Chapter one:

ABSTRACT
This study investigates the poverty estimates in Zanzibar using HBS data of 2004/05 on
12,617 heads of households. The log-linear regression model, the Foster- GreerThorecke poverty measure and the logistic regression model are used to analyze the data.
The main findings of this study are that Zanzibar households with large household size,
have many dependants, have heads with low or no education and are allocated in rural
areas are the more likely victims of poverty.
A number of recommendations are made on how to address the poverty issue in Zanzibar.
In particular, recommendations are on the level of education in the society and the
attention to be given to the rural poor in addressing poverty.
This study is divided into six chapters.
First chapter explains the overview of poverty in worldwide and ideas of different
scholars. It explains the poverty situation in Zanzibar and procedures taken in fighting
with the problem in order to improve the wellbeing of the society.
Second chapter explains the Zanzibar profile and overview of Zanzibar economy and
their prospects since the 90s.
Chapter three is the review of related literatures out of the study area and within Zanzibar
boundaries. The researchers’ ideas and critics used as the guides of this study.
The fourth chapter gives the problem statement, objectives, significance of the study and
models used in analysis of the data.
Fifth chapter is analysis and interpretation of the results. The analysis was guided by four
research questions.
Chapter six presents the Conclusions and Recommendations of the study.
iv
LIST OF ABBREVIATIONS
BOT
-
Bank of Tanzania
COICOP
-
Classification of Individual consumption according to purpose
CPH
-
Census of Population and Housing
FGT
-
Foster-Greer Thorecke
FIES
-
Family Income and Expenditure Survey
GDP
-
Gross Domestic Product
HBS
-
Household budget Survey
ICT
-
Information and Communication Technology
IFI
-
International Financial Institutions
MKUZA
-
Mpango wa Kupunguza Umasikini Zanzibar
MDGs
-
Millennium Development Goals
NGOs
-
Non Governmental Organizations
OLS
-
Ordinary Least Squares
SADC
-
Southern African Development Community
UNICEF
-
United Nations Children’s Fund
URT
-
United Republic of Tanzania
ZPRP
-
Zanzibar Poverty Reduction Plan
ZSGRP
-
Zanzibar Strategy for Growth and Reduction of Poverty
v
TABLE OF CONTENTS
Certification .......................................................................Error! Bookmark not defined.
Declaration and Copyright .................................................Error! Bookmark not defined.
Acknowledgement……………………………………………………………………
iv
Abstract .............................................................................................................................. iv
List of abbreviations ........................................................................................................... v
Table of contents ................................................................................................................ vi
List of tables ....................................................................................................................... ix
CHAPTER ONE ................................................................................................................ ix
1.0 introduction ..................................................................Error! Bookmark not defined.
1.1 Background of the problem .........................................Error! Bookmark not defined.
1.1.1. Poverty status and overview of Zanzibar Economy .......... Error! Bookmark not
defined.
1.1.2. Health condition ...................................................Error! Bookmark not defined.
1.1.3. Education .............................................................Error! Bookmark not defined.
1.1.4. Poverty condition .................................................Error! Bookmark not defined.
1.1.5. Income poverty and Inequality ............................Error! Bookmark not defined.
1.2.0 Statement of the problem ........................................................................................... 1
1.3 Objectives of the study.................................................................................................. 1
1.3.1 Main objective ....................................................................................................... 1
1.3.4. Significance of the study....................................................................................... 2
CHAPTER TWO ...............................................................Error! Bookmark not defined.
2.0. introduction .................................................................Error! Bookmark not defined.
2.1. Overview of Zanzibar Economy and Poverty status...Error! Bookmark not defined.
CHAPTER THREE ...........................................................Error! Bookmark not defined.
3.0. Literature review .........................................................Error! Bookmark not defined.
vi
3.1. Economic theory underlying determination of household welfareError! Bookmark
not defined.
3.2 Approaches of measuring poverty ...............................Error! Bookmark not defined.
3.3. Empirical Review........................................................Error! Bookmark not defined.
3.3.1. Related Studies on developing Countries ............................................................. 3
3.4. Studies on Tanzania ................................................................................................... 11
CHAPTER FOUR ............................................................................................................. 18
4.0 Research questions ...................................................................................................... 18
4.1. Research methodology ............................................................................................... 19
4.2. Model structure .......................................................................................................... 19
4.3. Study area and Population ......................................................................................... 24
4.4. Data sources and scope, processing and method of analysis ..................................... 24
4.4.1. Data source and Scope of the study .................................................................... 25
4.4.2. Methods of Data analysis .................................................................................... 25
CHAPTER FIVE .............................................................................................................. 26
5.0. Empirical analysis and Interpretation of results findings .......................................... 26
5.1. Introduction ................................................................................................................ 26
5.2. Interpretation of the results ........................................................................................ 27
5.2.1. Log linear model results...................................................................................... 27
5.2.2 Foster- Greer- Thorecke approach results........................................................... 30
5.2.3 Logistic regression model results........................................................................ 36
CHAPTER SIX ................................................................................................................. 48
Summary, Conclusions and recommendations ................................................................. 48
6.1. Summary of the study ................................................................................................ 48
vii
6.2. Conclusions ................................................................................................................ 48
6.3. Recommendations ...................................................................................................... 50
References ......................................................................................................................... 51
Appendices ........................................................................................................................ 54
viii
LIST OF TABLES
Table 1.0: Percentage of the population below the Basic Needs poverty Line ......... Error!
Bookmark not defined.
Table 2.0 Distribution of poor persons by type of poverty and District (2004/05) ... Error!
Bookmark not defined.
Table.3.0 Percentage of households below the basic needs poverty line, by region in
Tanzania Mainland, 2000/01. ......................................................................................... 14
Table 5.1 The model summary. ........................................................................................ 27
Table 5.2 Regression of Per capita expenditure (Y) On Independent variables(X) ......... 28
Table 5.3 Food Poverty level by Districts, Zanzibar ........................................................ 30
Table 5.4 Distribution of heads of Household by Districts and Basic needs Poverty
Status ................................................................................................................................. 32
Table 5.5 Distribution of heads of Household by Household Size and Poverty Status .... 33
Table 5.6 Distribution of heads of Household by Age and Poverty Status ...................... 34
Table 5.9 Model summary .......................................................................................... ….37
Table 5.10 Food poverty logistic regression results ....................................................... 37
Table. 5.12 Logistic regression of basic needs poverty line against Poverty correlates. 59
APPENDICES .................................................................................................................. 54
ix
x
1.2.0 Statement of the problem
Zanzibar, since independence in 1963 and union with the mainland in 1964, has been
faced with the issue of poverty and inequality, ignorance and disease and low health and
welfare standards.
Despite the existence of government poverty reduction strategies and NGOs fighting
against poverty in Zanzibar; many of our people are still poor. Levels of poverty are still
high and do not show signs of significant decline (URT 2002c). Many people in the
country are unable to secure an adequate diet or meet other basic needs (Aboud 2001).
In order to eradicate poverty in Zanzibar, it is imperative to study income and poverty
more closely and an income and poverty model would be very useful for this purpose.
The model results would be useful for policy makers on increasing number of programs
specifically on targeting the low income areas as a way to redistribute national wealth to
address the problem.
Although literature on poverty measurement is relatively developed, yet, studies about
the analysis of income and poverty in Zanzibar are scarce.
Therefore, this study is an attempt to analyze Zanzibar income data and subsequently
come out with poverty estimates that may be used to give advice on key variables to be
monitored for poverty reduction among households.
1.3 Objectives of the study
1.3.1 Main objective
1
The main objective of this study is to make income poverty estimates of Zanzibar’s
households as a way of assessing her Strategy for Growth and Reduction of Poverty
(ZSGRP)”
Specific objectives to this end are:
1. To analyze whether living conditions of households give the income situation and
predict poverty status of households.
2. To assess the poverty gaps between the ten districts in Zanzibar.
3. To determine the impact of demographic factors on poverty reduction in Zanzibar.
4. To investigate whether or not demographic factors are more likely to cause poverty in
Zanzibar.
1.3.4. Significance of the study
The study will enable to create the income and poverty estimates model for checking the
economic performance.
Also, the study will attempt to rank all ten districts of Zanzibar in their poverty
magnitudes.
The analysis of income and poverty estimates would provide a comprehensive
assessment of the income distribution and poverty profiles by households and district
level as well.
The findings will identify variables highly associated with poverty reduction and
pinpoint to the sectors that should be involved in poverty reduction strategies so as to
develop appropriate approaches in combating poverty in the country.
In other words, the study will come out with recommendations that will aid the strategic
allocation of resources to address the problem of poverty in Zanzibar.
2
Lastly, the study will serve to strengthen the cooperation between the academic
department and the statistical office. The office of the chief Government statistician
collects the data and the academic department should assist in analyzing this data in
trying to solve the prevailing problems in Zanzibar.
3.3.1. Related Studies on developing Countries
In building the models to predict household incomes, a number of studies were reviewed.
Barrios (1996) used multiple linear regression to generate average household income per
municipality in Negros Oriental and Laguna using the 1990 census of population and
Housing (CPH) and the 1991 Family Income and Expenditure survey (FIES).
The predicted household income is used to calculate the incidence of poverty per
municipality. The variables used as predictors of income are: literacy rate, percentage of
housing unit with weak roof, percentage of housing units with bamboo or makeshift
walls, percentage of households with television, percentage of households with no safe
supply of drinking water, and percentage of households with unsanitary type of toilet.
3
Tabunda et al (1999) describe a number of variables that correlate with poverty from the
1997 APIS survey. The variables are classified into four categories: location and
barangay characteristics (25 variables plus 4 interaction variables), dwelling
characteristics (9 variables plus 3 interaction variables), family characteristics (19
variables plus 4 interaction variables), and ownership of durable goods (22 variables plus
2 interaction variables). Apart from the description, the survey shows that the issue of
ownership is crucial and has to be considered in the way of explaining poverty status of a
particular household. Ownership will be incorporated in this study but only household
ownership.
The study by Balisacan et al (1997) also identified variables that are correlated with
poverty. The poverty correlates are: location of household residence (region and province
dummy), barangay characteristics (type and number of private and public
establishments), housing characteristics (type of roof, wall, toilet, water source), age
composition and schooling attainment of family members, sex and marital status of
household head, and ownership of durable goods (vehicle, electrical appliances). From
the above experience this study used some of the variables to predict poverty of the
household. The variables used are: location of household residence, age composition and
schooling attainment of family members, sex and marital status of household head, and
ownership of durable goods (vehicle, electrical appliances).
In the study of estimates of income and poverty for region IV; Nikkin et al (2003) used
multiple linear regression technique and Foster-Greer-Thorecke(FGT) class of additively
decomposable poverty measures. The census of population and housing (CPH) and the
Family Income and Expenditure Survey (FIES) were used. The objective of the study was
4
to generate income estimates at the lowest level of disaggregation, namely the household
and to rank the provinces according to their poverty incidence. The family Income or
expenditure was used as the dependent variable depending on the goodness of fit. The
results were that, both possibilities were explored and 17 out 23 domains have a better fit
using total expenditure. The adjusted R-squared of the final regression models range from
47.7% to 70.3%. The range was comparable to other results - Tabunda, Balisacam and
Barios as mentioned above; their adjusted R- squared was 55.2, 65.9 and 50.4 per cent
respectively. This study employed the FGT measure and multiple linear regression to the
Zanzibar HBS (2004/05} data to rank districts according to their poverty incidence.
Coulombe (1996) used multivariate analysis to analyze the determinants of poverty in
Mauritania based on household survey data for 1990. They estimated a multinomial logit
model for the probability of being in poverty depending on household-specific economic
and demographic explanatory variable. The authors found that low education, living in
rural areas and high burden of dependence significantly increase the probability of a
household being poor. This study investigated whether or not education and location of
household can be contributive factors to the household being poor in the experience of
Zanzibar.
Rodriguez (2000) used logistic regression model with the probability of household being
extremely poor as the dependent variable and a set of economic /demographic variables.
It was found out that, the variables that are positively correlated with the probability of
being poor are size of the household, living in rural areas, working in rural occupation
and being a domestic worker. Variables negatively correlated with the probability of
being poor are the education levels of the household head, his/her age and occupation.
5
This study uses FGT to get poverty estimates and determine the variables that explain
poverty in Zanzibar.
The study conducted in Malawi by National Economic council (2001) identified
important determinants of poverty status to be education levels, agriculture productivity,
type of employment, access to services, household size. A regression procedure was used
to derive a model where the dependent variable was the natural log of the household
welfare indicator (daily total per capital consumption and expenditure) and independent
variables were education of the household members. The study established that
determinant of poverty
differ from area to area in Malawi, and that agriculture
intervention are appropriate for poverty alleviation in the shorter term and at the same
time larger reduction in poverty should be sought through education.
Iceland (1997) observed that, “poverty spell observed among women and less education
individuals are also significantly more likely to be left censored. He demonstrated that
additional year of education lowers the odds of recording a left-censored observation by
18 per cent”. Furthermore, the study revealed that families which are dual headed and
have few children are less likely to be left censored.
Furthermore, in the way of explaining how different scholars and researchers have said
about poverty; Radhakrishna, et al. (2007) have this to say “a household is identified as
chronically poor if its income is below the poverty line and if its children are suffering
from malnutrition for a longer period of time,” The determination of chronic poverty
measured by considering economic, demographic and social factors. The other poor
households may be able to move out of chronic poverty because of their small household
size, as well as the more intensive use of labor, including child labor.
6
Recent research however, provides compelling evidence that fertility rates have an
important bearing on poverty. The conclusion is supported both by micro level evidence
which comes in a series of studies that have provided comprehensive estimates of
intergenerational flows, including those to and from children (Lee, 2000). The study
consistently finds that children are a financial burden in high fertility setting. The
hypothesis that parents are having more children because they will benefit financially is
not borne out by the facts. Children consume far more than they produce even in
traditional settings. Thus the birth of an additional child reduces the material standard of
living of other family members.
Again literature on how to analyze poverty correlates mentioned two standard techniques
using household data. One way is to estimate the probability of being poor using logit or
probit techniques with household characteristics as the explanatory variables. Another
way is to estimate household welfare function with OLS methods. Both methods are
helpful in understanding poverty and its causes. One researcher who employ the first
technique and using panel survey data in Cote dvore was Kanbur (1997), the study
reported that for urban households, human capital is the most important factor for
determining welfare level and welfare changes over time. Meanwhile, the same approach
employed in Ethiopia and finds that well educated households have greater chances of
improving their welfare compared to others, and those households with many dependants
are in worse position (Ibid, 1997).
Apart from the above, Mwambu et al. (2000) using both urban and rural data find that,
“mean household education and literacy are strongly and positively correlated with
consumption expenditure, while the household size is negatively correlated with per
7
capita consumption expenditure”. The common thing in both studies is education to be
important variable for household welfare. This study will differ with above study on per
capita consumption expenditure. Instead, this study will use per capita expenditure due to
the reason that the consumption expenditure according to Zanzibar data included
purchased and home produced items, as well as the item received as payment in kind or
gifts. For the purpose of this study of analyzing the poverty of household using per capita
expenditure would be the best to portray the income status of household.
Furthermore, World Bank (1996) initiated a study known as Small area estimation. The
study aimed at acquiring a reliable estimate of the household expenditure(Y), enables
calculation of more specific poverty measures linked to a poverty line. Log linear
regression model per capita expenditure using a set of explanatory variables (X) that are
common to both the integrated household survey and the census, (eg. .Household size,
education, housing and infrastructure characteristics and demographic variables). The
estimated coefficients from this regression are used to predict log per capita expenditure
from every household in the census.
In general, in the analysis, the following variables: household size, level of education, age
of head of household, housing characteristics and dummies plus interaction with certain
household level variables, turned out to be key variables chosen in most regressions. As
expected, household size had a negative correlation with household per capita
expenditure. The housing variables showed mixed results depending on strata. However,
the parameter estimates of the dependent variables cannot be interpreted as a causal
effects, but simply provide information on the direction of relation ship.
8
Again, a significance level criterion was chosen with no ceiling on the number of the
variables selected. All household survey variables that were significant at the five per
cent level were selected for the regression. Similarly, in this study the same procedure as
followed except for the marital status dummy’s variables due to its importance in
Zanzibar in explaining poverty situation.
A study conducted in Ukraine in 1999 tried to assess the variables which will make the
household to be more in the poverty risk Using Ukraine 1999 data. Buitano (2000)
addressed the question “Who are the poor in Ukraine? As in many other less developing
countries these are, in particular, families with children, elderly living alone, especially
pensioners with low labor supply, unemployed and increasingly the working poor.
Women appear to be especially at risk as single providers and survivors. Generally,
pensioners do not appear to be at significant risk of poverty.” The researcher of this study
is interested to know the position of household headed by women whether they are more
at risk as in Ukraine.
There is a strong correlation in poverty rates depending on the level of education of the
head of household.
Poverty rates are also correlated with the access to and the size of subsidiary household
plot, housing conditions and other characteristics although differences in rates are not
large.
The share of household expenditures on food is high for all types of households: on
average 65.6 per cent of household expenditures were spent on food, including 63.7 per
cent by non-poor and 70.8 per cent by poor and 72.5 per cent by extremely poor
9
households. 90 per cent of populations spend more than 60 per cent of their consumption
on food.

Key facts are as follows:

Poverty is not confined to those with low labor supply. Because of wage arrears
and low wage levels, there is a growing incidence of poverty among workers and
working families.

Single elderly above 70 years form a significant share of poor households.
However, more generally poverty among pensioners is tempered by the fact that a
large number of such people keep working--some 35% of people aged 55 to 70
work.

Households with children are likely to be poor: some 50% of such households fall
below the poverty line, and the likelihood of poverty increases with the number of
children. Single provider households, especially families headed by women, are
particularly exposed.

Poverty is correlated with lack of education and occupational skills: the higher the
education level of the head of household, the less likely the household is to be
poor.

Poor households are more likely to reside in rural areas. Moreover, poverty is
differentiated by regions and related to residence outside growth areas.

While owning a plot of land will mitigate poverty, plot size plays an important
role in determining poverty--plot size of poor households is some 50% less on
average than for the non-poor
10
This study wished to analyze the claims using the HBS data (2004/05) if they are
comparable with the Ukraine findings.
3.4. Studies on Tanzania
A number of Poverty studies have been done in Tanzania.
Rutasitara (2001) examined the impact of economic policy changes on rural poverty in
Tanzania. He found out that while poverty in Tanzania is an overwhelmingly rural
phenomena and a reality, there is still a need to make a critical examination of its
determinants. Also, he found out that quantitative segmentation of rural “poor” and “non
poor” groups obscures the fact that even amongst the poor, poverty is still relative and
rural differentiation is a matter of policy interest. The state of rural education attainment
and poor infrastructure for education and health imply delirious consequences for the
productive life of the rural people.
Again Rutasitara (2001) states, despite the commitment to fight poverty, ignorance and
disease since independence and the fact that Tanzania still ranks among the poorest
nations, continued discourse on the impact of the reform policies on poverty reduction is
imperative (pg.4).
Mkenda (2001) did a multivariate analysis of fishery resource and welfare in rural
Zanzibar. Using household budget survey data, he assessed poverty and inequality
between different socio economic groups, particularly artisan, fishermen and peasant
farmers in rural Zanzibar. Two empirical approaches were used, first, log linear model
and logit model with per capita expenditure as a dependent variable and a set of
explanatory variables that include the households source of income.
11
Also, Mkenda (2001) used Zanzibar HBS 1991 and found that, even after controlling for
a variety of factors, the households whose main source of income is artisan fishing were
better off than households whose main source of income is peasant farming. As
mentioned above, Mkenda (2001) looked at the determinants of household welfare in
Zanzibar but strictly to the artisan fishermen and peasant farmers in rural Zanzibar, in
contrast to this study which looks at the determinants of poverty among the households in
both rural and urban Zanzibar. The study further added that, “With respect to household
size, the coefficient estimates of size and size squared suggest that an increase in
household size decreases the household’s welfare, though at a decreasing rate. The
dummy for coral zone has a negative and significant coefficient, confirming the fact that
a household located in coral zone tends to have a lower welfare compared to the one
located in the non coral zone. The average age of adult members of households has a
positive impact on household welfare, although this impact decreases as the average age
increases.(pg,144 and 245).
Kirama (2004) did Ordinary Least squares (OLS) to estimate determinants of poverty
using HBS data conducted in Tanzania mainland only in 2001. The dependent variable is
income and multinomial logit models were fitted. Log linear involves regressing the log
of household per capita real consumption/net expenditure on a set of explanatory
variables. He found out that poverty status is highly correlated with level of education,
household size, location of the household and type of occupational activities. Household
size proved to be very important in explaining both the household net expenditure and
poverty status. The study differs with this study in the data set and the areas of
investigation. This study utilizes the logistic regression approach as the method of
12
measuring the poverty correlates. This study includes household size and age variables in
the analysis using Zanzibar HBS (2004/05) data.
Actually, household’s expenditure can portray the position of the family whether is low
or high income earners by just observing the allocation of the expenditure. According to
HBS report 2004/2005, the classification of individual consumption according to purpose
(COICOP) for SADC countries said that, “It is normally expected that households with
higher Incomes spend proportionately less on food compared to those with low Income”
(pp.77). In other words, the fraction of total expenditure that goes into food is inversely
proportional to the household Incomes.
Report on poverty mapping which provided estimates of income poverty by district in
Tanzania (2005) showed that, the proportion of households who are below the poverty
line differs greatly between regions and districts. Again, the report added that, most
districts have distinctive patterns of needs that should be met by sector specific
interventions. Spatial analysis can aid the setting of priorities for the sectors and for the
equitable deployment of financial and human resources.
The regression relation for the HBS was used to infer for each household in the census its
per capita consumption. The common variables between the census and HBS were
identified and those variables that were present in both the HBS and the 2002 population
census were picked. The results of regional poverty estimates are as indicated in the
Table 3.0 below:
13
Table.3.0 Percentage of households below the basic needs poverty line, by region in
Tanzania Mainland, 2000/01.
REGION
Percentage
Households
of
Std error
below
Percentage
of
Household
below
poverty line on HBS
poverty line simulated
Estimates
Estimates
Std error
Dodoma
34
5.5
32
3.1
Arusha/Manyara
39
7.0
32
3.1
Arusha
-
-
21
1.4
Manyara
-
-
43
2.4
Kilimanjaro
31
6.3
28
1.3
Tanga
37
5.8
26
1.3
Morogoro
29
3.0
28
1.9
Pwani
46
8.3
38
2.1
Dar es Salaam
18
2.7
19
1.2
Lindi
53
14.1
39
2.3
Mtwara
38
4.3
38
2.0
Ruvuma
41
8.3
37
2.1
Iringa
29
5.3
28
1.6
Mbeya
21
5.1
23
1.1
Singida
55
4.8
49
3.4
Tabora
26
3.7
40
2.1
Rukwa
31
3.9
36
2.0
Kigoma
38
3.7
38
2.3
Shinyanga
42
6.5
43
2.4
Kagera
29
8.9
29
2.0
Mwanza
48
6.3
43
1.7
Mara
46
8.4
50
2.6
Source. Human and Development Report 2005
Another set of factors considered potentially important in explaining household incomes
are, as justified by Logan (1996); and Lam (2003), “Political status and connections
14
though difficult to measure directly, but may be associated with the presence of a cadre
within the household. Here we focus on party membership, as a cadre status is often
attached to employment”. In so doing, location of residence affect income levels.
According to Msambichaka at al. (2003), poverty remains widespread and deep with 61
per cent of Zanzibaris living without basic needs of livelihood. In rural areas the
population is most hit, with 50 per cent of people being poor, Pemba performs worst,
with 64 per cent poor population compared to 59 per cent in Unguja. Apart from income
poverty, non income indicators also reveal that poverty is a serious problem in Zanzibar.
There is a low education achievement, high literacy, and low net enrolment at basic and
primary education. Girls are the most disadvantaged, with 60 per cent of all illiterates and
lower net enrolment rate (66.7 per cent compared to 67.3 per cent for boys).
Furthermore, the study explained that life expectancy is low only 48 years and infant
mortality is high at 83 per 1000 live births. Other quality of life indicators are not
convincing either, under five mortality rate is 114.3 per 1000 births and maternal
mortality is as high as 377 per 100,000. Pemba has high rate (406 compared to Unguja
367).
Malnutrition has high prevalence. About 36 per cent of children less than 5 years are
stunted, 12 per cent of all under five severely stunted and 6 per cent wasted; 26 per cent
of under five children are severely under weight. Pemba has the worst performance in all
aspects of malnutrition, health related interventions, water and sanitation have not been
adequate in Zanzibar. More than a half of the population has no access to safe drinking
water and less than 4 per cent of households in rural areas do not have access to sewage
system. The study further added that, more than 50 per cent of household in rural areas
15
do not have toilets and again Pemba has an alarmingly high proportion, 93 per cent.(pg,
103). Msambichaka (2008) proposed a solution of poverty reduction by saying, ‘people
are to participate more meaningfully in economic activities including plans, strategies and
policies in the public and private sectors. Sustainable national economic development
will become a reality in Tanzania only if it is built on the full potential of the people and
communities across the country”.
Again, the Zanzibar HBS (2004/05) report, tried to look at the relationship between
poverty and households’ demographic, social and economic characteristics, especially the
head of household. It also analyzed poverty status in relation to basic community
facilities. The results showed that, overall, 49 per cent of Zanzibar population lives below
the basic needs poverty line. Poverty is apparent among households of all sizes. The
gradual increase in headcount ratio, however shows that larger households are more
likely to be poor; for example two thirds of the population in 10+ size households are
reported to be poor compared to less than 5 per cent of the population in households of
size less than 3 members. This applies to both rural and urban areas, but more apparent
in rural areas. Also, poverty levels by area suggest a generally similar trend in both rural
and urban areas, namely high risk of poverty to those not working and those engaged in
agricultural activities, and low poverty risk to employed (pg. 9). In this study, different
variables that correlate with poverty are considered in order to observe their influence in
explaining poverty in Zanzibar.
Generally, Zanzibar is poor, not only by using the standard measures of income but even
other measures. Non-income indicators also reveal that poverty in Zanzibar is a serious
problem. Education achievement in Zanzibar is very low. The rate of illiteracy is high
16
and the majority of illiterates are women. Also, most of the households in Zanzibar have
very poor sanitation facilities. A fairly high proportion of households do not have toilets.
Therefore, identification of factors or variables that are related to and therefore can help
to explain poverty is not only important for policy intervention but is also important for
monitoring welfare changes.
17
CHAPTER FOUR
4.0 RESEARCH QUESTIONS
This study is guided by the following research questions.
Firstly, do living conditions of a Household indicate the income situation and predict the
poverty status of a household in Zanzibar?
Secondly, is there a difference in the poverty level between the ten districts in Zanzibar?
Thirdly, do demographic factors (age, household size) have influence on poverty status in
both rural and urban areas of Zanzibar?
Fourthly, are demographic factors the most likely factors to influence poverty in Zanzibar
than other factors?
Rationale of the first question is that in Zanzibar the living condition of a household
explains the poverty status to the particular household. This just because their pattern of
expenditures is based on the type of commodities and ownership of assets.
Rationale of the second research question is that MKUZA which is the poverty reduction
strategy with a national outlook may take quite a long time to reach its goal if indeed
there are remarkable poverty differences between districts in Zanzibar.
Rationale of the third research question is that there is a debate among people in Zanzibar
as to whether or not household sizes are the causes of poverty in many of the households.
A number of people claim that large households whose members are young work hard to
maintain their family needs and therefore may actually exit from poverty quickly.
The rationale of the fourth research question is that people in Zanzibar do not know the
factors that are most likely to cause poverty but suspect that demographic factors are.
This study will shed light on this issue.
18
4.1. RESEARCH METHODOLOGY
Kothari(1990) defines research methodology as a way to systematically solve the
research problem.
In this study, the log multiple linear regression
technique has been employed in
estimation of Income of Households. Regression analysis is the statistical methodology
for predicting values of one or more response (dependent) variables using a collection of
predictor (independent) variables. It is used for accessing the effect of predictor variables
on the responses.
Also, the study uses FGT model for measuring poverty incidence in Households. The
model is used to rank all ten districts in Zanzibar in terms of their poverty level.
The predictor variables were those found in the budget survey and population census
because they helped to cross check the reality. Freund (1979, pg378) tried to show how to
improve prediction by saying, “it stands to reason that predictions should improve if one
considers additional relevant information.”
Decomposition of the variation is used in measuring of goodness of fit. This is used to
avoid using residuals as a measure of good fit since residual value is not unit free.
This study used t-test to test the significance of the parameters associated with the
variables under the study.
4.2. MODEL STRUCTURE
The steps in generating income estimates can now be outlined. Using the Household
budget survey data, multiple regression is used to estimate a model to predict family
19
incomes. Household per capita Expenditures are regressed against family and housing
characteristics as well as interaction variables derived from these characteristics.
Nikkin (2003) used this approach and utilized survey data to estimate per capita
expenditure as a function of a variety of household characteristics.
The regression model is
LnYi =
α + β1Χ1i + β2Χ 2i + β3 Χ 3i + β4Χ 4i +. . . + β8Χ8i+ εi
Where Y is household income or expenditure, X is a vector of housing and family
characteristics, and ε is an error term.
According to Gujarati (2003, pg 224), “whenever you have a log-linear regression model
involving any number of variables the coefficient of each of the X variables measures the
(partial) elasticity of the dependent variable Y with the respect to that variable.
This study used per capita expenditure because it provides a monetary measure of
poverty, since it is more reliable than income data.
Actually, studies of income distribution household income is the common measure of
household welfare, although household per capita expenditure is better since it
automatically "corrects" for household size (Jacob and Datta, 1980).
This is the approach used to estimate models for rural and urban areas in Zanzibar.
Since the objective is to predict the poverty levels of all households in Zanzibar, the
variables are carefully selected so that they are present both in the 2004/05 Household
budget survey and the 2002 population census.
The variables are;
Χ1 - Food expenditure of household
Χ 2 – Clothing and foot wear of household expenditure
20
Χ 3 – Education expenditure in the household
Χ 4- Household size
Χ 5 – Transportation expenditure of household
Χ 6- communication expenditure of household
Χ 7 – Housing, Water, Fuel and Power Expenditure
Χ 8 – Health expenditure of household
Again, Foster-Greer-Thorecke (FGT) class of additively decomposable poverty measures
is used in this study. This approach was used in Philippine to rank the provinces
according to their poverty incidence. Therefore in this study, this measure used to rank all
ten districts of Zanzibar in order of their poverty magnitude using 2004/05 HBS data.
Poverty for person i at time t can be written as:
P
1
 z  yi 
ni 

n i  q  z 
=

Where z = the per capita threshold income defined by the HBS.
y i = the per capita income or expenditure of the ith house hold.
n i = the household size of the ith household.
q
= the subset of the population with y i <z.
n = total population.

=
 non- negative parameter reflecting the weight given to the distribution of
poverty in the measure of poverty.
The poverty incidence is obtained when α = 0, the percentage of households who are
below the poverty line defined poverty threshold.
21
The depth index (α =1) is the average shortfall of income of the poor house holds from
the threshold income expressed in proportion to the threshold income.
The severity index (α =2) is like the depth index but it takes into account the magnitude
of the difference of the Poor’s income relative to the threshold income.
The higher the difference of the poor family’s income from the threshold, the “more
severe” is the poverty than if it were lower. Hence, it can be said that the severity index is
a overtly distribution sensitive index, Balisacan et al. (1998)
Also, the study used logistic regression model, it is useful for the situation in which we
want to be able to predict the presence or absence of a characteristic or outcome based on
values of a set of predictor variables.
This methodology developed for this study is inspired by Mkenda (2001), Rodriquez
(2000), Geda et al. (2001) and Kirama (2003) approaches with the use of cross section
data analysis from Zanzibar household Budget Survey of 2004/2005.
Log Linear model is:
Log(y)
=
XB + ε
Where y is total net expenditure adjusted to adult equivalent scales, X is explanatory
variables and ε is random term. The estimation is done for Zanzibar only.
With the Logistic regression model, the dependent variable is binary, taking only two
values, 0 if the household is poor, and 1 if non poor. The Probability of being extremely
poor depends on set of variables x so that
Pr(Y=1)
= F ( B' x)
Pr(Y=0)
= 1-F ( B' x)
Using logistic distribution we have
22
Pr(y=1)
=
e  '
1  e  '
= Ө [B’X]
Where Ө represents the logistic cumulative distribution function. The probability model
is the regression
(Y/X)
= Ө {[1- F ( ' X)] + [F ( ' X)]}
Where ' is a vector of coefficient, Bis
X is a vector of independent variables, Xis
Dependent variables
For the Logistic model, the study uses two dependent variables. There is a binary variable
indicating whether a household is below the basic needs poverty line, given a value 0 if
poor and 1 if non poor. There is also a binary variable indicating whether a household is
below the food poverty line given a value 0 if poor and 1 if non poor.
Log of net expenditure adjusted to adult equivalent scales has been dropped from this
model purposely because it is used in the log linear multiple regression model and its
results have been indicated in Table 5.2 in data analysis section of this research.
Independent variables
Independent variables used in this study, are:
Household size.
A binary variable for the gender of household head 1= male, 0 = female
Age of the Household head.
A binary variable for the marital status of the household head 1=married and 0
0therwise.
23
Highest level of education completed by the household head.
A binary variable for the residence of the household 1= Rural and 0 = urban.
A dummy variable indicating whether the household head has completed primary
education or not.
A dummy variable indicating whether the household head has completed secondary
education or not.
A dummy variable indicating whether the household head has completed the tertiary
education or not.
4.3. Study area and Population
The study is on Zanzibar. It uses the HBS data of 2004/05 with 12,617 heads of
household.
The data provide baseline information on level of poverty, monitoring and evaluation
information for Zanzibar poverty reduction plan (ZPRP). The area has been chosen
because there are data collected by the office which researchers have not analyzed.
It is the belief of the researcher that the results and conclusions of this study will enable
people in Zanzibar to improve their understanding of the poverty problem and
consequently be motivated to use the recommendations as the tools to exit from poverty.
4.4. Data sources and scope, processing and method of analysis
The most important decisions in poverty research concern the choice of the research
method. As Hagernaars (1991,pg.134) points out, both the population of [the] poor and
the extent of their poverty are dependent to a large degree on the chosen definition. The
24
methodological implications are important, for instance, for the targeting of the poverty
alleviation programs (Ladorchini et al. 2003)
In this study, the household is used as a unit of analysis. The household is scrutinized for
level of Income and poverty status and variables correlated to household poverty in
Zanzibar.
4.4.1. Data source and Scope of the study
The study used the data from the Office of Chief Government Statistician through its
household budget survey (HBS) in 2004/05 and population census data in 2002.
The HBS in 2004/05 is the largest ever-household budget survey in Zanzibar. The
information was collected for both rural and urban areas for the whole country.
The HBS 2004/05 focuses its analysis on national and regional indicators of poverty like
household consumption, income poverty, and a range of non consumption measures
including those of the priority sectors of education, health and water. This information
was enough for this study.
4.4.2. Methods of Data analysis
This study on its analysis used three approaches according to the purpose of the study.
The three methods are:
Log linear regression model, FGT approach and Logistic regression approach.
The results of the analysis are shown in the output in the coming chapters.
25
CHAPTER FIVE
5.0. Empirical analysis and Interpretation of results findings
5.1. Introduction
The research is concerned with statistical analysis of Income and Poverty in Zanzibar. It
used per capita expenditure data regressed with variables that correlate with poverty.
The predictor variables used in this study were tested for strength of their correlation.
Correlation performed in pair wise cases. This part presents the results of the multiple
regression model using per capita expenditure of household head as a dependent variable
and independent variable that correlate with poverty in Zanzibar basing on the 2004/05
HBS. Also, the FGT model used to rank all ten Zanzibar districts. And also the logistic
regression used to measure poverty determinants of households in Zanzibar.
Variables were:
House ownership of household head, Main activities of head of
household, Education level of head of household, household size, food, marital status of
head of household, Location of head of household head, age of head of household,
Household with dependents above fifteen years and above.
In analysis, dummy variables were created for the dichotomous variables in order to
allow the whole of the analysis of the variance to be treated within the multiple linear
regression framework.
This study realizes that the dummy variables are useful because they allow us to control
for membership with a particular category or group. If we neglect to split a categorical
variable into several dummy variables when using it in a regression, we would get invalid
26
results because regression analysis assumes variables to be continuous unless told
otherwise.
5.2. Interpretation of the results
5.2.1. Log linear model results
In analysis, the variables selected have indicated strong correlation with per capita
expenditure of head of Household. (61.9%) the standard error of the estimate is approx.
0.34. as shown in the Table 5.1 below:
Table 5.1 The model summary.
Model
1
R
.787(a)
Source: Researcher.
R Square
.619
Adjusted R
Square
Std. Error of
the Estimate
.619
.33987
From the purely statistical view point, the estimated regression line fits the data. The Rsquared of 0.619 means that about 62 per cent of the variance in the head household per
capita expenditure is explained by the independent variables mentioned below.
The regression is run, with the dependent variables being the (log of) per capita
expenditure adjusted to adult equivalent and a list of independent variables being
household size, marital status, and education, age of head of household, number of
dependants in the household above fifteen years old, transportation, health and location of
households.
The results from log linear model are reported in Table 5.2, all coefficient estimates are
statistically significantly different from zero at 5 per cent level of significance.
Table 5.2 below, contains the estimated coefficients (under column heading B) and
related statistics from the log multiple linear regression model that predicts the
expenditure of household head from a constant and variables mentioned above.
27
Table 5.2
Regression of Log Per capita expenditure (Y) On Independent variables(X)
Model
Unstandardized Coefficients
B
1
(Constant)
Sex
dependents
Marital status
Std. Error
Beta
Standardized
Coefficients
Sig
t
10.271
.005
2058.651
.000
-.040
.002
-.030
-16.342
.000
.003
.001
.012
4.661
.000
.009
.001
.014
7.419
.000
Age (years)
-.002
.000
-.062
-39.597
.000
Household Size
-.146
.001
-.775
-269.650
.000
.006
.000
.538
315.488
.000
2.28E-5
.000
.148
99.869
.000
5.48E-6
.000
.100
64.389
.000
1.34E-5
.000
.100
63.283
.000
Area
.140
.002
.123
81.235
.000
education
.062
.001
.104
65.235
.000
Food
Health
Transportation
Education - fisher
adjusted
Source: Researcher
From Table 5.2 above, we see that in Zanzibar House budget survey in 2004/05 the
qualified impact of the individual independent variables on per capita expenditure of
heads of household were -0.04, 0.003,0.009, -0,002, -0.146, 0.006, 0.0000228,
0.00000548, 0.0000134, 0.14 and 0.062 respectively. In other words, over the period of
study, holding all the other independent variables in the model constant, the change of
sex of the head of household from female to male leads to on average of about a 4 per
proportional decrease in the household expenditure.
Similarly, holding all other independent variables in the model constant, an increase of
one dependant of fifteen years and above will result on average in 0.3 per cent
proportional increase in the household expenditure. The research results indicate that the
28
larger the number of dependants on fifteen years and above supported by a working adult,
the more likely the household is to fall beneath the poverty line.
The above model results indicate that a change of the marital status of the head of
household from unmarried to married holding other variables constant will lead on
average to a 0.9 per cent proportional increase in household expenditure.
Again, the results indicate that a change of the age of the household head, holding other
independent variables constant leads to 0.2 per cent proportional decrease in household
expenditure.
Also, holding all other variables constant an increase of household size by 1 person leads
on average to about 14.6 per cent proportional decrease in the household expenditure.
This has been justified in literature that,” families which are dual headed and have few
children are less likely to be left censored. (John, 1997).
The coefficient of expenditure on food in the model shows the proportional effect of
household head per capita expenditure to increase by 0.6 per cent while controlling with
marital status, education and Household size.
The expenditure on food is higher compared to other estimates in this analysis except on
education and area. This scenario has an economic interpretation. It is normally expected
that the households with higher incomes spend proportionately less on food compared to
those with low incomes. That is, the fraction of total expenditure that goes into food is
inversely proportional to the household income. This analysis tells us that, majority of
Zanzibar people are poor. Therefore, in order to improve the wellbeing of Society, the
Government should control the food price and their availabilities.
29
The above model results indicate that a change of education level of the head of
household from uneducated to educated holding other variables constant will lead on
average to a 0.00134 per cent proportional increase in household expenditure. Other
results are as shown in Table 5.2 above.
About a quarter of adults in Zanzibar were reported to have no education. The HBS
reports, corroborates this; “almost 76 per cent of adults can read and write in at least one
language. This is most frequently the case in urban than the rural areas.
Like wise, a proportional change of education facility leads to on average per day
household expenditures go up by about 6.2 per cent, holding other expenditures constant.
Again, when cost of transport per distance in one km increases contribute proportional
change to an average household expenditure increases by about 0.5 per cent holding all
other independent variables in the model constant.
5.2.2 Foster- Greer- Thorecke approach results
For the purpose of this study, the severity index (α =2) is used, it takes into account the
magnitude of the difference of the Poor’s income relative to the threshold income. The
statistic tries to show the poverty gaps between different districts in Zanzibar. The
research question asked; “is there a difference in the poverty level between the ten
districts in Zanzibar?”
Table 5.3 shows the poverty levels in the ten districts in Zanzibar.
Table 5.3 Food Poverty level by Districts, Zanzibar
Districts
severity
Kaskazini "A"
Poverty levels
6
Kaskazini "B"
Poverty levels
Kati
Kusini
Poverty levels
Poverty levels
3
5
1
Magharibi
Poverty levels
26
Mjini
Poverty levels
42
Wete
Poverty levels
4
30
Micheweni
Poverty levels
2
Chake Chake
Poverty levels
7
6
Mkoani
Poverty levels
Source: Researcher.
Table 5.3 above shows that, there is a big poverty gap from district to district and within
district; using the result of FGT model. When all ten Zanzibar districts are ranked on the
basis of the model, it is found there is indeed a difference in the poverty levels.
According to Model, Kusini Unguja district has the smallest poverty gap followed by
Micheweni in Pemba. These two districts resemble in climate and soil features. Their
main activities are fishing and agriculture.
The third district is Kaskazini B in Unguja followed by Wete in Pemba.
Again, The fifth district is Kati district in Unguja. While the sixth one is Kaskazini A in
Unguja and Mkoani in Pemba. The eighth district in order of poverty gap is Chake Chake
in Pemba. The ninth one is Magharibi district in Unguja and the tenth one is Mjini
district in Unguja.
Table 5.4 below, using the food poverty headcount and Basic needs poverty Headcount
shows clearly the poverty gap from one district to another.
According to HBS 2004/05, adjusting the food poverty line was done to allow for nonfood consumption to give the basic needs poverty line. This is done by calculating the
share of expenditure that goes on food in the poorest 25 per cent of households.
Multiplying the food poverty line by the inverse of this share inflates it to allow for nonfood consumption. The food share was 62 per cent, giving a basic needs poverty line of
20,185 Tshs. Another poverty line is food poverty line, which represents the amount of
money needed to sustain an adult for a month, is Tshs 12,573.
31
Table 5.3 above and 5.4 below show the poverty status in the different districts in
Zanzibar. This demonstrates that, there are poverty gaps in Zanzibar districts, and this
justifies a need to narrow down the gaps.
Table 5.4 Distribution of heads of Household by Districts and Basic needs Poverty
Status
Districts
.
Basic Needs Poverty Status
Non Poor
Poor
%
%
Total
%
Kaskazini "A"
54.3
45.7
100.0
Kaskazini "B"
56.6
43.4
100.0
Kati
62.9
37.1
100.0
Kusini
51.0
49.0
100.0
Magharibi
66.4
33.6
100.0
Mjini
68.6
31.4
100.0
Wete
34.7
65.3
100.0
Micheweni
32.5
67.5
100.0
Chake Chake
47.4
52.6
100.0
Mkoani
62.6
37.4
100.0
Total
56.6
43.4
100.0
Source: Researcher
The district which has many poor household heads from the Table 5.4 above is
Micheweni for about 67.5 per cent, in which only 32.5 per cent are non poor heads of
household. Wete also takes the second position; it has 65.3 per cent heads of household
who are poor. Chake Chake has only 52.6 per cent poor heads of household. Kusini
Unguja has 49 per cent followed by Kaskazini Unguja with 45.7 per cent. Again
Kaskazini B has 43.4 per cent while Mkoani district has 37.4 per cent. Another district is
Kati district which has 37.1 per cent heads of household who are poor and 33.6 per cent
for Magharibi district. The district which has minimum number of poor heads of
32
household is Mjini district in Zanzibar with only 31.4 per cent and the rest are non poor
with 68.6 per cent.
The third research question being studied is; do demographic factors (age, household
size) have influence on poverty status in both rural and urban areas?
Table 5.5 Distribution of heads of Household by Household Size and Poverty Status
Basic Needs Poverty
Status
Household
Size
Non Poor
Poor
%
%
Total
%
1 person
9.9
.8
5.9
2 persons
13.2
2.5
8.5
3 persons
16.3
5.7
11.7
4 persons
16.2
9.7
13.4
5 persons
14.9
13.3
14.2
6 persons
10.0
16.8
13.0
7 persons
7.2
15.4
10.8
8 persons
4.2
12.3
7.7
9 persons
3.4
9.0
5.9
10+ persons
4.7
14.4
8.9
100.0
100.0
100.0
Total
Source: Researcher.
Table 5.5 above, using basic needs poverty line, clearly indicates that the likelihood that
people are poor increases as household size increases. The households which have six
persons size are poorer in comparison with the household with fewer numbers of
household members or above.
In Zanzibar Households with one person account for 0.8 per cent of basic needs poor,
while households with two persons account for 2.5 per cent. Also, Table 5.5 above
indicates clearly that those households with six persons account for 16.8 per cent of basic
needs poor and account for 10.0 per cent of basic needs non poor. Households with 10+
persons account for about 14.4 per cent of basic needs poor.
33
At the household level a decline in the number of children per adult leads to higher per
capita income for the country and the household.
Table 5.6 Distribution of heads of Household by Age and Poverty Status
Basic Needs Poverty
Status
Total
Age of Head
of Household
15-20
Non Poor
Poor
%
%
%
.5
.2
.4
21-30
20.2
8.4
15.1
31-40
30.9
26.6
29.0
41-50
21.6
30.9
25.6
51-60
13.7
19.7
16.3
61+
Total
13.1
14.2
13.6
100.0
100.0
100.0
Source: Researcher.
Table 5.6 above shows the effect of age of head of household as the increasing age of
household head. The results in Table 5.6 indicate that the age 15- 20 years account for 0.5
per cent basic needs non poor households and for 0.2 per cent basic needs poor
households.
Age 21-30 years account to 8.4 per cent basic needs poor households and for 20.2 per
cent basic needs non poor households. Table 5.6 indicates that the number of poor
household due to age of household head decreases at the age of 61+ because the young
people are physically fit, able to work, looking for work in order to accomplish their
wishes. Some of their children have started working somewhere and supporting their
families’ income. More about poverty status in relation to age of household head is as
indicated in the Table 5.6 above.
34
Table 5.7 below indicates the rural area households whose household size is above four
have big numbers of poor persons. Households with one person account for 0.1 per cent
number of poor persons in rural areas and the same percentage in urban area. Also,
households with seven persons account for a 17.1 per cent number of poor persons in
rural areas in contrast to about 13.2 per cent in urban areas.
Table 5.7 also shows that households with ten persons and above in rural areas account
for smaller number of poor persons of about 19.4 per cent than those in urban areas that
account for 37.2 per cent.
Table 5.7. Distribution of poor persons by district and area, Zanzibar 2005
Area
Household
Size
1 person
2 persons
3 persons
4 persons
5 persons
6 persons
7 persons
8 persons
9 persons
10+
persons
Total
439
2,613
9,148
21,854
36,931
55,111
59,219
52,289
41,671
Urban
% of Poor
Persons
0.1%
0.5%
2.0%
3.7%
6.4%
11.9%
13.2%
12.9%
12.1%
19.4%
67,297
37.2%
61,327
100.0%
346,572
100.0%
165,048
% of Poor
Persons
0.1%
0.8%
2.6%
6.3%
10.7%
15.9%
17.1%
15.1%
12.0%
Rural
Number of Poor
Persons
Number of Poor
Persons
Source: Researcher.
Table 5.7 indicates that generally, rural areas have more number of poor persons than in
urban areas. The Table 5.7 clearly shows the number of poor persons in the different
household sizes in Zanzibar
Generally, the results indicates that the number of poor persons decrease as the household
size decreases.
The Table 5.8 below, indicates that the majority of heads of household in rural areas have
35
141
830
3,267
6,156
10,578
19,616
21,831
21,307
19,994
no education compared to heads of households in urban areas this also indicates the
increase of poor persons in rural than in urban areas. About 48.5 per cent household
heads in rural areas have no education, while in urban areas about 21.6 per cent of the
household heads have no education.
Similarly, rural areas have small number of household heads with education above
secondary level of about 17.9 per cent as compared to urban areas with 39.6 per cent.
Table 5.8 Distribution of Poverty by Education of Household Head and Area,
Zanzibar 2005.
Rural
Number of Poor
Persons
% of Poor
Persons
No Education
Adult Education
Basic Education
Above Basic
Education
Total
168,140
21,069
95,176
48.5
6.1
27.5
62,187
17.9
100
346,572
Urban
Number of Poor
Persons
% of Poor
Persons
35,705
2,882
61,107
21.6
1.7
37.0
39.6
100
65,354
165,048
Source: Researcher.
5.2.3 Logistic regression model results
Fourthly, are demographic factors the most likely factors to influence poverty in Zanzibar
than other variables?
To shed some light on this research question a logistic regression model is adopted.
The variables selected for this logistic regression according to the purpose of this study
are household size (hhsize), area (urb-urban and rur-rural), sex (P03- sex), marital status
(P06-MST) and education (P08E-EDU). Household size (1=small size) and household
size (2= medium), Urban (2=urban) and rural (1=rural). Sex (1=male), (2=female).
Marital status (1= never married), (2= married), (3= divorced), (4= separated), 5=widow)
36
and (6= living together) and education (0=no education). (1= primary), (2= secondary)
and (3= tertiary).
Table 5.9 Model summary
-2 Log
likelihood
6878.644(a
)
Source. Research
Step
1
Cox & Snell
R Square
Nagelkerke R
Square
.067
.146
Table 5.10 below, contains the estimated coefficients (under column heading B) and
related statistics from the logistic regression model that predicts the households that are
more likely to be poor from variables household size, area, sex, marital status and
education.
Table 5.10 Food poverty logistic regression results
B
Step
1(a)
S.E.
hhsize
hhsize(1)
Wald
529.8
37
484.7
88
44.28
4
110.2
91
.448
df
Sig.
Exp(B)
2
.000
1
.000
.204
1
.000
1.380
1
.000
1.463
1
.503
.962
-1.588
.072
.322
.048
.380
.036
-.038
.057
6.740
6
.346
P06_MST(1)
-3.006
5748.404
.000
1
1.000
.049
P06_MST(2)
-2.923
5748.404
.000
1
1.000
.054
P06_MST(3)
-3.059
5748.404
.000
1
1.000
.047
P06_MST(4)
-3.762
5748.404
.000
1
.999
.023
P06_MST(5)
-3.235
5748.404
.000
1
1.000
.039
P06_MST(6)
-2.731
5748.404
1
1.000
.065
.192
.035
1
.000
1.212
.610
P<0.5
5748.404
.000
30.59
1
.000
1
1.000
1.840
hhsize(2)
urb_rur(1)
P03_SEX(1)
P06_MST
P08E_EDU(1)
Constant
Source. Researcher.
From Table 5.10 above, each of the regression coefficients describes the size of the
contribution in the model. Given the coefficients, the logistic regression equation for the
37
probability of a head of households being poor given the poverty variables above can be
written as,
e  '
Probability (a household head to be poor) =
where log(y) is the equation given
1  e  '
by 0.610 – 1.588( small household size ) + 0.322 (large household size) +0.380
(household in rural area) – 0.038 (household headed by female) – 3.006 (household with
no married couple ) – 2.923 ( household with married couple) -3.059 (household with
divorced head of household) -3.762 (household with separated marriage) -3.235
(household with widow household head) -2.731(household with no marriage) + 0.192
(primary education of household head).
Applying this to a heads of household who have small household sizes, holding other
variables in the model constant, we find;
B'X = 0.61 – 1.588(1) = -0.978
Probability (a head of household to be poor due to small household size) =
e 0.978
=
1  e 0.978
0.376
= 0.273
1  0.376
The probability of head of household who has small household size to be poor is 0.273.
Since this probability is smaller than 0.5 then, we predict that the event will not occur. On
the other hand, the probability of head of household which does not have small household
size to be poor is 1- 0.273=0.7267
When the same model is applied to a household with medium household size, holding all
other variables in the model constant, we find;
B'X = 0.61 + 0.322(1) = 0.932
38
The probability (medium household size to be poor) =
2.53960
e 0.932
=
= 0.717
0.932
1  2.53960
1 e
In this case, we predict the medium sized household to be poor. In other words, the
household with less than that will not be poor since the probability of being poor is 10.717 = 0.2825. These results are comparable to the results recorded in the study
conducted in Ukraine from which Buitano (2000) shows that, “Poverty is much higher
among households having three and more children.
Further more, the probability of household being poor due to being in rural areas, holding
all other poverty variables in the model constant is estimated as follows;
B'X = 0.61 + 0.380(1) = 0.99
2.6912
e 0.99
Probability (rural household to be poor) =
=
= 0.7290 and the
0.99
1  2.6912
1 e
probability of rural household not to be poor is estimated to 1- 0.729 = 0.2709. This
justify that rural households have high poverty risk in comparison with urban households.
Thus, to be in rural areas, there is high possibility to be poor.
In analyzing the head of household whether there is poverty difference for household
which is headed by male or female, the model tells us the following results.
B'X = 0.61 – 0.038(1) = 0.572.
The probability of households headed by female being poor =
1.7718
=0.639
1  1.7718
e 0.5 7
1  e 0.5
2
7 2
=
on the other hand, the probability of being non poor for households
headed by female is 1- 0.639 = 0.3607.
Again, the household in which the head of household never married holding other
variables in the model constant, we find;
39
B'X = 0.61 – 3.006(1) = -3.006
0.04949
e 3.006
The probability of being poor is
=
= 0.04715.
3.006
1  0.04949
1 e
Also, when the model is applied to household with married household head, holding
other variables constant, we find;
B'X = 0.61 – 2.923(1) = -2.313
The probability of married head of household basing on food poverty line is
=
e 2.313
1  e 2.313
0.09896
= 0.09.
1  0.09896
And the probability of the divorced household head holding other poverty variables
constant, we find,
B'X = 0.61 – 3.059(1) = -2.449
Therefore, the probability of household to be poor given that the head of household head
0.08637
e 2.449
is divorced is
=
= 0.16588. Using the same model, the
 2.449
1  0.08637
1 e
probability of separated household heads holding other variables constant, we find;
B'X = 0.61 – 3.762(1) = -3.152
The probability of household to be poor given the head of household head whose marital
status is separated is
0.04276
e 3.152
=
= 0.041
3.152
1  0.04276
1 e
Also, applying the same model to the widowed household head holding other variables in
the model constant we find;
40
B'X = 0.61 – 3.235(1) = -2.625. Hence, the probability of widowed household head to be
0.07244
e 2.625
poor is
=
= 0.067547
 2.625
1  0.07244
1 e
Furthermore, applying the same model to household in which the household head has no
education holding other variables constant, we find;
B'X = 0.61 + 0.192(1) = 0.802
The probability of a household being poor in which the head of household has no
education is
e 0.802
=
1  e 0.802
2.22999
= 0.6904, because this is bigger than 0.5, the
1  2.22999
event is likely to occur. On other hand, the probability of a household being poor in
which head of household head has education above primary education is 1- 0.6904 =
0.30959.
Education of household head has positive influence on the poverty risk. This
model tells us that poverty increases when household head has no education or low level
of education and the household will be out of poverty when the household head has
completed high level of education.
The results from Table 5.10 above show that, there is a positive relationship between
medium household size, marital status, location of household and education. This means
increasing values of each of the variables increases the likelihood of household falling
into poverty.
However, variables such as small household size and dummy variable for marital status
have negative relation ship with poverty. This means that the decreasing the values of
each of these, increases the probability of being poor except for marital status. The signs
of the coefficients are as expected.
41
Generally, the most influential factors to poverty are household size, location of
household and education level of household head.
Furthermore, looking the odds ratio from Table 5.10 above, small household size is less
likely to be poor. Or small household size is only 20.4 per cent likely to be poor
compared to a large household size. When household head has medium household size,
the household is nearly 1.4 times more likely to be poor compared to a non medium
household size.
Also, it is predicted that if household is at rural area, the household is nearly 1.46 times
more likely to be poor than household in urban area. Therefore, households which are
located in urban areas are better off than the rural ones.
Table 5.10 seems to indicate that, the sex odds ratio are almost equal to one, hence, it
implies that the sex of the household head does not have much influence on food poverty
status of households.
The odds ratio for marital status indicates that all ratio are less than one, which means,
poverty of households decrease for all married, divorce, separated, etc, even though
statistics from the Table 5.10 indicated that divorced marriages are more likely to be poor
than married ones.
Again, households those head has primary education are 1.212 more likely to be poor
than households whose head has higher than primary education.
Again, a logistic regression was run taking basic needs poverty line as dependent variable
and with the same independent variables.
Table.5.11 Model summary
Step
-2 Log
likelihood
Cox & Snell
R Square
Nagelkerke R
Square
42
1
14737.439(
a)
.161
.218
Source: Researcher
Table 5.12, shows the summary of results.
Table. 5.12 Logistic regression of basic needs poverty line against Poverty correlates
B
Step
1(a)
S.E.
Wald
1675.632
2
Sig.
.000
hhsize(1)
-2.677
.035
1559.680
1
.000
0.068
hhsize(2)
.298
.030
97.285
1
.000
1.347
hhsize
urb_rur(1)
df
Exp(B)
.325
.021
236.540
1
.000
1.384
-.003
.036
.007
1
.935
.997
9.452
6
.150
P06_MST(1)
-3.119
5736.931
.000
1
1.000
.044
P06_MST(2)
-2.769
5736.931
.000
1
1.000
.063
P06_MST(3)
-2.834
5736.931
.000
1
1.000
.059
P06_MST(4)
-2.912
5736.931
.000
1
1.000
.054
P06_MST(5)
-2.812
5736.931
.000
1
1.000
.060
P06_MST(6)
-2.647
5736.931
.000
1
1.000
.071
.149
.023
43.598
1
.000
1.161
2.554
5736.931
.000
1
1.000
12.861
P03_SEX(1)
P06_MST
P08E_EDU(1)
Constant
Source: Researcher
P<0.5
Results in the Table 5.12 above, indicate that there is a positive relationship between
medium household size, primary education and location of household. This means
increasing values of each of the variable while holding other variables constant increases
the likelihood of falling into poverty. However, variables such as small household size
and dummy variables for marital status have negative relationships with poverty.
Again, as for the food poverty discussion,
43
Pr(y=1) =
e  '
is the probability that household is basic needs poor while the
1  e  '
probability that the household is non basic needs poor is Pr (y=0) = 1-F ( B' x). Applying
this to heads of household who have small household sizes, holding other variables in the
model constant, we find;
B'X = 2.554 - 2.677(1) = -0.123
Pr (small household size to be poor) is
0.884
e 0..123
=
= 0.469
 0.123
1  0.884
1 e
The probability of household to be basic needs poor due to small household size basing
is 0.469 since this probability is less than 0.5 then, we predict that the event is less likely
occur.
Applying the same model to household which has medium household size holding all
other variables constant, we find;
B'X = 2.554 + 0.298(1) = 2.852. The probability of household being poor due to medium
household size is
17.322
e 2.852.
=
= 0.945.
2.852
1  17.322
1 e
In this case, we predict the household to be poor. Indeed, this estimate indicates that as
the household size increases as the poverty risk increases. This probability indicates the
magnitude of poverty risk for those household with large household size. On the other
hand, when a household has smaller than this size the household will be non poor is 10.945 = 0.055.
Furthermore, for household in rural areas, holding all other poverty variables in the
model constant the model gives;
B'X= 2.554 + 0.325(1) = 2.879
44
Accordingly, probability (rural household to be poor) =
e 2.879.
=
1  e 2.879
17.796
= 0.9467
1  17.796
and the probability of non rural household to be not poor is estimated to be 1- 0.947 =
0.0543. This justify that rural households have high poverty risk compared to non rural
households. Thus, to be in rural areas amounts to high possibility to be poor.
In analyzing the head of household whether there is poverty difference for household
which is headed by male or female, the model tells us the following results.
Applying the model based on basic needs holding other variables in the model constant,
we find;
B'X = 2.554 - 0.003(1) = 2.551. The probability of household to be poor due to household
12.8199
e 2.551
headed by female household head is
=
= 0.9276
2.551
1  12.8199
1 e
On the other hand, the probability of being non poor by headed by female household head
is 1- 0.9276 = 0.0724.
Again, applying the same model, the household that will be poor due to household head
has never married giving other variables in this model constant, we find;
B'X = 2.554 – 3.119(1) = -0.565
The probability of being poor giving the household head has never married is
0.568
e 0.565
=
= 0.362. The household to be poor due to household headed by
 0.565
1  0.568
1 e
married household head basing on basic needs poverty line, holding other poverty
correlates on the model constant, we find;
B'X = 2.554 – 2.769(1) = -0.215.
45
The probability of married household head basing on basic needs poverty line is
0.806
e 0.215
=
= 0.446
 0.215
1  0.806
1 e
Applying the model to divorced household head holding other variables constant, we
find;
B'X = 2.554- 2.834(1) = -0.28. The probability of being poor due to be headed by
divorced household head is
0.756
e 0.28
= 0.4305.
 0.28
1  0.756
1 e
Applying the same model to separated household head holding other variables constant,
we find;
B'X = 2.554- 2.912(1) = -0.358 the probability of household being poor due to be headed
by separated household head is
0.699
e 0.358
=
= 0.411
 0.358
1  0.699
1 e
Also, for widowed household head holding other independent variables in the model
constant, we find;
B'X = 2.554- 2.812(1) = -0.258 hence, the probability of household to be poor due to be
headed by widow household head is
0.7725
e 0.258
=
= 0.4358
 0.258
1  0.7725
1 e
Meanwhile, applying the same model for the household whose household’s members are
living together no marriage, holding other poverty variables constant, we obtain;
B'X = 2.554- 2.647(1) = -0.093
The probability of poor households in which household members are living together;
there are no formal marriages is
0.911
e 0.093
=
=
 0.093
1  0.911
1 e
46
0.477
Applying the same model to household in which household head has no education
holding all other variables constant, we find;
B'X = 2.554 + 0.149(1) = 2.703. The probability of a household being poor due to
household head has primary education is
14.924
e 2.703
=
= 0.937
2.703
1  14.924
1 e
Indeed, the probability is bigger than 0.5, therefore, the event is most likely occur. On the
other hand, the probability of a household being poor in which head of household has
education above primary education is 1- 0.937 = 0.063
Furthermore, looking the odds ratio from Table 5.12 above, small household size is less
likely to be poor. Or small household size is only 6.8 per cent likely to be poor compared
to a large household size. Table 5.12 above also indicates that, when household has
medium household size, the household is nearly 1.347 times more likely to be poor
compared to non medium household size.
Also, it is predicted that if household is at rural, the household is nearly 1.38 times more
likely to be poor than household in urban areas Therefore, the household which are
located in urban areas are better off than the rural ones.
Table 5.12 seems to indicate that, the sex odds ratio are almost equal to one, hence, it
implies that the sex of household head does not have much influence on household basic
needs poverty status.
The odds ratio for marital status indicates that all ratios are less than one, which means,
poverty of households decrease for all married, divorce, separated and widowed
47
variables. This result fluctuates with study by Balisacan et al (1997) as explained in
Literature of this study.
Again, households whose head has primary education are 1.161 more likely to be poor
than households whose head has higher than primary education.
Generally, the results of this analysis are not conclusive but should scores as the starting
point for further research to understand the poverty problem in Zanzibar for the benefit of
the Zanzibar society at a large.
CHAPTER SIX
SUMMARY, CONCLUSIONS AND RECOMMENDATIONS
6.1. Summary of the study
The research has analyzed variables that are likely to cause poverty in Zanzibar. The
study has been motivated by the fact that since Zanzibar’s independence in 1961, the
magnitude of poverty has been increasing instead of decreasing despite the several
initiatives against it.
The study has considered a number of factors that are likely to be associated with
poverty and basing on a number of conclusions and hence recommendations may be
the analysis made to the Government on how to effectively address the poverty issue.
6.2. Conclusions
The main conclusions are as follows.
48
-
Household size, education and location of households have significant impact on
poverty in Zanzibar.
-
The probability of a household falling into poverty increases with household size,
number of children per household and increase of number of dependants. Hence,
demographic factors have influence on poverty status.
-
Food expenditure is high for rural households and is less for the urban
households.
-
Marital status whether the head of the household is married or otherwise, has not
shown any distinctive difference on the poverty condition of households in
Zanzibar.
-
Gender of household head does not seem to have much impact on the living
condition of households. Hence, the sex of head of household cannot tell the
poverty condition of households in Zanzibar.
-
The poverty gap is highly observed in Mjini district in Zanzibar and low poverty
gap is observed in Micheweni district in Pemba and Kati district in Zanzibar.
-
The research indicates that the larger the number of dependants supported by a
working adult, the more likely the household to fall beneath the poverty line.
-
Rural households are more likely to be poor than urban households in Zanzibar.
-
The analysis identifies five principal elements of a poverty reduction strategy for
Zanzibar. These include (1) increased investment in education, (2) subsidize food
for low price (3) adoption of measures to raise agricultural productivity, (4)
improved rural infrastructure, and (5) reduced numbers of dependants in the
households and household size in general.
49
6.3. RECOMMENDATIONS
It is recommended that, the Zanzibar Government: (1) Always consider education as an
important factor for household welfare and thus take steps to ensure that household heads
strive to have their children acquire better and higher levels of education, (2) continue
tirelessly to address rural poverty in order to raise the general welfare of the Zanzibar
people. (3) As a start, provide resources to rural residents for services such as medical
care and introduce policies that will give incentives to increase and improve rural
agriculture production. (4) Control food prices and ensure availability of food; (5) think
on reducing import tax to encourage importers to provide low priced necessary goods; (6)
address within district poverty gap among the individuals. (7) The Government should
improve its scope in social welfare and restructure the terminal benefit of retirees;
(8)consider subsidizing the cost of the medical care as a way of further reducing poverty
since, health expenditure also consumes the income of heads of household.
50
REFERENCES
Adolf Mkenda (2005), Fishery Resources and Welfare in Rural Zanzibar,
Kompendiet- goteborg -Sweden
Balisacan, A. M.,:R. P.Alonzo, T.C Monsod, G. M.Ducanes, and J. Esguerra (1998).
Bian, Y., and J.R. Logan(1996), “Market Transition and the Persistence of Poverty.
’The changing Stratification system in Urban China; American Sociological review
61(5); 739-58.Conceptual Framework for the Development of an Integrated Poverty
monitoring and Indicator system” National economic Development authority.
Khondker.pdf
BOT “Annual Report 2004/2005 in Tanzania Investment Centre Report , December
2004’.of economic situation and development plan for 2005/2006”General Government
Publisher, Dar es salaam
Buitano, M Mapi (2000), Poverty Reduction
51
Ukraine Press submitted on 26/06/2000
Cuillermo Cruces (2005), Distributional analysis Research program.
Houghton Street London WC 2AE
David Piachau (2002), Capital and the Determinants of Poverty and Social exclusion
Oxford University, Central European University Press
Glewwe, P. (1991),
“Investigating the determinants of Household Welfare in C`ote d`
Ivore, Journal of Development Economics, 35, 307-337
Iceland John (1997), The Dynamic of poverty spells and Issues of left-censoring.
London.
http005.://www.blackwell-synergy.com/doi/abs/10.1111/j.1475-4991.1980.tb00175.x
http//www.crefa.ecn.ulava/.ca/develp/
Johnson, R and Dean (1992), Applied multivariate Statistical analysis
Third addition Prentice-Hall, Inc.
Kumar, Krishna T. and Gore, Anil Anil P, and Sitaraman , V. (1996), “Some conceptual
and Statistical Issues on measurement of Poverty”
Journal of Statistical Planning Inference 49, 53-71
Lee, R. D (2000),
Intergenerational transfer and Economic life cycle;
A cross Cultural perspective. In sharing the wealth, Demographic
Change and Economic transfer between Generation.
Oxford University Press, pp. 17-56.
Rutasitara, L. (2003), Economic Research Bureau.
University of Dar es Salaam Press.
52
Rutasitara L. (2001), “Economic Policy and Rural Poverty in Tanzania’ Research on
Poverty Alleviation (REPOA),Dar es Salaam, Research paper No. 02.1
Kirama, S (2004), “Determinants of Poverty in Tanzania: Household Budget,
Survey analysis.
University of Dar es Salaam Press.
Malawi National Economic Council (2001), “The determinants of Poverty in Malawi”
An Analysis of the Malawi integrated Household Survey, 1997-98.
The International Food Policy Research Institute, DC, USA
Ministry of Finance and Economy (2006) “Review of Zanzibar economy and the
implementation of Zanzibar Development Plan,2005/2006” General Government
Publisher, Dar es salaam
Ministry of Finance and Economic Affairs: Book II (2005), “The Direction
Ministry of Finance and Economic Affairs Book 1 (2005), “Economic situation and
implementation of the plan 2004/2005”Revolutionary Government of Zanzibar (2004),
The 2004/2005 Budget Speech.
Msambichaka - Guardian News paper of March 29, 2008
Sen, A.
(1976),
Poverty: An ordinal approach to measurement. Econometrical
Zanzibar Poverty reduction plan vision 2020 of October,2001
World Bank (1991)
Making Adjustment work for poor.
A frame work for policy Reform in Africa World Bank.
Washington D. C.
Zinduka (to get rid of poverty) quarterly magazine October- December 2005, number
008
pp. 7
53
APPENDICES
Table A1: Distribution of Mean Per Capita Expenditure (28 Days) by Category of Item by Area,
Zanzibar 2005.
Area
Food & Non Alcoholic
Beverages
Alcoholic Beverages &
Tobacco
Clothing & Footwear
Total
Rural
Urban
%
%
%
59.9
51.3
56.6
.4
.3
.3
7.0
7.7
7.3
Housing, Water, Fuel &
Power
15.7
19.1
17.1
Furniture, Household
Equipment & Household
Maintenance
5.1
5.3
5.2
Health
2.3
2.2
2.3
Transportation
3.3
4.1
3.6
Communication
.3
1.2
.7
Recreation &
Entertainment
.4
.4
.4
Education
1.3
1.9
1.5
Restaurants & Hotels
2.0
3.1
2.4
Miscellaneous Goods &
Services
2.2
3.3
2.6
100.0
100.0
100.0
Total
Source: Researcher.
54
Table 9.7: Mean Annual Income Per Earner by Education Level and District, Zanzibar 2005.
Kaskazini "A"
Kaskazini "B"
Kati
Kusini
Magharibi
Mjini
Wete
Micheweni
Chake Chake
Mkoani
Total
No Education
TShs.
245,455
355,018
397,926
214,135
371,782
529,900
231,579
276,014
433,013
454,339
335,114
Education of Household Head
Primary /
Adult
Basic
Education
Education
Secondary
TShs.
TShs.
TShs.
447,610
213,089
679,504
576,253
349,824
818,735
304,446
295,161
587,309
382,711
221,938
608,551
570,275
512,967
1,073,400
879,882
631,176
1,049,575
430,290
245,620
587,345
472,975
221,748
555,445
743,720
473,407
1,035,739
805,067
485,634
904,874
552,575
405,611
909,855
55
Total
Tertiary
TShs.
46,000
85,436
870,174
932,068
1,050,725
1,975,724
989,359
626,339
2,002,993
1,961,718
1,364,055
TShs.
272,813
426,709
369,444
297,876
672,384
808,206
337,688
304,784
592,288
566,689
514,581