 0.555 0.171 0.071 Round 2 0.503 0.140 0.054 Round 3 0.694 0.242 0.108
Less than class 5 Round 1 0.413 0.098 0.034 Round 2 0.356 0.074 0.023 Round 3 0.565 0.164 0.066
Fifth standard Round 1 0.477 0.115 0.047 Round 2 0.313 0.064 0.022 Round 3 0.508 0.153 0.063
Between 5th and 9th standard Round 1 0.355 0.115 0.050 Round 2 0.264 0.063 0.021 Round 3 0.484 0.127 0.048
Higher level education Round 1 0.182 0.044 0.014 Round 2 0.127 0.012 0.002 Round 3 0.259 0.062 0.021
Note: Calculations based on upper poverty lines
Poverty profiles in Table 5.3c show that poverty increases with age and then falls
after age 30-44. Household heads between the ages 30-44 appear to experience the
highest poverty levels; the poverty experienced by heads above the age of 65 is low. It is
important to remember that these characteristics are of the household head only. In the
89
data households headed by an older member also have larger numbers of working
members. More working members to contribute to the pool that enables consumption
smoothing.
Table 5.3c: Age and Gender of the Household Head and Poverty Measures
Headcount Poverty Gap Squared Poverty Gap Age category
Age 20-29 Round 1 0.466 0.113 0.036 Round 2 0.561 0.181 0.072 Round 3 0.604 0.201 0.080
Age 30-44 Round 1 0.556 0.174 0.074 Round 2 0.672 0.212 0.088 Round 3 0.702 0.244 0.109
Age 45-64 Round 1 0.434 0.123 0.049 Round 2 0.505 0.141 0.058 Round 3 0.514 0.157 0.066
Age 65 or more Round 1 0.419 0.123 0.050 Round 2 0.486 0.126 0.053 Round 3 0.514 0.141 0.057
Gender of the Head
Male Round 1 0.476 0.138 0.056 Round 2 0.576 0.169 0.068 Round 3 0.603 0.193 0.082
Female Round 1 0.758 0.284 0.134 Round 2 0.714 0.291 0.154 Round 3 0.630 0.291 0.160
Note: Calculations based on upper poverty lines
Table 5.3c shows that female-headed households have greater incidence, depth
and severity of poverty than male-headed households. Headcount poverty falls from 76
to 63 percent among female-headed households over the three survey round but continues
to be higher than for male-headed units. Only 4 percent of the households are female
90
headed in the data, precluding separate analysis for these particularly interesting
households from a poverty perspective. This is in line with the fact that rural women are
classified as one of the most vulnerable groups.
5.3 Stochastic Dominance (First-order) Test
A first-order stochastic dominance test is performed to check the robustness of the
changes in the calculated poverty indices (Coudouel et al. 2002). This is done by
comparing cumulative distribution of per capita consumption across different situations
(the three rounds in this case). These curves are called as poverty incidence curves and
they test if the choice of the poverty line affects the poverty results (World Bank 2005).
If poverty analysis is sensitive to the choice of the poverty lines then the poverty
measures calculated are not robust and slightest change in poverty line could affect our
results.
If there is first order dominance then the poverty incidence curves do not intersect.
The curve lying completely above the others is poorer than the other curves. Figure 5.1
reiterates that consumption expenditure of the households fell over the survey period.
The green curve corresponds to the cumulative distribution of household consumption in
round 3 which is above both round 1 and round 2. Whatever be the poverty line chosen
this result will hold (see Figure 5.1). Results become ambiguous when these curves
intersect when it may not be possible to conclude if poverty levels have increased or
decreased while comparing different periods.
91
Figure 5.1 Stochastic Dominance Curve
5.4 Receiver Operating Characteristic (ROC) Analysis
Given that the objective of the research is to identify the poor and target policies
to reduce their deprivation, a useful but not commonly applied tool in poverty analysis is
Relative Operating Characteristics (ROC) analysis. ROC curves are used to access how
well the indicators perform in predicting the probability of being poor (Estache et al.
2002). This analysis assesses the performance of indicators based on errors of inclusion
and exclusion. This methodology is based on logit regression and Wondon (1997) in his
study using the ROC curves defines errors of inclusion and exclusion as follows.
Box 5.1: ROC Definitions Predicted Status Actual Status Nonpoor Poor Predicted nonpoor SP = NP-/( NP-+ NP+) 1 - SE = P-/ (P+ + P-) Predicted Poor 1 - SP = NP+/( NP-+ NP+) SE = P+/ (P+ + P-)
0
.2
.4
.6
.8
1
0 1000 2000 3000 Round 1 Round 2Round 3
92
where P is the number of poor, P- is the number of poor classified as non-poor (negative
outcome), P+ is the number of poor classified as poor (positive outcome), NP is the
number of non-poor, NP- is the number of non-poor classified as non-poor and NP+ is the
number of non-poor classified as poor by a model. The sensitivity and specificity
measure which are then calculated, being defined as follows (Coudouel et al. 2002)
Sensitivity = SE = P+/ (P+ + P-) = P+/ P is the fraction of poor households classified as
poor
Specificity = SP = NP-/ (NP-+ NP+) = NP-/ NP is the fraction of non-poor households
classified as non-poor.
Therefore, errors of inclusion are 1 – SP (type I error) and errors of exclusion are
(type II) 1 – Sensitivity. Errors of inclusion include identifying the poor as non-poor and
errors of exclusion include identifying the non-poor as poor. The predictive value in a
logit or probit regression for poverty classifies the households according to their poverty
status. Statistical packages use one-half as the cut-off point implying households above
this cut-off point are poor. The ROC curves are very similar to Lorenz curves with 1-
Specificity on the horizontal axis 1- Specificity and Sensitivity on the vertical axis for
alternative values of cut-off points. Sensitivity is zero and specificity is one at the origin
where the cut-off is 1. At this point no one is classified as poor and the probability of
Specificity error is zero and of Sensitivity error is one. The probability of the non-poor
being classified as poor is zero and the probability of the poor being classified as non-
poor is 1. Similarly, in the upper right side of the curve where the cut-off is equal to one,
93
Sensitivity is equal to one and Specificity is equal to zero; hence the probability of
Specificity error is one and that of SE error is zero (Wondon 1997).
The area below the ROC curve indicates the explanatory power of the variable
and not the direction of the relationship. Predictive power of .5 would exactly coincide
with the 45-degree line, which would imply no predictive power, and an area of one
would imply perfect explanatory power. That is, a higher ROC curve means better
predictive power of the variable. The dependent variable used in the model is headcount
index. This gives an overall predictive power of 0.7902 which is good and Figure 5.2
indicates the same.
0.00
0.25
0.50
0.75
1.00
Sen
sitiv
ity
0.00 0.25 0.50 0.75 1.001 - Specificity
Area under ROC curve = 0.7902
Figure 5.2: ROC Curve for Poverty Models
94
Table 5.4: Areas Under the ROC Curve for Individual Poverty Indicators and Overall Model Using Upper Poverty Line
Household Characteristic Area Under the ROC Curve Household size (log) 0.5467 Dependency ratio 0.6525 Working members 0.5502 Location
Derai 0.5173 Madaripur 0.5010 Mohammadpur 0.5127 Muladi 0.5244 Saturia 0.5114 Shibpur 0.5374 Shahrasti 0.5188
Physical assets Land owned (area) 0.6557
Financial assets Credit availability 0.5393 Social assistance 0.5893 Remittances 0.5210
Household Head Characteristics Age 0.5553 Education
No education 0.5706 Less than 5th standard 0.5201 Fifth standard 0.5005 Between 6th to 9th standard 0.5311 Higher than 9th standard 0.5449
Primary occupation28 On-farm agricultural work (on-farm) 0.5106 Off-farm agricultural work 0.5125 Industrial sector 0.5346 Trade sector 0.5343 Transportation sector 0.5069 Construction sector 0.5009 Self-employed and service sector 0.5267 Out of the labor force 0.5180
Overall Model 0.7902
28 There are nine occupation categories including on-farm work (agricultural work on farm, supervising agricultural work, agricultural wage labor, share cropper), off-farm work (fisherman, fish culture, livestock, poultry, growing fruits, off-farm wage activity), industrial enterprise (processing crops, tailoring, sewing, pottery, blacksmith, goldsmith, repairing manufactured products, carpenter, mechanic, other wage labor), trade (small retail shop, wholesale trader, contractor, employee, employer), transport (rickshaw pulling, boat, wage labor in transport, other transport work, driver, helper), construction work (mason, helper, construction worker, helper, house repairing), self-employed profession (doctor, kabiraj, advocate, barber, washerman, house tutor, deed writer, Purohit, Dhatri, handicrafts), miscellaneous services (service, pension, working in NGO, servant), other (income from hats, income rent, household work, child, student, beggar, unemployed, disabled).
95
Table 5.4 presents the individual and overall explanatory power of household
characteristics on poverty. Dependency ratio and land ownership show the highest levels of
explanatory power. Social assistance which includes sources of revenue from NGOs and
government also performs well. Better results are obtained when combinations of
characteristics are used.
5.5 Econometric Analysis: Poverty Dynamics
5.5.1 Methodology I
Based on the McCulloch and Baulch (1999) approach, Table 5.5 categorizes the
households into three independent groups. The first group is comprised of households whose
expected value of per capita consumption over time is always below the poverty line. This
group is considered chronically poor. The second group includes households whose expected
value of income over time is above the poverty line but annual per capita consumption falls
below the poverty line at least once during the three rounds (sometimes poor); and the final
category then includes households with per capita consumption above the poverty line in all
three periods (never poor). The IFPRI-FMRSP 1998-99 Bangladesh data used for the
analysis is a short panel. We take advantage of the panel set up in the poverty categorization
which is based on the consumption expenditures of households at all three time periods.
Care is taken that regional poverty lines are used after correcting for differences in prices
using the consumer price index.
The multinomial logit not only distinguishes between the two chronic and transient
groups but also brings out the difference between these poverty categories and those
households who have not experienced poverty during the survey period. Based on the
96
definition of poverty groups in Chapter 4, Table 5.5 reports that of the 727 households in the
panel, 276 (37.96 percent) are chronically poor and 167 (22.97 percent) transient poor if the
lower poverty line is considered. Using the upper poverty line, 400 (55.02 percent) and 159
(21.87 percent) of households are chronically and transient poor, respectively. The figures
here are higher than the national figures. The CPRC (2004) reports 31.4 percent to be in
chronic poverty and 43.6 percent as transitory poor. However, the short panel data set used
here, was collected immediately after a major flood which slid the Bangladeshi population
deeper into poverty.
Table 5.5: Number of Poor in Bangladesh by Poverty Categories Poverty Status Lower Poverty Line (ZL) Upper Poverty Line (ZU) Always poor 276 (37.96) 400 (55.02) Sometimes poor 167 (22.97) 159 (21.87) Never poor 284 (39.06) 168 (23.11)
Total 727 727
The model estimated in table 5.6 includes household size (log); dependency ratio; age;
educational level and occupation of the household head29; household asset ownership (land);
number of working members in the household (binary); access to credit (binary variable);
social assistance (binary variable); remittances (binary variable) and geographical location of
the household30 as explanatory variables. Transient and chronic poor households are
29 There are nine occupation categories including on-farm work (agricultural work on farm, supervising agricultural work, agricultural wage labor, share cropper), off-farm work (fisherman, fish culture, livestock, poultry, growing fruits, off-farm wage activity), industrial enterprise (processing crops, tailoring, sewing, pottery, blacksmith, goldsmith, repairing manufactured products, carpenter, mechanic, other wage labor), trade (small retail shop, wholesale trader, contractor, employee, employer), transport (rickshaw pulling, boat, wage labor in transport, other transport work, driver, helper), construction work (mason, helper, construction worker, helper, house repairing), self-employed profession (doctor, kabiraj, advocate, barber, washerman, house tutor, deed writer, Purohit, Dhatri, handicrafts), miscellaneous services (service, pension, working in NGO, servant), other (income from hats, income rent, household work, child, student, beggar, unemployed, disabled). 30 Geographical regions included in the analysis are Derai, Madaripur, Mohammadpur, Muladi, Saturia, Shahrasti, and Shibpur.
97
compared with those who are never poor during the period that data were collected. Social
assistance includes transfers made to the household by the government and NGOs.
Government transfers were made through programs including Gratuitous Relief, Vulnerable
Group Feeding, Food for Work, Test Relief and Vulnerable Group Development. Detailed
information was collected on food grains (wheat and rice) and cash transfers made to the
household, but it is not possible to differentiate within the household as to who received the
transfers. But data are available on every loan taken by the household in each round, and
information was collected that identifies the receiver of the loan. Table 5.6 reports the means
and standard deviations of variables used in the model for non-poor, chronically poor and
transient poor households. The final sample used in the estimation consists of 706
households.
Compared to other categories, the average consumption level for the chronic poor is
the lowest. Average consumption of the never-poor is twice that of the chronic poor and the
transient poor group, on average, has a 38 percent higher consumption level than the chronic
poor. This indicates the extent of deprivation of the chronically poor population. There is
only a marginal difference between the poverty categories in terms of household size, ranging
between 5.22 (transient poor) to 5.52 (non-poor). Examining the average number of working
members in the household, the transient poor households have the highest average number of
workers and the chronically poor are at the lowest end. On the other hand, the dependency
ratio among the chronically poor is the highest. Fewer working members and a higher
dependency ratio could be stressors for chronically poor families.
Table 5.6 shows that 35 percent of the non-poor, 63 percent of the chronic poor and
52 percent of the transient poor are not educated. On the other hand, 18 percent of the non-
98
poor, 3 percent of the chronic poor and 7 percent of the transient poor household heads have
more than a 9th grade education. Examination of land ownership by households reveals that
the transient poor own almost twice the amount than non-poor own and three times more land
than the chronic poor. Those in persistent poverty also have limited access to human and
physical capital. This negatively affects their income-earning capacity and also limits their
consumption smoothing abilities. With respect to occupation, 54 percent of the transient poor
heads are engaged in the agricultural sector and around 47 percent of the non-poor and
chronically poor belong to this sector. In addition, workers in the industrial sector, off-farm
agricultural labor and the unemployed are most like to head households that are chronic poor.
The trade, industry and self-employed sectors see the presence of transient poor.
A greater proportion (79 percent) of the heads in chronically and transient poverty
participate in the credit market while only 70 percent among the non-poor seem to have taken
loans. Table 5.6 also reveals that 39 percent of chronically poor households avail social
assistance. This is lower for the transient poor (26 percent) and even lower for the non-poor
(17 percent). Remittances data shows that around 8 percent of the poor (transient and
chronic) receive these transfers as opposed to 13 percent of the non-poor. Descriptive
analysis reveals that government and NGO transfers and credit seem to be reaching the poor
as targeted. Remittances received from family abroad are associated with higher income or
consumption levels. It is observed from Table 5.6 that the average age of the household head
is highest (49 years) among non-poor households. It is 46 years among the transient poor and
43 years among the chronic poor. Younger heads are more likely to be poor. As pointed out
previously, this may be true because households with older heads are found to have more
working members. Regions chosen by IFPRI to collect data are diverse in terms of levels of
99
poverty and flood severity experienced. Shibpur, Shahrasti and Madaripur have the highest
number of non-poor households. Shibpur and Shahrasti are non-poor thanas and Shahrasti
and Madaripur were moderately affected by the 1998 floods. Households that are chronically
poor are uniformly distributed across regions, with Derai (17 percent), Muladi (16 percent)
and Saturia (15.25 percent) with the highest percentages. There is more variation among the
transient poor within regions with Mohammadpur having the highest percentage.
Preliminarily we find that the poor are not homogeneous. A high dependency ratio,
fewer working household members, a younger household head, lower educational attainment
of the head and land ownership are among the characteristics more strongly associated with
chronic poverty than transient poverty. Transient poor with higher consumption levels are
clearly better-off. There is also a regional pattern among the seven locations in the data.
100
Table 5.6: Characteristics of Sample Households in Rural Bangladesh
Characteristicsa Non-poor Chronic Poor Transient poor
Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.
Household Characteristic
Average consumption (Taka) 1169.64 357.68 491.03 111.45 805.58 144.61
Household size 5.52 2.39 5.47 1.77 5.22 2.12
Dependency ratio 0.85 0.71 1.33 0.80 0.87 0.59
Working members 1.69 1.16 1.39 0.94 1.78 1.16
Location (%)
Derai 5.95 0.24 17.25 0.38 11.95 0.33
Madaripur 16.67 0.37 13.25 0.34 14.47 0.35
Mohammadpur 11.90 0.32 13.25 0.34 20.75 0.41
Muladi 14.29 0.35 16.00 0.37 10.69 0.31
Saturia 13.10 0.34 15.25 0.36 12.58 0.33
Shibpur 21.43 0.41 10.75 0.31 18.24 0.39
Shahrasti 16.67 0.37 14.25 0.35 11.32 0.32
Physical Assets (decimalb)
Land owned 211.39 274.66 70.94 99.47 137.41 164.62
Financial Assets (%)
Credit availability 70.24 0.46 78.75 0.41 78.62 0.41
Social assistance 17.26 0.38 38.75 0.49 25.79 0.44
Remittances 13.10 0.34 8.00 0.27 8.18 0.27
Household Head Characteristics
Age 48.81 12.97 43.38 11.40 45.55 13.53
Education (%)
No education 35.12 0.48 63.25 0.48 52.20 0.50
101
Less than 5th standard 14.88 0.36 12.00 0.33 11.32 0.32
Fifth standard 9.52 0.29 7.50 0.26 11.32 0.32
Between 6th to 9th standard 20.83 0.41 8.50 0.28 13.84 0.35
Higher than 9th standard 18.45 0.39 3.00 0.17 6.92 0.25
Primary occupation (%)
On-farm agricultural work 47.62 0.50 47.25 0.50 54.09 0.50
Off-farm agricultural work 2.38 0.15 7.25 0.26 4.40 0.21
Industrial sector 5.36 0.23 15.25 0.36 8.18 0.27
Trade sector 16.07 0.37 8.00 0.27 13.84 0.35
Transportation sector 4.17 0.20 6.75 0.25 3.77 0.19
Construction sector 2.38 0.15 3.75 0.19 3.14 0.18
Self-employed and service sector 17.26 0.38 4.50 0.21 8.81 0.28
Out of the labor force 4.17 0.20 6.75 0.25 3.77 0.19
Source: Based on own calculations using IFPRI-FMRSP Survey. a: Means of variables are based on Round one. b:1 acre = 100 decimal.
102
5.5.1.2 Results from Multivariate Analysis
Multinomial logit results are shown in Table 5.7. The coefficient, marginal
effects and the level of significance are presented. The three independent categories in
the analysis are non-poor, chronically poor and transient poor households. Non-poor
households are the reference category. The variables included in the model are log of
household size, dependency ratio, and number of working members in the household,
location of the household, land ownership, participation in credit market, social assistance
and remittances received by the household, age category, educational levels and
occupation of the household head.
Determinants of Chronic Poverty
Household size: The multinomial logit models reveal that household size is a significant
predictor of chronic poverty. Bigger households are more likely to be chronically poor.
This is true among households that have limited access to resources and assets.
McCulloch and Baulch (2000), Jalan and Ravallion (2000), Haddad and Ahmed (2003)
and Aliber (2003) in their study of Pakistan, China Egypt and South Africa, respectively,
found this to be true.
Dependency ratio: Higher dependency ratio increases the probability of being chronically
poor in reference to non-poor. Studies have found with all things equal, households with
a greater number of children, more members above the age of 60 and disabled members
are more likely to be chronically poor (Aliber 2003).
Working members: The multinomial logit model indicates that more the working
members in the family less the chance of being chronically poverty. These households
103
have very low consumption levels to begin with and any addition to the family resources
would affect the poverty status of the household.
Location: Results show that compared to Muladi thana, Madaripur, Mohammadpur and
Shibpur are less likely to be chronically poor. Interestingly, Madaripur, Mohammadpur
and Shibpur are all regions that are closer to Dhaka and Muladi is further away. Being
located near Dhaka which is the capital of the country and is also hub of activities could
potentially give residents of these regions greater access to resources.
Land ownership: Households with greater land holdings are less likely to experience
chronic poverty. Lack of physical assets is associated with chronic poverty (McCulloch
and Baulch 2000; Aliber 2003). Assets such as livestock and land help poor households
not only generate income but are also a form of investment.
Social Assistance: The estimates and marginal effects reflect that social assistance
increases the likelihood of being chronically poor.
Education: Higher educational levels are strongly associated with lower chronic poverty
levels. Studies have found that an increase in number of years of education decreases the
probability of being chronically poor (McCulloch and Baulch 1999; Jalan and Ravallion
1999; Aliber 2003; McCulloch and Calandrino 2003).
Occupation: Our results confirm that compared to being employed in the agricultural
sector, employment in trade and self-employment reduces the chances of being
chronically poor. Our descriptive analysis revealed that these sectors had one of the
highest percentages of non-poor households.
104
Determinants of Transient Poverty
Household size: Household size seems to have opposite effect on the transient poor.
Bigger households are less likely to be transient poor. This result is not unusual as Jalan
and Ravallion (2000) find higher transitory poverty among smaller Chinese households.
Dependency ratio: Table 5.7 shows that fewer dependents in the household increase the
likelihood of being transitory poor. The marginal effects indicate such household is 8
percent less likely to be transient poor when compared to never-poor category.
Working members: Households with higher number of working members are more likely
to be transient poor. Agriculture and related activities employ majority if population in
rural Bangladesh. These households cannot escape from seasonality of employment.
Even when members are seemingly employed they may not be earning consistently
throughout the year and fall in and out of poverty.
Land ownership and credit: Both these variables have a positive impact on transient
poverty. The chances of falling into transient poverty are actually higher with increase in
land and credit.
Social assistance, location, occupation, remittances, age and education of the
household head are not significantly associated with transient poverty.
105
Table 5.7: Estimates and Marginal Effects from Multinomial Logistic Regression: Persistent and Sometimes Poor Chronic Poor Transient Poor Characteristics Coefficient Marginal Effects z Coefficient Marginal Effects z Household Characteristic
Household size (log) 2.268 0.519 5.30*** 0.274 -0.276 -3.69*** Dependency ratio 0.493 0.131 3.00*** -0.065 -0.084 -2.29** Working members -0.301 -0.091 -3.17*** 0.119 0.066 3.00*** Location(reference: Muladi thana)
Derai 0.498 0.030 0.30 0.590 0.039 0.45 Madaripur -1.287 -0.245 -2.74*** -0.518 0.066 0.79 Mohammadpur -0.452 -0.212 -2.35** 0.639 0.211 2.43** Saturia 0.049 0.018 0.19 -0.044 -0.015 -0.20 Shibpur -2.094 -0.397 -5.45*** -0.676 0.114 1.33 Shahrasti -0.708 -0.116 -1.19 -0.440 0.012 0.14
Physical assets Land owned (Decimal) -0.009 -0.002 -6.94*** -0.002 0.001 4.59***
Financial Assets Credit availability 0.120 -0.036 -0.64 0.482 0.071 1.70* Social assistance 1.027 0.162 3.24*** 0.571 -0.044 -1.05 Remittances 0.399 0.093 1.12 0.021 -0.053 -0.82
Household Head Characteristics Age category (reference: less than 30 years)
Age between 30-44 -0.260 -0.058 -0.59 -0.043 0.029 0.38 Age between 45-64 -1.165 -0.155 -1.54 -0.955 -0.020 -0.27 Age 65 plus -1.068 -0.230 -1.95* -0.221 0.102 0.92
Education(reference: no education) Less than 5th standard -0.721 -0.077 -1.04 -0.798 -0.053 -0.98 Fifth standard -0.680 -0.145 -1.69* -0.169 0.062 0.86 Between 6th to 9th standard -1.568 -0.260 -3.72*** -0.997 -0.002 -0.04 Higher than 9th standard -1.910 -0.316 -3.66*** -1.221 -0.023 -0.31
Primary occupation (reference: on-farm Agricultural work)
Off-farm agricultural work 0.387 0.041 0.36 0.353 0.009 0.10 Industrial sector -0.337 -0.009 -0.11 -0.563 -0.057 -0.89 Trade sector -1.633 -0.319 -4.63*** -0.528 0.098 1.41
106
Transportation sector -0.667 -0.059 -0.53 -0.875 -0.071 -0.92 Construction sector -0.179 -0.010 -0.07 -0.247 -0.021 -0.21 Self-employed and service sector -1.580 -0.286 -3.42*** -0.751 0.043 0.55 Out of the labor force 0.783 0.192 1.95* -0.131 -0.122 -1.73*
Log likelihood -533.46 Pseudo –square 0.249 Number of observations 706
Note: Never poor is the reference category. *, **, *** represent significance at 10, 5 and 1 percent level.
107
5.5.2 Methodology II: Censored Quantile Regression
Following Jalan and Ravallion (2000), aggregate intertemporal poverty levels are
calculated for each household. The welfare indicator used is squared poverty gap index
using the upper poverty line. Household total poverty and chronic poverty is calculated
using consumption levels in all three rounds. This classification is important in order to
identify the policy that would be most effective in targeting different poverty categories.
Table 5.8 reports total, chronic and transient poverty across regions. As in the
previous analysis more households are found to be chronically poor. Around 75 percent
of total poverty is attributed to chronic poverty and chronic poverty is found to be higher
in all the regions. Muladi, followed by Shahrasti have the highest percent of chronic
poverty and Shibpur has lowest proportion of chronic poverty. Shibpur, followed by
Madaripur have highest percent of transient poverty.
Table 5.8: Total, Chronic and Transient Poverty by Region Region Total Poverty Chronic Poverty Transient Poverty Absolute Value Absolute Value Absolute Value
Derai 0.118 (100)
0.092 (77.91)
0.026 (22.09)
Madaripur 0.062 (100)
0.044 (70.27)
0.019 (29.73))
Mohammadpur 0.069 (100)
0.049 (71.10)
0.020 (28.90)
Muladi 0.069 (100)
0.055 (80.31)
0.014 (19.69)
Saturia 0.065 (100)
0.046 (71.71)
0.018 (28.29)
Shibpur 0.042 (100)
0.028 (67.52)
0.014 (32.48)
Shahrasti 0.081 (100)
0.064 (79.31)
0.017 (20.70)
Total 0.072 (100)
0.054 (74.86)
0.018 (25.14)
Note: Percent value in parentheses.
108
Having defined chronic and transient poverty at household level using squared
poverty gap index the next step is to examine if factors influence chronic and transient
poverty in a similar manner. With this objective, separate models of censored quantile
regression are estimated for both transient and chronic poverty at the household level. As
suggested in Jalan and Ravallion (2000), we use upper poverty line and 85th quantile to
circumvent the high-degree of censoring. We use Qreg command in STATA to estimate
these models. The Pseudo R-square indicates that the model predicts chronic poverty
better. More variables are significantly different from zero in chronic poverty regression.
The estimates in table 5.9 show indicate that household characteristics like household
size and dependency ratio and demographic characteristics including education of the
household head seem to be more important for chronic poverty than transient poverty.
Like human capital, increased land ownership is associated with lower chronic poverty.
Increased age of the household head is associated with lower levels of both kind of
poverty but this relationship is stronger for transient poor household. Households located
in Madaripur, Mohammadpur, Shibpur are associated with lower chronic poverty levels,
and Derai, Madaripur and Mohammadpur are associated with higher transient poverty.
While household heads’ occupation in the trade sector decreases chronic poverty, being
in industrial sector increases transient poverty and being out of the labor force is likely to
increase chronic poverty. Among the financial assets variables increased social
assistance is associated with chronic poverty and credit availability and remittances are
positively associated with transient poverty.
109
Table 5.9: Censored Quantile Regression Results (85th quantile) Chronic Poverty Transitory Poverty
Coefficient T-statistics Coefficient T-statistics Household Characteristic
Household size (log) 0.111 3.02*** -0.013 -1.54 Dependency ratio 0.057 3.67*** -0.002 -0.62 Working members -0.009 -0.81 0.001 0.38 Location(reference: Muladi thana)
Derai -0.005 -0.14 0.023 3.04*** Madaripur -0.067 -1.87* 0.024 3.36*** Mohammadpur -0.060 -1.66* 0.026 3.42*** Saturia -0.031 -0.80 0.013 1.56 Shibpur -0.121 -3.42*** 0.008 1.12 Shahrasti -0.050 -1.32 0.010 1.38
Physical Assets Land owned (decimal) 0.000 -3.03*** 0.000 1.01
Financial Assets Credit availability (Formal and Informal) -0.023 -1.02 0.009 1.68* Social assistance 0.062 2.86*** 0.002 0.34 Remittances 0.006 0.17 0.013 1.74*
Household Head Characteristics Age category (reference: less than 30 years)
Age between 30-44 -0.014 -0.37 -0.032 -3.66*** Age between 45-64 -0.060 -1.45 -0.035 -4.12*** Age 65 plus -0.094 -1.97** -0.032 -3.00***
Education(reference: no education) Less than 5th standard -0.072 -2.51*** -0.005 -0.90 Fifth standard -0.072 -1.98** 0.003 0.33 Between 6th to 9th standard -0.100 -3.23*** -0.009 -1.42 Higher than 9th standard -0.115 -3.15*** -0.001 -0.09
Primary occupation (reference: on-farm Agricultural work)
Off-farm agricultural work 0.030 0.67 0.000 0.01 Industrial sector 0.042 1.26 0.014 2.05** Trade sector -0.094 -2.87*** 0.008 1.10 Transportation sector -0.039 -1.02 0.005 0.55 Construction sector -0.010 -0.18 0.008 0.71 Self-employed and service sector -0.045 -1.19 0.006 0.73 Out of the labor force 0.119 3.13*** 0.002 0.24
Constant 0.197 3.33*** 0.069 4.77*** Pseudo R-square 0.262 0.094 Number of observations 706 706
*, **, *** represent significance at 10, 5 and 1 percent level.
110
5.6 Conclusion
Decomposition of poverty profiles and the poverty measures calculated suggest
that household welfare declined over the survey period. Large proportions of the
households were found to be in poverty and proportion of chronically poor households is
greater than transient poor. Still around twenty percent of the households are moving in
and out of poverty. The multinomial logit models and the censored quantile regression
analysis indicate slightly different results. On one hand, multinomial logit models
distinguish between different poverty categories and differentiate the poor from the non-
poor. On the other, quantile regressions identify the factors that affect each kind of
poverty-ridden households.
The multivariate analyses show that compared to the non-poor, household-level
characteristics like household size, dependency ratio and working members in the
household have differential impacts on chronically and transient poor households. A
larger household size and higher dependency ratio seem to increase chronic poverty and
more working members reduce the likelihood of being chronically poor. Long-term
investments in human and physical assets clearly help households out of chronic poverty.
Credit access and remittances explain transient poverty better.
111
Chapter 6
Household Expenditure and Credit as Bargaining Measure
6.1 Introduction
Poor households in rural areas not only face higher risks (i.e., more vulnerable to
shocks) but also have lower capacity to cope with these risks and shocks. One of the
factors responsible for lower ability of the poor to absorb shocks is lower asset
ownership of these households (Fafchamps 1999). One of the risk management
strategies used by the poor is the distribution of resources among the members of the
household. Families with different access to resources have different welfare response
to shocks (Heitzmann et al. 2001).
Increasingly, there is a shift from unitary to collective household models.
Households under the unitary framework are assumed to act as a single decision-
making unit. On the other hand, collective models are where the household utility
function is disaggregated and the models take into account different preferences of each
member of the household (Haddad et al. 1997; Quisumbing 2003; Mendoza 1997).
Distribution of resources and power within a household is mostly in the favor of men
and beneficial household outcomes are reported when resources are controlled by
women, especially child and nutritional outcomes (Quisumbing 2003). The husband
and wife use their bargaining power to decide how the household resources are
allocated within the household.
The objective of this research is also to study how individuals interact and operate
within a family or household. To improve the well-being of individuals, development
policies not only have to take into account how resources are allocated within the
112
family or household but also consider the impact of this resource allocation on
individuals. With this objective, intrahousehold dynamics (e.g., variations in household
bargaining behaviors) are examined with a focus on the household’s expenditure
patterns. Participation in credit market are taken as the measure of bargaining between
the head and the spouse. Combined formal and informal credit received by the
household is considered in the analysis. The household bargaining model is used to
analyze the effects of credit participation on consumption choices within poor
households. The focus is on the household’s expenditure patterns.
The intrahousehold bargaining literature has used various measures of bargaining
including unearned income, inherited assets, assets at marriage, current assets and
credit (see Chapter 2 for a detailed literature review). Pitt et al. (2003) studied the
impact of participation of men and women in credit programs in Bangladesh on
empowerment of women. They found that participation in microcredit programs indeed
improved the position of women in the household and tended to increase spousal
communication. Women were found to communicate freely to their husbands about
family planning and parenting. Swaminathan (2003) used access to credit as an
indicator of resource control in household decision-making. In this dissertation, credit
receipt of the head and spouse are used as the bargaining measures. Formal and
informal sources of credit cannot be differentiated as discussed previously.
6.2 Credit Availability
United Nations Resolution 53/197 proclaimed 2005 as the International Year of
Microcredit in recognition of its contribution to poverty eradication, empowerment of
113
women and for social and human development (United Nations 1999). Availability of
credit to the poor is an important poverty eradication strategy and could be instrumental
in achieving the first MDG (United Nations 2005)31. Therefore, a poverty alleviation
policy which incorporates development of the financial sector of a country including
microfinance has immediate and direct effects on the disadvantaged.
Among formal credit, microfinance plays a very important role in Bangladesh.
Numerous non-governmental, government and private institutions came forward to help
the poor to rehabilitate after the disaster flood of 1998. Microfinance and credit
availability increases household incomes, increases employment, diversifies income
sources, enables consumption smoothing, improves food intake and empowers women
and thereby reducing poverty and vulnerability of individuals (United Nations 2005).
Hashemi et al. (1996) in their study of Grameen Bank and Bangladesh Rural
Advancement Committee find that credit programs do empower women and are cost-
effective means of resource transfer. It gives women the ability to voice their opinion
and make decisions despite the patriarchal society they live in. Formal and informal
sources of credit are not distinguished here and we argue that whatever the source of
credit it influences woman’s decision-making abilities and has positive outcomes for the
household. The latter is tested.
31 Millennium Development Goal 1 (from United Nations 2005)
Target 1: Halve, between 1990 and 2015, the proportion of people whose income is less than $1/ day. Target 2: Halve, between 1990 and 2015, the proportion of people who suffer from hunger.
114
6.3 Results
Following Quisumbing and de la Briere (2000) and Swaminathan (2003), this
research considers agricultural household with two members (head and spouse) in the
household and applies the Nash-bargaining model to study the role of credit receipt in
determining the bargaining power. The following expenditure equations are estimated:
∑ =+++
++
++=
K
1k 16
54
321
cdemographi credit) spouses'log(
credit) heads'log(area) land household total(
size) household totallog( e)expenditur household totallog(
εδα
αα
ααα
kk
jb
[6.1]
where bj is the expenditure share of the jth good, independent variables include
demographic compositions of the household, location variables and ε is the error term.
Separate equations for food and each of the non-food categories are estimated,
controlling for household size, location of household and the round in which data were
collected. Households which are monogamous and whose family structure has not
changed across all three rounds of data collection were identified and included in the
analysis.
OLS is used to estimate the food share and personal care share equations, and
Tobit models are used to estimate the 10 other non-food budget share equations. The
Tobit model is used when the dependent variable is zero for a non-trivial fraction of the
population and is otherwise continuously distributed (Wooldridge 2006). Food and
personal care are the only shares which are continuous and non-zero across all
households. Not all households are found to be spending on all non-food expenditure
categories32. Potential endogeneity of total expenditure with budget shares is taken into
32 In the data, 87.32 percent of the households do not spent on social activities, 10.25 percent don’t spend on cigarettes and beetel (tobacco), 17.34 percent do not spend on adult clothing, 31.05 percent do not spend on children’s clothing, 67.51 percent don’t spend on durable goods, 21.69 percent do not spend on
115
consideration and tests for endogeneity are conducted using the regression-based
Hausman test. As in Swaminathan (2003), if the shares are endogenous then we use
multiple instruments in two stage least square estimator (2SLS) and simultaneous-Tobit
models. Access to sanitized toilets, material used in building walls of a house, source of
water for washing, electricity availability and availability of fixed garbage disposal
techniques are the instruments used in the 2SLS models. After performing the
endogeneity test, we find that expenditure shares in cigarettes, fuel and health are not
endogenous. Since a significant portion of the household expenditure is in food, we
classify food expenditure into spending on cereals, animal-based and plant-based
products in order to get a clearer picture of impact of credit on food categories. Cereals
include consumption of rice and wheat only.
6.3.1 Descriptive Statistics
Table 6.1 indicates the numbers of households in which only men, only women
and both men and women access credit. The data show that more men compared to
women participate in the credit market. It is interesting to note that the number of
households with both husband and wife accessing loans increased from 23 to 40 percent
over the period of one year. The number of households with only men or only women
incurring loans only marginally changed.
Table 6.2 presents borrowings by the head and the spouse in a household. NGO
credit includes loans taken from BRAC, Jagorani, ASA, Grameen Bank, Proshika, BRDB
(Bangladesh Rural Development Board), GKT (Gano Kallyan Trust), Save the Children
education, 1.29 percent do not have fuel expenditure, 6.49 percent do not have health expenditure, 0.30 percent do not have personal care expenditure, 64.44 percent do not have repair expense and 45.27 percent do not have travel expense.
116
and the Government of Bangladesh Landless Cooperative. Households also received
loans from banks such as government banks, commercial banks, Sonali bank and Krishi
Bank. Informal loans were received from neighbors, land owners, relatives and money
lenders. Uniformly across all rounds, women received maximum loans from NGOs.
Nearly 49 percent of the borrowings came from NGOs in the first round following the
flood and there was only a marginal decline to 42 percent in the second and third rounds.
On the other hand, for men, informal sources seem to be more important with 90 percent
of the credit coming from informal sources. It appears easier for women to access NGO
credit whereas patriarchal society promotes men to take loans from neighbors, money
lenders and other informal sources.
Table 6.1: Number of Households in Which Men, Women or Both Take Loans a Round 1 Round 2 Round 3 Men 407 412 379 Women 58 61 58 Both 47 75 82
Note: a total number of households = 673
The mean value of loans taken by the household head is higher than that of the
spouse in the case of bank and institutional credit in all three rounds. The average
amount received by women from NGOs is close to what is received by men in all three
rounds. Curiously, amount borrowed from informal sources is lower for women in the
first round and in the subsequent rounds they catch up with men. Small and in-kind loans
were the highest in the first round and higher for men than women.
117
Table 6.2: Formal and Informal Loans and Amounts by Head and Spouse Round 1 Round 2 Round 3 Head Spouse Head Spouse Head Spouse Type of credit (percentage)
NGOs 2.4 48.7 2.8 42.3 3.21 41.8 Bank/other institutional credit 7.2 11.7 7.9 6.5 3.81 7.5 Informal sources 89.6 36.0 88.9 48.8 92.0 50.7 Small loans/ in-kind loans 0.8 3.6 0.5 2.4 0.8 0.0
Amount of credit (mean)
NGOs 5531.3 5791.7 5700.0 4963.55 5562.5 5500.0 Bank/other institutional credit 8406.4 6021.7 10209.2 6000.0 6579.0 4045.5 Informal sources 3029.1 1724.4 2613.2 2429.2 3068.5 3352.2 Small loans/ in-kind loans in the last 4 weeks 526.4 61.0 105.0 75.7 78.8 0.0
02,
000
4,00
06,
000
Cre
dit A
mou
nt
Her credit His credit
Head and Spouse Credit by Region
Derai Madaripur Mohammadpur MuladiSaturia ShibpurShahrasti
Figure 6.1: Credit Receipt of Household Head and Spouse
Figure 6.1 again indicates that women borrow lower amounts compared to men.
A significant amount of regional variation is also observed. As shown by the poverty
118
dynamics analysis, Saturia region has one of the lowest poverty levels. It is also the
region with the highest borrowings by women. Men in Shahrasti are found to have the
maximum borrowings. Interestingly, this region also had high poverty rates during the
period after the 1998 flood.
Figures 6.2 and 6.3 present the distribution of expenditure shares of all
commodities by region. Figure 6.2 compares only food expenditures across regions and
Figure 6.3 compares all non-food expenditures. The food share is the highest expenditure
share among all regions but there is also not much variation in food expenditures by
region. However, there is some evidence of variation in non-food expenditures. Figure
6.3 indicates that among all regions expenditures on adult goods, personal care and travel
are found to be highest in Shahrasti. Spending on education and fuel is highest in
Muladi. Derai, which is severely poor, has the highest spending of all regions on
cigarettes/beetel and health. Higher spending on health and housing could be reflective
of reconstruction attempts by households after the 1998 floods. Spending on children’s
goods and education was low but uniform across regions with Derai being on the lower
end.
119
Figure 6.2: Food Expenditure Share by Region
Figure 6.3: Non-food Expenditure Shares by Region
0 .2 .4 .6 .8 Food Expenditure Share
Shibpur
Shahrasti
Saturia
Muladi
Mohammadpur
Madaripur
Derai
Food Expenditure by Region
0 .02 .04 .06
0 .02 .04 .06
0 .02 .04 .06
Adult goods Children's goods Cigarettes/beetel Durable goods
Education Fuel Health Housing
Personal Care Social Activities Travel
Derai MadaripurMohammadpur MuladiSaturia ShibpurShahrasti
Graphs by Expenditure
120
Tables 6.3 and 6.4 show why men and women took loans. For each loan type,
individuals were asked the reason for requesting the loan. Table 6.3 shows the
distribution of informal credit and Table 6.4 refers to formal credit. A substantial portion
of informal credit is found to be going towards food for both men and women although
this falls over subsequent survey rounds. The next two important reasons for informal
loans are medical payments and farming. More women give medical payments as the
reason and more men give farming as the reason for requesting a loan. Repayment of
loans is also given as an important reason and this percentage increases in round 3.
Money also seems to be borrowed for professional development like for business, self-
employment and going abroad. It is also interesting to note that loans were taken for
social events including marriage in the third round, indicating return of normalcy after the
flooding in 1998.
Table 6.3: Use of Informal Credit (%)
Reason Round 1 Round 2 Round 3 Men Women Men Women Men Women
Food (including crop) 58.58 57.14 44.75 49.21 36.93 44.59 Education 1.45 0.00 1.03 1.59 1.51 5.41 Doctor/medicine/health 6.32 14.29 6.54 15.87 10.58 14.86 Farming (crop) 6.58 5.19 17.73 3.17 7.56 6.76 Farming (fish) 0.17 0.00 0.00 0.00 1.51 0.00 Farming (livestock & poultry) 1.96 1.30 0.17 0.00 1.51 0.00 Cottage industry 0.09 0.00 0.00 0.00 0.22 0.00 Business 4.61 1.30 3.27 3.17 9.29 0.00 Self-employment 3.07 6.49 2.24 1.59 1.08 0.00 Repayment of loan 2.99 3.90 2.24 4.76 8.86 12.16 Marriage expenses 2.22 0.00 2.24 1.59 4.54 4.05 Dowry 0.26 0.00 0.34 0.00 1.30 0.00 Purchase of land 1.28 2.60 1.72 0.00 1.51 0.00 Agricultural equipment purchase 0.00 0.00 0.52 0.00 0.00 0.00 Going abroad to work 2.90 0.00 3.27 4.76 4.54 1.35 Mortgage in land 0.17 0.00 0.17 0.00 0.43 0.00 Other 7.34 7.79 13.77 14.29 8.64 10.81 Total 100.00 100.00 100.00 100.00 100.00 100.00
Note: Based on own calculations using the IFPRI-FMRSP Bangladesh data 1998-99.
121
According to Table 6.4, food seems to be an important reason for taking formal
credit but not as important as in the case of informal loans. Since formal credit included
loans from banks and NGOs, individuals are under some kind of obligation to
productively use the loan amount. Borrowing towards farming including crop, fish and
livestock & poultry are high and definitely higher than in the case of informal loans.
Investment in income-generating activities including business, self-employment and
cottage industry is higher for formal credit and this is true for both men and women.
Table 6.4: Use of Formal Credit (%) Round 1 Round 2 Round 3
Reason Men Women Men Women Men Women Food (including crop) 18.64 15.38 10.14 11.67 11.43 5.56 Education 1.69 0.00 0.00 0.00 2.86 0.00 Doctor/medicine/health 0.85 2.56 1.45 1.67 2.86 2.78 Farming (crop) 20.34 11.97 30.43 20.00 14.29 6.94 Farming (fish) 0.00 0.85 0.00 0.00 2.86 0.00 Farming (livestock & poultry) 6.78 10.26 0.00 8.33 5.71 13.89 Cottage industry 0.85 1.71 0.00 3.33 0.00 1.39 Business 13.56 26.50 27.54 15.00 20.00 25.00 Self employment 4.24 4.27 0.00 1.67 2.86 0.00 Repayment of loan 12.71 11.97 8.70 15.00 25.71 22.22 Marriage expenses 2.54 1.71 0.00 0.00 0.00 4.17 Dowry 0.85 0.00 0.00 1.67 2.86 0.00 Purchase of land 4.24 0.00 1.45 0.00 0.00 1.39 Agricultural equipment purchase 1.69 0.00 1.45 0.00 0.00 0.00 Going abroad to work 0.85 0.00 5.80 1.67 0.00 0.00 Mortgage in land 0.00 1.71 0.00 0.00 0.00 1.39 Other 10.17 11.11 13.04 20.00 8.57 15.28 Total 100.00 100.00 100.00 100.01 100.00 100.00
Note: Based on own calculations using the IFPRI-FMRSP Bangladesh data 1998-99
Table 6.5 includes the means and standard deviations of the dependent and
independent variables used in the expenditure share equations. Food expenditure
accounts for 71.6 percent of household expenditures. This large share is expected in a
low-income rural economy. Among the non-food items, higher amounts are spent by
122
households on health, housing and adult goods. In total, 3 and 2.8 percent of
consumption expenditures are attributed to cigarettes and fuel use, respectively.
Table 6.5: Means and Standard Deviations Variable Mean Std. Dev. Dependent variables
Food 0.716 0.156 Cigarettes/beetel 0.030 0.026 Adult goods 0.036 0.041 Children's goods 0.015 0.023 Durable goods 0.003 0.008 Education 0.020 0.035 Fuel 0.028 0.042 Health 0.047 0.074 Personal care 0.025 0.014 Housing 0.041 0.111 Travel 0.020 0.037 Social activities 0.009 0.051
Independent variables Log per capita expenditure 8.143 0.600 Log household size 1.638 0.359 Household land area (acres) 1.334 1.951 Wife's amount of credit borrowing 788.34 2745.72 Husband's amount of credit borrowing 3770.22 10642.54
Demographics (years of age) Female share, 0-5 0.087 0.129 Female share , 6-10 0.066 0.117 Female share, 11-15 0.073 0.109 Female share, 16-64 0.284 0.146 Female share, 64 plus 0.016 0.059 Male share, 0-5 0.083 0.120 Male share, 6-10 0.082 0.117 Male share, 11-15 0.073 0.114 Male share, 16-64 0.284 0.146 Male share, 64 plus 0.021 0.077
Note: Based on own calculations using the IFPRI-FMRSP Bangladesh data 1998-99.
6.3.2 Credit and Household Expenditure
The results of the 2SLS and Tobit random effect models are presented in Table
6.6 and Table 6.7. Only food and personal care shares are estimated as OLS models
whereas the Tobit model is estimated for other non-food expenditures. Our endogeneity
123
test found only cigarettes, fuel and health to be exogenous. Results presented in Table
6.6 and Table 6.7 differ in the way in which the credit variable is defined. In Table 6.6,
the total amounts of loans taken by the head and spouse are incorporated as independent
variables. In the subsequent table, credit is measured as a dichotomous variable. If the
individual has borrowed from formal or informal sources then the variable is coded as 1,
otherwise zero.
6.3.3 Amount of Credit and Household Expenditure
Table 6.6 shows that after controlling for location, household demographic shares,
household size, land owned by the household and panel round, the credit provided to the
head or spouse influences household expenditures. This is especially evident when actual
loan amount is used in the analysis, since more estimated coefficients in this model are
significantly different from zero.
Credit does not have any impact on expenditures on social activities, cigarettes,
fuel, health and personal care. Amount borrowed by the head has effects on food
expenditure, adult goods and educational expenditure. Amount of credit taken by the
household head negatively affects food expenditure and positively effects the share spent
on adult goods. Adult goods include clothes and footwear expenditure of both men and
women in the household. Negative and significant impacts on educational expenditures
are also observed. The negative effect on food expenditure has important policy
implications related to nutritional intake of children in the household. Women and girls
in the household may also suffer from resultant nutritional deficiencies.
124
Women’s use of credit has positive and significant effects on expenditures on
children’s goods, durable goods, education and housing. The results show that resources
in the hands of women have implications for improvement in child outcomes, especially
educational outcomes. These results importantly reflect the findings of other studies.
The data were collected immediately after a major flood event, hence repairs and
reconstruction were common. NGOs, private and government institutions were actively
involved in these activities, providing households with income and food. The positive
and significant impact of spouse’s credit on the housing share indicates that resources in
the hands of women also go towards improvement in household and related outcomes.
Table 6.6 indicates that credit received by the spouse has a negative effect on the travel
expense share. This is in line with the fact that women in rural Bangladesh primarily
perform household tasks and look after livestock within the homestead. In Bangladesh,
women are not found to be traveling for work.
Taking into account that food expenditure is an especially important component
of the total expenditure among the households, we disaggregate food share into cereal
share (rice and wheat), plant shares and animal shares. Plant shares include food groups
such as bread, pulses (lentils), oils, vegetables, fruits, spices, sugar and other drinks.
Animal shares include meat, eggs, milk and fish. The random effect models show that
credit has no influence on cereals or food grain consumption. However, negative effects
are observed for plant and animal food products when credit is received by men. That is,
men are less likely to spend their formal and informal credit on plant and animal food
products. It is surprising that we find at this disaggregated level that women are less
likely to spend on animal food products. The round dummy variable coefficients in the
125
food expenditure, education expenditure and personal care expenditure equation show
that households are more likely to spend in round 2 and round 3 than in round 1. This
may be because of the external transfers and assistance the households received from the
government and from NGOs. Within food, households are more likely to spend on
cereals in round 1. On the other hand, households tend to spend more on adult goods,
children’s goods, durable goods, fuel, health and housing in round 1 than in round 2 or 3.
With the immediate impact of floods waning by rounds 2 and 3 households may have
overcome their initial setback and be able to spend on other non-food items.
126
Table 6.6: Effect of Credit Amount on Household Expenditure Shares, OLS and Tobit Estimates Fooda Social Activities Cigarettes/Beetel Adult Goods Coefficient z-statistic Coefficient z-statistic Coefficient z-statistic Coefficient z-statistic Log total monthly expenditure -0.132 -18.93*** -0.005 -0.75 -0.014 -10.95*** 0.017 2.82*** Log total household size 0.108 8.28*** 0.008 1.18 0.003 1.14 -0.014 -2.70*** Household land area (acres) 0.005 2.24** 0.001 0.75 0.000 0.72 0.002 2.69*** Wife's amount of credit borrowing 0.000 -1.18 0.000 0.15 0.000 -0.52 0.000 -1.39 Husband's amount of credit borrowing 0.000 -3.54*** 0.000 1.42 0.000 0.33 0.000 2.50** Location variables
Madaripur -0.020 -1.39 0.007 1.48 -0.010 -3.72*** 0.005 1.32 Mohammadpur -0.015 -1.08 0.003 0.80 -0.017 -6.31*** 0.008 2.16** Muladi -0.018 -1.28 -0.004 -0.85 -0.008 -3.10*** -0.002 -0.65 Saturia -0.022 -1.50 0.005 1.18 -0.013 -4.58*** 0.017 4.48*** Shibpur 0.036 2.51** 0.002 0.36 -0.011 -4.05*** 0.002 0.40 Shahrasti 0.009 0.64 0.008 1.66* -0.007 -2.64*** -0.004 -0.99
Demographics Female share, 0-5 -0.075 -2.49** 0.024 2.39** -0.016 -2.94*** 0.009 1.16 Female share, 6-10 -0.031 -0.88 0.040 3.31*** -0.017 -2.58*** 0.001 0.12 Female share, 11-15 -0.028 -0.87 0.041 3.89*** 0.006 1.01 -0.028 -3.25** Female share, 16-64 -0.019 -0.67 0.080 8.00*** 0.003 0.64 0.010 1.27 Female share, 64 plus -0.079 -1.29 0.035 1.76* -0.003 -0.27 0.026 1.61 Male share, 0-5 -0.040 -1.22 0.007 0.63 -0.007 -1.16 -0.009 -0.98 Male share, 6-10 0.058 1.76 0.011 1.07 -0.013 -2.08** -0.013 -1.46 Male share, 11-15 -0.009 -0.26 0.037 3.27*** 0.000 -0.04 -0.019 -2.06** Male share, 64 plus 0.066 1.40 0.056 3.60*** 0.003 0.29 0.008 0.63
Round variables Round 2 0.063 9.44*** 0.003 1.20 0.002 1.96** -0.011 -5.13*** Round 3 0.064 9.26*** 0.009 3.20*** 0.000 0.36 -0.007 -3.20***
Constant 1.593 28.88*** -0.005 -0.09 0.149 14.82*** -0.078 -1.89* Sigma_u 0.064 0.009 2.83*** 0.014 19.97*** 0.013 9.97*** Sigma_e 0.118 0.048 50.82*** 0.020 50.93*** 0.035 50.99*** Rho 0.226 0.034 0.318 0.130 Log likelihood ratio statistic 3154.08 4594.21 3686.28 P-value 0.000 0.000 0.000
a: OLS models; Derai, males share 16-64 and round 1 are the reference categories in location, demographic and round variables. *, **, *** represent significance at 10, 5 and 1 percent levels.
127
Table 6.6: Effect of Credit Amount on Household Expenditure Shares, OLS and Tobit Estimates (continued)
Children's Goods Durable Goods Education Fuel Coefficient z-statistic Coefficient z-statistic Coefficient z-statistic Coefficient z-statistic Log total monthly expenditure 0.010 2.91*** -0.001 -1.09 0.047 8.98*** 0.001 0.26 Log total household size -0.004 -1.38 0.001 1.03 -0.023 -4.96*** -0.008 -2.03** Household land area (acres) -0.001 -1.50 0.000 3.19*** -0.003 -3.82*** -0.002 -3.76*** Wife's amount of credit borrowing 0.000 1.90* 0.000 2.56** 0.000 1.94* 0.000 -0.48 Husband's amount of credit borrowing 0.000 -0.62 0.000 -0.47 0.000 -2.33** 0.000 0.54 Location variables
Madaripur 0.004 1.81* 0.002 2.49** 0.006 1.64* 0.013 3.31*** Mohammadpur 0.004 1.88* 0.003 3.73*** 0.017 5.20*** 0.000 0.11 Muladi 0.003 1.52 0.001 1.26 0.021 6.20*** 0.026 6.55*** Saturia 0.007 3.14*** 0.001 1.44 0.017 4.83*** 0.002 0.53 Shibpur 0.002 1.10 0.002 2.60*** 0.003 0.83 -0.006 -1.46 Shahrasti 0.000 0.05 0.002 2.77*** 0.007 1.96** 0.004 0.90
Demographics Female share, 0-5 0.007 1.61 0.004 2.29** -0.005 -0.69 0.017 1.98** Female share, 6-10 0.018 3.27*** -0.001 -0.38 0.006 0.70 -0.006 -0.60 Female share, 11-15 0.022 4.60*** -0.002 -0.90 0.009 1.25 0.007 0.81 Female share, 16-64 -0.015 -3.41*** -0.002 -1.16 -0.020 -2.98*** 0.001 0.14 Female share, 64 plus -0.010 -1.06 0.007 2.02** -0.007 -0.46 0.009 0.49 Male share, 0-5 -0.003 -0.49 0.001 0.56 -0.008 -0.99 0.018 1.95* Male share, 6-10 0.003 0.55 -0.002 -1.12 -0.003 -0.42 -0.013 -1.40 Male share, 11-15 0.001 0.27 0.001 0.64 0.014 1.74* -0.008 -0.79 Male share, 64 plus -0.019 -2.61*** 0.001 0.55 -0.022 -2.01** -0.012 -0.87
Round variables Round 2 -0.003 -2.19** -0.004 -7.79*** 0.006 3.92*** -0.008 -3.69*** Round 3 -0.006 -4.43*** -0.004 -7.71*** 0.008 4.75*** -0.010 -4.69***
Constant -0.055 -2.41** 0.013 1.53 -0.333 -9.15*** 0.038 2.31** Sigma_u 0.003 1.93* 0.000 0.00 0.016 18.22*** 0.014 9.21*** Sigma_e 0.022 50.91*** 0.008 62.69*** 0.026 51.09*** 0.038 51.09*** Rho 0.023 0.000 0.286 0.118 Log likelihood ratio statistic 4685.72 6647.74 4130.15 3514.68 P-value 0.000 0.000 0.000 0.000
a: OLS models; Derai, males shares 16-64 and round 1 are the reference categories in location, demographic and round variables. *, **, *** represent significance at 10, 5 and 1 percent level.
128
Table 6.6: Effect of Credit Amount on Household Expenditure Shares, OLS and Tobit Estimates (continued)
Health Personal Carea Housing Travel Coefficient z-statistic Coefficient z-statistic Coefficient z-statistic Coefficient z-statistic Log total monthly expenditure 0.011 3.18*** 0.003 1.16 0.024 1.50 0.018 3.18*** Log total household size 0.000 0.01 -0.003 -1.34 -0.046 -3.28*** -0.008 -1.65* Household land area (acres) -0.002 -1.81* 0.000 -0.12 -0.002 -0.94 0.000 -0.56 Wife's amount of credit borrowing 0.000 -0.66 0.000 0.12 0.000 3.04*** 0.000 -2.10** Husband's amount of credit borrowing 0.000 0.67 0.000 -0.55 0.000 0.72 0.000 -0.23 Location Variables
Madaripur -0.004 -0.56 0.004 2.52** -0.004 -0.37 0.008 2.15** Mohammadpur -0.016 -2.55** 0.006 4.01*** -0.010 -1.07 0.014 4.07*** Muladi -0.006 -0.98 0.004 3.00*** -0.018 -1.84* 0.009 2.74*** Saturia -0.028 -4.11*** 0.006 4.15*** -0.004 -0.42 0.017 4.73*** Shibpur -0.009 -1.45 0.003 1.65** -0.021 -2.01** 0.006 1.73* Shahrasti -0.009 -1.33 0.002 1.59 -0.006 -0.56 0.003 0.70
Demographics Female share, 0-5 0.009 0.66 0.007 2.34** 0.000 -0.01 0.010 1.29 Female share, 6-10 0.005 0.32 -0.002 -0.52 -0.014 -0.53 -0.008 -0.92 Female share, 11-15 -0.003 -0.18 0.001 0.33 -0.014 -0.61 -0.006 -0.73 Female share, 16-64 0.021 1.54 0.008 2.45** -0.042 -1.96** -0.018 -2.42** Female share, 64 plus 0.014 0.47 0.008 1.25 -0.004 -0.10 0.004 0.24 Male share, 0-5 0.038 2.41** 0.000 -0.04 -0.010 -0.42 0.002 0.20 Male share, 6-10 -0.011 -0.69 -0.006 -1.63 -0.004 -0.18 -0.003 -0.39 Male share, 11-15 0.010 0.62 -0.005 -1.46 -0.012 -0.49 -0.007 -0.80 Male share, 64 plus -0.012 -0.56 -0.011 -2.16** -0.046 -1.36 -0.011 -0.90
Round variable Round 2 -0.042 -11.22*** 0.003 4.48*** -0.012 -2.04** 0.003 1.50 Round 3 -0.031 -8.13*** 0.004 4.35*** -0.029 -4.78*** 0.002 0.89
Constant -0.018 -0.66 0.000 0.00 -0.036 -0.34 -0.115 -2.97*** Sigma_u 0.019 5.90*** 0.005 0.01 0.031 7.30*** 0.013 10.17*** Sigma_e 0.067 51.11*** 0.012 0.01 0.096 50.59*** 0.033 51.12*** Rho 0.072 0.141 0.18 0.094 0.132 Log likelihood ratio statistic 2466.85 1718.27 3800.08 P-value 0.000 0.000 0.000
a: OLS models; Derai, males shares 16-64 and round 1 are the reference categories in location, demographic and round variables. *, **, *** represent significance at 10, 5 and 1 percent levels.
129
Table 6.6: Effect of Credit Amount on Household Expenditure Shares, OLS and Tobit Estimates (continued)
Cereal Plant Products Animal Products Coefficient z-statistic Coefficient z-statistic Coefficient z-statistic Log total monthly expenditure -0.211 -13.44*** 0.025 1.61 0.076 7.18*** Log total household size 0.179 12.45*** -0.034 -2.40** -0.048 -4.97*** Household land area (acres) 0.006 3.13*** -0.002 -0.80 -0.002 -1.16 Wife's amount of credit borrowing 0.000 -0.43 0.000 -0.21 0.000 -1.71* Husband's amount of credit borrowing 0.000 -0.15 0.000 -2.76*** 0.000 -2.81*** Location variables
Madaripur -0.016 -1.55 -0.003 -0.27 -0.011 -1.66* Mohammadpur 0.014 1.45 -0.031 -3.12*** -0.003 -0.42 Muladi -0.036 -3.58*** 0.015 1.51 -0.011 -1.63 Saturia -0.003 -0.23 -0.022 -2.09** -0.002 -0.34 Shibpur 0.008 0.71 0.011 0.98 0.010 1.40 Shahrasti -0.049 -4.63*** 0.032 3.05*** 0.012 1.71*
Demographics Female share, 0-5 -0.071 -3.13*** -0.031 -1.40 0.020 1.27 Female share, 6-10 -0.039 -1.42 0.022 0.85 0.004 0.22 Female share, 11-15 0.009 0.36 -0.029 -1.25 -0.012 -0.76 Female share, 16-64 0.028 1.30 -0.018 -0.86 -0.026 -1.74* Female share, 64 plus -0.050 -1.10 0.003 0.08 -0.018 -0.60 Male share, 0-5 -0.047 -1.83* -0.008 -0.34 0.019 1.10 Male share, 6-10 0.068 2.79*** -0.005 -0.20 0.002 0.13 Male share, 11-15 0.052 2.04** -0.040 -1.62 -0.032 -1.85* Male share, 64 plus 0.098 2.79*** 0.005 0.15 -0.040 -1.69*
Round variable Round 2 -0.031 -5.33*** 0.081 15.79*** 0.018 4.56*** Round 3 -0.048 -7.80*** 0.069 12.35*** 0.048 11.13***
Constant 1.761 16.21*** 0.098 0.91 -0.431 -5.90*** Sigma_u 0.036 0.041 0.021 Sigma_e 0.102 0.084 0.074 Rho 0.109 0.195 0.077
a: OLS models; Derai, males shares 16-64 and round 1 are the reference categories in location, demographic and round variables. *, **, *** represent significance at 10, 5 and 1 percent levels.
130
Apart from these two variables of interest, regional dynamics are clearly evident.
Keeping in mind the high poverty levels in Derai, the Tobit model estimated for cigarette
shares shows that all other regions are significantly less likely to spend on this
commodity. Similarly, compared to Derai, all regions except for Shibpur are more likely
to spend on education. The survey round variables indicate that households are more
likely to spend on food in round one compared to rounds two and three, although
expenditure on cereal indicates the opposite. This is true for education and personal care
as well. However, for durable goods, adult goods, children’s goods, fuel, health and
repair, households are more likely to spend in round one.
6.3.4 Credit Participation and Household Expenditure
Finally, Table 6.7 presents results that are similar to those presented in Table 6.6.
Borrowing done by men is observed to have negative effects on expenditures on food,
cigarettes and children’s goods. Within the food category, plant and animal products
show significant negative effects. The table also shows that spending on social activity,
health and housing are positively linked to the male head’s credit receipt. Women’s
borrowing has positive and significant effects on durable goods and housing. Use of
dichotomous variables as bargaining measures is not as effective as using actual credit
amounts.
131
Table 6.7: Effect of Credit (Dichotomous) on Household Expenditure Shares, OLS and Tobit Estimates
Fooda Social Activities Cigarettes/Beetel Adult Goods Coefficient z-statistic Coefficient z-statistic Coefficient z-statistic Coefficient z-statistic Log total monthly expenditure -0.136 -19.64*** -0.004 -0.58 -0.014 -10.99*** 0.017 2.93*** Log total household size 0.107 8.17*** 0.008 1.19 0.003 1.08 -0.014 -2.70*** Household land area (acres) 0.005 2.18** 0.001 0.73 0.000 0.64 0.002 2.82*** Wife's credit (dummy) -0.011 -1.23 0.004 1.26 -0.001 -0.88 -0.004 -1.54 Husband's credit (dummy) -0.022 -3.52*** 0.009 3.80*** -0.002 -1.67* -0.002 -0.96 Location variables
Madaripur -0.022 -1.56 0.007 1.63 -0.010 -3.77*** 0.005 1.28 Mohammadpur -0.021 -1.48 0.005 1.24 -0.017 -6.41*** 0.008 2.09** Muladi -0.021 -1.45 -0.003 -0.74 -0.008 -3.10*** -0.002 -0.58 Saturia -0.028 -1.88* 0.007 1.46 -0.013 -4.68*** 0.017 4.37*** Shibpur 0.033 2.28** 0.003 0.58 -0.011 -4.15*** 0.001 0.32 Shahrasti 0.006 0.40 0.008 1.74* -0.007 -2.65*** -0.003 -0.83
Demographics Female share, 0-5 -0.070 -2.33*** 0.023 2.34** -0.016 -2.92*** 0.009 1.10 Female share, 6-10 -0.030 -0.86 0.039 3.28*** -0.016 -2.54** 0.002 0.19 Female share, 11-15 -0.029 -0.88 0.041 3.91*** 0.006 1.00 -0.028 -3.24*** Female share, 16-64 -0.023 -0.81 0.082 8.16*** 0.003 0.56 0.009 1.16 Female share, 64 plus -0.082 -1.33 0.037 1.84* -0.003 -0.28 0.027 1.62 Male share, 0-5 -0.038 -1.15 0.007 0.64 -0.007 -1.14 -0.009 -0.97 Male share, 6-10 0.060 1.83* 0.011 1.03 -0.013 -2.08** -0.013 -1.52 Male share, 11-15 -0.007 -0.21 0.036 3.22*** 0.000 -0.05 -0.019 -2.12** Male share, 64 plus 0.065 1.38 0.059 3.80*** 0.002 0.18 0.006 0.43
Round Variables Round 2 0.063 9.33*** 0.004 1.60 0.002 1.68* -0.011 -5.45*** Round 3 0.062 8.82*** 0.011 3.78*** 0.000 -0.05 -0.008 -3.59***
Constant 1.635 29.76*** -0.023 -0.47 0.150 14.99*** -0.077 -1.92* Sigma_u 0.064 0.009 2.99*** 0.014 20.08*** 0.013 9.90*** Sigma_e 0.118 0.048 50.88*** 0.020 50.93*** 0.035 50.98*** rho 0.229 0.036 0.320 Log likelihood ratio statistic 3159.91 4595.60 P-value 0.0000 0.0000
a: OLS models; Derai, males shares 16-64 and round 1 are the reference categories in location, demographic and round variables. *, **, *** represent significance at 10, 5 and 1 percent levels.
132
Table 6.7: Effect of Credit (Dichotomous) on Household Expenditure Shares, OLS and Tobit Estimates (continued)
Children's Goods Durable Goods Education Fuel Coefficient z-statistic Coefficient z-statistic Coefficient z-statistic Coefficient z-statistic Log total monthly expenditure 0.009 2.87*** -0.001 -1.00 0.047 9.04*** 0.001 0.32 Log total household size -0.004 -1.40 0.001 0.99 -0.023 -4.97*** -0.008 -2.06** Household land area (acres) -0.001 -1.53 0.000 3.10*** -0.003 -3.78*** -0.002 -3.79*** Wife's credit (dummy) 0.000 -0.24 0.001 2.64*** 0.002 0.84 -0.003 -1.08 Husband's credit (dummy) -0.002 -1.89* 0.000 0.37 -0.001 -0.71 0.000 0.08 Location Variables
Madaripur 0.004 1.80* 0.002 2.40** 0.006 1.64* 0.013 3.36*** Mohammadpur 0.003 1.70* 0.003 3.67*** 0.017 5.07*** 0.001 0.16 Muladi 0.003 1.50 0.001 1.28 0.021 6.10*** 0.026 6.57*** Saturia 0.007 3.26*** 0.001 1.47 0.017 4.86*** 0.003 0.63 Shibpur 0.002 1.05 0.002 2.60*** 0.003 0.85 -0.006 -1.45 Shahrasti 0.000 0.09 0.002 2.66*** 0.007 1.85* 0.004 0.97
Demographics Female share, 0-5 0.007 1.59 0.004 2.27** -0.005 -0.65 0.017 1.97** Female share, 6-10 0.018 3.31*** -0.001 -0.42 0.006 0.69 -0.006 -0.57 Female share, 11-15 0.022 4.61*** -0.002 -0.93 0.009 1.26 0.008 0.82 Female share, 16-64 -0.016 -3.47*** -0.002 -1.08 -0.021 -3.00*** 0.001 0.10 Female share, 64 plus -0.010 -1.14 0.007 2.01** -0.008 -0.54 0.008 0.48 Male share, 0-5 -0.003 -0.55 0.001 0.52 -0.008 -1.00 0.019 1.96** Male share, 6-10 0.003 0.60 -0.002 -1.11 -0.003 -0.38 -0.013 -1.40 Male share, 11-15 0.002 0.37 0.001 0.64 0.015 1.80* -0.008 -0.78 Male share, 64 plus -0.020 -2.81*** 0.002 0.60 -0.022 -2.03** -0.013 -0.93
Round variables Round 2 -0.003 -2.48** -0.004 -7.60*** 0.006 3.94*** -0.008 -3.74*** Round 3 -0.007 -4.76*** -0.004 -7.45*** 0.008 4.67*** -0.010 -4.66***
Constant -0.050 -2.26** 0.011 1.40 -0.329 -9.21*** 0.037 2.30** Sigma_u 0.003 2.02** 0.000 0.00 0.017 18.31*** 0.014 9.28*** Sigma_e 0.022 50.95*** 0.008 62.69*** 0.026 51.04*** 0.038 51.09*** rho 0.024 0.000 0.288 0.119 Log likelihood ratio statistic 4684.75 6647.02 4124.80 P-value 0.000 0.000 0.000
a: OLS models; Derai, males shares 16-64 and round 1 are the reference categories in location, demographic and round variables. *, **, *** represent significance at 10, 5 and 1 percent levels.
133
Table 6.7: Effect of Credit (Dichotomous) on Household Expenditure Shares, OLS and Tobit Estimates (continued)
Health Personal Carea Housing Travel Coefficient z-statistic Coefficient z-statistic Coefficient z-statistic Coefficient z-statistic Log total monthly expenditure 0.011 3.24*** 0.003 1.62 0.026 1.70* 0.017 3.19*** Log total household size 0.001 0.14 -0.003 -1.72* -0.047 -3.32*** -0.008 -1.63 Household land area (acres) -0.002 -1.64* 0.000 -0.30 -0.002 -1.12 0.000 -0.49 Wife's of credit (dummy) 0.004 0.83 0.001 0.84 0.012 1.79* -0.004 -1.52 Husband's credit (dummy) 0.008 2.42** -0.001 -1.47 0.011 2.16** 0.000 0.14 Location variables
Madaripur -0.003 -0.48 0.003 2.59*** -0.002 -0.22 0.008 2.17** Mohammadpur -0.015 -2.31** 0.005 4.23*** -0.007 -0.76 0.014 4.06*** Muladi -0.006 -0.96 0.004 3.35*** -0.017 -1.73* 0.009 2.72*** Saturia -0.027 -4.07*** 0.006 4.35*** 0.000 0.04 0.017 4.62*** Shibpur -0.008 -1.29 0.002 1.62 -0.019 -1.83* 0.006 1.73* Shahrasti -0.008 -1.30 0.002 1.53 -0.005 -0.46 0.003 0.69
Demographics Female share, 0-5 0.009 0.64 0.007 2.45*** -0.002 -0.07 0.010 1.31 Female share, 6-10 0.004 0.26 -0.002 -0.72 -0.014 -0.54 -0.009 -0.93 Female share, 11-15 -0.003 -0.18 0.001 0.28 -0.013 -0.59 -0.006 -0.71 Female share, 16-64 0.023 1.68* 0.006 2.07** -0.039 -1.83* -0.019 -2.45** Female share, 64 plus 0.016 0.54 0.007 1.25 -0.005 -0.12 0.004 0.25 Male share, 0-5 0.038 2.43** 0.000 -0.13 -0.011 -0.46 0.002 0.23 Male share, 6-10 -0.011 -0.74 -0.006 -2.10** -0.004 -0.16 -0.003 -0.41 Male share, 11-15 0.009 0.55 -0.006 -1.88* -0.010 -0.42 -0.007 -0.83 Male share, 64 plus -0.008 -0.37 -0.012 -2.59*** -0.046 -1.36 -0.010 -0.86
Round variables Round 2 -0.041 -10.86*** 0.003 4.80*** -0.011 -1.87* 0.003 1.56 Round 3 -0.029 -7.54*** 0.004 4.57*** -0.027 -4.40*** 0.002 0.98
Constant -0.026 -0.98 -0.004 -0.26 -0.064 -0.60 -0.112 -2.97*** Sigma_u 0.018 5.50*** 0.005 0.031 7.40*** 0.013 10.02*** Sigma_e 0.067 51.04*** 0.012 0.097 50.61*** 0.033 51.10*** rho 0.067 0.140 0.095 0.130 Log likelihood ratio statistic 2469.41 1716.23 P-value 0.000 0.000
a: OLS models; Derai, males shares 16-64 and round 1 are the reference categories in location, demographic and round variables. *, **, *** represent significance at 10, 5 and 1 percent levels.
134
Table 6.7: Effect of Credit (Dichotomous) on Household Expenditure Shares, OLS and Tobit Estimates (continued)
Cereal Plant Products Animal Products Coefficient z-statistic Coefficient z-statistic Coefficient z-statistic Log total monthly expenditure -0.211 -14.19*** 0.021 1.41 0.073 7.04*** Log total household size 0.180 12.87*** -0.033 -2.39** -0.047 -4.89*** Household land area (acres) 0.006 3.22*** -0.002 -0.81 -0.002 -1.19 Wife's of credit (dummy) 0.003 0.40 -0.007 -1.02 -0.008 -1.61 Husband's credit (dummy) -0.001 -0.24 -0.012 -2.66*** -0.010 -2.91*** Location Variables
Madaripur -0.017 -1.65* -0.003 -0.34 -0.012 -1.77* Mohammadpur 0.014 1.39 -0.033 -3.39*** -0.005 -0.78 Muladi -0.037 -3.69*** 0.014 1.38 -0.012 -1.78* Saturia -0.004 -0.39 -0.024 -2.26** -0.005 -0.71 Shibpur 0.007 0.69 0.009 0.88 0.009 1.21 Shahrasti -0.049 -4.75*** 0.030 2.91*** 0.011 1.49
Demographics Female share, 0-5 -0.072 -3.22*** -0.029 -1.32 0.021 1.34 Female share, 6-10 -0.039 -1.46 0.022 0.84 0.003 0.16 Female share, 11-15 0.008 0.33 -0.029 -1.25 -0.013 -0.77 Female share, 16-64 0.028 1.31 -0.019 -0.93 -0.028 -1.85* Female share, 64 plus -0.049 -1.11 0.001 0.01 -0.021 -0.67 Male share, 0-5 -0.047 -1.87* -0.008 -0.34 0.019 1.11 Male share, 6-10 0.067 2.78*** -0.003 -0.14 0.003 0.15 Male share, 11-15 0.050 2.01** -0.038 -1.55 -0.031 -1.81* Male share, 64 plus 0.098 2.86*** 0.004 0.13 -0.040 -1.68*
Round Variables Round 2 -0.031 -5.18*** 0.081 15.49*** 0.018 4.47*** Round 3 -0.048 -7.55*** 0.068 11.80*** 0.047 10.69***
Constant 1.763 17.07*** 0.134 1.29 -0.401 -5.60*** Sigma_u 0.032 0.041 0.023 Sigma_e 0.105 0.084 0.072 rho 0.083 0.195 0.091
a: OLS models; Derai, males shares 16-64 and round 1 are the reference categories in location, demographic and round variables. *, **, *** represent significance at 10, 5 and 1 percent levels.
135
Chapter 7
Conclusions
7.1 Introduction
The Human Poverty Index reported by the Human Development Report (HDR
2005) places Bangladesh at the 86th position among 103 developing countries. The report
ranks Bangladesh in the 105th position for Gender-related Development Index (out of 140
countries) and in the 78th position for the Gender Empowerment measure (among the 80
countries). In contrast, the United States is ranked 8th and 12th respectively for the
Gender-related Development Index and the Gender Empowerment measure. Despite its
progress in the last 35 years after its independence, Bangladesh has a long way to go in
tackling the problem of both poverty and gender empowerment. Apart from high poverty
levels and low gender empowerment rates, the country also faces yearly natural disasters
in the form of floods. Issues related to poverty and the environment become critical, as
vulnerable households seek to cope with constantly reoccurring environmental disasters
such as floods.
Given this scenario, this dissertation recognizes the multi-dimensionality and
heterogeneity of the poor. It first analyzes issues relating to chronic and transient poverty
following a major catastrophic (flood) event using a short panel of household data from
Bangladesh. The International Food Policy Research Institute’s Food Management and
Research Support Project (IFPRI-FMRSP) household survey of rural Bangladesh for the
years 1998-99 is used for the analysis. The households were interviewed in three waves
including approximately 750 households in seven flood-affected thanas (administrative
units). The data were collected between the 3rd week of November and the 3rd week of
136
December 1998, between April and May 1999 and finally, collected exactly a year after
the first round (November-December 1999). Bangladesh experienced the largest floods
of the century in 1998. An increase in private borrowing was one of the medium-term
impacts of the floods. Borrowing occurred by men and women, having potentially
differential impacts on poor households seeking to cope with this significant natural
disaster.
The traditional method of consumption expenditure is used to measure poverty
and longitudinal data are useful for studying movements into and out of poverty.
Households are differentiated on the basis of their poverty experience using the FGT
poverty measures. Then the characteristics that distinguish between those who are able to
eventually escape poverty following the flood (the transient poor) versus those unable to
leave poverty (chronic poor) are identified. The study uses cost-of-basic-needs (CBN)
poverty lines calculated by the World Bank for Bangladesh for the year 2000.
Two approaches are used to categorize the poor. First, the McCulloch and Baulch
(1999) method is used to categorize households into three mutually exclusive groups:
never poor, chronically poor and transitory poor based on mean household expenditure
levels and poverty lines. A household is defined as chronically poor if its mean
expenditure is below the poverty line across all periods and transitory poor if its mean
expenditure is above the poverty line but total per capita household expenditure is not
above the poverty line for all periods. If the total household expenditure is above the
poverty line in all rounds then the household is defined as never poor. Given these three
mutually exclusive groups, we use multinomial logit models to study the determinants of
chronic and transient poverty comparing them to those households that were never poor.
137
Independent variables come from the data from the first round as most are time-invariant
except the financial-asset variable. This method aims to distinguish the chronic and the
transient poor from non-poor households.
The second approach uses the Jalan and Ravallion (2000) method of classifying
the poor into total, chronic and transient poor. This approach involves calculating an
aggregate inter-temporal poverty measure for each household. We calculate the squared
poverty gap index. Households who have mean consumption below the poverty line and
whose household consumption is below the poverty line in all periods are defined as
chronically poor. The transient poor are those who have mean consumption levels below
the poverty line but are not poor in all periods. Their consumption expenditure scould be
above the poverty line in some rounds. In the calculations of poverty status, there are
some households that do not experience any poverty according to the squared poverty
gap index. Households who are non-poor generate ‘bunches’ of zero values in the
dependent variables are poor and thus there is the need to use a censored model. The
study uses Censored Quantile Regression models which are robust to heteroscedastic and
non-normality assumptions. This method attempts to identify the correlates of each kind
of poverty.
We find that mean consumption expenditure declined and poverty levels
increased between round 1 and round 3 of the IFPRI-FMRSP household survey which
implies that household consumption levels were boosted by government and NGO
transfers and aid in round 1. This is evident from first-order stochastic dominance tests
which show that household welfare deteriorated over the survey period. Both approaches
138
show that the majority of households in the data are chronically poor33. Using the upper
poverty line, 400 (55.02 percent) and 159 (21.87 percent) of households are chronically
and transient poor, respectively. Household size, dependency ratio, number of working
members, land ownership, location, social assistance and education characterize the
chronically poor. Ownership of physical and human capital make households less likely
to be chronically poor. Larger household size and dependents in the household push
families towards chronic poverty. Increase in number of working members in the family
bring in more income and reduce the chances of household being chronically poor.
Given that Bangladesh is an agrarian society and faces yearly floods it is not surprising
that households with heads employed in trade and self-employment sector are less likely
to be chronically poor compared to those in the agricultural sector. . Long terms
investments in human and physical assets clearly help households out of chronic poverty.
Apart from household size, dependency ratio, number of working members, land
ownership transient poor are characterized by credit access. Credit access and
remittances explain transient poverty better. Our models are not able to characterize the
transient poor as well as it does the chronically poor. This has found to be true in other
studies as well (Haddad and Ahmed 2003).
After having studied the poor and their characteristics, the research next studies
how individuals interact and operate within a family or household. Realizing that to
improve the well-being of individuals, development policies not only have to take into
account how resources are allocated within the family or household but also consider the
impact of this resource allocation on individuals. This would go a long way in achieving
33 Around 55 percent and 37 percent of the households are chronically poor according to the upper and lower poverty line.
139
the third millennium development goal of empowerment of women. In order to return to
their original level of consumption, households adopt various consumption-smoothing
strategies. Formal and informal borrowing is one of the most popular means of
consumption smoothing in Bangladesh especially after the 1998 floods. Our qualitative
study reiterates this as well.
For this objective, we looked at intrahousehold dynamics (e.g., variations in
household bargaining behaviors) with a focus on the household’s expenditure patterns.
Receipt of credit is taken as the measure of bargaining between the head and the spouse.
The dissertation does not distinguish between formal and informal sources of credit and
we argue that whatever the source of credit it influences woman’s decision-making
abilities and has positive outcomes for the household. The latter is tested.
The dissertation followed Quisumbing and de la Briere (2000) and Swaminathan
(2003) and considered agricultural household with two members (head and spouse) in the
household and used the Nash-bargaining model to study the role of credit receipt in
determining bargaining power. We also restrict our analysis to male-headed households
since there were too few female-headed households in our data. Food and non-food share
equations were individually estimated using random effect OLS and Tobit models to test
if participation in the credit markets influenced the food and non-food expenditure shares.
Endogeneity corrections were incorporated whenever tests indicated an endogenous
relationship between total household consumption and a particular expenditure share.
2SLS models and simultaneous Tobit models were used to correct for endogeneity
between the expenditure shares and total expenditure.
140
Non-food categories used in the analysis included cigarettes/beetel, adult goods,
children's goods, durable goods, education, fuel, health, personal care, housing, travel and
social activities. The data provided information on each loan type in every survey round
which enabled identification of the individual who took the loan in the household and 655
male-headed households were included in the analysis. Independent variables used in the
analysis included household size, location of household and the round in which data were
collected.
Our results indicated that more men compared to women participate in the credit
market. As is typical of any rural-developing economy, household expenditure share is
highest for food34. Amount borrowed by the head has effects on food expenditure, adult
goods and education expenditure. Amount of credit taken by the household head
negatively affects food expenditure and positively affects share spent on adult goods. The
negative effect on food expenditure has policy implications related to nutritional intake of
children in the household. Women and girls in the household may also suffer from
resultant nutritional deficiencies. Women’s use of credit has positive impacts on expense
on children’s goods, durable goods, education and housing. The results show that
resources in the hands of women have implications for improvement in child outcomes,
especially educational outcomes. The data were collected immediately after a major
flood event, hence repairs and reconstruction were common. NGOs, private and
government institutions were actively involved in these activities, providing people with
income and food. The positive and significant impact of spouse’s credit on housing share
indicates that resources in the hands of women go towards improvement in household
and related outcomes. At the same time, we find that credit receipt by women has a 34 We find 70 percent of the total expenditure is on food.
141
negative effect on travel expenses. This is in line with the fact that women in rural
Bangladesh primarily perform household tasks and look after livestock within the
homestead. In Bangladesh, they are not found to be traveling for work..
Finally, qualitative field work in Bangladesh was undertaken in February 2005 to
provide a more solid understanding of the country and more perspective on poverty and
intrahousehold bargaining situations within Bangladeshi households. The focus group
discussions, which centered around the coping strategies employed by households during
and after floods and decision-making in households, helped to more clearly interpret the
empirical results. The field research proved to be a very important research tool.
7.2 Policy Implications
This research shows that poverty and gender issues go hand in hand, with both
issues being policy relevant from a disaster-management point-of-view. Engendering of
poverty issues has important implications for economic growth, individual welfare, and
the coping strategies that households can use. Identification of the poor and their
characteristics suggests that the poor are heterogeneous and substantially differ from each
other. We find that transient poor, although better off than the chronically poor both in
terms of human and physical capital accumulation, yet are difficult to identify. They are
more vulnerable to shocks as their consumption levels are close to the poverty line. On
the other hand, the chronically poor are easily defined according to our analysis and have
extremely low consumption levels. These findings call for different approaches to help
different poverty groups. Slight modification of poverty-alleviating policies to target
different poverty groups could have huge impacts on the success of these policies.
142
Standard poverty policies aim at increasing the mean utility of the poor. Our study shows
that policies aiming at increasing mean utility would help the chronically poor who have
lower consumption levels whereas policies that aim to reduce variance of household well-
being would benefit the transient poor who move in and out of poverty.
We also find that women’s receipt of credit has positive implications for child
outcomes in general and educational outcomes in particular. This is in line with other
studies on intrahousehold allocation where empowerment of women in the household has
beneficial effects on children, health and nutritional attainment. A gender mainstreaming
approach to poverty eradication should be adopted. Our study looks at the combined
effect of both formal and informal credit on the household. Despite the presence of
micro-finance institutions in Bangladesh, the amount of formal credit going to the
household is small. Informal borrowing from moneylenders, friends and family is still
the most common method of consumption smoothing. It is recommended that
microfinance programs be further supported so that their prominence among poorest of
the poor be increased. This mandates improvement in credit access in rural Bangladesh,
especially that targeting women. This is important because studies show that gender,
poverty and development are closely linked. We cannot deal with these issues in
isolation.
7.3 Future Research
This research is based on traditional income and consumption measures of
poverty. There is an increasing emphasis in the literature on adopting a multidimensional
approach by considering other measures such as educational attainment, nutritional intake
143
and ownership of assets in the analysis (McKay and Lawson 2002). Given the richness
of IFPRI-FMRSP Bangladesh survey and availability of anthropometric information, it
would be possible to extend the analysis to use these non-traditional measures.
Recent developments in research show that space matters. Our results indicate
differences in outcomes due to place of residence of the households. Proximity to Dhaka
has an important role to play in securing access to resources and opportunities. However,
the IFPRI-FMRSP Bangladesh survey only has thana-level information of the residence
and smaller units are not available to apply any spatial econometric models to capture the
impact of space in the bargaining models.
Finally, extended family structure is common in developing economies. Taking
into account the bargaining power of other members of the family may be important in
understanding intrahousehold allocation outcomes (Quisumbing 2003). Exploring the
household expenditure pattern resulting from bargaining between other members of the
family (such as between daughter-in-law and mother-in-law) could be an interesting
extension.
144
Reference
Alderman, H. 1996. Saving and Economic Shocks in Rural Pakistan. Journal of Development Economics 51(2): 343-365. Alderman, H. and P. Gertler. 1997. Family Resources and Gender Differences in Human Capital Investments: The Demand for Children's Medical Care in Pakistan. In Intrahousehold Resource Allocation in Developing Countries: Models, Methods and Policy, eds. Haddad. L., J. Hoddinott and H. Alderman. Baltimore: Johns Hopkins University Press, for the International Food Policy Research Institute. Aliber, M. 2003. Chronic Poverty in South Africa: Incidence, Causes and Policies. World Development 31(3): 473–490. Apps, P.F. and R. Rees. 1997. Collective labor-supply and household production Journal of Political Economy 105: 178-190. Bamberger, M., M. Blackden, L. Fort and V. Manoukian. 2002. Gender. Poverty Measurement and Analysis. PRSP Sourcebook, World Bank, Washington D.C. Available at: http://povlibrary.worldbank.org/files/4221_chap10.pdf Bane, T.J. and D.T. Ellwood. 1983. Slipping Into and Out of Poverty: The Dynamics of Spells. NBER Working Paper No. 1199, National Bureau of Economic Research, Cambridge, MA. Bangladesh Bureau of Statistics (BBS). 2001. Preliminary Report of Household Income and Expenditure Survey 2000. Dhaka, Bangladesh. Baulch, B. and J. Hoddinott. 2000. Economic Mobility and Poverty Dynamics in Developing Countries. Journal of Development Studies 36(6): 1-24. Basu, K. 2001. Gender and Say: A Model of Household Behaviour with Endogenously – Determined Balance of Power. CAE Working Paper-01-01, Cornell University, USA. Becker, G. 1965. A Theory of the Allocation of Time. Economic Journal 75: 492-517. Beck, T. 2005. Learning Lessons from Disaster Recovery: The Case of Bangladesh. Working Paper Series No. 11. The World Bank, Washington D.C. Behrman, J. 1997. Intrahousehold Distribution and the Family. In Handbook of Population and Family Economics, eds. Mark Rosenzweig and Oded Stark. Amsterdam. New York: Elsevier. Benson, C. and Edward Clay. 2002. Bangladesh Disasters and Public Finance. Disaster Risk Management Working Paper Series No. 6. The World Bank.
145
Bloch, F. and V. Rao. 2002. Terror as a Bargaining Instrument: A Case Study of Dowry Violence in Rural India. The American Economic Review 92(4):1029-1043. Browning, M., F. Bourguignon, P.A. Chiappori and V. Lechene. 1994. Income and Outcomes: A Structural Model of Intrahousehold Allocations. Journal of Political Economy 102(6): 1067-1096. Browning, M. and P.A. Chiappori. 1998. Efficient Intra-household Allocations: A General Characterization and Empirical Tests. Econometrica 66(6): 1214-1278. Carter, M.R and E. G. Katz. 1997. Separate Spheres and the Conjugal Contract: Understanding the Impact of Gender-biased Development. In Intrahousehold Resource Allocation in Developing Countries: Models, Methods and Policy, eds. Haddad. L., J. Hoddinott and H. Alderman. Baltimore: Johns Hopkins University Press, for the International Food Policy Research Institute. Cartwright, J., S. R. Khandker and M.M. Pitt. 2003. Does micro-credit empower women : evidence from Bangladesh. Working Paper Series 3641. World Bank, Washington D.C. Chang, C. O., S. M Chen and T. Somerville. 2003. Economic and Social Status in Household Decision-making: Evidence Relating to Extended Family Mobility. Urban Studies 40(4): 733-746. Charitonenko, S., and S.M. Rahman. 2002. Commercialization of Microfinance - Bangladesh. Asian Development Bank, Manila, Philippines. Chiappori, P.A. 1988. Rational Household Labor Supply. Econometrica 56 (1): 63-89. _____. 1992. Collective Labor Supply and Welfare. Journal of Political Economy 100(3): 437-467. Coudouel,A., J. S. Hentschel and Q.T. Wodon. 2002. Poverty Measurement and Analysis. PRSP Sourcebook, World Bank, Washington D.C. Available at: http://povlibrary.worldbank.org/files/5467_chap1.pdf. Cruces, G. and Q. T. Wodon. 2003. Transient and chronic poverty in turbulent times: Argentina 1995−2002. Economics Bulletin 9(3): 1−12. Datt, G., D. Jolliffe and M. Sharma. 1998. A Profile of Poverty in Egypt: 1997. Food Consumption and Nutrition Division Discussion Paper NO. 49. International Food Policy Research Institute. Washington, D.C. U.S.A. del Ninno, C. 2001. Coping Strategies in Bangladesh. Survey Documentation. (November 1998-December 1999). FMRSP data Documentation Report No.1. FMRSP Bangladesh. International Food Policy Research Institute.
146
del Ninno, C., P. A. Dorosh, L. C. Smith and D. K. Roy. 2001. The 1998 Floods in Bangladesh Disaster Impacts, Household Coping Strategies, and Response. Research Report 122. International Food Policy Research Institute Washington, D.C. del Ninno, C., P. A. Dorosh and L. C. Smith. 2003. Public Policy, Markets and Household Coping Strategies in Bangladesh: Avoiding a Food Security Crisis Following the 1998 Floods. World Development 31(7): 1221-1238. Dercon, S. 2001a. Assessing Vulnerability to Poverty. Paper prepared for the DFID. Available at: http://www.economics.ox.ac.uk/members/stefan.dercon/. 2001b. Income Risk, Coping Strategies and Safety Nets. Working Paper, Centre for the Study of African Economies. Oxford University. Dercon, S. and P. Krishnan. 2000. Vulnerability, Seasonality and Poverty in Ethiopia. Journal of Development Studies 36(6): 25-53. Doss, C.R. 1996. Intrahousehold Resource Allocation in an Uncertain Environment: Evidence from Ghana. Ph.D. diss., University of Minnesota, USA. Ermisch, J. 2003. An Economic Analysis of the Family. Princeton University Press. Estache, A., V. Foster and Q. Wodon. 2002. Accounting for Poverty in Infrastructure Reform Learning from Latin America's Experience. WBI Development Studies. The World Bank, Washington, D.C. Fafchamps, M. 1999. Rural poverty, Risk and Development. Center for the Study of African Economies. Oxford University. Report Submitted to the Food and Agriculture Organization. Fafchamps, M. and Agnes R. Quisumbing. 1999. Social Roles, Human Capital, and the Intrahousehold Division of Labor: Evidence from Pakistan. FCND Discussion Paper No. 73, International Food Policy Research Institute, Washington D.C. Foster, J., J. Greer, and E. Thorbecke. 1984. A class of decomposable poverty measures. Econometrica 52 (3): 761–765. Greene, W.H. 2000. Econometric Analysis. New Jersey: Prentice-Hall Inc. Haddad, L. and A. Ahmed. 2003. Chronic and Transitory Poverty: Evidence from Egypt, 1997–99. World Development 3(1): 71-85. Haddad, L., J. Hoddinott and H. Alderman. 1997. Introduction: The Scope of Intrahousehold Allocation Issues. In Intrahousehold Resource Allocation in Developing Countries: Models, Methods and Policy, eds. Haddad. L., J. Hoddinott and H. Alderman.
147
Baltimore: Johns Hopkins University Press, for the International Food Policy Research Institute. Hashemi S. M., S. R. Schuler and A. P. Riley. 1996. Rural credit programs and women's empowerment in Bangladesh. World Development 24(4): 589-791. Hashemi, S. and R. Rosenburg. 2006. Graduating the Poorest into Microfinance: Linking Safety Nets and Financial Services. Focus Note. Consultative Group to Assist the Poorest (CGAP). Available at: http://www.microfinancegateway.org/files/31760_file_FocusNote_34.pdf Hazell, P. and L. Haddad. 2001. Agricultural Research and Poverty Reduction. Food, Agriculture, and the Environment Discussion Paper 34. International Food Policy Research Institute Technical Advisory Group (TAC), Consultative Group on International Agricultural Research. Heitzmann, K., R. S. Canagarajah and P. B. Siegel. 2001. The Sources of Vulnerability: A Rationale and Guideline for the Assessment of Risks and Risk Responses. Paper prepared for the Social Protection Discussion Paper Series. Human Development Network. The World Bank. Hsiao, C. 2003. Analysis of Panel Data. Second Edition. New York: Cambridge University Press. Hulme, D. and A. Shepherd. 2003. Conceptualizing Chronic Poverty. World Development 31(3): 403-423. Human Development Report (HDR). 2005. International cooperation at a crossroads: Aid, trade and security in an unequal world. United Nations Development Programme, New York, USA. Available at: http://hdr.undp.org/reports/global/2005/pdf/HDR05_frontmatter.pdf International Labour Organization (ILO). 2003. Working Out of Poverty. Report of the Director-General. International Labour Conference. 91st Session. International Labour Office, Geneva. Available at: http://www.ilo.org/public/english/standards/relm/ilc/ilc91/pdf/rep-i-a.pdf Jalan, J. and M. Ravallion. 2000. Is Transient Poverty Different? Evidence for Rural China. Journal of Development Studies 36(6): 83-99. Jose, S. 2003. Gender Bias in Resource Allocation in India: Where Do Household Models and Empirical Evidence Intersect? Indian Journal of Gender Studies 10 (3): 405-29. Kabeer, N. 2001. Conflicts over Credit: Re-Evaluating the Empowerment Potential of Loans to Women in Rural Bangladesh. World Development 29(1):63-84.
148
Kantor, P. 2003. Women's Empowerment through Home-Based Work: Evidence from India. Development and Change. 34(3): 425-45. Kedir, A. M. and A. McKay. 2003. Chronic Poverty in Urban Ethiopia: Panel Data Evidence. Kochar, A. 1995. Explaining Household Vulnerability to Idiosyncratic Income Shocks. The American Economic Review 85(2): 159-164. Littlefield, E., J. Morduch, and S. Hashemi. 2003. Is Microfinance an Effective Strategy to Reach the Millennium Development Goals? Focus note # 24. CGAP. Available at: http://www.cgap.org/docs/FocusNote_24.pdf Lipton, M. and M. Ravallion. 1995. Poverty and Policy. In Jere Behrman and T.N. Srinivasan (Eds), Handbook of Development Economics, Volume 3, Amsterdam: North-Holland. Long, J. Scott. 1997. Regression Models for Categorical and Limited Dependent Variables. Thousand Oaks: Sage Publications. Lundberg, S. J. and R. Pollak. 1993. Separate Spheres Bargaining and the Marriage Market. Journal of Political Economy 101(6): 988-1010. Lundberg, S. J. and R. Pollak. 2004. Bargaining in Families. Discussion Papers in Economics and the University of Washington 0065, Department of Economics at the University of Washington. Lundberg, S.J., R. Pollak and T.J.Wales. 1997. Do Husbands and Wives Pool Their Resources? Evidence from the United Kingdom Child Benefit. Journal of Human Resources 32 (3) 463-480. Manser, M. and M. Brown. 1980. Marriage and Household Decision-making: A Bargaining Analysis. International Economic Review 21(1): 31-44. Martinelli, C. and S. W. Parker. 2003. Should Transfers to Poor Families Be Conditional on School Attendance? A Household Bargaining Perspective. International Economic Review 44(2): 523-44. McElroy, M.B. 1990. The empirical content of Nash-bargained household behavior. Journal of Human Resources 25(4): 559-583. McElroy, M.B. and M.J. Horney. 1981. Nash-bargained Household Decisions: Toward a Generalization of the Theory of Demand, International Economic Review 22(2): 333-349.
149
McCulloch, N. and B. Baulch. 1999. Distinguishing the Chronically from the Transitorily Poor: Evidence from Rural Pakistan. Working Paper No. 97, Institute of Development Studies, University of Sussex. McCulloch, N. and B. Baulch. 2000. Simulating the Impact of Policy Upon Chronic and Transitory Poverty in Rural Pakistan. Journal of Development Studies 36(6):100-130. McCulloch, N. and M. Calandrino. 2003. Vulnerability and Chronic Poverty in Rural Sichuan. World Development 31(3):611-628. McKay, A. and D. Lawson. 2002. Chronic Poverty in Developing and Transition Countries: Concepts and Evidence. CREDIT Research Paper (No. 02/27). Centre for Research in Economic Development and International Trade, University of Nottingham. Mendoza, L. 1997. Essays on Decision Making in Rural Households: A Study of Three Villages in the Cordillera Region of the Philippines. Ph.D. diss., Katholieke Universiteit Leuven. Morduch, J. 1999. The Microfinance Promise. Journal of Economic Literature 37(4):1569-1614. Muller, C. 1997. Transient and Chronic Poverty of Peasants: Evidence from Rwanda. Working Paper Series 97-8. Centre for the Study of African Economies, University of Oxford 11: 503-541. Muller, C. 2002. Censored Quantile Regressions of Chronic and Transient Seasonal Poverty in Rwanda. Journal of African Economies 11: 503-541. Newman, C. 2002. Gender, Time Use, and Change: The Impact of the Cut Flower Industry in Ecuador. World Bank Economic Review 16(3): 375-95. Oduro, A.D. 2002. Poverty Dynamics. Paper prepared for the Advanced Poverty Training Programme Organized by SISERA and WBI. Centre for Policy Analysis, Accra, Ghana. OECD. 2001. In the Face of Poverty: Meeting the Global Challenge through Partnership. DAC Guidelines on Poverty Reduction. Available at: http://www.oecd.org/dataoecd/18/19/1849018.pdf Pitt, M. 1997. Specification and estimation of the Demand for Goods within the Household. In Intrahousehold Resource Allocation in Developing Countries: Models, Methods and Policy, eds. Haddad. L., J. Hoddinott and H. Alderman. Baltimore: Johns Hopkins University Press for the International Food Policy Research Institute.
150
Quisumbing, A. and B. de la Briere. 2000. Women’s Assets and Intrahousehold Allocation in Rural Bangladesh. FCND Discussion Paper No. 86. International Food Policy Research Institute, Washington D.C. Quisumbing, A. 2003. Power and Resources within the Households: Overview. In Household Decisions, Gender and Development. A Synthesis of Recent Research, eds. A. R. Quisumbing. Published by International Food Policy Research Institute, Washington D.C. Ray, D. 1998. Development Economics. Princeton: Princeton University Press. Sadoulet, E and A. de Janvry. 1995. Quantitative Development Policy Analysis. Baltimore, London: The Johns Hopkins University Press. Sahn, D. E. and D. C. Stifel. 2002. Parental Preferences for Nutrition of Boys and Girls: Evidence from Africa. Journal of Development Studies 39(1): 21-45. Schultz, T. P. 1990. Testing the Neoclassical Model of Family Labor Supply and Fertility. Journal of Human Resources 25 (4): 599-634. Schultz, T. P. 2001. Women's Role in the Agricultural Household: Bargaining and Human Capital. In Handbook of Agricultural Economics Volume. A., eds. B. Gardner and G. Rausser, Amsterdam, London, New York: Elsevier Science. Singh, I., L. Squire and J. Strauss. 1986a. The Basic Model: Theory, Empirical Results, and Policy Conclusions. In Agricultural Household Models: Extensions, Applications and Policy, eds. Singh, I., L. Squire and J. Strauss, Baltimore: Johns Hopkins University Press. Singh, I., L. Squire and J. Strauss. 1986b. Methodological Issues. In Agricultural Household Models: Extensions, Applications and Policy, eds. Singh, I., L. Squire and J. Strauss, Baltimore: Johns Hopkins University Press. Strauss, J. and K. Beegle. 1996. Intrahousehold Allocation: A Review of Theories, Empirical Evidence, and Policy Issues. Working Paper 62. East Lansing: Department of Agricultural Economics and Department of Economics, Michigan State University. Suen, W., W. Chan and J. Zhang. 2003. Marital Transfer and Intra-household Allocation: A Nash-Bargaining Analysis. Journal of Economic Behavior and Organization 52(1): 133-46. Swaminathan, H. 2003. Bargaining and Intrahousehold Resource Allocation: An Analysis of the Impact of Credit and Land in Malawi. Ph.D. diss., Penn State University. University Park.
151
Swaminathan, H. and J. L. Findeis. 2003. Access to Credit and Women's Work Decisions: An Empirical Study in Rural Malawi. Population Research Institute Working Paper. Working Paper 03-04. Penn State University, University Park. Thomas, D. 1990. Intrahousehold Resource Allocation: An Inferential Approach. Journal of Human Resources 25(4): 635-664. Ulph, D. 1988. A General Noncooperative Nash Model of Household Consumption Behavior. Discussion Paper 88–205. Bristol, U.K.: University of Bristol, Department of Economics. United Nations. 1993. General Assembly Resolution A/RES/48/183 of 21 December 1993. Available at: http://ods-dds-zy.un.org/doc/UNDOC/GEN/N94/135/85/PDF/N9413585.pdf?OpenElement United Nations. 1999. General Assembly. Fifty-third session. Agenda Item 98. Available at: http://www.uncdf.org/english/microfinance/yom/UNGA-YoM_eng.pdf UNCDF. 2005. Building Inclusive Financial Sectors to Achieve the Millennium Development Goals. Concept Paper. International Year of Microcredit 2005. Available at: http://www.uncdf.org/english/microfinance/yom.php United Nations Development Programme (UNDP) Poverty Report. 2000. Overcoming Human Poverty. Available at: http://www.undp.org/povertyreport/ United Nations. 2005. Microfinance and Millennium Development Goals. Published by UN Department of Public Information in Co-operation with the Year of Microcredit Secretariat DPI/2357C-455499. Available at: http://www.yearofmicrocredit.org/docs/mdgdoc_MN.pdf Wooldridge, J. 2002. Econometric analysis of cross section and panel data. Massachusetts Institute of Technology Press, USA. Wooldridge, J.M. 2006. Introductory Econometrics: A Modern Approach. Third Edition. Thomson South-Western. Woolley, F. 1988. A Noncooperative Model of Family Decision Making. Discussion Paper No. TIDI/125. London: London School of Economics. World Bank. 2000. Attacking Poverty. World Development Report 2000/2001. Oxford University Press. World Bank. 2001a. Poverty Trends and Voices of the Poor. Fourth edition. Poverty Reduction and Economic Management. Human Development. Development Economics. The World Bank Group.
152
World Bank. 2001b. World Development Report 2002. Building Institutions for Markets. Published for the World Bank. Oxford University Press. World Bank. 2002a. Poverty and Vulnerability in South Asia. Washington D.C. The World Bank. World Bank. 2002b. Poverty in Bangladesh: Building on Progress. Poverty Reduction and Economic Management Sector Unit. South Asia Region, World Bank. Document of the World Bank and Asian Development Bank. World Bank. 2002c. Integrating Gender into the World Bank’s Work:A Strategy for Action. The World Bank, Washington D.C. World Bank. 2003a. South Asia Region (SAR) - Bangladesh. Country Brief. Available at:http://siteresources.worldbank.org/BANGLADESHEXTN/Resources/CountryBrief2003.pdf World Bank. 2003b. Bangladesh: A National Strategy for Economic Growth, Poverty Reduction and Social Development. Economic Relations Division, Ministry of Finance, Government of the People’s Republic of Bangladesh. Available at: http://poverty.worldbank.org/files/14223_Bangladesh_IPRSP.pdf. Accessed on: July 2, 2004. World Bank. 2004. World Development Indicators 2004. Available at: http://www.worldbank.org/data/wdi2002/ World Bank. 2005. Introduction to Poverty Analysis. Poverty Manual. World Bank Institute. The World Bank, Washington D.C. Yaqub, S. 2002. Chronic poverty: scrutinizing estimates, patterns, correlates, and explanations. Working Paper No. 21. Chronic Poverty Research Centre. Zaman, H. 2004. The scaling-up of microfinance in Bangladesh: determinants, impact, and lessons. World Bank Policy Research Working Paper 3398. World Bank. Washington D.C.
Anuja Jayaraman
308 Armsby, The Pennsylvania State University, University Park, PA 16802.
Ph: (814) 8638248 E-mail: axj902@psu.edu
EDUCATION
• Ph.D. Agricultural, Environmental and Regional Economics and Demography, The Pennsylvania State University, Expected August 2006
o Dissertation Title: Poverty Dynamics And Household Response: Disaster Shocks In Rural Bangladesh
• M.A. Economics, Delhi School of Economics, Delhi University, India, 1997-1999
• B.A. Economics, University of Delhi , India, 1994-1997
ADDITIONAL PROFESSIONAL TRAINING
• Population Policy Fellow, Population Reference Bureau (P.R.B), Washington, DC 2006-2007
AREAS OF SPECIALIZATION
• Development Economics • Agricultural and Environmental Economics • Demography/Population
EXPERIENCE
• Graduate research assistant, Pennsylvania State University 2000-2006