Determinant of Poverty in Ethiopia

Teshome Kebede Deressa and M. K. Sharma1

Abstract

Poverty has turned out to be a great global social and economic problem.
In Ethiopia, it is multifaceted and deep rooted. This study attempts to
analyze the impact of socioeconomic and demographic characteristics of
households on poverty in Ethiopia, using the latest Household Income,
Consumption and Expenditure Survey (HICES) 2010-11. The study employs
a logistic regression model to identify determinants of wellbeing of the
household by considering per capita consumption as a dependent variable.
Different households are classified as either poor or non-poor on the basis
of absolute per capita consumption of Birr 3781. Results show owner of
agricultural land, head (self-employed or employed in formal sector) are
more likely to exit from poverty line. The results also reveal that female
headed households, large family size and high dependency ratio are
adversely affected.

Keywords: Poverty, Household, Per Capita Consumption, Determinants, Logistic

Regression.
JEL classification: C8

1
Addis Ababa University, Addis Ababa, Ethiopia
Email: - [email protected]
Acknowledgement: We are thankful to referees for their valuable comments and
suggestions.
Teshome Kebede and M. K. Sharma: Determinant of Poverty in Ethiopia

1. Introduction

Achieving sustainable economic growth with a focus on combating poverty

has become the key development goal for governments around the world, as
reflected in the Millennium Development Goals, Goal 1; “eradicate extreme
poverty and hunger”. In this objective, analysis of poverty aroused the
interest of researchers, public authorities and international organizations. The
specificities of developing economies, in particular the dualism between the
urban and rural areas incites to identify the determinants of poverty with a
view of designing policies and strategies to alleviate poverty that persists in
most of these countries. Poverty in Ethiopia has many manifestations. The
Human Development Index (HDI) for 2014 (based on estimates of 2013),
which takes life expectancy, adult literacy, primary schooling and per capita
income is 0.435 which is low in comparison of Rwanda, Uganda and Sub-
Saharan Africa and rank of Ethiopia is 173 out of the 177 countries.
According to most recent multidimensional poverty index estimation, 88.2%
of the populations of Ethiopia are multi-dimensionally poor while an
additional 6.7% are near multidimensional poverty. The MPI, which is the
share of the population that is multidimensional poor, adjusted by the
intensity of deprivations, is 0.537. Rwanda and Uganda have MPI is of 0.352
and 0.359 respectively.

By any standard, the majority of people in Ethiopia are among the poorest in
the world (Dercon and Krishnan, 1998; IMF, 1999; Rahmato and Kidnanu,
1999; World Bank, 2001). Poverty seems to persist in large sections rural
society as well as urban sections with little hope for a substantial
improvement of the living conditions of the rural poor as well as urban poor
in the near future. In order to combat such debilitating poverty considering
very scarce financial resources available to be allocated for the purpose, we
have to understand the determinants of poverty in rural and urban Ethiopia.
For this, the poor must be properly identified and an index taking the
intensity of poverty suffered by the poor into account needs to be
constructed.

114
 Ethiopian Journal of Economics Vol. XXIII No 1, April 2014

Analytical work that scrutinizes poverty profiles is best scanty. Even the
available ones are mostly descriptive focusing on explaining the extent of
poverty and mostly associated with studies that relate to food entitlement
failure (see Webb et al., 1992; Webb and Von Braun, 1994). Among those
studies, Beevan and Joireman (1997) adopt a sociological approach towards
the measurement of poverty on the meaning and use of different
measurements.

Using micro level panel data from villages in rural Ethiopia, Dercon (2001)
analyses the determinants of growth and changes in poverty during the initial
phases of the economic reform (1989-1995) making use of a standard
decomposition of income and poverty changes. His empirical results indicate
that overall, consumption grew and poverty fell substantially during the
period under consideration and that on average the poor have benefited more
from reforms than the non-poor households, even though the reforms did not
deliver similar benefits to all the poor. He argues that the main factors driving
changes are relative price changes, resulting in changes in the returns to land,
labour, human capital and location. Bogale et al. (2005) investigated the
determinants of rural poverty in Ethiopia on the basis of survey data of three
districts namely Alemaya, Hitosa and Merhabete and found that nearly 40% of
the sample households live below poverty line with an average gap of 0.047.

The Ethiopian Ministry of Finance and Economic Development (MoFED)

assessed the 1999/2000 Household Income and Consumption Expenditure
(HICE) and welfare Monitoring Survey results and concluded that the
incidence of poverty is higher in rural than in urban areas with poverty head
count ration of 45.4 and 36.9%, respectively (MoFED). However, as
compared to 1995/96 level, poverty incidence increased by 11.4% in urban
areas and declined by 4.42% in rural areas in 1999/2000.

Most of these studies aim to assess the extent of poverty and explain relative
changes which occur in the incidence of poverty due to policy changes. The
article aims to add discussion by examining the socio- economic and
demographic characteristic of households on poverty in rural and urban

115
Teshome Kebede and M. K. Sharma: Determinant of Poverty in Ethiopia

Ethiopia. We analyze the latest Household Income, Consumption and

Expenditure Survey (HICES) 2010-2011 and estimate determinants of
poverty using a maximum likelihood binary logistic regression model
considering whether a household is poor or non-poor as a response variable.
This allows us to derive further meaningful insight about various poverty –
generating factors that determine the persistence of poverty in Ethiopia and
the relevance those specific policies can play in alleviating poverty.

2. Data Source and Research Methodology

2.1. Data Source

The data used in this study has been taken from the 2010-11 Household
Income, Consumption and Expenditure Survey (HICES) for Ethiopia. The
survey covered both rural and urban areas of the country which was
conducted from 8 July 2010 through 7 July 2011.

For the purpose of representative sample selection, the country was divided
into three broad categories, i.e., rural, major urban centers and other urban
areas categories. Therefore, each category of a specific region was
considered to be a survey domain for which the major findings of the survey
are reported. However, Harari and Dire Dawa have rural and urban
categories, only; while Addis Ababa has only urban areas divided into10
sub-cities considered as survey domain or reporting levels.

2.2. Poverty Line

Large literature exists on approaches to assess poverty. However, the

question still remains as where to draw the poverty line. Ideally, the poverty
line should be based on a basket of goods and services including food and
nutrition, as well as clothing, housing and health care and education that can
be considered basic needs (Baffoe, 1992). Greer and Thorbecke (1986) apply
the cost of food consumption corresponding to the recommended daily

116
 Ethiopian Journal of Economics Vol. XXIII No 1, April 2014

allowance of calories and provide the profile and decomposition of food

poverty among Kenyan smallholders.

Economic theories suggest that per capita expenditure is the best indicator of
welfare, but this presupposes that households, as consumers, maximize a
continuous utility function defined over commodities (Glewwe, 1987).
Bevan and Joierman (1997) employed personal wealth ranking, community
wealth ranking and consumption poverty, and concluded that none of the
indicators applied identifies the poor on a convincing way.

The most popular method of poverty measurement have used the nutritional
norm and defined poverty in terms of minimum calorie requirements
(Dandekar and Rath, 1971; Osmani, 1982; Greer and Thorbecke, 1986;
Ahmed et al.1991; Ercelawn, 1991; Ravallion and Bidani, 1994).

In the absence of an invariably acceptable national poverty line for Ethiopia,

we decided to use the official poverty line constructed by MoFED in
2010/2011. That is, a household is deemed as living in poverty if the per
capita consumption is less than equal Birr 3781 otherwise the household will
be considered as non-poor.

2.3. Definitions of Variables Used in the Study

The dependent variable of our study is binary variable i.e it takes value 1 for
poor and 0 for non-poor. To know the impact of independent variables on
poverty we have considered the following socio- economic and demographic
variables i.e independent variables.

117
Teshome Kebede and M. K. Sharma: Determinant of Poverty in Ethiopia

Table 1: List of Variables and their Description

Independent (Explanatory) Variables
SEX Sex of the Household Head (0 = Male, 1 = Female)
AGE Age of the Household Head (in year)
FSZ Number of Household Members (Family Size)
FSZSQ Family Size Squared
AREA Place of Residence of Household (0 = Urban, 1 = Rural)
NWOR Number of Working Members/Productive Age (between 15 and 64 years inclusive)
AGRL Household Having Agricultural Land (0 = No, 1 = Yes)
DEPR Dependency Ratio
People of (Age 14 and Below  Age 65 and Above)

People Above Age of 15 and Below Age of 64
Head of the Household Has No Education (NSCH = 0)**
EDLEV Head of the Household Completed Elementary School (CMPE = 1)
Head of the Household Completed Secondary School (CMPS = 2)
Head of the Household Completed College/University & Above (CCUA = 3)
Head of the Household is Single (SINGLE= 0)**
MARST Head of the Household is Married (MARRIED = 1)
Head of the Household is Divorced/Widowed (DIVSEW = 2)
Head of the Household is Employed in Informal Sector (INFOE = 0)**
EMPST Head of the Household is Employed in Formal Sector (FORME = 1)
Head of the Household Head is Self-Employed (SELFE = 2)
Note: ** implies reference category

2.4. Method of Data Analysis

The data set was analysed using bivariate and multiple logistic regression
analyses. Bivariate analysis was done in order to identify which
characteristics independently related to socioeconomic status (poverty level)
were using Pearson’s chi-square tests of associations as given below.
r c


i 1 j 1
( O ij  E ij ) 2
 2
 ~  2 ( r  1 )( c  1 )
E ij
(1)
Where: Oij is the observed value in the ith row and jth column
Eij is the expected value of the ith row and jth column cell

118
 Ethiopian Journal of Economics Vol. XXIII No 1, April 2014

r = is number of row and c = is number of column.

Given the dependent variable of main interest that a household may be

classified as poor or non-poor, a binary Logistic regression model is useful
when the outcome (dependent) is binary, meaning zero or one, with one
being success. Suppose in multiple logistic regression case, a collection of p
explanatory variables be denoted by x'  (1, x1 , x 2, ..., x p ) . Furthermore, let

 i denotes the conditional probability that the i th household is below the

poverty line. Thus, the model for which the outcome variable is binary, can
be written as:

yi   i   i ; i  1, 2,..., n
(2)
Where:
exp( z i )
i  (3)
1  exp( z i )
β

with z i   0  1 x1i   2 x2i  ...   p x pi  X' . Here y is n×1 vector of

response having yi = 0 if the household is not-poor and yi = 1 if the
household is poor, X is an n× (p+1) design matrix of explanatory variables, β
ε

is a (p+1) ×1 vector of parameters, is also an n×1 vector of unobserved

random errors. The quantity  i is the probability for the ith covariate
satisfying the important requirement 0   i  1 . Then, the log-odds of
having y = 1 for given x is modeled as a linear function of the explanatory
variables as:
  
E (y / x)  ln i    0  1 x1i   2 x2i  ...   p x pi
1  i  (4)

119
Teshome Kebede and M. K. Sharma: Determinant of Poverty in Ethiopia

β
exp( X' )

β
The function  i  is known as logistic function. The most
1  exp( X' )
commonly used method of estimating the parameters of a logistic regression
model is the method of Maximum Likelihood (ML) instead of Ordinary
Least Square (OLS) method.

3. Results and Discussion

3.1 Model Selection for National Data

The model at National level with all variables is shown in the Table 2.

Table 2: Logistic Estimate of Poverty at National Level (Full Model)

Parameter Standard Wald Pr > Odds
Variable
Estimate Error Chi-Square Chi-Square Ratio
INTERCEPT -7.245 0.1527 2250.574 0.0001
DEPR 0.186 0.0368 25.706 0.0001 1.205
NWOR -0.095 0.0384 10.878 0.0010 0.910
FSZ 1.017 0.0384 702.694 0.0001 2.766
FSZSQ -0.050 0.0025 389.424 0.0001 0.951
AGE -0.007 0.0011 48.636 0.0001 0.992
SEX 0.471 0.0401 137.893 0.0001 1.602
AGRL -0.297 0.0503 34.925 0.0001 0.743
AREA 1.655 0.0359 2128.093 0.0001 5.234
DIVSEW 0.443 0.0789 31.521 0.0001 1.557
MARRIED 0.409 0.0448 83.274 0.0001 1.505
CMPE -0.168 0.0400 17.724 0.0001 0.845
CMPS -0.413 0.0650 40.342 0.0001 0.388
CCUA -0.669 0.0860 60.528 0.0001 0.515
SELFE -0.445 0.0495 81.040 0.0001 0.640
FORME -0.208 0.0820 6.350 0.0120** 0.812
** Significant at 5% level of significance
The calculated value e of the likelihood ratio test statistic is
G 2  32854 .561  23431 .280  9423 .281

120
 Ethiopian Journal of Economics Vol. XXIII No 1, April 2014

For   0.05, we have  02.05 (15)  24.996 , since

G 2  9423 .281  24.996 and the p-value for the test is

 
P  2 (15)  9423.281  0.000 which is significant at the α = 0.05 level.
Thus, we reject the null hypothesis and conclude that at least one and
perhaps all 15 coefficients are different from zero.

The Table 3 exhibits AIC, SC, R-Square and number of parameters to be

estimated for null and full model, respectively.

Table 3: Model Selection Criteria

Model AIC SC R-Square No. of Parameters
Null 32855.561 32859.005 - 1
Full 23463.280 23502.392 0.3991 16

The Table 3 reveals that the full model have the lowest value in both
criterion (AIC and SC) which is the indication of a better fit model by
adjusting for the number of explanatory variables and the number of
observations. The R-square value for the full model is 39.91%. The
estimated logit model for household level determinants of poverty at
National level is as given below:

^
logit( π i )   7.25  0.186depr  0.095nwor  1.02fsz  0.05fszsq  0.007age  0.471sex
 0.297agrl  1.66area  0.443divse w  0.41marrie d  0.168cmpe  0.413cmps  0.669ccua
 0.4376self e  0.21forme .Model  I 

Another way to analyze the effects of independent variables to know the

probability of being poor is the change of odds ratio as the independent
variables change. The odds ratio is defined as the probability of being poor
divided by the probability of not being poor. Table 2 (the last column) shows
the odds ratios for each independent variable at National level.

121
Teshome Kebede and M. K. Sharma: Determinant of Poverty in Ethiopia

As it is evident from Table 2 that the variables dependency ratio (DEPR),

family size (FSZ), household head being female (SEX), marital status
(DIVSEW and MARRIED) and living in the rural area (AREA) have odd
ratios greater than one, which means that these variables are positively
correlated with the probability of being poor. On the contrary, the variables
number of working members (NWOR), agricultural landholding (AGRL),
family size squared (FSZSQ), age of household head (AGE), having
completed elementary education (CMPE), having completed secondary
education (CMPS), having college education and above (CCUA), being self-
employed (SELFE) and being household head employed in formal sector
(FORME) all have odd ratios lower than one, which means that these
variables are negatively correlated with the probability of being poor.

3.4. Model Diagnostics

In model diagnostics we are concerned with goodness fit of the model.

3.4.1 Goodness-of-Fit of the Models (H-L Tests)

The goodness-of-fit measures how effectively the model describes the

response variable. Now, we can test the reduced model to see if it is a good
fit. The following table gives H-L test Statistics (2000) (summary for Model
(National)).

Table 4: Hosmer and Lemeshow Goodness-of-Fit Tests

National
^
H-L test statistic ( C ) 2.223
P-value 0.9734
No. of observations 27827

For   0.05, we have  0.05 (8)  15.507.

2

122
 Ethiopian Journal of Economics Vol. XXIII No 1, April 2014

^
Since C National  2.223 are all less than the tabulated value 15.507, we do
not reject H 0 , and conclude that the fitted models fit the data adequately
well. Thus, the goodness-of-fit test with p-values 0.9734 indicates that there
is insufficient evidence to claim that the models do not fit the data
adequately. If the p-value is less than our accepted α-level (5% in this case),
the test would reject the null hypothesis of an adequate fit. So our models fit
the data well.

3.4.2 Classification Table

In order to assess the predictive power of the models, a classification table of

correct and incorrect predictions was constructed, based on the predicted
probability of being poor for each data. A probability equal or greater than
0.5 was interpreted as a prediction of a household being poor, while a
probability lower than 0.5 was interpreted a prediction of a household being
non- poor. Table 5 shows the classification for the models. In this table, “D”
represents the number of poor households in the sample while “~D”
represents the number of non- poor cases in the sample. The symbol “+”
represents the number of households predicted as poor by the model while
“˗” represents the number of no- poor cases predicted by the model.

As it can be seen in the Table 5, the models sensitivity rate (percent of poor
cases correctly predicted by model) are 55.3%, 27.26% and 77.79%, while
the models specificity rate (percent of non-poor cases correctly predicted by
the model) are 90%, for National.

The false positive rate for households classified as poor by the model at
National level is 31.6 percent, which means that 31.6 percent of the number
of households predicted as poor by the model are in fact non-poor. The false
negative rate for households classified as non- poor by the model is 15.95
percent, which means that 15.95 percent of households predicted as non-poor
by the model are in fact poor.

123
Teshome Kebede and M. K. Sharma: Determinant of Poverty in Ethiopia

Table 5: Classification Table of Correct and Incorrect Predictions for

National [Urban] Rural
True
Classified
D ~D Total

+ 4272 [613] 4252 1974 [1001] 1859 6246 [1614] 6111

- 3443 [1636] 1214 18138 [14257] 2995 21581 [15893] 4209

Total 7715 [2249] 5466 20112 [15258] 4854 27827 [17507] 10320

National

Sensitivity 55.30%

Specificity 90.10%

Positive predictive value 68.40%

Negative predictive value 84.05%

False + rate for true ~D 9.82%

False – rate for true D 44.63%

False + rate for classified + 31.60%

False – rate for classified - 15.95%

Correctly classified 80.53%

The positive predictive value rate of the National model is 68.4 percent,
which means that 68.4 percent of the total number of predicted poor
households is in fact poor. Negative predictive rate is 84 percent, meaning
that 84 percent of the total number of non-poor cases predicted by the model
is in fact non-poor. As a whole, the National model correctly predicts 80.53

124
 Ethiopian Journal of Economics Vol. XXIII No 1, April 2014

percent of cases. Sensitivity and specificity rely on a single cut point to

classify a test result as positive.

3.5 Diagnostic Plots

One way of looking at the model adequacy is to graph studentized Pearson

and deviance residuals against predicted probabilities. The studentized
Pearson residuals and deviance residuals are plotted against the estimated
logistic probability respectively and in all case, the lower smooth
approximates a line having zero slope and intercept. Any significant
departure from this suggests that the model may be inadequate and potential
outliers may have dramatic impact on the fit of the model (Sarkar et al
2011).

It is to be mentioned here that the reduced model also passes the model
checking procedure.

4. Discussions of the Results

The data has been analyzed at National level. Ethiopia, like other
developing countries is subject to the threat of high population growth rate.
This high growth accompanied by the high unemployment rate and low
female labor force participation rate poses a serious threat to the households.
High dependency ratio (DEPR) and larger family size (FSZ) contribute
positively to the probability of becoming a poor household for national level.
The coefficients for both of these variables are positive and significant at 5%
level of significance. The coefficient of family size squared (FSZSQ) is
however negative and significant, controlling for the fact that very large
families can also have potential earners and can reduce the poverty through
larger participation in the work force. However, this situation is not highly
desirable due to the fact that the odds ratio are at a very low level of less than
1% in reducing the probability of being poor for national level. The odd
ratio of the variable dependency ratio (DEPR) shows a contribution of 20.5%

125
Teshome Kebede and M. K. Sharma: Determinant of Poverty in Ethiopia

in increasing the likelihood of being poor where as family size (FSZ)

contributes 76.6%.

Number of working household members/productive age (NWOR) has a

potential in reducing the probability of remaining in the poor household
category. Sex of household head i.e being female (SEX) positively affects
the likelihood of remaining poor. Several studies have discussed the
phenomenon of the feminization of poverty, which is said to exist if poverty
is more prevalent among female-headed households than among male-
headed households. This situation might be due to the presence of
discrimination against women in the labor market, or it might be due to the
fact that women tend to have lower education than men and they are paid
lower salaries. Using a probit model, Meron (2003) found that female-
headed households are poorer and more vulnerable to poverty than male-
headed households in Urban Ethiopia.

Looking at the results of logistic regression estimated national level; we

reach at the same conclusion as Meron (2003) since the sign for sex of the
head (SEX) is positive and statistically significant at 5% level of
significance. Moreover, the odd of being poor for female-headed households
are 1.602 (OR = 1.602 with p-value = 0.0001) times in comparison to male-
headed households. Conversely, we can say that the odd of being poor for
those male-headed are 0.6242 (OR = 0.6242, given by the reciprocal of
1.602) times for those headed by female.

It is argued that poverty increases at old age as the productivity of the

individual decreases and the individual has few savings to compensate for
this loss of productivity and income. This is more likely to be the case in
developing countries like Ethiopia, where savings are low because of low
income. However, the relationship between age and poverty might not be
linear, as we would expect that incomes would be low at relatively young
age, increase at middle age and then decrease again. Therefore, according to

126
 Ethiopian Journal of Economics Vol. XXIII No 1, April 2014

life-cycle theories we would expect to find that poverty is relatively high at

young ages, decreases during middle age and then increases again at old age.

For the case of Ethiopia based on 1999/00 Household Income, Consumption

and Expenditure Survey (HICES) and Welfare Monitoring Survey (WMS),
Tassew (2008) finds that age of the household head (AGE) is relevant in
explaining poverty. Using the 2010/11 HICES and the methodology
developed above we reach at the same conclusion as Tassew (2008) and age
of the head is statistically significant in explaining poverty at National level.
Various researchers have identified the linkages between education and
poverty. A base hypothesis is that higher education negatively affects
poverty. That the coefficients on the educational attainment of household
head (CMPE, CMPS and CCUA) are all negative and statistically significant
at 5% level of significance for the three models.

The level of education is grouped into four categories ranging from illiterate
to higher education (college and above). The odds of being poor with
education level elementary school (CMPE), secondary school (CMPS) and
college and above (CCUA) was found to be 0.845, 0.388 and 0.515 times
that of illiterate (no schooling-reference category) respectively, implying that
household head with higher educational attainment (CCUA) exhibited a
lower chance to be poor as compared to the illiterate household head for
National.

The result of marital status for national level indicates that

divorced/widowed and married clients are 55.7% and 50.5% more likely to
be poor respectively than single (never married-reference category) clients,
implying clients whose marriage ended because of death of a partner or due
to some disagreement are found to have a significantly high likelihood of
being poor in Ethiopia.

Employment status of the household head is one of the determinants of

household’s poverty status. Self-employed household head (SELFE) are

127
Teshome Kebede and M. K. Sharma: Determinant of Poverty in Ethiopia

about 64% less likely to be poor than those employed in informal sector
(INFOE) which is the reference category. Household’s which are headed by
the one who is employed in formal sector (FORME) are about 81% less
likely to be poor than those works in informal sector (INFOE) at country
level (Model-I).

Last but not certainly the least, the ownership of agricultural land of
household (AGRL) significantly help in lowering the possibility of being
poor. The results show that households having agricultural (farming) land
have 74.3%, 89.3% and 63.2% less chances to be remain as a poor. The
possible reason might be that the most of the population majorly employed
in agricultural sector; the agricultural sector therefore is a big sector of
employment in rural area especially as compared to urban area of the
country.

128
 Ethiopian Journal of Economics Vol. XXIII No 1, April 2014

References

Ahmed, A. U., H. A. Khan and R. K. Sampath. (1991). Poverty in Bangladesh:

Measurement, Decomposition and International Comparison. In: Journal of
Development Studies 27 (4) : 48-63.
Baffoe, J. K. (1999). Change in poverty in Uganda 1992-1997. Working Paper No.
99.22. Centre for Study of African Economics, University of Oxford,
Oxford.
Bevan, P. and S. F. Joirmann. (1997). The Perils Profile in Ghana 1987-1988. In:
African Development Review 4(1): 1-28.
Central Statistical Agency [Ethiopia] and ICF International. (2012). Ethiopia
Demographic and Health Survey 2011. Addis Ababa, Ethiopia and
Calverton, Maryland, USA: Central Statistical Agency and ICF
International.
Dandekar, V. M. and Rath. (1971). Poverty in India. Indian School of Political
Economy, Pune.
Dercon, S. (2001). Economic Reform, Growth and the poor: Evidence from Rural
Ethiopia. Working Paper Series 2001-8. Center for study of African
Economics, Oxford.
Ercelawn, A. (1991). Absolute Poverty as Risk of Hunger: Norms, Incidence and
Intensity for Rural and Urban Pakistan. Applied Economics Research
Centre, University of Karachi.
Greer, J. and E. Thorbecke. (1986). Food Poverty Profile in Applied to Kenyan
Smallholders. In: Economic Development and Cultural Change 35(1): 115-
141
Hosmer, D. and Lemeshow, S. (2000). Applied Logistic Regression, 2nd Edition,
John Wiley & Sons Inc., New York.
IMF (International Monetary Fund). (1999). Ethiopia: Recent Economic
Development. IMF Staff Country (Eds.). Hand book of Development
Economics (Vol.3), Elsevier Ecience, Amsterdam.
Meron Assefa. (2003). Female-Headed Households and Poverty in Urban Ethiopia,
MA Thesis, Addis Ababa.
Ministry of Finance and Economic Development (MoFED). (2012). Ethiopia's
Progress Towards Eradicating Poverty: An Interim Report on Poverty
Analysis Study (2010/11), Addis Ababa.

129
Teshome Kebede and M. K. Sharma: Determinant of Poverty in Ethiopia

Sarkar, S., Midi, H. and Rana, S. (2011). Detection of Outliers and Influential
Observations in Binary Logistic Regression. Journal of Applied Sciences 11
(1): 26-35.
Tassew Woldehanna. (2008). Correlates of Poverty in Urban and Rural Ethiopia.
Ethiopian Journal of Agricultural Economics, 7 (1): 49-81.
Webb, P., J. von Braun and Yishak Yohannes. (1992). Famine in Ethiopia: Policy
Implications of Policy Failures at National and Household Level. Research
Report 92. International Food Policy Research Institute, Washington, D.C.
Webb, P. and J.von Braun. (1994). Famine and Food Security in Ethiopia: Lessons
for Africa. John Wiley and Sons, Chichester.
World Bank. (2001). "World Development Report 2000/01: Attacking Poverty",
New York: Oxford University Press.

130