Master in E conomic Development and Growth
Determinants of Poverty in Rural Ethiopia: A Household
Level Analysis
Ahmed Mohammed Awel
ahmed_moham med.awel.166@st udent .lu.se
A bstract: This paper investigates the dynamics of poverty in rural E thiopia during
the period from 1994 to 2009. In order to explore factors that decisively affect the
possibility of falling into and exiting out of poverty, the paper uses six rounds of
data and employs alternative dynamic probit model which handles the problem of
serial correlation, unobserved individual heterogeneity, state dependence and the
initial conditions problem. The estimation result shows that the likelihood of falling
in to poverty in any round is a direct function of previous experience in poverty
suggesting strong evidence for the existence of true state dependence.
Socioeconomic variables like land size, oxen and other tropical livestock units have
tremendous role in reducing the probability of falling into poverty. Additionally,
while demographic characteristics and drought has significant effect in the northern
part, cash crop production plays a vital role for households in southern Ethiopia.
Finally, the paper draws important policy implications that can be helpful for policy
making and enlighten appropriate intervention areas.
Key words: poverty dynamics, state dependence, transitory shocks
E KH R92
Master thesis (15 credits E CTS)
June 2013
Supervisor: E rik Green
E xaminer: Christer Gunnarsson
Website www.ehl.lu.se
Table of Contents
1.
Introduction ....................................................................................... 5
1.1
Research Question........................................................................... 6
1.2
Outline of the Thesis........................................................................ 8
2.
Theory................................................................................................ 8
2.1
Review of Previous Studies ............................................................... 8
2.2
Theoretical Framework .................................................................. 12
2.2.1
Individualistic Theory of Poverty............................................... 14
2.2.2
Cultural Theory of Poverty ....................................................... 14
2.2.3
Geographical Theory of Poverty ................................................ 15
2.2.4
Structural Theory of Poverty .................................................... 17
3.
Data and Methodology ...................................................................... 19
3.1
Data Source .................................................................................. 19
3.2
Sub-dividing the Samples .............................................................. 21
3.3
Methodology .................................................................................. 22
3.3.1
Setting Poverty Line ................................................................. 22
3.3.2
The Model ................................................................................ 23
3.3.3
The Wooldridge CML Estimator ................................................ 25
4.
Results ............................................................................................. 26
4.1
Descriptive Statistics ..................................................................... 26
4.2
Regression Result .......................................................................... 28
4.3
Discussion..................................................................................... 34
5.
Conclusion and Policy Implication .................................................... 37
Reference .................................................................................................. 39
Appendix .................................................................................................. 44
2
List of Tables
Table 1. Percentage of households by poverty status: 1994-2009 ............... 26
Table 2. Descriptive statistics for rural households (Northern Ethiopia) ..... 27
Table 3. Descriptive statistics for rural households (Southern Ethiopia) ..... 28
Table 4. Regression result (North) .............................................................. 31
Table 4. Regression result (South) ............................................................. 33
Table A1. Characteristics of the sample sites ............................................ 44
Table A2. Food basket composition used for poverty lines (per month) ....... 44
Table A3. Nutrition (calorie) based equivalence scales ............................... 45
Table A3. Nutrition (calorie) based equivalence scales ............................... 45
3
Acknowledgement
First of all I pay my whole hearted gratitude to the Almighty ALLAH that
endowed me with an innate ability to recognize and acknowledge His
existence. Without His grace and wish, I won’t be hear where I am now and
obviously this work can’t be accomplished successfully. May ALLAH send
His praises upon the holly and beloved prophet, Mohammad bin Abdullah,
who brought the message of peace and happiness to all creatures.
After that I would like to extend my heartfelt gratitude to my thesis
supervisor Erik Green. By his constructive criticism, guidance and countless
fruitful discussions, he used to bring me back to the track, recharge and
stimulate my enthusiasm for pressing on further. Undoubtedly, his
innovative comments greatly improved the content of this thesis. I would
also like to extend my thanks
to my examiner Professor Christer
Gunnarsson for his valuable comments.
Next, I would like to thank all individuals and institutions who helped me in
many ways. Especially, I would like to acknowledge the invaluable and
generous financial support from the European Commission throughout my
stay in UK and Sweden. Above all, I am greatly indebted to my parents and
my wife for their constant support, encouragement and prayers during my
study. I hereby am dedicating my thesis to my sons Abdurahman Ahmed
and Abduraheem Ahmed.
4
1. Introduction
In recent years, many countries in Africa have experienced extraordinary
rebound in economic growth. Have poor individuals benefited from this
growth? This is a policy debate and controversial issue in recent literature.
Over the past three decades, there has been substantial number of works on
the analysis of poverty worldwide. The research interest in poverty analysis
is further intensified with the decision of many countries, including Ethiopia,
to adopt the United Nations Millennium Declaration during the 2000
Summit and to exert as much effort as possible to achieve the Millennium
Development Goals (MDGs). The first goal in MDGs is ‘eradication of extreme
poverty and hunger’. This goal has three specific targets: (a) to halve the
proportion of people living on less than one dollar a day; (b) achieve decent
employment for men, women, and young people; (c) to halve the proportion
of people who suffer from hunger. However, recent evidences reveal that
despite considerable progress in reducing poverty in some regions over the
past decades-remarkably in East Asia- still nearly 1.4 billion people are
living on less than US$1.25 a day, and about 1 billion people are suffering
from hunger (IFAD, 2011).
Poverty continues to be the main challenge in developing countries,
especially in sub-Saharan Africa (SSA). Three fourths of the poor in the
developing world live in rural areas, and rural poverty remains high and
persistent-51 percent in SSA-while the absolute number of poor people
increased since 1993 (World Bank, 2008). In fact, the burden of poverty in
SSA is disproportionately borne by rural dwellers and women (UNECA,
2012).
Nowadays, across SSA rural infrastructure has almost deteriorated, farming
has languished, food systems have stagnated, and inequalities have
deepened (UNDP 2012). While the rapid growth and quick reduction in
poverty continue to be witnessed in Eastern Asia, growth in SSA could not
be
fast
enough
to
eradicate
extreme
poverty.
Despite
the
recent
improvements, majority of SSA countries have very low Human Development
5
Index (HDI). In the year 2011, the 15 lowest ranked countries with HDI are
from SSA; and of the bottom 30 countries, only Afghanistan and Haiti are
outside SSA (UNDP, 2012). This poor level of achievement is manifested in
all dimensions of the HDI: income, health, and education.
Specifically, poverty is widespread in Ethiopia as a large proportion of its
population lives below one dollar a day. Despite rapid economic growth in
the past decade, averaged 10.1 percent for the last nine years, poverty is still
prevalent in Ethiopia that makes the country among the poorest in the
world. According to UNDP (2012), Ethiopia is ranked 174th out of 187
countries in terms of HDI. Similar to in other developing countries, majority
of the poor in Ethiopia live in rural areas (Alemu et al., 2011) where 83
percent of the total population lives (World Bank, 2012).
Poverty, underdevelopment and backwardness in Ethiopia are not confined
to destiny. What matters is taking relevant reforms that can relax the
binding constraint. The poor in Ethiopia are not fated to be malnourished
and face misery as long as the government moves determinedly to introduce
appropriate reforms, policies and support mechanisms. Indeed, proper
policies might necessarily differ based on the inherent features of poverty in
the target population. However, if one finds the relative influence of different
correlates of poverty, government policies can easily focus on those
determinants and take appropriate measures to combat poverty.
1.1 Research Question
After decades of political instability, civil war and economic decline,
economic reform in Ethiopia began in the late 1980s. The initial phase of the
reform program, which took place following the downfall of the military
(Derge) regime in 1991, focussed mainly on liberalization of food markets
(Dercon, 2000; Dercon, 2006). As of 1994, with enormous supports from the
World Bank and IMF, Ethiopia implemented a structural adjustment
program and took several reforms related to investment and trade
liberalization,
exchange
rate
determination
and
removal
of
fertilizer
subsidies. This was followed by economic recovery after 1996 (Dercon, 2000;
6
Dercon, 2006) which led to a reduction of poverty in the country (IMF, 1999;
Demery, 1999; Dercon, 2000; Dercon, 2006).
In 2002, the government instigated a comprehensive poverty reduction
strategy (i.e. the Sustainable Development and Poverty Reduction Program)
with four building blocks: (a) Agricultural Development Led Industrialization
(ADLI), (b) justice system and civil service reform, (c) decentralization and
empowerment, and (d) capacity building in public and private sectors
(MoFED, 2002). Since the livelihood of the majority of the population in
Ethiopia is based on the agricultural sector, poverty reduction policies in the
country have targeted at strategies to increase agricultural productivity
through provision of credit and input supply services, access to better
extension packages, expansion of infrastructure facilities, mainly water
supply and rural roads, and expansion of health care services and primary
education (MEDaC, 1999; Dercon, 2000; FDRE 2000). The poverty reduction
strategies have been complemented with food transfers, food-for-work and
cash-for-work programs primarily to alleviate short-term food insecurity,
and also to finance public investments such as schools, clinics, rural roads
and irrigation facilities (MEDaC, 1999; FDRE 2000). Regardless of the
implementation of various reform programs and poverty reduction strategies
and despite the rapid economic growth in the past decade, poverty is still a
widespread phenomenon in Ethiopia.
The literature that analyse poverty dynamics in Ethiopia is at best scanty.
Majority of available studies predominantly focus on poverty profile which
describes the pattern of poverty. To date, to the best of my knowledge, there
have not been much rigorous studies on the determinants of household
poverty in Ethiopia. Even among those do exist, some focus specifically on
urban areas (e.g. Tadesse, 1999; Kedir, 2005; Bigsten & Shimeles, 2011;
Gebremedihin & Whelan, 2008; Alem, 2011), where only small proportion of
the Ethiopian poor lives in, and others are based on cross sectional data and
very limited sample size (e.g. Alemu et al., 2011; Oumer & de Neergaard,
2011; Regassa & Stoecker, 2012; Uraguchi, 2012). Static analysis of poverty
using cross sectional data gives the picture for a particular point in time,
7
and hence, shows a meagre analysis of the evolution of poverty. In order to
tackle poverty, analysing poverty dynamics using longitudinal data and
exploring factors that determine the possibility of falling into poverty is
indispensable. Examining and understanding factors that determine the
situation of rural poor in Ethiopia helps to draw clear direction for policy
making and enlightens appropriate intervention areas.
Therefore, the prime aim of this research is to answer the following specific
questions.
1. What are factors that decisively affect the likelihood of rural
households to fall into poverty in Ethiopia?
2. Do those factors that affect the likelihood of rural households to fall
into poverty vary across regions?
3. What lessons or policy implications can be drawn from the findings, if
any?
1.2 Outline of the Thesis
This paper is divided in to five parts. The first part discusses an introduction
to the topic and clarifies the research question that is going to be thoroughly
addressed in the paper. The second part deals with the theory. It, in the
beginning, discusses exiting empirical works on the topic, and later on
presents the theoretical framework. The third part describes the data and
explains the methodology used for the entire analysis. The fourth part
presents the results and discusses thoroughly by comparing the north with
the southern part of the country. Finally, the fifth part concludes the paper
and sheds light on important policy implications.
2. Theory
2.1 Review of Previous Studies
There is limited but highly growing literature on poverty dynamics. The
beginning of pragmatic works on poverty dynamics is attributed to Bane and
Ellwood (1986). They use Panel Study of Income Dynamics (PSID), a
8
longitudinal survey of a representative sample of the United States (US)
individuals and families, for the periods between 1970 and 1981 in their
study. They argue that poverty dynamics is properly understood as long as
it is defined in terms of poverty spells that allows estimating the degree of
falling in to and exiting out of poverty due to variations in income and
changes
in family structure.
They found majority of the poor are
characterized by longer spells of poverty. Besides, nearly two-fifth of the
spells of poverty begun due to a decline in the earnings of the household
head while three-fifth of the spells end as a result of a rise in the earnings of
the household head.
Bane and Ellwood (1986), further, examine female-headed households
excluding the male-headed ones and come to know that changes in
household structure are fairly important, though not important as earnings.
They find that a quarter of female-headed households having children exit
out of poverty when they shift their family structure in to a male-headed
household. Stevens (1994) updated the work of Bane and Ellwood (1986)
using PSID but by extending the period through 1987 and finds that during
the period under consideration, female-headed households are less likely to
move out of poverty than their male-headed counterparts.
Jalan and Ravallion (1998) use panel data for China and apply components
approach to decompose the poor in to transient and chronic poor. They also
employ the censored conditional quintile regression model to investigate the
process behind transient and chronic poverty. They find that physical assets
as important determinant of transient poverty, wealth holdings decreases
the amount of transient poor while demographic characteristics and
education level of the household are less likely to affect transient poverty.
However, chronic poverty is highly influenced by household demographic
characteristics, high variance of wealth holding and size of cultivated land.
In general, they find that the determinants of transient and chronic poverty
are different except for life cycle effects and physical asset holding. They
recommended that poverty reduction strategies require policy instruments
9
like seasonal credit schemes, public works, insurance options, and buffer
stocks for the poor that can smooth the consumption variability.
Glewwe, Gragnolati and Zaman (2002) use the decomposition method and
multinomial logit regression models to investigate factors driving the change
in poverty status of a household between years 1993 and 1998 in Vietnam.
Their results show that the common drivers of poverty dynamics are
demographic
characteristics,
education
of
household
head,
type
of
employment, ethnicity, access to infrastructure and location. Litchfield and
Justino (2004) use similar dataset to examine factors affecting the rural
poverty dynamics in Vietnam. Their result confirmed the findings of Glewwe,
Gragnolati and Zaman (2002) that the major drivers of poverty dynamics are
education of household head, type of employment, access to infrastructure
and location.
Woolard and Klasen (2005) employ multivariate analysis method on a panel
household data from a populous province in South Africa, Kwazulu-Nata,
and find a quite high degree of mobility, in contrast to developing economies.
Their analysis shows that employment changes and demographic changes
are the principal determinants of mobility. These factors relay on high
unemployment resulted from labour market volatility as well as on
demographic changes resulted from rapidly shifting household boundaries
(i.e. changes in fertility and mortality). The authors also explore four poverty
traps that obstruct the advancement of the poor. These obstructions are low
level of assets, poor initial education, large household size, and poor labour
market participation. In contrast, having more job opportunities, smaller
household size and better education offer them chance to be better-off.
However, when we come specifically to Ethiopia, despite the rampant
poverty in the country, the literature on poverty dynamics is at best scanty.
This may be because of the demanding nature of the longitudinal data to
analyze the dynamics. Most of existing studies of poverty in Ethiopia are
attributable to Dercon and Krishnan (1998; 2000), Dercon (2000; 2004;
10
2006), Bigsten, et al (2003; 2005) and Bigsten and Shimeles (2003; 2008;
20011).
Dercon and Krishnan (1998) examine rural poverty in Ethiopia using the
longitudinal data from the Ethiopian Rural Household Survey (ERHS) i.e.
rounds of 1989, 1994, and 1995. They use consumption per adult
equivalent as welfare indicator and observed significant decline of poverty
between 1989 and 1994 but remained almost unchanged between 1994 and
1995. They also found that households with better access to roads and
towns, and considerable human and physical capital have lower poverty
levels and have higher possibility to be better-off over time. Besides, access
to roads and towns and having substantial human capital also reduce the
variations in poverty across the seasons.
Dercon (2004) and Dercon, Hoddinott and Woldehanna (2005) explore that
shocks significantly affect rural households in Ethiopia. The most prominent
types of shocks that distress the welfare of households are drought, crop
pests, shocks on price of inputs and outputs, crime, death and serious
illness. In addition, Dercon (2006) analyzes rural poverty changes and
determinants of growth during the initial phases of the economic reform in
Ethiopia (1989–1995). His result indicates that generally, there was a
substantial
reduction
in
poverty
and
considerable
improvement
in
consumption during the period under consideration. Moreover, he noted
that on average the poor were better-off than the non-poor households,
although the benefits from the reforms are not evenly distributed among all
the poor. He also finds that shocks led to changes in the returns to human
capital, land, labor and location. This implies that, besides the short-run
poverty
impact,
shocks
in
Ethiopia
have
serious
negative
growth
implications.
Bigsten and Shiemeles (2003) use ERHS 1994-1997 and employ the spells
and component approach to analyze the dynamics of poverty in Ethiopia.
They noticed that transient poverty dominating rural households and found
a modest decline in poverty for the rural areas. They also found that factors
11
that affect the probability of moving into poverty are dependency ratio and
age of the household head. Besides, factors that significantly reduce the
likelihood of falling in to poverty are education of the household head, size of
cultivated land, type of crops planted, value of crop sales, and access to
local markets.
2.2 Theoretical Framework
Poverty continues to attract global attention particularly in programmes that
concerns development since it is a lifelong phenomenon that plagued
mankind in our efforts on the way to development. It is difficult to define
poverty mainly due to its multidimensionality. Poverty is usually taken as
the lack of necessities though what is a necessity to one individual may not
be for the other. Necessities are relative to what is possible usually based on
social characterization and past experience (Sen, 1999). Poverty is also a
social phenomenon which goes further than economic spheres and
encompasses inability of individuals to participate in social life and political
milieu. One way of defining poverty is by letting the poor to explain their
own poverty. It is allowing individuals or groups who are practically facing
poverty to define what represents their basic requirements in life. However,
the most commonly used definition is the one defined by the World Bank
(2000) as “the economic condition in which people lack sufficient income to
obtain certain minimal levels of health services, food, housing, clothing and
education generally recognized as necessary to ensure an adequate standard
of living”.
According to the World Bank (2000), poverty is pronounced deprivation in
well-being. It is possible to look well-being in three different dimensions: (a)
as the command over commodities in general, (b) as an ability to obtain
specific type of consumption good, or (c) as a “capability” to function in
society (World Bank, 2005). In the first approach of looking poverty (wellbeing), the prime interest is whether households have sufficient resources to
satisfy their needs. Accordingly, poverty is measured in monetary terms by
comparing household’s income or consumption against specified threshold
12
level below which they are considered as poor. The second approach goes
beyond monetary measures to look detail nutrition, health and education of
individuals under consideration. The third approach to well-being is
articulated by Sen (1987), who argues that lack of key capabilities,
inadequate income, inadequate education, poor health, low self confidence,
insecurity, freedom of speech, and sense of powerlessness leads people
towards poverty.
Of the three approaches, the money-metric approach (i.e. using income or
consumption as welfare indicator) is a dominant approach mainly due to the
fact that one can analyse the individual characteristics
and
other
socioeconomic conditions that are correlated with poverty (Bigsten et al.,
2005). Particularly, consumption is usually viewed as the better indicator of
poverty measurement than income (Ravallion 1994; Lipton & Ravallion 1995;
Deaton 1997). There are two crucial reasons for preferring consumption to
income (Coudouel et al., 2002). First, consumption is considered to be a
better indicator of outcome than income. Actual consumption indicates the
ability of a household to meet its basic needs, while income is only one of
the basic elements (there are others like availability and access) that
influence levels of consumption. Therefore, it implies that a standard of
living of individuals is better reflected by consumption data than purely by
income. Second, consumption data can be better measured than income
mainly due to seasonality of income among rural households, and
underreporting of their income than their actual consumption. For these
reasons, consumption expenditure is the main indicator of welfare to
categorise households as poor and non-poor.
Up on discussing the pertinent way of measuring poverty, the next step is to
look poverty theories which provide comprehensive explanation of why
people are poor. Recent literature acknowledges various theories that
explain poverty. This review presents a brief description of individualistic,
cultural, geographical and structural theories of poverty.
13
2.2.1 Individualistic Theory of Poverty
The individualistic theory explains poverty as a result of the characteristics
that are intrinsic in the individual and that consists the personal ability like
intelligence and the character of the person. This theory states that the poor
people become poor due to their lack of ability to compete with others for
resources. This theory perceives the poor as if they are born with it (i.e. born
being disabled like crippled, blind, or deformed) and for that reason they
cannot do anything to change the situation in which they are living
(Rainwater, 1970). Furthermore, the individualistic theory perceives that
poverty is resulted due to acquired personality traits like character and
actions of individuals. The idea here is that some individuals who are born
being lazy do not voluntarily participate in tasks that have meaningful effect
in their life. However, this theory fails to realize the ability of those born
disabled to do something that can drive them out of poverty. Asen (2002)
argue that any individual can succeed by hard work, and that persistence
and motivation are all that are required to be successful.
In favour of the idea of individualistic theory, the neoclassical economics
advocates that the poor are poor because of their decisions. The assertion is
due to the fact that individuals seek to maximize well being by making their
own choices and investments. When some individuals choose low-payoff and
short term returns, economic theory holds those individuals largely
responsible for their choices, for instance to forego the adoption of
production process that will boost output or to forego education that will
lead to better paying jobs in the future.
2.2.2 Cultural Theory of Poverty
The second theory is cultural theory of poverty which is primarily originated
from the culture of poverty. It is the theory developed by an anthropologist
Oscar Lewis in 1959 based on his experience of Mexico. This theory
advocates that poverty is caused by the spread over generations of a set of
skills, values, and beliefs that are socially created but individually held
(Lewis, 1959). The culture of poverty is a syndrome that develops in some
14
specific situations. It occurs in an economic setting with low wages, high
rate of unemployment, and people with low skills. In the absence of
deliberate support from the government, the low-income population have a
tendency to build up the culture of poverty against the prevailing ideology of
expanding the middle class. The poor understand that they have a negligible
position within an individualistic and highly stratified capitalistic society,
which does not give them any hope for upward mobility (Lewis, 1959). As a
result, the poor create survival strategy by developing their own subculture
and institutions, and finally come to embody a common pattern of behaviour,
norms and values. The subculture developed by the poor is characterized by
pervasive
feelings
of
dependency,
helplessness,
marginality,
and
powerlessness (Lewis, 1959).
Nevertheless, the cultural theory of poverty and the way in which it is
understood and applied to society was not far from flaws and criticisms. The
main critics comes due to the fact that the culture of poverty takes for
granted that culture itself is unchanging and relatively fixed, i.e. once a
population
falls
within
the
culture
of
poverty,
poverty
alleviation
interventions will not change the behaviours and cultural attitudes
embodied in that population. Thus state support and public welfare
assistance to the poor cannot eliminate poverty for the reason that poverty
is integrated in the culture of the poor. Due to this reasoning, the cultural
theory of poverty shifts the blame for poverty from economic and social
conditions to the poor people themselves (Bourgois, 2001). Though the
theory acknowledges basic factors that led to the initial state of poverty
(such as lack of sufficient social services, substandard housing and
education, persistent racial discrimination, and lack of job opportunities), it
primarily focuses on the cause of current poverty as the attitudes and
behaviours of the poor.
2.2.3 Geographical Theory of Poverty
The third theory is geographical theory of poverty which corresponds to
spatial characterization of poverty. This theory states that poverty is severe
15
in certain areas than in the other due to the fact that individuals, cultures,
and institutions in some areas are deficient in the objective resources
essential to generate income and well being. Recent explanations include
proximity to natural resources, disinvestment, density, and other similar
factors (Morrill and Wohlenberg, 1971). The theoretical perspective on
geographical theory of poverty comes
from the economic theory of
agglomeration. The economic theory of agglomeration is used to characterize
the emergence of industrial clusters, the concentration of firms in proximate
area so as to benefit from internal and external economies (Bradshaw, King,
and Wahlstrom, 1999). In the same way, the geographical theory of poverty
describes that the proximity of poverty and favourable conditions leading to
poverty generate more poverty. For example, the poor usually live in areas
where there is more crime and inadequate social services. These places have
commonly low housing prices and this attracts more poor individuals to the
area.
The other theoretical insight of geographical theory of poverty is from central
place theory that traces the flows of capital as well as knowledge. For
example, rural areas are most of the time the last stop of technologies, and
competitive pricing and low wages dominate production (Hansen, 1970). The
lack of social infrastructure limits economic activity and places left behind
experience the largest competition (Lyson and Falk, 1992). Therefore,
privileged areas stand to grow more than underprivileged areas even during
the time of general economic growth with some “trickle-down” but not lead
to equalizing effects as classical economists assert (Rural Sociological
Society, 1990; sited in Bradshaw, 2007). The geographical theory of poverty
connotes that responses need to be focussed to solving the key dynamics
that create deprivation and economic decline in disadvantaged areas while
other
areas
are growing (Bradshaw, 2007).
Instead
of focusing on
individuals, governments, businesses, cultural processes, or welfare systems,
the geographical theory guides community developers to emphasize at
depressed areas. The prime reason is that the evils of poverty are highly
16
reinforced by the geographical environment of the slum districts where the
poor are concentrated.
2.2.4 Structural Theory of Poverty
Finally, the structural theory is a progressive social theory. This theory does
not blame the victim for his/her own poverty as individualistic and cultural
theories do, but it look to the social, political, and economic system which
causes individuals to have inadequate resources with which to realize their
income and well being. The standards of living and social relations of
individuals in a society are shaped by educational facilities, labour market
opportunities, and economic growth. The inherent structures in the society
including social relations such as gender, race, power and class determines
the fate of individuals (Bradshaw, 2007). This implies that it is the
malfunction of the structures that causes poverty in the society.
Therefore, using structural theory in explaining poverty helps to target on
factors that perpetuates poverty. It can be made without changing the poor
themselves, rather by changing the condition of the poor by means of
adjusting the restrictive socioeconomic structures that aggravate poverty.
This
theory
advocates
that
elimination
of
structural
barriers
and
implementing a wide range of socioeconomic policies generates substantial
numbers of successes in reducing poverty. The range of socioeconomic
policies that can be adjusted to realize poverty reduction include raising
wages, providing jobs, assuring effective access to medical care, expanding
the safety net, and coordinating social insurance programs (Bradshaw,
2007).
The conclusion to be drawn from the discussion of the poverty theories, in
so far as this research is concerned, is that all the individualistic, cultural,
geographical and structural theories seek to identify the various reasons of
falling in to poverty. Nevertheless, all poverty theories are divergent and do
not add to a single consistent theory of explaining poverty. No one theory
has appeared that either invalidates or subsumes the others (Blank, 1997).
For example, some individuals/households can be poor due to their lack of
17
ability (for instance, due to old-age or gender bias) to compete with others
for resources. Others can be poor because they are born being lazy and do
not voluntarily participate in tasks that have meaningful effect in their life.
Such factors can be well explained by individualistic theory. But this
explanation is partial since it does not describe the whole neighbourhood.
Part of the community might be hardworking, but still stay in poverty due to
disadvantaged settlement in less fertile and drought prone areas. For such
neighbourhoods geographical theory is well suited than the rest. It might
also be explained with structural theory if it is the social, political, or
economic system which causes individuals to settle on such neighbourhoods
and have inadequate resources. Hence, the conceptual framework in this
research does not solely depend on one particular poverty theory. It is
sensible to combine the different theories since, as Duncan (1984) notes, a
framework with a complete explanation of why the poor become poor would
require several interrelated theories of poverty.
Based on the above discussed theories and empirical literature, it is possible
to summarize the various factors that affect the likelihood of households to
experience poverty. Poverty may arise due to household or individual level
characteristics. It may also arise due to factors that are external to the
household, i.e. due to community level, and/or regional level characteristics.
It is possible to disaggregate household level characteristics in to two broad
categories as demographic and socioeconomic characteristics. Indicators of
demographic characteristics that may be associated with poverty are
household size and structure, dependency ratio, the age of the household
head and the gender of household head.
The size and composition of the household is usually different for the poor
and non-poor, as the poor tend to live in larger households (Lanjouw &
Ravallion 1995; Deaton & Paxson 1998; Jalan & Ravallion, 1998). It is also
possible to argue that the rich may have many children than the poor as the
rich can afford the cost of raising a child. Finding an evidence for these
arguments can have policy implications, either to incorporate population
policy or implement demographically contingent interventions for fighting
18
poverty. Households headed by women and/or with high dependency ratio
tend to be poorer (World Bank, 2005; Jalan & Ravallion, 1998).
The most familiar socioeconomic characteristics that explain poverty are
household asset and household employment. The ownership of tangible
goods, livestock units and financial assets affects the income flow of a
household. The employment status, the type of work and the length of hours
an individual works also highly matters. Typically, in rural areas, the
cropping system of the household can affect the income obtained from
farming activities. Cash crop farmers may generate higher income and,
therefore, be less poor than food crop farmers irrespective of the amount of
inputs and the size of the cultivated land.
There are various community level characteristics that might be related with
poverty for certain neighbourhood. Infrastructure is the core determinant at
this level. It includes access to electricity, proximity to paved roads, access
to market, access to schools and health care service centres. In addition,
inadequate social service provision, social exclusion and discrimination are
associated with chronic poverty (Grant & Marcus, 2009). At the regional
level, poverty might be associated with several features. Grant and Marcus
(2009) identify remoteness (geography) as structural factor associated to
chronic poverty. Generally, poverty is higher in areas characterized by low
resource base, geographical isolation, rainfall deficit, and other harsh
climatic conditions (World Bank, 2005).
Despite the prime focus of this research is on household level characteristics,
some important community and regional level characteristics like drought
and access to market are also made part of the analysis in the research.
3. Data and Methodology
3.1
Data Source
Most researches on Poverty are typically constrained by lack of adequate
data on various indicators of households. Recently, in many developing
countries including Ethiopia, governments and development partners placed
19
a high concern and started to develop relatively reliable longitudinal data for
poverty analysis. This study uses a unique longitudinal household dataset
from the Ethiopia Rural Household Survey (ERHS)1. The ERHS started in
1989, when a survey was commenced with 450 households in 6 Peasant
Associations2 (or Kebeles) specifically in Central and Southern Ethiopia. The
survey was further expanded in 1994 to increase the sample size from 450
to 1477 households from a total of 15 Peasant Associations. The selection of
Peasant Associations took in to account the diversity of the farming systems
found in Ethiopia. Additionally, stratified sampling within each village was
made to include a representative sample of female-headed and landless
households. The survey addressed a wide range of characteristics which
include
household
characteristics,
food
consumption,
livestock
and
agriculture information, health, sewage and toilet facilities, electricity and
water, production and marketing, wages, education, and health services
(Dercon & Hoddinott 2011, for further details about ERHS).
In order to create a longitudinal data, additional consecutive rounds was
conducted in the late 1994, 1995, 1997, 1999, 2004 and 2009. In all rounds
(except in 1989) of the survey, the questions asked were identical, or very
similar, and the data were processed in comparable ways. Therefore, due to
the fact that the 1989 round had relatively smaller sample size with a
narrow set of questions, this study will consider the data of later rounds
(1994-2009). However, there are still caveats in this dataset to be considered
as nationally representative since it does not include urban dwellers and
pastoral households which constitute 17 and 12 percent of the total
population respectively.
1
These data have been made available by the Economics Department of Addis Ababa
University (AAU), the Centre for the Study of African Economies (CSAE) at the University of
Oxford and the International Food Policy Research Institute (IFPRI). Funding for data
collection was provided by the Economic and Social Research Council (ESRC), the Swedish
International Development Agency (SIDA) and the United States Agency for International
Development (USAID); the preparation of the public release version of these data was
supported, in part, by the World Bank. AAU, CSAE, IFPRI, ESRC, SIDA, USAID and the
World Bank are not responsible for any errors in these data or for their use or
interpretation.
2 Peasant Association or Kebele is the lowest administrative unit in Ethiopia consisting of a
number of villages.
20
3.2 Sub-dividing the Samples
Analysing poverty dynamics at the aggregate level shows the overall picture
at country level and this aggregation hides the stark contrast of living
conditions in different regions and may overlook some important features
which are specific to some places/regions. Since Ethiopia is endowed with
diverse agro-ecological zones which vary in terms of topography, climate,
rainfall patterns, soil types, farming system and living arrangements, a onefits-all approach does not help much (Alemu, Nuppenau and Bolland 2009),
and hence it will be worthwhile to scrutinize poverty dynamics at
disaggregated level.
The data I am using in this research (i.e. ERHS) is collected from four main
administrative regions in Ethiopia, namely Tigray, Amhara, Oromia and
SNNP3. According to CSA (2010), these four regions cover about 60 percent
of the total land area and constitute more than 86 percent of the total
population. Tigray and Amhara are located in the northern part while
Oromia and SNNP are in Southern part of the country. The northern part is
characterized
by
subsistence
agriculture,
rugged
topography,
land
degradation, rainfall variability and drought, and higher population pressure
(Bewket, 2009). However, the southern part, which includes Oromia and
SNNP, is endowed with diverse natural resources, has fertile soil, rich for its
abundant surface and ground water, is a region of relatively high rainfall by
Ethiopia standards, and the source of major cash crops such as coffee and
khat4 (USAID, 2005; FDRE, 2011). As a result, the samples are sub-divided
in to two, north and south constituting 620 and 840 sample households
respectively and the underlying features of poverty dynamics is analyzed for
both regions separately.
Southern Nations, Nationalities, and People's Region
Khat is a flowering plant native to East Africa, especially Ethiopia and Somalia, which is
chewed as a stimulant, for excitement and euphoria
3
4
21
3.3 Methodology
3.3.1 Setting Poverty Line
The three basic steps in poverty analysis are choosing a welfare indicator,
establishing a poverty line and aggregating poverty data (Ravallion, 1994;
Deaton, 1997). As discussed in section 2.2.1, consumption expenditure is
the main indicator of welfare to categorise households as poor and non-poor
in this study. It is figured in monthly percapita terms and deflated by using
the Food Price Index (FPI) with 1994 base year.
In order to set poverty lines, the research employs the cost of basic needs
approach (CBN) and uses a bundle of food items from the 1994 data that
would provide 2300 Kcal per person per day, which is the minimum calories
required for an adult to lead an average physical life under normal
conditions based on estimation of the Ethiopian Nutrition and Health
Research Institute (EHNRI). Therefore, a household is considered to be living
in poverty provided that the percapita daily household food energy intake
goes below this threshold (2300 kcal).
Though many combinations of food items could yield the required 2300 kcal,
care has to be taken while selecting the bundle of food items to consider the
actual consumption pattern of the poor. At this stage, one cannot know who
precisely are poor and non-poor to define the reference basket as the poverty
line has not yet been set. Ravallion and Bidani (1994) take the poorest
fifteen percent of the population in Indonesia to construct the reference
bundle for their study. In Ethiopia, Dercon and Krishnan (1998) and Bigsten
et al. (2003) focus on the poorest fifty percent of their sample households to
set the reference food basket for their studies. Similarly, this research takes
the poorest half of the sample households in constructing the reference
bundle. Following the method used in Ravallion and Bidani (1994), the
research adds the non-food basket and finally the resulting bundle is
converted to monetary values so as to set the poverty line.
22
3.3.2 The Model
Investigating the dynamics of poverty is an important way to capture the
interaction between past poverty history of a household and its persistence
over time. Poverty persistence may arise either due to transitory shocks, or
because of unobserved characteristics, or due to state dependence of poverty.
State dependence is a situation when poverty propagates itself due to the
fact that households who have a long history of being poor are less likely to
leave the state of poverty (Duncan et al., 1993; Biewen, 2006). Therefore, to
acquire the precise measure of true state dependence, models of poverty
dynamics should account for effects of transitory shocks and unobserved
heterogeneity. Many empirical studies used the parametric (i.e. proportional
hazard models and logistic regression) and non-parametric models (i.e.
Kaplan–Meir survival function) to examine the dynamics of poverty. Even
though these parametric and non-parametric models give consistent
estimates of hazard rates, they are not paramount to properly model the
true state dependence (Cappelari & Jenkins, 2002; Devicienti, 2003, cited in
Bigsten & Shimeles, 2008). In order to explore factors that decisively affect
the possibility of falling into and exiting out of poverty, this paper uses a
dynamic
probit
model
which
handles
the
problem
of
unobserved
heterogeneity, state dependence and serial correlation. Finally, following
Stewart (2006), the latent variable specification of the model takes the
following form:
∗
=
where
+
∗
′
+ α + ℰ ,
= 1,2, …, ; = 2,3, …,
is the latent dependent variable,
variable which is defined as:
and where
characteristics),
=
1
0
ℎ
( 1)
is observed binary outcome
∗
≥0
stands for a vector of explanatory variables (observable
is a vector of parameters to be estimated,
corresponds to
the state dependence that shows a condition in which facing poverty in one
period leads to a higher possibility of continuing to be poor, also taken as a
measure of a poverty trap (Chay et al., 1998, cited in Bigsten & Shimeles,
23
2008),
represents
household-specific
time-invariant
unobserved
determinants of poverty (these might be factors like ability, intelligence,
general attitude or motivation of household members), and ℰ is the error
term with ℰ
subscript
∼
( 0,
ℰ)
. The subscript
indexes households and the
indexes rounds of observations. Despite the serial-independence
of ℰ , the composite error term, i.e.
=
+ ℰ , will be correlated over time
due to the household-specific time-invariant terms, i.e.
equicorrelation between the
(
=
ℰ
, = 2,3, …, ; ≠
ℰ
Since
(
in any two (different) periods:
) =
,
is a binary variable, for convenience,
= 1) . Given
at time is then given by:
Pr (
) = Ф
,
,
( 2)
ℰ
is normalized to one
, and ℰ is normally distributed, the transition probability
for household
|
. This implies
The presence of both
+
′
and
+ α (2
− 1)
( 3)
in equation (3), which in many cases is
correlated, will create the “initial conditions problem”. It occurs because the
start of the process (poverty) does not coincide with the start of the first
observation (round one in 1994 in this case). Households found to be poor
or non-poor in the first observation may be poor or non-poor due to prior
history of poverty or as a result of observed and/or unobserved features
affecting their poverty status. Therefore, using the standard panel probit
model to estimate equation (3) will result in inconsistent estimates. So as to
take care of this problem, recent empirical works suggest using other
alternative estimators, i.e. Heckman (1981) two-step estimator, Orme (1997,
2001) two-step estimator, and Wooldridge (2005) conditional maximum
likelihood
(CML)
estimator.
The
simulation
experiments
by
Arulampalam and Stewart (2009) suggest that these three estimators
provide similar results and none of the three estimators dominates the other
two in all cases. However, among these three estimators, the Wooldridge
CML estimator is straight forward to use in standard econometric software
24
like Stata. Therefore, the Wooldridge CML estimator is used in this paper
and the way how this estimator works is elaborated as follows.
3.3.3 The Wooldridge CML Estimator
The Wooldridge CML estimator is a method that takes care of the initial
conditions problem of the ordinary dynamic non-linear panel data models. It
basically works through the distribution of
,
exogenous variables and the initial period value
dependent variable is written in sequence as
,
conditional on
, …,
. The joint density of the
(
,
,
, …,
|
, , )
.
Wooldridge specifies an alternative approximation for the density of the
time-invariant unobserved individual specific term
initial value of the dependent variable
individual specific term
conditional on the
. He also integrates unobserved
out from the equation and suggests the following
specification:
│
≈
,
Where
=
+
+
+
′
+
′
,
( 4)
+
( 5)
Equation (5) avoids the correlation between the time-invariant unobserved
individual specific term and the initial observation ( and
a new unobservable term
the dependent variable
) and results in
which is uncorrelated with the initial value of
.
Finally, substituting equation (5) into equation (3) gives
Pr (
= 1|
,
) = Ф
+
′
+
+
′
+
Accordingly, the likelihood function for household
=
∗(
Ф
+
′
+
+
′
+
(2
− 1)
− 1)
(2
is specified as:
∗(
)
) in equation (7) is the normal probability density function of
unobservable term which is introduced in equation 5).
25
( 6)
( 7)
(the new
4. Results
4.1 Descriptive Statistics
The poverty status of panel households is presented in table 1 below. The
proportion of households that have never been poor accounts only 12
percent in the north while it constitutes about 15 percent in the south. At
the same time, it can be seen that the proportion of households who have
always been poor (throughout the period under consideration) constitutes 8
percent in the north and about 6 percent in the south. The result further
shows significant evidence that a great deal of households do not experience
poverty continuously, they rather fall in to it for some period and exit out of
it during some other periods, which makes the analysis of poverty dynamics
of paramount importance. Unlike urban households who usually get their
income from labour market where salary income and nominal wage
increases modestly over time (Duncan, 1984), households in rural areas
have fragile income since the lion share of their earning comes from
agricultural produces. Agricultural production is usually affected by rainfall
variability, pests and diseases, drought, flood and other factors that lead to
harvest failure. Therefore, households in rural areas do often move from one
income level to another over time.
Table 1. Percentage of households by poverty status: 1994-2009
Region
Poverty status
Never
Once
Twice
Thrice
Four times
Five times
Always
poor
poor
poor
poor
poor
poor
poor
North
12.32
18.12
20.25
17.11
13.13
11.24
7.83
South
14.82
19.95
19.46
16.22
12.53
10.66
6.36
Source: Author's computation
Table 2 and 3 present the demographic and socioeconomic characteristics of
households by poverty status. Some variables show distinct differences
among households, especially between households who have always been
poor and those who have never been poor.
26
For example in northern Ethiopia, the mean household size for households
who have never been poor is only 5.2, while it rises to 7.8 for households
who have been always poor. Other demographic variables like age of
household head and mean age of the household show distinct differences
between households. Similarly, socioeconomic variables unveil significant
differences among households across poverty status.
For instance, there is
1.5 hectare differential on the average land size between households who
have never been poor and who have been always poor. Ownership of oxen
and other tropical livestock units (TLU) show distinct differences between
households.
Table 2. Descriptive statistics for rural households (Northern Ethiopia): 1994-2009
Variable
Never
poor
Once
poor
Twice
poor
Thrice
poor
Four times
poor
Five times
poor
Always
poor
Household size
5.2
6.2
6.4
6.5
6.9
7.3
7.8
Age of household head
42
44
45
48
48
49
49
26.83
22.61
21.52
18.22
17.45
15.96
15.28
Female headed (%)
11
19
18
16
14
12
10
Land size (hectare)
2
1.8
1.6
1.2
1
0.7
0.5
375
322
315
281
242
185
166
Off-farm employment (%)
36
41
35
24
26
22
18
Cash crop production (%)
10
8
6
6
4
2
2
TLU
4.5
2.8
2.4
1.8
1.6
1
0.6
Number of oxen ow ned
2.1
1.4
1.2
0.8
0.7
0.6
0.4
M ean age
Asset value (Eth. birr)
Source: Author's computation
In the case of southern Ethiopia, households under different poverty status
do not have big differences in household size and age of household head
unlike the households in northern Ethiopia, but the striking difference in
the south is on cash crop production. 76 percent of households under the
‘never poor’ category produce cash crops while this figure is only 6 percent
for households under the ‘always poor’ category. Similarly, land size, TLU
and ownership of oxen show distinct differences across categories.
27
Table 3. Descriptive statistics for rural households (Southern Ethiopia): 1994–2009
Variable
Never
poor
Once
poor
Twice
poor
Thrice
poor
Four times
poor
Five times
poor
Always
poor
Household size
6.3
6.6
6.8
6.7
7.1
6.9
7.2
Age of household head
44
43
45
44
47
48
48
27.22
24.51
22.84
21.21
19.54
16.21
16.16
Female headed (%)
17
16
19
15
13
11
9
Land size (hectare)
2.3
2.1
1.8
1.6
1.2
0.8
0.6
Asset value (Eth. birr)
411
374
342
305
261
196
170
Off-farm employment (%)
32
36
33
26
22
24
18
Cash crop production (%)
76
49
31
18
10
8
6
TLU
4.2
3.4
2.3
1.6
1.2
0.9
0.4
2
1.6
1.2
0.7
0.6
0.4
0.2
M ean age
Number of oxen ow ned
Source: Author's computation
The result shows some evidence on the relationship between variables and
poverty status in rural Ethiopia. One can also note from this result that
factors that explain the poor are not one and the same across the country.
There are some significantly different characteristics among households in
the north and in the south.
4.2 Regression Result
The econometric model specified in section (3.3.2) is estimated to analyse
the nature of poverty dynamics in rural Ethiopia. The key variables included
to model the probability of falling into poverty are household size, age of the
household head, mean age of the household and its square, gender of the
household head, land size, total value of household asset, participation in
off-farm employment, cash crop production, tropical livestock units, number
of oxen owned, drought and access to market as potential determinants of
poverty.
I start the estimation with a simple static probit model that takes the binary
outcome dependent variable (being in poverty or not) as a function of a
number of regressors [column 1]. I then estimate a dynamic model with
28
random effects probit estimator [column 2]. Finally, I used the Wooldridge
(2005) conditional maximum likelihood estimator that controls for state
dependence, unobserved household heterogeneity and serial correlation
[column 3]. In fact, both models in column 1 and 2 simplify the
determination of initial states and at the same time assume that the
unobserved household-specific characteristics are independent of the other
observed correlates. Consequently, the coefficients estimated in these
models are inconsistent for reasons stated in Section (3.3.2). I still report the
results so as to compare with the model in column 3 that deals with those
problems and show the magnitude of the bias. The results for the north and
the south are reported separately.
Column 1 in table 4 below presents the simple static probit model estimates
for households in northern Ethiopia. Having larger household size, being
headed by female and drought raises the probability of falling into poverty.
On the other hand, having less dependents, land size, participation on offfarm employment, ownership of oxen and other livestock and access to
market reduces the probability of falling in to poverty. Age of the household
head and elderly members have very small effects though both are
statistically significant.
Columns 2 present the results from the random effects dynamic probit
model where the state dependence (lagged dependent variable) is included as
part of the explanatory variables discussed above. The estimated true state
dependence (lagged dependent variable) is statistically and economically
significant. As compared with the results from the static probit model in
column 1, the results of the dynamic random effects model in column 2
show the inclusion of the lagged dependent variable has a significant effect
on other covariates as well. For instance the estimated coefficients for offfarm employment and tropical livestock units have declined by almost 50%.
On the other hand, the estimated coefficients for number of oxen and
drought are more than doubled.
29
Finally, column 3 reports the results from the Wooldridge CML estimator.
This is the result which is relatively compelling since the model controls for
state
dependence,
unobserved
household
heterogeneity
and
serial
correlation. It also shows a remarkable improvement in the fit of the model,
as indicated by the log likelihood. One of the important features of the
results is that the coefficient of the true state dependence (lagged dependent
variable) rose significantly once I controlled for the persistence of the error
term, also sometimes referred to as transitory shocks. The implication is
that the magnitude of the state dependence would have been understated
because of the effects of transitory shocks as well as measurement errors.
Positive and statistically significant coefficient of true state dependence
implies
that
even
after
controlling
for
observed
household
specific
characteristics and unobserved time-invariant terms, past experience was
associated with a higher risk of future poverty. This means that households
who have been poor in the previous year have higher risk of staying in
poverty than other households who were not poor in the preceding year. The
marginal effects5 computed for the Wooldridge CML estimator, for example,
show that being poor in the preceding round increases the probability of
falling in to poverty in the subsequent round by about 36 percentage points.
Among the demographic characteristics of households in northern Ethiopia,
household size and presence of elderly members raises the probability of
falling into poverty. However, the magnitude of the effect of household size is
declined as compared to other estimators (column 1 and 2). The effect of the
mean age within the household is rather higher and significant in this case
(column 3). It is also evidenced that households headed by female have
higher chance of falling into poverty. The computed marginal effect shows
that being headed by females increases the probability of falling in to
poverty by about 11 percentage points.
5
One cannot interpret the coefficients of a probit regression in any standard way. It is
necessary to interpret the marginal effects of the regressors, i.e. how much the probability
of the dependent variable changes when the value of a regressor changes, holding all other
regressors constant at their mean or median. The marginal effect for the Wooldridge CML
estimator is presented in the appendix.
30
Table 4. Regression result (North)
Simple Static Probit
estimator
[1]
RE Dynamic Probit
estimator
[2]
W ooldridge’s
CM L estimator
[3]
Coeff.
p-value
Coeff.
p-value
Coeff.
p-value
Lagged povert y
-
-
0.385* *
0.000
0.462* *
0.000
Household size
0.125* *
0.000
0.162* *
0.000
0.048*
0.032
Age of household head
0.001*
0.030
0.012*
0.042
0.003
0.150
M ean age
-0.015
0.162
-0.023
0.080
-0.148* *
0.000
M ean age sq.
0.003* *
0.000
0.012* *
0.000
0.069* *
0.000
Female headed
0.028* *
0.001
0.015*
0.060
0.132*
0.020
Land size
-0.210*
0.012
-0.184* *
0.000
-0.231* *
0.000
-0.044* *
0.000
-0.012*
0.018
-0.052
0.083
Off-farm employment
-0.141
0.152
-0.071
0.098
-0.063*
0.040
Cash crop product ion
-0.008
0.125
-0.011*
0.033
-0.002*
0.012
TLU
-0.215*
0.032
-0.105*
0.021
-0.198* *
0.000
Oxen
-0.112* *
0.000
-0.251*
0.012
-0.253* *
0.000
Drought
0.121* *
0.000
0.325* *
0.000
0.195* *
0.000
Access t o market
-0.152*
0.011
-0.009* *
0.000
-0.004*
0.045
Const ant
-1.520*
0.014
-0.665* *
0.000
-0.651
0.115
0.025
0.054
-0.392*
0.032
Asset
-
0.000
AR 1
Number of observat ion
3720
3720
3720
Log Likelihood
-1462
-1435
-1421
Source: Author's computation
** significance at 1%
* significance at 5%
The other interesting result that appears from socioeconomic variables is the
role of land size, oxen and other tropical livestock units in reducing the
probability of falling in to poverty. Holding all other regressors constant at
their mean, an increase in land holding by one hectare reduces the chance
of falling into poverty by 25 percent. Off-farm employment, cash crop
production and ownership of durable assets have less effect though they are
statistically significant. Drought is the other factor that significantly affects
households in northern Ethiopia. Agricultural households are often mainly
vulnerable, since weather shocks like climate change and rainfall variability
31
can destroy their farm and wipe out a large proportion of their annual
income. Finally, the coefficient of the serially correlated auto regressive error
term is less than unity (-0.392) and statistically significant implying that
even after controlling for first order state dependence and unobserved
heterogeneity, there is a negative transitory shock that affect poverty
persistency which stay longer than one year but its effect deteriorate over
time.
Similar regression techniques are applied for the households in southern
Ethiopia. The results are reported in Table 5 below. As was the case with
sample households in northern Ethiopia, the results for the south show that
the coefficient of the true state dependence (lagged dependent variable)
increased significantly once I controlled for the persistence of the error term,
also sometimes referred to as transitory shocks. Nevertheless, the results
show that households in southern Ethiopia display a smaller degree of true
state dependence than households in the north. This indicates that a
household in the north that experienced poverty in the preceding year faces
higher risk (about twofold) of staying in poverty than a household in the
south.
In the case of the Wooldridge CML estimator that controls for state
dependence, unobserved household heterogeneity and serial correlation, one
of the striking features of the results for the south is that demographic
variables like having larger household size, age of the household head and
being headed by female are less important and statistically insignificant.
However, the effect of the mean age within the household is higher and
significant implying that the higher the mean age of the household, the
smaller the number of dependents and the lower will be the chance of falling
into poverty. On the other hand, land size, participation on off-farm
employment, ownership of oxen and other livestock units and access to
market reduces the probability of falling in to poverty.
32
Table 5. Regression result (South)
Simple Static Probit
estimator
[1]
RE Dynamic Probit
estimator
[2]
W ooldridge’s
CM L estimator
[3]
Coeff.
p-value
Coeff.
p-value
Coeff.
p-value
Lagged povert y
-
-
0.196* *
0.000
0.213* *
0.000
Household size
0.001* *
0.002
0.023*
0.011
0.005
0.091
Age of household head
0.005*
0.030
0.002*
0.042
0.001
0.150
M ean age
-0.005
0.322
-0.013
0.130
-0.101*
0.042
M ean age sq.
0.011
0.240
0.174
0.412
0.041*
0.049
Female headed
0.008*
0.021
0.054
0.100
0.006
0.118
Land size
-0.112*
0.045
-0.184* *
0.000
-0.211* *
0.000
-0.052* *
0.000
-0.074*
0.022
-0.058
0.083
Off-farm employment
-0.184
0.141
-0.062
0.114
-0.054*
0.010
Cash crop product ion
-0.158* *
0.000
-0.251* *
0.000
-0.345* *
0.000
-0.285*
0.042
-0.239*
0.011
-0.192* *
0.000
-0.188* *
0.001
-0.147*
0.022
-0.185*
0.020
0.008
0.120
0.015
0.099
0.052
0.125
Access t o market
-0.184*
0.041
-0.059* *
0.000
-0.096*
0.015
Const ant
-1.852
0.521
-0.710* *
0.000
-0.395*
0.022
0.031
0.065
-0.311*
0.025
Asset
TLU
Oxen
Drought
-
0.000
AR 1
Number of observat ion
5040
5040
5040
Log Likelihood
-2481
-2462
-2458
Source: Author's computation
** significance at 1%
* significance at 5%
The other striking features of the result are the remarkable role played by
cash crop production and the negligible influence of drought in southern
Ethiopia. Cash crop production, though statistically significant, plays very
little role in reducing poverty in the north. However, cash crop production,
especially coffee and khat, plays substantial role in the south. The computed
marginal effect shows that, holding all other regressors constant at their
mean, being cash crop producer decreases the probability of falling in to
poverty by about 35 percentage points. Finally, the coefficient of the serially
correlated auto regressive error term is less than unity (-0.311) implying
33
that there is considerable effect of negative transitory shocks in poverty
persistency. As compared to the north, the effect of transitory shocks in
poverty persistency is less strong in southern Ethiopia.
4.3 Discussion
The finding shows that the likelihood of falling in to poverty in any round is
a direct function of previous experience in poverty, in both northern and
southern regions of Ethiopia, suggesting that individuals who experience
poverty are more likely to experience poverty in future periods. This means
that households with the experience of poverty in the previous year have
higher risk of staying in poverty than other households who were not poor in
the preceding year. There are various mechanisms that might explain such a
relationship between past experience of poverty and present state of being
poor. One explanation is that low earnings from farm and non-farm
activities by a rural household may possibly be associated with adverse
incentives which make it worthless for the household to be engaged in any
income generating activities. The other mechanism through which past
poverty history may increase the risk of experiencing poverty is through
depreciation of human capital, loss of motivation or demoralization. These
phenomena may lead to engaging in less productive agricultural activities
and a series of low quality jobs which in turn increases the risk of staying in
or returning back to poverty. Besides, experiencing poverty and depressed
socioeconomic conditions weaken the welfare of households and leads to
health problems. Rural households who typically live from harvest to
harvest do not have much room for health and other unfavourable shocks.
The cost of medication, if they opt for it, takes part of their income. Most
importantly, there might be a significant loss in the household income if the
workforce is particularly the victim of the health problem.
The finding also provides strong evidence on the role of demographic
characteristics of households. Among the demographic characteristics of
households in northern Ethiopia, household size, mean age and being
headed by females raises the probability of falling into poverty. This means
34
that people living in larger and younger families are typically poorer. The
higher the mean age of the household, the smaller the number of
dependents and the lower will be the chance of falling into poverty, and vice
versa. The result also shows that households head by females in northern
Ethiopia have higher chance of facing poverty. The structures of the female
headed households usually differ in predictable ways from the male headed
household. For example, female headed households have fewer people in
their households due to the absence of the spouse caused by widowhood or
divorce. Female headed households are typically disadvantaged regarding
the access to productive resources. They are also discriminated against by
cultural norms and suffering from high dependency burdens and economic
immobility. However, in the southern part, most of the demographic
characteristics like household size, age of household head and being headed
by females
are rather
less
important and
statistically insignificant.
Nonetheless, the effect of the mean age within the household is higher and
significant implying that the higher the mean age of the household the lower
will be the chance of falling into poverty. Therefore, what matters most is not
the size; it is rather the number of children and elderly dependents.
Socioeconomic variables like land size, oxen and other tropical livestock
units have tremendous role in reducing the probability of falling in to
poverty. These are factors that commonly affect households throughout the
country. Since the livelihood of rural households in Ethiopia is mainly
dependent on agriculture, land is one of the most important inputs in
explaining the welfare of the people. Other things remained the same, the
higher the size of cultivated land the higher will be the output. Households
with bigger plots of land have an option to cultivate varieties of crops which
in turn help them diversify the risk in periods of crop failure. Besides,
livestock ownership plays an important role in reducing poverty. They
provide important services like ploughing and hauling. Especially, oxen are
used for ploughing land and its ownership creates significant differences
among households in the study area. Moreover, livestock serves as a source
35
of food (e.g., meat, milk and eggs), and a means of saving and generating
additional income especially in periods of shocks and harvest failure.
The other feature of the result is on the role played by cash crop production
and the influence of drought in the country. The major cash crops which are
produced by small holder farmers in southern Ethiopia (especially coffee and
Khat) are not common in the northern part mainly due to the unsuitable
rainfall pattern and the nature of agro ecological zones. Although cash crop
production has negligible effect in reducing poverty in the north, it plays a
remarkable role in the south. Cash crop production allows farmers to earn
more money in order to fulfil their needs and thus enhance their capacity to
achieve
food
security.
Additionally,
producing
cash
crops
enables
households to acquire resources for other purposes than cash crop
production. For example, making money by producing cash crops opens up
access to inputs for use on other food crops. With constrained access to
farm credit, intensifying food crop production may depend on participation
of households in cash crop schemes. Participation in commercialized crop
scheme allows the use of improved seed, pesticides, fertilizer, herbicides and
machine services for both cash crops as well as food crop production.
Drought is a factor that severely affects households in northern Ethiopia.
Agricultural households are often vulnerable, since weather shocks like
climate change and rainfall variability can destroy their farm and wipe out a
large proportion of their annual income. A reduction in agricultural products
usually results in increased prices for food and high unemployment.
Drought aggravates the death of livestock which are in some cases a means
of production, i.e. oxen used for ploughing, and a source of additional
income especially in periods of shocks and harvest failure. Therefore, apart
from reducing agricultural outputs, drought in one period will have
prolonged impact on household wellbeing during the subsequent years.
As far as the correlates of poverty are concerned, the results discussed
above shows important distinctions between the north and south. This
peculiarity across regions may arise due to a number of reasons. It could be
36
due to geographical differences that result in differences with agro ecological
zones, rainfall distribution and soil fertility that lead to different farming
systems. It could also be due to differences in ethnicity and culture with
different set of skills, values and beliefs that are socially created within each
region. Knowing precisely why factors that affect the probability of falling
into poverty vary across regions needs further investigation and has
considerable scope for further research. What is very important at this stage
is to recognize that a “one-fits-all” approach in designing poverty alleviation
strategies and overall policy setting does not help much.
5. Conclusion and Policy Implication
This paper investigates the dynamics of poverty in rural Ethiopia during the
period from 1994 to 2009. In order to explore factors that decisively affect
the possibility of falling into and exiting out of poverty, the paper uses six
rounds of data and employs alternative dynamic probit model which handles
the problems of serial correlation, unobserved individual heterogeneity, state
dependence and the initial conditions problem.
The estimation result shows that the likelihood of falling in to poverty in any
round is a direct function of previous experience in poverty, in both
northern and southern regions of Ethiopia, suggesting strong evidence for
the existence of true state dependence. This means that households with the
experience of poverty in the previous year have higher risk of staying in
poverty than other households who were not poor in the preceding year.
Nevertheless, households in the south display a smaller degree of true state
dependence than households in the north. This indicates that a household
in the north that experienced poverty in the preceding year faces higher risk
of staying in poverty than a household in the south.
The result also provides strong evidence on the role of demographic
characteristics like household size, higher dependency ratio and being
headed by females in northern Ethiopia. Socioeconomic factors that have
tremendous role in reducing the probability of falling in to poverty are land
size, ownership of oxen and other tropical livestock units. The other striking
37
feature of the result is on the role played by cash crop production and the
influence of drought in the country. Although cash crop production has
negligible effect in reducing poverty in the north, it plays a remarkable role
in the south. Drought is severe in the north and rural households are
mainly vulnerable, since weather shocks like climate change and rainfall
variability can destroy their farm and wipe out a large proportion of their
annual
income.
Additionally,
the
result
confirms
the
presence
of
considerable effect of negative transitory shocks in poverty persistency in
both regions.
Even though identifying the various factors that affect the probability of
falling into poverty does not in itself assist in its alleviation, it gives a
framework upon which poverty alleviation strategies may be implemented to
address poverty from different perspectives. For this reason, based on the
findings of the paper, important policy implications can be drawn to
highlight
a
direction
for
policy
making
and
enlighten
appropriate
intervention areas. First, the existence of true state dependence in both
regions has a key message that past history of poverty determines its future
path. This implies that protecting households from falling into poverty is an
important prevention strategy in dealing with both short-term and long-term
poverty in the country. Thus, it is essential to pay attention for effective
poverty reduction strategies like providing income-generating schemes,
insurance schemes, safety net programs, and other interventions targeted at
the poor. Second, the fact that cash crop production plays a positive role in
poverty reduction in southern Ethiopia appears to be useful for policy
makers to design scaling-up strategies and other interventions like providing
inputs,
extension
services,
and
creating
and
facilitating
market
opportunities. Finally, the findings that drought is an important factor that
affect the likelihood of falling into poverty in northern Ethiopia implies that
the region requires special attention from policy makers. It needs policy
responses targeted at agricultural adaptation, such as adoption of drought
resistant varieties and enhancing small-scale irrigation projects that can
avoid reliance on rain-fed agriculture.
38
Reference
ALEM, Y. 2011. The dynamics and persistence of urban poverty: Panel data
evidence from Ethiopia. Tech. rep., University of Gothenburg, Sweden.
ALEMU, B. A., NUPPENAU, E. & BOLLAND, H. 2009. Technical Efficiency
Across Agro-Ecological Zones in Ethiopia: The Impact of Poverty and
Asset Endowments. Agricultural Journal, 4, 202-207.
ALEMU, D., BEWKET, W., ZELEKE, G., ASSEFA, Y. & TRUTMANN, P. 2011.
Extent and Determinants of Household Poverty in Rural Ethiopia: A
Study of Six Villages. Eastern Africa Social Science Research Review,
27, 21-49.
ARULAMPALAM, W. & STEWART, M. B. 2009. Simplified Implementation of
the Heckman Estimator of the Dynamic Probit Model and a
Comparison with Alternative Estimators*. Oxford bulletin of economics
and statistics, 71, 659-681.
ASEN, R. 2002. Visions of poverty: Welfare policy and political imagination,
MSU Press.
BANE, M. J. & ELLWOOD, D. T. 1986. Slipping into and out of poverty: The
dynamics of spells. Journal of Human Resources 21, 1-23.
BANK, W. 2007. World development report 2008: Agriculture for development,
World Bank.
BANK, W. 2009. World Development Report 2009, World Bank Washington^
eDC DC.
BEWKET, W. Rainfall variability and crop production in Ethiopia: Case
study in the Amhara region. Proceedings of the 16th International
Conference of Ethiopian Studies, 2009. 823-836.
BIEWEN, M. 2006. Who are the chronic poor? An econometric analysis of
chronic poverty in Germany. Research on Economic Inequality, 13, 3162.
BIGSTEN, A., KEBEDE, B., SHIMELES, A. & TADDESSE, M. 2003. Growth
and poverty reduction in Ethiopia: Evidence from household panel
surveys. World Development, 31, 87-106.
BIGSTEN, A., KEBEDE, B., SHIMELES, A. & TADDESSE, M. 2005. Theory
and Methods of Poverty Analysis. In: BIGSTEN, A., KEBEDE, B. &
SHIMELES, A. (eds.) Poverty, Income Distribution and Labour Markets
in Ethiopia. Sweden: Nordiska Afrikainstitutet.
BIGSTEN, A. & SHIMELES, A. 2003. Dynamics of poverty in Ethiopia.
WIDER Conference on Inequality, Poverty and Human Well-being.
United Nations University (UNU).
BIGSTEN, A. & SHIMELES, A. 2008. Poverty transition and persistence in
Ethiopia: 1994–2004. World Development, 36, 1559-1584.
39
BIGSTEN, A. & SHIMELES, A. 2011. The persistence of urban poverty in
Ethiopia: A tale of two measurements. Applied Economics Letters, 18,
835-839.
BLANK, R. M. 1997. It takes a nation: A new agenda for fighting poverty ,
ERIC.
BOURGOIS, P. 2001. Culture of poverty. International encyclopedia of the
social and behavioral sciences, 17, 11,904-11,907.
BRADSHAW, T. K. 2007. Theories of poverty and anti-poverty programs in
community development. Community Development, 38, 7-25.
BRADSHAW, T. K., KING, J. R. & WAHLSTROM, S. 1999. Catching on to
Clusters. Planning, 65, 18-21.
COUDOUEL, A., HENTSCHEL, J. S. & WODON, Q. T. 2002. Poverty
measurement and analysis. A Sourcebook for poverty reduction
strategies.
DEATON, A. 1997. The analysis of household surveys: a microeconometric
approach to development policy , Johns Hopkins University Press.
DEATON, A. & PAXSON, C. 1998. Economies of scale, household size, and
the demand for food. Journal of political economy, 106, 897-930.
DEMERY, L. 1999. Poverty dynamics in Africa: an update. ronéo, unité PRSD,
Banque mondiale, Washington, DC.
DERCON, S. 2000. Growth and Poverty in Ethiopia in the 1990s: an
Economic Perspective. University of Oxford: Centre for the Study of
African Economies.
DERCON, S. 2004. Growth and shocks: evidence from rural Ethiopia.
Journal of Development Economics, 74, 309-329.
DERCON, S. 2006. Economic reform, growth and the poor: Evidence from
rural Ethiopia. Journal of Development Economics, 81, 1-24.
DERCON, S. & HODDINOTT, J. 2011. The Ethiopian rural household
surveys: Introduction. International Food Policy Research Institute,
Washington, DC Photocopy.
DERCON, S. & KRISHNAN, P. 1998. Changes in poverty in rural Ethiopia
1989-1995: measurement, robustness tests and decomposition, Centre
for the Study of African Economies, Institute of Economics and
Statistics, University of Oxford.
DERCON, S. & KRISHNAN, P. 2000. Vulnerability, seasonality and poverty
in Ethiopia. The Journal of Development Studies, 36, 25-53.
DUNCAN, G. J. & CENTER, U. O. M. S. R. 1984. Years of poverty, years of
plenty: The changing economic fortunes of American workers and
families , Survey Research Center, Institute for Social Research,
University of Michigan.
40
DUNCAN, G. J., GUSTAFSSON, B., HAUSER, R., SCHMAUSS, G.,
MESSINGER, H., MUFFELS, R., NOLAN, B. & RAY, J.-C. 1993.
Poverty dynamics in eight countries. Journal of Population Economics,
6, 215-234.
FDRE 2000. Ethiopia-Interim Poverty Reduction Strategy Paper 2000/012002/03. Addis Ababa: Federal Democratic Republic of Ethiopia.
FDRE 2008. Summary and Statistical Report of the 2007 Population and
Housing Census–Population Size by Age and Sex. Addis Ababa:
Federal Democratic Republic of Ethiopia Population Census
Commission
FDRE 2011. Oromia National Regional State Program of Plan on Adaptation
to Climate Change, Technical and Financial Support. Addis Ababa:
Environmental Protection Authority.
GEBREMEDHIN, T. A. & WHELAN, S. 2008. Prices and Poverty in Urban
Ethiopia1. Journal of African economies, 17, 1-33.
GLEWWE, P., GRAGNOLATI, M. & ZAMAN, H. 2002. Who Gained from
Vietnam's Boom in the 1990s?*. Economic Development and Cultural
Change, 50, 773-792.
GRANT, U. & MARCUS, R. 2009. Chronic poverty and PRSPs: A desk study.
Chronic Poverty Research Centre Working Paper.
HANSEN, N. 1970. Poverty and the Urban Crisis .
HAUSMANN, R., TYSON, L. D. & ZAHIDI, S. The global gender gap report
2008. 2008. World Economic Forum.
IFAD 2011. Rural Poverty Report 2011. IFAD (International Fund for
Agricultural Development) Rome, Italy.
IMF 1999. Ethiopia: Recent Economic Developments. IMF Stuff Country
Report 00/ 37 (Washington DC March 2000).
JALAN, J. & RAVALLION, M. 1998. Determinants of transient and chronic
poverty: evidence from rural China, World Bank Publications.
KEDIR, A. M. 2005. Understanding urban chronic poverty: crossing the
qualitative and quantitative divide. Environment and Urbanization, 17,
43-54.
LANJOUW, P. & RAVALLION, M. 1995. Poverty and household size. The
Economic Journal, 1415-1434.
LIPTON, M. & RAVALLION, M. 1995. Poverty and policy. Handbook of
development economics, 3, 2551-2657.
LITCHFIELD, J. & JUSTINO, P. 2004. Welfare in Vietnam during the 1990s:
poverty, inequality and poverty dynamics. Journal of the Asia Pacific
Economy, 9, 145-169.
LYSON, T. A. & FALK, W. W. 1992. Forgotten Places: Uneven Development in
Rural America. Rural America Series , ERIC.
41
MEDAC 1999. Survey of Ethiopia Economy. Addis Ababa: Ministry of
Economic Development and Cooperation.
MOFED 2002. Ethiopia: Sustainable Development and Poverty Reduction
Program. Ministry of Finance and Economic Development.
MORRILL, R. L. & WOHLENBERG, E. H. 1971. The geography of poverty in
the United States , McGraw-Hill New York.
OSCAR, L. 1959. Five families: Mexican case studies in the culture of poverty ,
Basic Books.
OUMER, A. M. & DE NEERGAARD, A. 2011. Understanding livelihood
strategy-poverty links: empirical evidence from central highlands of
Ethiopia. Environment, Development and Sustainability, 13, 547-564.
RAINWATER, L. 1970. Neutralizing the disinherited: Some psychological
aspects of understanding the poor. Psychological factors in poverty , 928.
RAVALLION, M. 1994. Poverty comparisons , Taylor & Francis US.
REGASSA, N. & STOECKER, B. J. 2011. Household food insecurity and
hunger among households in Sidama district, southern Ethiopia.
Public Health Nutrition, 15, 1276.
SEN, A. 1976. Poverty: an ordinal approach to measurement. Econometrica:
Journal of the Econometric Society , 219-231.
SEN, A. 2001. Development as freedom, Oxford Paperbacks.
SEN, A. & HAWTHORN, G. 1988. The standard of living, Cambridge
University Press.
STEVENS, A. H. 1994. The dynamics of poverty spells: updating Bane and
Ellwood. The American Economic Review, 84, 34-37.
STEWART, M. B. 2006. Maximum simulated likelihood estimation of
random-effects dynamic probit models with autocorrelated errors.
Stata Journal, 6, 256.
TADDESSE, M. 1999. Determinants and Dynamics of Urban Poverty in
Ethiopia. Ethiopian Journal of Economics, 8, 61-82.
UNDP 2012. Africa Human Development Report 2012: Towards a Food
Secure Future. New York, USA: Regional Bureau for Africa (RBA).
UNECA 2012. Assessing Progress in Africa toward the Millennium
Development Goals , United Nations Economic Commission for Africa.
URAGUCHI, Z. B. 2012. Rural Income Transfer Programs and Rural
Household Food Security in Ethiopia. Journal of Asian and African
Studies, 47, 33-51.
USAID 2005. Ethiopia Southern Nations, Nationalities and Peoples Region
(SNNPR) Overview of Livelihood Profiles. Chemonics International Inc.
42
WOOLARD, I. & KLASEN, S. 2005. Determinants of income mobility and
household poverty dynamics in South Africa. Journal of Development
Studies, 41, 865-897.
WOOLDRIDGE, J. M. 2005. Simple solutions to the initial conditions
problem in dynamic, nonlinear panel data models with unobserved
heterogeneity. Journal of applied econometrics, 20, 39-54.
WORDBANK 2000. World development report 2000/2001: attacking poverty.
New York, Oxford University Press), table, 19, 311.
WORLDBANK 2005. Introduction to Poverty Analysis: The World Bank
Institute. Washington DC: The World Bank.
WORLDBANK 2008. World development report 2008: Agriculture for
development, World Bank.
WORLDBANK 2012. World Development Indicators 2012. Washington, DC:
The World Bank.
43
Appendix
Table A1. Characteristics of the sample sites
Survey site
Location
Region
M ain Crops
Rainfall (mm)
Haresaw
Geblen
Dinki
Debre Berhan
Yetmen
Shumsha
Sirbana Godeti
Adele Keke
Korodegaga
Turfe Kechemane
Imdibir
Aze Deboa
Addado
Gara Godo
Doma
Tigray
Tigray
Nort h Shoa
Nort h Shoa
Gojam
Nort h Wollo
Shoa
Hararghe
Arsi
Sout h Shoa
Shoa (Gurage)
Shoa (Kambat a)
Sidamo (Dilla)
Sidamo (Wolayt a)
Gamo Gofa
Tigray
Tigray
Amhara
Amhara
Amhara
Amhara
Oromia
Oromia
Oromia
Oromia
SNNP
SNNP
SNNP
SNNP
SNNP
Cereals
Cereals
M illet , t eff
Teff, barley, beans
Teff, w heat , beans
Cereals
Teff
M illet , maize, coffee, khat
Cereals
Wheat , barley, t eff, pot at oes
Enset , khat , coffee, maize
Enset , coffee, maize, t eff
Coffee, Enset
Barley, Enset
Enset , maize
558
504
1664
919
1241
654
672
748
874
812
2205
1509
1417
1245
1150
Source: Dercon and Krishnan (1998)
Table A2. Food basket composition used for poverty lines (per month)
Items
Teff
Barley
W heat
M aize
Sorghum
Horse beans
Cow peas
Chick peas
M ilk
Coffee
Sugar
Salt
Oil
Spices
Potatoes
Enset
Onions
Cabbage
Quantity
M easurement unit
1.70
4.85
3.15
4.48
2.67
1.29
0.23
0.69
0.55
0.10
0.10
0.70
0.15
0.25
1.51
0.19
0.20
0.38
kg
kg
kg
kg
kg
kg
kg
kg
lit res
kg
kg
kg
lit res
birr
kg
kg
kg
kg
Source: Dercon and Krishnan (1998)
44
Table A3. Nutrition (calorie) based equivalence scales
Age range (years)
M en
W omen
0-1
1-2
2-3
3-5
5-7
7-10
10-12
12-14
14-16
16-18
18-30
30-60
60 +
0.33
0.46
0.54
0.62
0.74
0.84
0.88
0.96
1.06
1.14
1.04
1.00
0.84
0.33
0.46
0.54
0.62
0.70
0.72
0.78
0.84
0.86
0.86
0.80
0.82
0.74
Source: Dercon and Krishnan (1998)
Table A1. Marginal effects of the Wooldridge CML estimator
Variable
North
South
Lagged poverty
Household size
Age of household head
M ean age
M ean age sq.
Female headed
Land size
Asset
Off-farm employment
Cash crop production
TLU
Oxen
Drought
dy/ dx
0.358
0.036
0.001
-0.198
0.041
0.112
-0.255
-0.001
-0.027
-0.003
-0.214
-0.282
0.192
p>│z│
0.045
0.002
0.395
0.033
0.007
0.030
0.000
0.075
0.040
0.012
0.042
0.011
0.034
dy/ dx
0.172
0.001
0.002
-0.121
0.008
0.004
-0.202
-0.002
-0.024
-0.351
-0.208
-0.198
0.001
p>│z│
0.031
0.145
0.251
0.004
0.015
0.325
0.044
0.225
0.020
0.000
0.018
0.035
0.425
Access to market
-0.002
0.041
-0.018
0.036
Source: Author's computation
45