ADDIS ABABA UNIVERSITY

OFFICE OF GRADUATE STUDIES

DEPARTMENT OF STATISTICS

DETERMINANTS OF FAMILY PLANNING

PRACTICE IN ETHIOPIA

By
Nibret Alene

A Thesis submitted to the Office of Graduate Studies of Addis

Ababa University in Partial fulfillment of the requirement for
the Degree of Masters of Science in Statistics

June 2010
Addis Ababa, Ethiopia
 ADDIS ABABA UNIVERSITY
OFFICE OF GRADUATE STUDIES
DEPARTMENT OF STATISTICS

DETERMINANTS OF FAMILY PLANNING

PRACTICE IN ETHIOPIA

By
Nibret Alene

A Thesis submitted to the Office of Graduate Studies of Addis

Ababa University in Partial fulfillment of the requirement for
the Degree of Masters of Science in Statistics

Advisor: Professor Eshetu Wencheko

June 2010
Addis Ababa, Ethiopia
 ADDIS ABABA UNIVERSITY
OFFICE OF GRADUATE STUDIES
DEPARTMENT OF STATISTICS

DETERMINANTS OF FAMILY PLANNING

PRACTICE IN ETHIOPIA

By
Nibret Alene

Approved by the Board of Examiners:

_____________________________ ------------------------
Department Head Signature

_____________________________ -------------------------
Examiner Signature

_____________________________ -------------------------
Examiner Signature

_____________________________________________________________
Tel. +251-11-1239465 P.O.Box 1176 Ababa Addis, Ethiopia
 DECLARATION
I, the undersigned, declare that the thesis is my original work, has not been presented for
degrees in any other University and all sources of materials used for the thesis have been
duly acknowledged.

Name: Nibret Alene

Signature: ------------------------

Place of submission: Department of Statistics, College of Natural Sciences, Addis

Ababa University

Date of submission: June, 2010

This thesis has been submitted for examination with my approval as a University advisor.

Prof. Eshetu Wencheko ---------------------------------

Advisor’s Name signature
 TABLE OF CONTENTS
Page
ACKNOWLEDGEMENT...............................................................................i
ABSTRACT....................................................................................................ii
CHAPTER ONE
1 INTRODUCTION……………………….…………………….....……....1
1.1 Background……………………………………………………………….…….....1
1.2 Statement of the problem………………………………………………….....…...5
1.3 Objective of the study………………………………………………...……..........6
1.4 Significance of the study …………………………………...………………....….7
1.5 Limitation of the study ………………………………………………...…………7
CHAPTER TWO
2 LITERATURE REVIEW…….…………….…..………….………………..……8
2.1 Overview of family planning…...………………….…………………………...…8
2.1.1 Situation of family planning in the world……………………………...…..…9
2.1.2 The situation of family planning in developing countries…………………..10
2.1.2.1 Socio-economic factors……………………………………………...12
2.1.2.2 Demographic factors …………………………………….…….....…14
CHAPTER THREE
3 DATA AND METHODOLOGY…………….…………………….…………..16
3.1 Data source…………………………………….………………………..…….....16
3.2 Variables of the study………………………….….…………………….……….17
3.2.1 The response variable…………………………………….………………..17
3.2.2 Explanatory variables/factors….…………………………………………..17
3.3 The Methodology………………………………………….……….……..……..19
3.3.1 The logistic regression analysis……..……...…………….....….…….........19
3.3.1.1 Test of goodness of fit……………...……………..…………....…….22
3.3.2 The multiple logistic regression analysis………………….………...…...23
3.3.3 Multilevel logistic regression model……..………...……....….…….……24
 3.3.3.1 Two-Level model……………………….……….……………….….24
3.3.3.2 Heterogeneous proportion………………..….……..….….....…..…..26
3.3.3.3 The empty logistic regression model ……………..……….………..28
3.3.3.4 The random intercept logistic regression model………………...…..29
3.3.3.5 The random coefficient logistic regression model……………....…..30
3.3.3.6 Estimation and testing technique………………………..…..…..…..32
CHAPTER FOUR
4 STATISTICAL DATA ANALYSIS AND RESULTS …………...………34
4.1 Major socio-economic, demographic and proximate characteristics of family
planning practice status of couples………….………………….…………..……34
4.2 Determinants of family planning practice among couples in Ethiopia: logistic
regression analysis………………………………………………..…..…….……38
4.2.1 Determinants of family planning practice in urban areas………….………38
4.2.2 Determinants of family planning practice in rural areas ……….……...…40
4.2.3 Determinants of family planning practice at national level……………..…42
4.2.4 Goodness of fit and model diagnostics …………………………..……..…44
4.3 Determinants of family planning practice: A multilevel logistic
regression analysis…………..…………………………………………..……….45
4.3.1The empty logistic regression model ……………………….………..….…46
4.3.2 Random intercept model and fixed explanatory variables…..…….….……47
4.3.3 Random coefficient model …………………………………..……...……..48
CHAPTER FIVE
5 DISCUISSION, CONCLUSION AND RECOMMEDATION………...53
5.1 Discussion………………………...………………...………..……....…...……53
5.2 Conclusion……………………...…………………...…………….......…......…55
5.3 Recommendations……………...…………….………...………..……..…..…56
REFERENCES ………………………………..………………….…….…57
APPENDIX…………………………...…………..………...…………….…...….…..62
 LIST OF TABLES

Page
Table 3.1 Description of the variables and coding……………………….……………18

Table 4.1 Distribution of couples by socio-economic, demographic and proximate

related characteristics and by region …………………………..………….36
Table 4.2 Selected important variables in the model for urban areas……...….….……39

Table 4.3 Important selected variables in the model for rural areas….………….……..41

Table 4.4 Important selected variables in the model at national level………………....43

Table 4.5 Classification table………………………………………………..…………44

Table 4.6 Estimates for the empty model ……………………………………..…..…..46

Table 4.7 Estimates for random intercept and fixed coefficient model………………..47

Table 4.8 Results for fixed and random effects of random coefficient model……..….48

Table 4.9 Level-2 covariance matrix of the random coefficient………….……………49

Table 4.10 Level-2 correlation matrix of the random coefficient……….….….………50

Table 4.11 Test of the significance of correlation……………………………………..50

Table 4.12 Parameter estimates for regions……………………………………..……..51

 ACKNOWLEDGEMENT
First and foremost, I would like to express my profound gratitude to my respectable
advisor, Professor Eshetu Wencheko, for his exemplary guidance, monitoring and
constant encouragement throughout the course of this thesis work. His stimulating
supports helped me in preparing my thesis within the time-frame.

I would like to thank the Central Statistical Agency (CSA) for providing me the 2005
EDHS data on family planning.

Special thanks are due to my dear brothers, Birhanu, Damtie, Yohannes and Habtamu
Alene and my sister Amarech Alene for their kind assistance by providing Laptop.

I record with appreciation the help rendered by my dear classmates encouraging me to

work hard. My best wishes to all who helped directly or indirectly in ensuring a better
thesis write-up. Finally, I acknowledge the help of Dire Dawa University for funding my
MSc study.

i
 ABSTRACT
This study is an attempt to identify socio-economic, demographic and proximate
predictors of couples practice of family planning in Ethiopia. In this study the data source
is the Ethiopian Demographic and Health Survey conducted in 2005(EDHS 2005) by the
Central Statistical Agency (CSA). From the analysis of the data it was found that “visit
by family planning (FP) workers in the last 12 months”, “discussing FP with partner”,
“husbands/partners approval of a method”, “exposure to information from mass media”,
“educational level of women”, “partners education level”, “number of living children”,
“age of a woman”, “occupation of a women and that of her partner” are the most
important factors that influence the practice of FP by couples at national level.
In demographic studies, large-scale surveys often follow a nested structure of data
because information is collected at different levels of hierarchy. It is thus necessary to
elicit the inherent variations, which occur between different levels. In order to explore
this idea, multilevel logistic regression analysis has been employed.
The random coefficient model for the selected few predictor variables (“Place of
residence”, “visit by family planning (FP) workers in the last 12 months”, “discussing FP
with partner”, “husbands/partners approval of a method” and “exposure to FP
information from mass media”) are significant determinants for practice of family
planning, and these vary from region to region. The analysis showed there is difference in
the status of couples on FP practice that depend on place of residence.

Key words: Family planning, couples practice, multilevel logistic regression model.

ii
 CHAPTER ONE
1 INTRODUCTION
1.1 Background
The continuing growth of the world population has become an urgent global problem. Most of
this growth is occurring in developing countries where the fertility rate is very high (Bandura
2002; Merrick 2002; Potts 2000; Ross & Winfrey 2002; WHR 2005). Ethiopia, like most
countries in sub-Saharan Africa, is characterized by high fertility and rapid population growth. It
stands third after Nigeria and Egypt in this respect (Fitaw et al. 2003). Total fertility rate (TFR)
declined by one birth per woman (or about 15 percent) from 6.4 births in 1990 to 5.4 births in
2005, but that rate remains critically high in a country already burdened by severe poverty and
resource depletion. Under-five childhood mortality declined by about 25 percent, from 166 per
1,000 in 2000 to 123 in 2005. Maternal mortality declined by about 23 percent from 871 per
100,000 in 2000 to 673 in 2005(Mizanur et al., 2007).

In most developing countries in general, and in sub-Sahara Africa in particular, the problem of
population growth and reproductive health challenges include - high maternal mortality, high
population growth rate, total fertility rate and much unmet need for family planning in the world.
The situation in Ethiopia is still much worse than most African countries (Ethiopia Trend Report
(ETR; 2007)). In spite of the fact that family planning service delivery facilities and other
supplies have increased in number in the country, a significant portion of the Ethiopian
population is still living in abject poverty.

The high population growth creates a hindrance to economic development of the country.
Urbanization in the country is at a very low stage with 15 percent of the population residing in
urban areas and the rest living in rural areas. A substantial segment of the population is under
poverty line. The population living on less than USD1 per day accounts for 31 percent of the
population (MOH, 2001).

Family planning is the practice of a couple to prevent or avoid unwanted birth and control the
spacing between child birth to help create a small and planned family. It is the best way to
control the rapidly and massively growing population. So, family planning contributes to

1
promote the health and welfare of the family and thus contribute effectively to the social
development of a country. The health of mothers is not only affected by nutrition status but also
by early marriage, frequent pregnancies, early motherhood, abortion etc. Moreover, the health of
a child is also affected by the mother’s health status. In this context family planning provides
advices and methods to avoid the above events.

Family planning is essential to the well-being of women, men, adolescents, and the community
at large. It offers women opportunities to plan and space pregnancies in order to achieve personal
goals and self-sufficiency.

The controversy over abortion has influenced family planning programs and also had
implications for women’s health. To appreciate how the controversy has affected family
programmes, some background on the relationship between family planning programmes and
abortion is useful. From the earliest days of family planning programs, prevention of abortion to
reduce related maternal mortality and morbidity has been an important part of the health
rationale for promoting contraception (Seltzer J., 1998). Abortion is illegal in Ethiopia; studies
indicate that abortion is widespread and generally performed by untrained persons. The work of
Planned Parenthood Federation of America (PPFA) in Ethiopia focuses on reducing maternal
death and disability by preventing unintended pregnancies and increasing the availability of
affordable, safe abortion services. Despite the decriminalization of abortion, many medical
providers are not offering services due to discomfort, lack of knowledge about the law, or lack of
skills. According to Planned Parenthood, Ethiopia has among the highest fertility and maternal
death rates in the world. Approximately 1 out of every 7 women die from pregnancy or abortion
related complications. Ministry of Health (MOH, 1996) of Ethiopia provide family planning
services to 8.5 million women across the country during the next years. If the program fully
implemented, the plan will definitely have a positive ripple effect across the Ethiopian society.

Although family planning tools are available in Ethiopia, access to them has been a major
hindrance for the majority of the women. Planned Parenthood states that only 13% of Ethiopian
women - and only 4% in rural areas - use modern contraception. This is despite the fact that, as
studies show, approximately 60% of the women in the country approve family planning.
Undoubtedly, improved access to family planning and other reproductive health services in the

2
country could significantly combat the incidence of maternal mortality and improve the state of
women. Comprehensive, high-quality sexual reproductive services can help to prevent the
unnecessary deaths of women in the country. "Traditionally, women in Ethiopia have been
consigned to strict societal roles, based on cooking, raising children, and a muted voice in
decisions affecting them," says World Bank (2001).

Family planning methods vary according to their convenience, cost, effectiveness, side effects,
risks, and benefits for the individual. Family planning users are best able to evaluate the relative
importance of these factors based on their preferences; their desired family size; stage of life
which is period in the development of progress; goals of delaying, spacing, or limiting future
pregnancies; health status; relationship status; and living conditions. Family planning costs are
often treated as a minor issue in parents' decision making. In the developing world, 137 million
women who do not want to get pregnant are not practicing contraception. The key cause of this
unmet need for contraception is that contraception is often quite costly to individuals in terms of
commodities (pills, condoms, IUDs, etc.), transportation, and provider fees for contraceptives
and health care services, even when subsidies are provided by the government. In addition, there
are significant non-economic costs, such as health concerns, social disapproval and spousal
resistance, as well as unnecessary medical barriers (e.g., requiring a doctor instead of a nurse or
other trained health care workers to provide contraceptives). This unmet need is in turn
responsible for most of the 76 million unplanned pregnancies that occur each year. About half of
these pregnancies end in abortion and the other half end in births; both contribute unnecessarily
to health risks for mothers and children, to the cost of raising families and to the adverse impact
of population growth (Bongaart et al., 2009).

More than 40 years ago, Congress authorized the US Agency for International Development to
start working on family planning and population issues. Over the ensuing decades, while
contraceptive use in the developing world increased from 10 to 60 percent of married couples
and average fertility declined from about six to three children per woman. USAID was the source
of the majority of the money, information and ideas in the field of family planning (Levine.,
2007). High rates of population growth are largely the result of frequent childbearing or high
fertility often corresponding with a large unmet need for family planning (FP). In Ethiopia,

3
women have, on average, about five children and surveys show that the unmet need for family
planning services is high (35 percent of married women of reproductive age want to space or
limit births but are not currently using any method of family planning) . If access to family
planning services was increased, this unmet need could be met, therefore slowing population
growth and reducing the costs of meeting the MDGs (USAID, 2007).

Many countries have introduced family planning programs since the 1960s. The resources
devoted to them and hence their fertility impacts have varied widely. The aggregate effect of all
these efforts has been substantial: Fertility declined in the developing world from more than six
to fewer than four births per woman between 1960–1965 and 1985–1990, and almost half of that
decline 43% (i.e. in the year between 1960-1965 it is at least six births but in between 1985-1990
it is fewer than four births) is attributable to family planning programs. Voluntary family
planning programs are intended to reduce the number of unplanned pregnancies, but they also
legitimize and diffuse the idea of smaller families, thus accelerating the transition to lower
fertility. This is important in countries where women still want large families. In such
populations, there will have to be declines in the number of children desired before sustained
fertility decline can occur. Desired family size is highly responsive to improvements in human
development, in particular in female education and child survival (Bongaart et al., 2009).

The benefits of family planning extend beyond slowing the pace of population growth. By using
contraception, women can avoid the high risk of poorly timed pregnancies that jeopardize their
health and that of their children. For instance, children spaced three to five years apart are more
than twice as likely to survive to age five than are children born within two years of a sibling.
The poorest countries, beset by high maternal and child mortality rates, have the most to gain
from family planning’s health benefits. One estimate suggests that meeting the unmet need for
modern contraception among women in less developed countries could prevent 1.4 million infant
deaths and 142,000 maternal deaths each year (Population Reference Bureau, November 2004).

There is a safe and effective family planning method for every woman that can enable her to
protect her health and that of her children. More than half of all couples in the developing world
are using family planning to delay, space, or limit future pregnancies, yet the need for family

4
planning keeps increasing as the number of women of reproductive age continues to grow. An
estimated 137 million women worldwide have an unmet need for family planning—they are not
using any method and report that they want to avoid a pregnancy (Rhonda et al., 2009).

1.2 Statement of the Problem

Many demographers and scholars believe and recommend the need to conduct in-depth studies
on the various aspects of family planning (FP) practice of couples in different demographic,
economic and socio-cultural settings in developing countries. This study attempts to explore the
major socio-economic, demographic, cultural and environmental factors that affect awareness
and practice of women and men about family planning in Ethiopia.

A major debate in the recent literature is that a major justification for the less than ideal family
planning performance in sub-Saharan Africa is the neglect of men in that endeavor. Almaz
Terefe and Larsen (1993) have shown that modern contraceptive use improved significantly
when husbands were involved in family planning interventions in Ethiopia. In fact, in many
cultures men often have more power than women in decision making with regard to use of
contraceptive and the number of children that the couple will have (McCauley et al., 1994).
Husband’s approval of family planning is an influential factor to women’s contraceptive use
behavior. Many women do not use contraceptive since their husbands disapprove of using
contraception. In seven sub-Saharan countries, contraceptive use among women whose husbands
disapprove of family planning averaged only one third as much as among women whose
husbands approve of it (Robey et al., 1996).

Recently family planning programs and providers are seeing that involving men in addition to
women in family planning results in an improved program effectiveness. The 1994 International
Conference on Population and Development also encouraged family planning programs and
providers to consider both men and women jointly. This new interest in men is in consideration
that although most reproductive health burden is born by the women, the majority of the
decisions that affect both women and men reproductive health are made by men or by men and
women jointly. And also if men are involved they may be a potential partner in and advocates for
good reproductive health rather than bystanders, barriers, or adversaries. According to the

5
guidelines for family planning services in Ethiopia, family planning is a means of promoting the
health of women and families. Accordingly, all individuals male or female who can conceive or
cause conception regardless of age or marital status are eligible for family planning services in
the country. Also this study shares the idea and the main reason behind the need to study the
socio-economic, demographic and cultural determinants of couples to practice family planning
methods.
So far, there are very few studies conducted on the practice of men and women jointly about
family planning methods in Ethiopia particularly on the effects of economic, demographic and
socio-cultural factors. There is still a need to educate and motivate couples and improve family
planning services to achieve more effective and appropriate use of contraceptives and to arrest
the trend towards increase in population.

1.3 Objective of the study

General objective of the study

The general objective of this study is to examine factors that influence practice of various
available family planning methods by married couples (men and women) in Ethiopia. Also, the
study examines the regional differences about status of family planning practice.

Specific objectives of the study

1) To identify socioeconomic, demographic and cultural determinant of couples practice and
awareness about family planning methods based on DHS (2005) for rural, urban and national
level.

2) To select the most important factors that are associated with couples practice of the family
planning methods based on EDHS.

3) To analyze within and between regional level differences by identifying the determinant
factors for the married couples to practice the family planning methods in Ethiopia.

6
 1.4 Significance of the Study
The findings or results obtained from this research could be useful in many ways. Governmental
and non-governmental organizations could take intervention measures and set appropriate plans
to reduce and improve the existing level of awareness and practice of family planning by
identifying and giving priority for the areas which have low and poor practice.

The findings could also be helpful for policy making, monitoring and evaluating the activities for
the government and different concerned agencies. And it helps individuals (women and men) to
have enough knowledge about the practice of family planning. It is hoped that this study could
contribute to the improvement of family planning services in the country through appropriate
service delivery approaches and strategies.

1.5 Limitation of the study

The data used here being secondary may have a number of limitations on the outcome of this
study.
 The study is based on couples status of FP practice at the time of survey.
 Limited literature on Ethiopia related to the subject.
 In the analysis of data using multilevel logistic regression, a student version of
LISREL software is used which is restricted to handle few explanatory variables.
Because of this, only a limited number of predictor variables are used in multilevel
logistic regression analysis.
 The study did not identify the status of family planning practice of women and men
separately.

7
 CHAPTER TWO

2 LITERATURE REVIEW
2.1 Overview of Family Planning
The widespread adoption of family planning represents one of the most dramatic changes of the
20th century. The growing use of contraception around the world has given couples the ability to
choose the number and spacing of their children and has had tremendous life saving benefits. Yet
despite these impressive gains, contraceptive use is still low and the need for contraception is
high in some of the world’s poorest and most populous places (Rhonda et al., 2009).
Family planning is defined as birth spacing, preventing unwanted pregnancies or secure wanted
pregnancy (WHO, 1995). Family planning is adopted voluntarily through the practice of
contraception or other methods of birth control on the basis of knowledge, attitude and
responsible decision by individuals and couples, in order to promote the health welfare of the
family and contribute to the social and economic development of the country.

Comparative studies conducted in DHS-surveyed countries during 1986-1989 have shown that
despite the relatively high fertility rate in Africa, women’s feeling about desired future fertility
indicate that in several countries a decline in fertility could possibly happen. Indeed, in some
African countries (e.g. Kenya, Mali, Togo and Uganda) the projected declines equal or exceed
those in lower fertility areas in which family planning services and information are already much
better established, such as in North Africa and Latin America (Sinding, 1994).

Shane (1997) and Bandarage (1997) have pointed out that the majority of people in developing
countries are reluctant to use modern contraceptives even when they have a widespread
knowledge to the benefits obtainable from family planning due to many social, economic, and
cultural reasons. This indicates that couples in developing countries are unable to regulate
fertility because of various social, economic, cultural, psychological and demographic reasons.
Thus, at present development programs that focus on women empowerment as a poverty
reduction strategy are accepted as a quick fix solution of population growth and economic
development United Nations Fund for Population Activities (UNFPA, 1999).

8
In most countries surveyed by the Demographic and Health Surveys (DHS), between 20% and
30% of all married women of reproductive age have an unmet need for family planning. Outside
sub-Saharan Africa, most women with an unmet need do not want to have any more children. In
sub-Saharan countries surveyed, most unmet need is for spacing births. Most couples want to
plan their families, when they have the opportunity to do so. Women everywhere want to avoid
unintended pregnancies. Young and unmarried people want to avoid the consequences of
unprotected sexual relations, both pregnancy and sexually transmitted diseases (Rhonda et al.,
2009).

2.1.1 Situation of family planning in the world

The era of modern contraception began in the 1960s, when both the birth control pill and
intrauterine contraceptive device (IUCD) became available. These effective and convenient
contraceptive methods resulted in widespread changes in birth, fertility and demography in the
United States. Since 1972, the average family size of United States had leveled off at
approximately two children, with increasing safety, efficacy, diversity, accessibility and use of
contraceptive methods (Forrest, 1994).
China established a one-child policy in 1979 in order to keep the total population at fewer than
1.2 billion by 2000. The Chinese population in mid-2000 is estimated at 1.265 billion
(Population Reference Bureau, 2000). Despite an official national policy stating that family
planning program is voluntary, not compulsory, and that people are persuaded but not forced to
practice birth control, a system of mandatory fertility control was instituted in China. Couples
had little choice about whether they practiced birth control, how many children they would have,
and the timing of births. Family planning contracts that include a one-child pledge were required
of couples. Births were either approved as part of the local government’s birth plan or they were
forbidden and outside the official birth plan. A practice of requiring use of specific contraceptive
methods was also instituted in which women with one child had to use an IUD, couples with two
children had to have one partner sterilized, and those with unauthorized pregnancies had to
undergo abortion. Thus the primary methods of fertility regulation used to implement the
program were sterilization, IUD, and abortion, and women or couples had little choice of which
method to use.

9
Throughout the 1980s, the Chinese program was characterized by regulations, and guidelines
were established at the national level and then “prescribed to provincial and lower levels of
government and ultimately to couples of childbearing age” (Hardee-Cleaveland and Banister,
1988). Family planning was rarely considered a personal matter. China is the prime example of a
program driven to achieve collective welfare by reducing population growth, but which
sacrificed individual rights and welfare through an essentially involuntary program. While the
one-child policy was not uniformly implemented in all areas of China—particular exceptions
pertain to rural areas and ethnic minorities—it was nevertheless the driving force of the family
planning program (Hardee-Cleaveland and Banister, 1988).

Between 1990 and 1994, the global average contraceptive use by married women of reproductive
age, that is ages from 15 to 49, rose from 57% to 60% (Hamilton, 1997). The introduction of
combined oral contraceptive pill also brought about the sexual revolution in the West, where it
was possible for sex without any fear of pregnancy. Worldwide, however, there are still unmet
needs especially in developing countries, where a scarcity of resources and information, cultural
and political barriers, and societal attitudes or misconceptions, conspire to exact a heavy toll on
all women health, with unwanted pregnancies, unsafe abortions, maternal mortality and HIV
infection still leading causes of death of women. Even in developed countries, the situation is far
from ideal and policies and provision of services vary considerably within each country.
Unwanted side effects, inconvenience of the chosen method, and media scares about safety of
modern contraceptives are some of the issues that limit their acceptability. Poor contraceptive
use is further compounded by ignorance among users and providers of a wide range of methods
available now and likely to be so in the future. Giving women reproductive autonomy through
comprehensive and up-todate information about all methods is vital for successful and long-term
use of contraception (Kubba et al., 2000).

2.1.2 The situation of family planning in developing countries

The history of family planning programs in the developing countries partly originates with
concern about a “world population problem”. In the late 1940s and 1950s, the phenomenon of
rapid population growth, resulting from the gap between declining mortality and continuing high

10
fertility, was emerging in some South and East Asian countries. The results from postwar
censuses of the early 1950s provided the initial evidence that population growth could be an
impending problem. By the mid-1960s, more countries, including a number in Latin America
and the Middle East, were experiencing unprecedented rates of population growth of more than 3
percent annually. The implementation of voluntary family planning programs has been the
principal policy initiative pursued by governments in the developing world that wish to reduce
population growth. The major justification for this programmatic approach is an unfulfilled
demand for family planning that presumably exists in many populations (Bongaarts, 1990).

The effectiveness of family planning was premised on the assumption that couples in developing
countries wanted fewer children, and that to achieve smaller families, they would practice
contraception. The evidence supporting this assumption has come from surveys that have been
carried out since the 1960s. The surveys of women and couples provide information about their
attitudes regarding desired family size, whether they wanted to have additional children and if so
when, and contraception. Over more than 35 years, these surveys have shown that a large
proportion of couples had favorable attitudes toward contraception and that many couples
wanted no more children. The surveys also showed that among those not wanting no more
children, many were not practicing family planning methods. The discrepancy between women’s
stated preferences for children (both the number and the timing) and their actual contraceptive
use is referred to as the “Unmet need” for family planning. Reducing or filling this unmet need
has been an objective of family planning programs for many years (Seltzer, 1998).

In Ethiopia family planning was initiated in 1960’s. However even after such a long period of
time the level of family planning in Ethiopia is amongst the lowest in Africa with contraceptive
prevalence rate (CPR) 15% and unmet need for family planning is very high 36% (CSA, 2005).
Several factors had been incriminated for the low coverage of family planning services. The
reasons include desire to have more children, lack of knowledge about contraceptive use and
where to find contraceptives, health concerns, religious prohibition, husband opposition and low
involvement of men (Tularo et al., 2006).

11
 2.1.2.1 Socio-economic factors
The factors associated with family planning practice can be divided into personal, demographic,
socio-cultural, religious, economic, and health services. Among the personal factors associated
with family planning practice are knowledge of family planning methods and influence of family
members and friends, especially those who have experience in family planning methods.
Demographic factors such as parity, age, marital status, religion of a woman, husband’s
education, husband’s occupation, monthly family income, and woman’s occupation are also
known to be associated with family planning practice.
Among the socio-economic factors that may affect the practice of family planning methods by
married couples are place of residence, work status/occupation, education level of women,
partner’s education level, visited by FP worker in the last 12 months are considered to be
important (Daniel, 1995).

High fertility and rapid population growth have an impact on the overall socio-economic
development of a country in general and maternal and child health in particular. Maternal and
child mortality are two of the major health problems challenging healthcare organizations,
especially in developing countries. The majority of maternal deaths are the direct result of
complications encountered during pregnancy and arising from unsafe terminations (Gaym 2000;
Merric 2002; Population Reports 1999). The World Health Report (WHR 2005) noted that
unwanted, mistimed and unintended pregnancy is the most common cause of maternal mortality
in developing countries. The Ethiopian demographic and Health survey (DHS 2000) identified
that one in four deaths among Ethiopian women in the period 1994–2000 was due to a pregnancy
or pregnancy-related cause. One of the reasons for this is lack of skilled healthcare personnel
attending births. The WHR (2005) reported that in 2000 only 6% of births in Ethiopia were
attended by an appropriately skilled person. However, there is also evidence to suggest that up to
100,000 maternal deaths could be avoided each year if women who did not want children used
effective contraception (Marston and Cleland 2003). Children with many siblings are less likely
to enter and remain in school and more likely to have poor health and high mortality rates as
compared with children who have fewer siblings (Asefa et al. 2000; Merrick 2002; Population
Reports 1999). This problem in the developing world is disproportionately higher than in the
developed world. For example, in Ethiopia, mortality rates under the age of 5 years for the 1996–

12
2000 period were 166 per 1000 live births (WHR 2005). The issue of family planning clearly has
implications for the planning and development of educational and healthcare policies and
strategies. Pregnancy, poor health and nutritional status, communicable diseases, high workload,
early marriage, high fertility, inadequate access to and underutilization of health services, and the
low status of women in the society are among the many underlying causes of maternal mortality
(MOH, 1996).

Birth control is a priority in Ethiopia and many programmes to increase contraceptive practice
have been implemented by the Ethiopian National Population Policy since 1993. However, the
problem of high fertility and low contraceptive practice remains unresolved. The total fertility
rate of the country for the year 2000 was found to be 5.9 children per woman and the
contraceptive prevalence was only 8%. This is too low to affect the fertility levels significantly
(DHS 2000; Fitaw et al. 2003), and the fertility rate of the country actually increased to 6.1
children per woman for the year 2003 (WHR 2005). High population growth prevents the long-
term socio-economic development needed to alleviate poverty and to meet the immediate basic
needs of the burgeoning population (Bandura 2002; Merrick 2002). Accurate and specific data
about the reason behind the low contraceptive practice in a country should be available in order
to develop an effective and relevant family planning strategy.

In most developing countries, especially in Sub-Saharan Africa, promoting family planning

through radio and television is an important means of raising awareness, improving knowledge,
and motivating use of modern contraceptive methods (ETR, 2007). In most countries regular
exposure to mass media has a positive effect on use of contraceptive. In relation to Ethiopia, a
study based on DHS (2005) indicated that family planning information is largely disseminated
through radio with limited dissemination through television or print media. According to EDHS
(2005) 29 percent of women heard about family planning on radio compared with only 11
percent of women who get family planning information from television and 8 percent who got
such information from newspapers or magazines.

13
 2.1.2.2 Demographic factors
Various studies identified different demographic variables as influential in the practice of family
planning methods by couples in Ethiopia. These variables among others include age, number of
living children and lack of exposure to the risk of pregnancy. The social cognitive theory for
social change was used as a theoretical framework because it helps to understand and explain
how social and cultural norms, values, people’s beliefs, behaviors, gender roles and social
networks influence people’s choices about family planning (Bandura, 2002).

According to Bandura (2002), one of the important components of social cognitive theory is
efficacy belief. It plays an important role in the adoption of change and regulates human
functioning through cognitive, motivational, affective and decisional processes. Bandura (2002)
states that among the mechanisms of self influence for change, none is more central or pervasive
than beliefs in one’s efficacy to exercise control over one’s functioning and events that affect
one’s life. This core belief system is the foundation of human motivation and accomplishments.
Unless people believe they can produce desired effects by their actions they have little incentive
to act or to preserve in the face of difficulties whatever other factors serve as guides and
motivators, they are rooted in the core belief that one has the power to effect changes by one’s
actions. However, in many spheres of life, people do not live their lives independently. They
depend on those who have power and resources, and they act according to the desires of the
powerful (Bandura 2002; Hogan et al. 1999; Nagase, 2003).

Furthermore, low socio-economic, occupational and educational status, and cultural and
community norms affect behavior through their influence on people’s sense of control over their
own lives (Bandura 2002; Hogan et al. 1999). This is particularly true in the use of
contraception, which is subject to strong social and normative constraints. Therefore, assessment
of the various socio-economic, demographic and cultural variables that contribute to low
contraceptive practice is essential for promoting the use of contraception and lowering the birth
rate.

14
According to Sayed et.al (1986), the need for family planning is positively related with the
number of surviving children. The percent in need increases with the number of living children.
For instance in Egypt, out of those women with no living children, only 3 percent were in need
of family planning while 35 percent of the women with seven or more living children were in
need . In the case of Mali the same result was observed. The intention among non-users to start
contraception soon rose from 7.7 percent for those with no children to 14.7 percent for those with
four surviving children (Bongaarts, 1990). Similar results were recorded by Sathara and
Chictambara (1984). For African and Asian countries, some studies have found relatively highest
contraceptive use mostly among women with four or more children (Sathara and Chictambara,
1984). With regard to Ethiopia, a study based on 2005 EDHS indicated contraceptive use is
highest among currently married women with one or two children (17%) and lowest among
women with no children (12%).

15
 CHAPTER THREE
3 DATA AND METHODOLOGY
3.1 Data source
The source of data is the 2005 Ethiopia Demographic and Health Survey (EDHS) which is
obtained from Central Statistical Agency (CSA). It is the second comprehensive survey
designed to provide estimates for the health and demographic variables of interest for the
following domains: Ethiopia as a whole, urban and rural areas of Ethiopia (each as a separate
domain), and all geographic areas (nine regions and two city administrations), namely: Tigray,
Affar, Amhara, Oromiya, Somali, Benishangul-Gumuz, Southern Nations, Nationalities and
Peoples (SNNP), Gambela and Harari regional states and two city administrations, that is, Addis
Ababa and Dire Dawa. In the 2005 EDHS a representative sample of approximately 14,500
households from 540 clusters was selected. The sample was selected in two stages. In the first
stage, 540 clusters (145 urban and 395 rural) were selected from the list of Enumeration Areas
(EA) from the 1994 Population and Housing Census sample frame. In the second stage, a
complete listing of households was carried out in each selected cluster. Between 27 and 32
households from each cluster were then systematically selected for participation in the survey.

All women age 15-49 who were either permanent residents of the households in the 2005 EDHS
sample or visitors present in the household on the night before the survey were eligible to be
interviewed. In addition, in a sub-sample of half of all the households selected for the survey, all
men age 15-59 were eligible to be interviewed if they were either permanent residents or visitors
present in the household on the night before the survey.
From among the 14,500 households, 14,717 women were identified as eligible for the individual
interview. Interviews were completed with 14,070 women, yielding a response rate of 96
percent. Of the 6,778 eligible men identified in the selected sub-sample of households, 89
percent were successfully interviewed. Response rates were higher in rural than in urban areas,
with the rural-urban difference in response rates most marked among eligible men. Thus, the
analysis presented in this study on couples (women and men) about practice of family planning
method is based on the 25,240 couples in Ethiopia.

16
 3.2 Variables of the Study
3.2.1 The Response Variable
The response variable of this study is status of couples about family planning methods practice.
For the analysis purpose practice of a couple about FP method covers modern, traditional and
folkloric methods of contraception.

Among the methods, ‘modern method’ includes methods like Pill, IUD, Injectables, condom,
LAM etc, ‘traditional method’ includes periodic abstinence, withdrawal etc and ‘folkloric
method’ includes use of herbs etc. For our study purpose the response variable ‘status of a couple
about FP practice’ is recoded as follows: those couple who are currently practicing any of the
above three methods are coded as 1 and those who do not practice any method are coded as 0.
The response variable for the ith couple is represented by a random variable Yi with two possible
values coded by 1 and 0. In view of this, the response variable of the ith couple Yi was measured
as a dichotomous variable.

1, if the i th couple practice any FP method

Yi  
0, otherwise
3.2.2 Explanatory Variables/factors
The explanatory variables considered for this analysis include current age of women, number of
living children, education level of women, religion of women, place of residence, region,
occupation of women, occupation of male partner/husband, partner’s education level, current
martial status, economic status of household, knowledge of FP method, visited by FP worker
during the last 12 months, husband approves FP, discussed FP with partner and exposure to any
mass media.

17
Table 3.1: Description of the variables and coding
The description of socioeconomic, demographic and other related variables with the practice of
couples about family planning methods are presented below.
The Response variable
Variable Representation Factor categories
of variable

Status of a couple about FP Y 0= Do not practice a FP method

method practice (Practice)
1= Practice a FP method

Explanatory variables
No. Factors/ variables Categories

1 Age of a woman (AGEW) 1=15-19

2=20-34
3=35-49
2 Place of Residence (RESIDE) 1= Urban
2= Rural
3 Region (REGION) 1=Tigray
2=Affar
3=Amahra
4=Oromiya
5=Somali
6=Ben-Gumuz
7= SNNP
8=Harari
9=Gambela
10=Dire Dawa
11= Addis Ababa
4 Occupation of a woman (OCCPW) 0=Not working
1=Non-Agriculture
2=Agric-employee

5 Religion group of a woman 1=Orthodox

(RELIGION) 2=Protestant
3=Muslim
4=Others

18
 6 Current martial status (CMS) 1=Married
2=Living together

7 Women’s education level (EDUW) 0=No education

1=Primary
2=Secondary and higher
8 Occupation of a man (OCCPM) 0=Non-agriculture
1=Agric-employee

9 Exposure to any mass media (MEDIA) 0=No

1= Yes

10 Partner’s education level (PartEdu) 0=No education

1=Primary
2=Secondary and higher
11 Economic status of household 0=Poor
(WEALTH) 1=Medium and higher

12 Knowledge of a couple on a FP method 0= Knows no FP method

(KnowMethd) 1= Knows a FP method
13 Number of living Children (Numchild) 0= No children
1= 1 or 2 children
2= Three or more
14 Discuss FP with partner (DiscPart) 0=Never
1=Once or more than
15 Visited by FP worker during the last 12 0= No
months (VisitedFP) 1= Yes
16 Male partner/Husband approves FP 0=Disapproves
(HusApprov) 1=Approves
2=Don’t know

3.3 The Methodology

3.3.1 The Logistic Regression Analysis

Logistic regression is a popular modeling approach when the dependent variable is dichotomous
or polytomous as well. This model allows one to predict outcomes, from a set of variables that
may be continuous, discrete, dichotomous, or a mix of any of these. Hosmer and Lemeshow
(2000) has described logistic regression focusing on its theoretical and applied aspect.

Logistic model, as compared to its competitor, the probit model, is less sensitive to outliers and
easy to correct a bias (Copas, 1988). In instances where the independent variables are categorical

19
or a mix of continuous and categorical, logistic analysis is preferred to discriminant analysis
(Agresti, 1996). The assumptions required for statistical tests in logistic regression are far less
restrictive than those for ordinary least squares regression. There is no formal requirement for
multivariate normality, homoscedasticity, or linearity of the independent variables within each
category of the response variable. However, the assumptions that apply to logistic regression
model include: meaningful coding, inclusion of all relevant and exclusion of all irrelevant
variables in the regression model, low error in the explanatory variables, no outliers and
sampling adequacy.

In the terminology of logistic regression analysis the odds of a success is defined to be the ratio
of the probability of a success to the probability of a failure. Hence if p is the true success
probability the odd of a success is p/ (1-p).

Let Y be a dichotomous outcome random vector with categories 1 (a couple practices a FP

method) and 0 (a couple does not practice a FP method). Let X be an n x (p+1) matrix denotes
the collection of P-predictor variables of Y, i.e.

1 x11 .... x1 p 
 
1 x21 ... x2 p 
X  . ........... . 
 
. ............ . 
1 x n1 .... xnp 
 

X without the leading column of 1s, is termed as predictor data matrix. Then, the conditional
probability that the ith individual couple practice any of family planning methods given the
individual characteristics Xi is given by:

pi  P Yi  1 | xi 

In logistic regression analysis, it is assumed that the explanatory variables affect the response
through a suitable transformation of the probability of the success. This transformation is a
suitable link function of pi, and is called the logit-link, which is defined as:

20
  p 
logit(pi) = log i  (3.1)
 1  pi 

The transformed variable logit (pi) is related to the explanatory variables as:

logit (pi) = β0 + β1X1i + β2X2i + …+ βpXpi = Xi/β. (3.2)

where β = (β0, β1, β2, …, βp)/ are the model parameters and Xi =( x0i,x1i,…xpi)/

with x0i =1,i=1,…,n

The probability of success can be expressed as

/
i
eX
pi = P (Yi =1x1i, x2i, …, xpi) = /
(3.3)
1  e X i

With further rearrangement we obtain the odds of success

pi /
Odds (Yi = 1) = =eX i  (3.4)
1  pi

The above three equations give suitable representations of log-odds, the success probability,
and odds, respectively. Indeed, these representations facilitate interpretations of parameter
estimates. The estimated logistic coefficients ˆ j 's are interpreted as the change in the log-

odds for every unit increase/decrease (depending on the variable change in xi) holding other
predictors constant. Hence, the odds of being in the category of interest for the ith subject,
pˆ i / ˆ
namely = e X i  = e j represents the multiplicative factor by which the odds change
1  pˆ i 
for every unit change in xi controlling for the other predictor variables, where Xi = (0, 0,…,1,
0,….,0) with the 1 in the jth position. For predictor variables having L levels (L ≥ 2),
interpretation can be made by making one of the L-levels as a reference category.

Regression methods have an integral component of any data analysis concerned with
describing the relationship between a response variable and one or more explanatory variables.
The estimated coefficients tell us the increased or decreased chance of status of couples
practice about FP given a set of level of the determinant factors while controlling for the

21
effects of other variables in the model. When the outcome variable is binary the logistic
regression model can be used mainly for two reasons. The first is from a mathematical point of
view, it is extremely flexible and easily used function. The second is, it leads itself to
meaningful interpretation.
Multiple and multilevel logistic regression analysis is a methodology for analysis of data with
complex patterns of variability, with a focus on nested sources of variablities on multiple
categories. For this study the variation in FP practice by couples within regions means that not
only unexplained variation between couples but also unexplained variation between regions.
This can be expressed by statistical models with random coefficients.

3.3.1.1 Test of Goodness of Fit

Once a model has been developed, we would like to know how effective the model is in
describing the outcome variable. This is referred to as goodness-of-fit. In testing the hypothesis
that the model fits the data, the two common approaches are Pearson’s X2 statistic and the
likelihood-ratio statistic G2; see details in Agresti (1996).
The Hosmer-Lemeshow test is another alternative to check model fit. In this approach, data are
divided into 10 groups. From each group, the observed and expected number of events will be
computed. Then, the Hosmer-Lemeshow test statistic is given by
g

 (O k  E k) 2
Cˆ  k 1
, (3.5)
Vk

where E k  nPk , Vk  nPk (1  Pk ) , g is the number of group, Ok is observed number of

events in the kth group, E k is expected number of events in the kth group, and Vk is a variance

correction factor for the kth group. If the observed number of events differs from what is
expected by the model, the statistic Ĉ will be large and there will be evidence against the null
hypothesis that the model is adequate to fit the data. This statistic has an approximate chi-
square distribution with (g-2) degrees of freedom.
If the calculated value of the Hosmer-Lemeshow goodness-of-fit test statistic is greater than
0.05, we will not reject the null hypothesis that there is no difference between observed and
model-predicted values, implying that the model estimates are adequate to fit the data at an
acceptable level.

22
The Wald statistic is also an alternative test which is commonly used to test the significance of
the individual logistic regression coefficients for each independent variable (that is, to test the
null hypothesis in logistic regression that a particular logit (effect) coefficient is zero i.e. H0:
 i  0 against  i  0 ). For a dichotomous independent variable, the Wald statistic (W) is

ˆi
2
W= (3.6)
var(ˆi )
For large sample size this statistic has an approximate chi-square distribution with one degree
of freedom.

3.3.2 The Multiple Logistic Regression Analysis

Multiple logistic regression analysis is used to study the effects of each independent variable
controlling other factors on the response variable, which is status of a couple to practice family
planning.
An important aspect in multiple levels of analysis is the partitioning of unexplained variability
over the various levels. The logistic regression can have an arbitrary number of parameters and
terms in the model representing qualitative variables, quantitative variables and interaction
terms in order to model a dichotomous or categorical outcome variable. When explanatory
variables are included to model probabilities, a problem is that probabilities are restricted to the
domain 0 and 1whereas a linear effect for an explanatory variable could take the fitted value
outside this interval.
Three different models are fitted in this study to see the basic determinant factors for couples
about practice of family planning methods in Ethiopia.

The first model is fitted to identify the socio-economic and demographic factors for couples to
practice FP in urban Ethiopia. The second model demonstrates the socio-economic and
demographic factors related to the practice of FP among couples in rural Ethiopia. Similarly,
the third model identifies the factors which contribute for couples to practice FP practice at
national level both in urban and rural Ethiopia.

Consider a collection of k independent variables which will be noted by the vector

23
X= ( x1 , x2 ,..., x k ) . Let the conditional probability that the outcome is present be denoted by

P(Y=1| X) = p(X). Then the logit of the multiple logistic regression is given by the equation
g(X) = β0 + β x + β x + …+ β x
1 1 2 2 k k

And the odds in favor of success for the multiple logistic regression will be

 p 
log it    log e  
g ( x)
= β + β x + β x + …+ β x (3.7)
1  p  0 1 1 2 2 k k

in which case

g (x)  0  1 x1   2 x2 ...  k xk

p(x) = e g (x)
= e  0  1x1   2 x2 ...  k xk
, (3.8)
1 e 1 e
where a model parameters  i will be interpreted as the change in the log-odds for a one unit

increase in xi, holding all the other predictor’s constant, or after adjusting for the other
predictors. In order to estimate the parameters, maximum likelihood-based iteration algorithms
can be employed. Inspite of the attractive properties of the logit function, it is by no means the
only suitable function for transforming probabilities to arbitrary real values. The general term
for such a transformation function is the link function, as it links the probabilities (or more
generally, the expected values of the dependent variable) to the explanatory variables.

3.3.3 Multilevel Logistic Regression Model

3.3.3.1 Two-Level Model
Basically, the two-level logistic model is equivalent to model (3.2) except for the notation in
the outcome variable. Let yij be the binary outcome variable, coded ‘0’ or ‘1’, associated with
level-one unit i nested within level two unit j. Also let pij be the probability that the response
variable equals 1, and pij = Pr(yij = 1). Here, yij follows a Bernoulli distribution. Like the
logistic regression the pij is modeled using the link function, logit. The two-level logistic
regression model can be written as,
 p 
log ij  =  0 +  1xij + u0j (3.9)
 1  pij 
 

24
where uj is the random effect at level 2. Without uj, Equation (3.9) can be considered as a
standard logistic regression model. Therefore, conditional on uj, the yij’s can be assumed to be
2
independently distributed. Here, u0j is a random quantity and follows N(0,  u ).

The model (3.9) can be written as follows splitting up into two models: one for level 1 and the
other for level 2.
 pij 
log it ( Pij ) = log  =  0j +  1xij [model: level 1]
 1 p 
 ij 
and
 0j =  0 + u0j [model: level 2]
The multilevel logistic regression model cannot be derived in the way the simple logistic
regression model is derived.
A multilevel logistic regression model as a hierarchical model, can account for lack of
independence across levels of nested data (i.e., individuals nested within groups). Conventional
logistic regression assumes that all experimental units are independent in the sense that, for
example, any variable affecting dropout from contraceptive use, has the same effect in all
groups. Multilevel modeling relaxes this assumption and allows the effects of these variables to
vary across groups.

Consider a population having two-levels. The basic data structure of two-level logistic
regression is a collection of N groups (units at level-two (regions)) and within group j (j=1,
2,…, N) a random sample of nj level-one (couples) units. The outcome variable is dichotomous
and denoted by Yij (i=1,2,…,nj, j=1,2,…,N) for level-one unit i in group j. The outcomes are
supposed to be coded 0 and 1: 0 for ‘failure’, 1 for ‘success’ or vice versa. The total sample
N
size is M=  n j . If one does not take explanatory variables into account, the probability of
j 1

success is assumed constant in each group. Let the success probability in group j be denoted
by pj. The dichotomous outcome variable for the individual i in group j, Yij; which is either 0
or 1 can be expressed as the sum of the probability in group j, pj (the average proportion of j
levels in group j, E (yij) = pj) plus some individual-dependent residual  ij , that is, Yij = pj+  ij

the residual term is assumed to have mean zero and variance, Var (  ij ) = p j (1  p j ).

25
Since the outcome variable is coded 0 and 1, the group sample average is the proportion of
successes in group j given by

nj
1
pˆ j 
nj
Y .
i 1
ij (3.10)

p̂ j is an estimate for the group-dependent probability p j . Similarly, the overall sample average

is the overall proportion of successes, p̂. , and is given by

N nj
1
Pˆ .   Y . ij (3.11)
M j 1 i 1

This is an estimate for the overall probability of success, p. An estimator of the variance of p j

can be obtained by using

S 2 within
ˆ 2  S 2 between  , (3.12)
n~

where

N
 2

1   nj 

n~ 
j 1
M  
N 1 M 
 

Pˆ .(1  Pˆ .) ˆ
S 2 between  ~ C , where Ĉ is as given by equation (3.5), and
n ( N  1)

1
S 2 within   n P (1  P ) .
j j j
M N

3.3.3.2 Heterogeneous Proportion

For the proper application of multilevel analysis the first logical step is to test heterogeneity of
proportions between groups. Here we present two commonly used test statistics that are used to
check for heterogeneity. To test whether there are indeed systematic differences between the

26
groups, the well-known chi-square test for contingency table can be used. In this case the chi-
square test statistic is

2
N( Pˆ j  Pˆ .) 2
X  nj . (3.13)
j 1 Pˆ .(1  Pˆ .)

This statistic follows approximately chi-square distribution with N-1 degrees of freedom. The
approximation is valid if the expected numbers of success and of failures in each group, n j Pˆ j

and n j (1  Pˆ j ) , respectively, are at least 1 while 80 percent of them are at least 5 (Agresti,

1990). This condition will not always be satisfied, and the chi-square test then may be seriously
in error. For a large number of groups the null distribution of the test statistic of the chi-square
can be approximated by a normal distribution with the correct mean and variance (McCullagh
and Nelder, 1989).

A second test of heterogeneity of proportions was proposed by Commenges and Jacqmin

(1994). The proposed test statistic is

 n 
(n j Pˆ j  Pˆ .) 2  MPˆ .(1  Pˆ .)
2
j
j 1
Z . (3.14)
N
Pˆ .(1  Pˆ .) 2 n j (n j  1)
j 1

The statistic, Z, follows the standard normal distribution for large value of M. Thus, large
calculated values of this statistic are indication of heterogeneous proportions.

2
In the statistic Z the numerator contains a weight of n j where chi-square test uses a weight n j .

This shows that the two tests combine the groups in different ways. Hence, when the group
sizes n j are different, it is possible that the two test lead to different outcomes. The test statistic

Z is shown to have high power over the chi-square test and it can be applied whenever there
are many groups, even with small group sizes, provided that no single group dominates
(Snijders and Bosker, 1999).

27
Multilevel logistic regression can be employed in the simplest case without explanatory
variables, (usually called empty model) and also with explanatory variables by allowing only
the intercept term or both the intercept and the slopes (regression coefficients) to vary
randomly. Mainly the normal (multivariate normal) distribution is assumed for the varying
coefficients. To keep the discussion on multilevel logistic regression models simple and taking
into account the data to be analyzed in this study, we concentrate on the case of two-levels. The
extensions to the case of three or higher levels is straight forward (Snijders and Bosker, 1999).

3.3.3.3 The Empty Logistic Regression Model: The empty two-level model
for a dichotomous outcome variable refers to a population of groups (level-two units (regions))
and specifies the probability distribution for group-dependent probabilities pj in Yij = pj+  ij

without taking further explanatory variables into account. We focus on the model that specifies
the transformed probabilities f ( p j ) to have a normal distribution. This is expressed, for a

general link function f ( p ) , by the formula

f ( p j ) =  0 U0 j , (3.15)

where  0 is the population average of the transformed probabilities and U0j the random
deviation from this average for group j. If f ( p ) is the log it function, then f ( p j ) is just the

log-odds for group j. Thus, for the log it link function, the log-odds have a normal distribution
in the population of groups, which is expressed by

log it ( p j ) =  0  U 0 j , (3.16)

For the deviations U 0 j it is assumed that they are independent random variables with a normal
2
distribution with mean zero and variance  0 .

This model does not include a separate parameter for the level-one variance. This is because
the level-one residual variance of the dichotomous outcome variable follows directly from the
success probability, as indicated by equation Var (  ij ) = p j (1  p j ).

Denote by  0 the probability corresponding to the average value  0 , as defined by

28
 f ( 0 )   0

For the log it function, the so-called logistic transformation of  0 , is defined by

exp(  0 )
 0  log istic (  0 )  . (3.17)
1  exp(  0 )

Note that due to the non-linear nature of the log it link function, there is no a simple relation
between the variance of the deviations U 0 j . However, there is an approximate formula which

is valid when the variances are small and is given by

2
var( p j ) =(  0 (1-  0 ))2  0 . (3.18)

Note that an estimate of population variance var( p j ) can be obtained by replacing sample
2
estimates of  0 and  0 . The resulting approximation can be compared with the

nonparametric estimate, ˆ 2 , given in equation (3.12).

3.3.3.4 The Random Intercept Logistic Regression Model: In the

random intercept logistic regression model the intercept is the only random effect meaning that
the groups differ with respect to the average value of the response variable. But the relation
between explanatory and response variables can differ between groups in more ways.

The random intercept model expresses the log-odds, i.e. the log it of pij , as a sum of a linear

function of the explanatory variables. That is,

 p 
log it ( Pij ) = log  ij  =  0 j   1 x1ij   2 j x2 ij  ...   k x kij
1  pij 

k
  0 j    h xhij , (3.19)
h 1

where the intercept term  0 j is assumed to vary randomly and is given by the sum of an

average intercept  0 and group-dependent deviations U 0 j , that is

29
 0 j  0 U 0 j

As a result

k
log it ( pij )   0    h xhij  U 0 j , (3.20)
h 1

Solving for pij

k
0    h X hij U 0 j
pij  e h 1
k . (3.21)
1 e
 0   h X hij U 0 j
h 1

Thus, a unit difference between the Xh values of two individuals in the same group is
associated with a difference of  h in their log-odds, or equivalently, a ratio of exp(  h ) in their

odds. Equation (3.20) does not include a level-one residual because it is an equation for the
probability pij rather than for the outcome Yij . The level-one is already included in (3.19).

Note that the first part of the right-hand side of (3.20), incorporating the regression
k
coefficients,  0    h xhij is the fixed part of the model, because the coefficients are fixed. The
h 1

remaining part, U 0 j , is called the random part of the model. It is assumed that the residual, U 0 j ,
2
are mutually independent and normally distributed with mean zero and variance  0 .

3.3.3.5 The Random Coefficient Logistic Regression Model: In logistic

regression analysis, linear models are constructed for the log-odds. The multilevel analogue,
random coefficient logistic regression, is based on linear models for the log-odds that include
random effects for the groups or other higher level units.

Consider explanatory variables which are potential explanations for the observed outcomes.
Denote these variables by X1, X2,…, Xk. The values of Xh (h=1, 2,…, k) are indicated in the
usual way by Xhij. Since some or all of these variables could be level-one variables, the success
probability is not necessarily the same for all individuals in a given group. Therefore, the
success probability depends on the individual as well as the group, and is denoted by P ij.

30
Now consider a model with group-specific regressions of log it of the success probability,
log it ( Pij ), on a single level-one explanatory variable X,

 pij 
log it ( Pij )  log     oj  1 j x1ij , (3.22)
1  pij 

The intercepts  0 j as well as the regression coefficients, or slopes, 1 j are group-dependent.

These group-dependent coefficients can be split into an average coefficient and the group-
dependent deviation:

0 j  0 U 0 j
1 j  1  U 1 j

Substitution into (3.20) leads to the model

 Pij 
log it ( Pij )  log    (  0  U 0 j )  ( 1  U 1 j ) x1ij
1  Pij 

  0  1 x1ij  U 0 j  U 1 j x1ij , (3.23)

There are two random group effects, the random intercept U 0 j and the random slope U 1 j . It is

assumed that the level-two residuals U 0 j and U 1 j have means zero given the value of the

explanatory variable X. Thus 1 is the average regression coefficient like  0 is the average

intercept. The first part of equation (3.23),  0  1 x1ij , is called the fixed part of the model and

the second part, U 0 j  U 1 j x1ij , is called the random part.

The term U 1 j x1ij can be regarded as a random interaction between group and X. This model

implies that the groups are characterized by two random effects: their intercept and their slope.
These two group effects U 0 j and U 1 j will not be independent, but correlated. Further, it is

assumed that, for different groups, the pairs of random effects ( U 0 j , U 1 j ) are independent and

31
identically distributed. Thus, the variances and covariance of the level-two random effects
( U 0 j , U 1 j ) are denoted as follows:

2
var(U 0 j )   00   0
2
var(U 1 j )   11   1
var(U 0 j ,U 1 j )   01

The model for a single explanatory variable discussed above can be extended by including
more variables that have random effects. Suppose that there are k level-one explanatory
variables X1, X2,…, Xk , and consider the model where all X-variables have varying slopes and
random intercept. That is

 pij 
log it ( Pij ) = log   =  0 j  1 j x1ij   2 j x2ij  ...   kj x kij . (3.24)
1  pij 

Letting  0 j   0  U 0 j and  hj   h  U hj , h = 1, 2,…, k.

We have from (3.24)

 pij  k k
log it ( Pij ) = log   =  0    x
h hij  U 0j   U hj xhij , (3.25)
1  pij  h 1 h 1

k
The first part of equation (3.25),  0    h xhij , is called the fixed part of the model. The
h 1

k
second part, U 0 j  U hj xhij , is called the random part.
h 1

3.3.3.6 Estimation and Testing Technique

Parameter estimation for multilevel logistic model is not straightforward like the methods for
logistic regression. The most common methods for estimating multilevel logistic models are
based on likelihood. Among the methods, Marginal Quasi Likelihood or MQL [Goldstein
(1991), Goldstein and Rasbash (1996)] and Penalized Quasi Likelihood or PQL [Laird (1978)
and Breslow and Clayton (1993)] are the two prevailing approximation procedures. Both MQL

32
and PQL are based on Taylor series expansion to achieve the approximation. Based on the
usage of first and second term of Taylor expansion, MQL and PQL are often known as first-
order MQL and second-order MQL, first-order PQL and second-order MQL respectively. After
applying these quasi likelihood methods, the model is then estimated using iterative
generalized least squares (IGLS) or reweighted IGLS (RIGLS) [Goldstein (2003)].
Besides, there are other estimation methods: Maximum Likelihood Method (several simulation
based; McCulloch (1997)), Bayesian methods using Markov Chain Monte Carlo (MCMC), and
the Iterative Bootstrap method. Using MCMC simulation technique has come to the forefront
of statistical research over the last one and half decade [Gelfand et al. (1990)] and also it is
being used with greater extent in multilevel modeling recently. An important part of modeling
involves testing parameters and models to see which parts of the multilevel model are
statistically important. In fixed coefficients simple logistic regression tests about parameters
are done using the Wald test.

Parameter estimation in hierarchical generalized linear models is more complicated than the
hierarchical linear models. The most frequently used kind of approximation method used are
based on a first-order or second-order Taylor series expansion of the link function. There are
different methods of parameter estimation which are implemented in various softwares like
MLwiN and LISREL among others. In this study, the multilevel data analysis is supported by
the software LISREL.

33
 CHAPTER FOUR
4 STATISTICAL DATA ANALYSIS AND RESULTS
4.1 Major Socio-economic, demographic and proximate characteristics of
family planning practice status of couples

The major socio-economic, demographic and proximate background characteristics of the

respondents (couples) are presented in Table 4.1. The total number of couples covered in the
study is 25,240. Among these, 19,340 (76.6 percent) did not practice FP methods whereas
5,900 (23.4 percent) practiced a FP method.

The proportion of couples who do not practice FP, as can be seen in Table 4.1, differs by type
of place of residence: urban and rural. Accordingly, more than eighty percent of couples who
do not practice a method (86.8 percent) reside in rural areas. A small number of those couples
who are not practicing a FP method (31.2 percent) reside in urban areas. The status of couples
on a FP practice varied from one region to the other. For example, the highest status of
couples not practicing the method was recorded in Somali (96.7 percent) followed by Affar
(87.1 percent) as opposed to the lowest status which was recorded in Addis Ababa (16.3
percent) followed by Dire Dawa (55.1 percent). Moreover, region-wise and urban-rural
difference showed that in almost all regions, the highest status of couples who do not practice
family planning method was observed in rural parts.
Age of women was found to be an important and determinant variable of family planning
practice. The highest number of couples not practicing the method was observed among
women age group of 15-19 years (83.9 percent) followed by age group 35-49 years (79.7
percent) and age group 20-34 years (74.0 percent).
The educational status of a woman and her male partner under consideration is also an
important variable. Among the respondents (couples), those who have secondary or higher
educational level were less likely to not practice any FP method (34.7and 34.5 percent) than
primary level (74.8 and 74.9 percent) and no education (88.4 and 88.1 percent), respectively,
for a woman and male partner/husband. This indicates that completing at least primary level of
school would help to decrease the number of couples who do not practice the FP methods.

34
The status of not practicing a FP method among couples from poor households (79.2 percent)
is higher than couple from medium and higher economic status households (75.0 percent). As
can be seen from Table 4.1, the occupation of a woman and that of her partner has a
contribution not to practice a FP. Women who are agriculture employees (82.6 percent) and not
working (76.3 percent) were not using any family planning method. When we look at the
figures for the male partners/husbands who are agricultural employee (88.2 percent) do not
practice a FP method whereas those who are non-agriculture employee (39.1 percent) are not
using any method.

With regard to the number of living children, the greater number of couples who do not
practice FP were those who have three and more children (78.3 percent) followed by couples
with no living children (76.8 percent). Among couples with 1-2 living children 26.2 percent
practiced a family planning method. It is believed that exposure to any kind of family planning
methods through the mass media like radio, TV and newspapers and magazines enhance
practice of family planning. Couples who are exposed to any kind of mass media (56.0 percent)
are less likely not to practice a FP method than those who are not (87.3 percent). Couples with
no knowledge about a family planning method are not using family planning method. Those
couples who have knowledge about at least one family planning method (77.1 percent) do not
practice FP. “Access to health facility”, “visited by FP worker” and “discuss FP methods with
partners” are important variables. About 77.4 percent of couples who are not visited by a FP
worker during the last 12 months did not practice a FP method, while 42.9 percent of couples
who were visited by a FP worker during the last 12 months did not used a method. Discussion
of couples about FP methods helped reduce the number by 44.9 percent. When men did not
participate in FP issues it was observed that 82.1 percent of couples did not practice FP.
The status of FP practice of couples whose women’s were agriculture employee (82.6 percent)
did not practice FP methods is greater than that of couples whose women’s do not have job
(76.3 percent) and non-agriculture employees(52.9 percent). When we look at male partners
who are agricultural employee in a couple about (88.2 percent) do not practice FP than that of
non- agriculture employees (39.1 percent).
The number of couples who do not practice FP method is higher for those women’s who are
followers of catholic, traditional and other (93.8 percent) followed by muslim, protestant and

35
orthodox, 84.4 percent,78.6 percent and 67.5 percent, respectively. As can be seen from Table
4.1, husbands approval of a method is important determinant factor for FP practice of a couple.
For approval of husband about a method (51.6 percent) couples are less likely not to practice
FP methods than for disapproval of husband (88.5 percent).
Table 4.1 Distribution of Couples by Socio-economic, demographic and proximate related
characteristics and by region
Background Urban Rural Total
Characteristic

Number of

practice a
who do not
who do not

who do not
FP method

FP method

FP method
Number of

Number of
number of
practice a

practice a
couples

couples

couples

couples
Percent

Percent

Percent
Total

Total

Total
Residence
Urban 1440 31.2 4620
Rural 17900 86.8 20620
Region
Tigray 100 41.7 240 1380 80.2 1720 1480 75.5 1960
Affar 40 40.0 100 1180 90.8 1300 1220 87.1 1400
Amhara 60 25.0 240 3040 86.4 3520 3100 82.4 3760
Oromiya 60 20.0 300 3780 86.7 4360 3840 82.4 4660
Somali 140 77.8 180 1620 98.8 1640 1760 96.7 1820
Ben-Gumuz 100 62.5 160 1680 84.8 1980 1780 83.2 2140
SNNP 80 36.4 220 2960 84.6 3500 3040 81.7 3720
Gambela 120 66.7 180 1180 81.9 1440 1300 80.2 1620
Harari 200 45.5 440 500 89.3 560 700 70.0 1000
Dire Dawa 280 29.2 960 580 96.7 600 860 55.1 1560
Addis Ababa 260 16.3 1600 260 16.3 1600
Age of a
woman
15-19 140 50.0 280 1420 89.9 1580 1560 83.9 1860
20-34 780 26.7 2920 10400 85.4 12180 11180 74.0 15100
35-49 520 36.6 1420 6080 88.6 6860 6600 79.7 8280
Educational
level of women
No education 480 63.2 760 12680 89.8 14120 13160 88.4 14880
Primary 380 33.3 1140 4440 83.8 5300 4820 74.8 6440
Secondary and 580 21.3 2720 780 65.0 1200 1360 34.7 3920
higher
Male partner’s
education level
No education 460 57.5 800 13100 89.7 14600 13560 88.1 15400
Primary 380 35.8 1060 4040 83.5 4840 4420 74.9 5900
Secondary and 600 21.7 2760 760 64.4 1180 1360 34.5 3930
higher
Occupation of
a woman
Not working 620 31.6 1960 6320 88.5 7140 6940 76.3 9100
Non-agriculture 560 33.3 1680 1100 75.3 1460 1660 52.9 3140
Agric-employee 260 26.5 980 10480 87.2 12020 10740 82.6 13000

36
Male partner's
occupation
Non-agriculture 1260 29.2 4320 1060 65.4 1620 2320 39.1 5940
Agric-employee 180 60.0 300 16840 88.6 19000 17020 88.2 19300
Number of
living children
No children 320 39.0 820 1860 92.1 2020 2180 76.8 2840
1 to 2 children 460 26.1 1760 5780 86.3 6700 6240 73.8 8460
Three or more 660 32.4 2040 10260 86.2 11900 10920 78.3 13940
Knowledge of a
couple about
any FP method
Knows no FP 100 100.0 100 4301 100.0 4301 4401 100. 4401
method 0
Knows a FP 1340 29.6 4520 13599 83.3 16319 14939 71.7 20839
method
Visited by FP
worker last 12
months
No 1360 31.1 4380 17740 87.4 20300 19100 77.4 24680
Yes 80 33.3 240 160 50.0 320 240 42.9 560
Discussed FP
with partner
Never 1300 37.4 3480 16360 90.8 18020 17660 82.1 21500
More than once 140 12.3 1140 1540 59.2 2600 1680 44.9 3740
Partners/husba
nds approval
of FP
Disapproves 460 56.1 820 7220 91.9 7860 7680 88.5 8680
Approves 560 17.5 3200 4160 70.0 5940 4720 51.6 9140
Don’t know 420 70.0 600 6520 95.6 6820 6940 93.5 7420
Exposure of a
couple to any
media
No 540 49.1 1100 13960 90.1 15500 14500 87.3 16600
Yes 900 25.6 3520 3940 77.0 5120 4840 56.0 8640
Religion of a
woman
Orthodox 600 21.6 2780 6760 83.3 8120 7360 67.5 10900
Protestant 120 30.0 400 2820 84.4 3340 2940 78.6 3740
Muslim 640 47.8 1340 7500 90.4 8300 8140 84.4 9640
Others 80 80.0 100 820 95.3 860 900 93.8 960
Wealth index
Poor 497 33.0 1505 7194 87.6 8209 7691 79.2 9714
Medium or 943 30.3 3115 10706 86.3 12411 11649 75.0 15526
higher
Marital status
of a couple
Married 1440 32.6 4420 17839 86.8 20559 19279 77.2 24979
Living together 0 0.0 200 61 100.0 61 61 23.4 261
Total
25,240

37
4.2 Determinants of family planning practice among couples in Ethiopia:
Logistic Regression Analysis
Multiple logistic regression analysis was used to examine the effect of an independent variable
in the model to study the status of FP practice methods, while controlling the other independent
variables. Three different models are fitted in this study to identify the basic determinants of
family planning practice. The first model for couples not practicing FP in the urban areas, the
second model is for rural areas, and the third model is for both rural and urban (national) level.

4.2.1 Determinants of family planning practice in urban areas

The most important explanatory variables of status of family planning practice in urban
couples are identified using forward selection in logistic regression. In the model for urban
areas, region of residence, age of a woman, religion of a woman, educational level of women
and men, occupation of a female, visited by FP worker in the last 12 months, discussion about
FP with partners, husband’s approval of a method, exposure to mass media and number of
living children were found to be significant. Results of all variables in the logistic regression
are presented in Appendix B using enter method.

The urban sample showed that, odds of couples practice a FP method compared to those who
do not practice a method is decreased by a factor of 0.06 by being a couple in Somali rather
than Addis Ababa, controlling for other variables in the model. When age of a woman
increases one unit for age group 20-34 years rather than for age group 35-49 years, the odds
that couples practice of family planning increase by a factor of 2.327, when other variables are
controlled. “Visited by FP worker during last 12 months before the survey” is also another
important variable. We can say that odds of practicing the FP methods of couples compared to
those who do not practice a method are increased by a factor of 2.498 by being visited by FP
worker in the last 12 months rather than those who are not visited, controlling for other
variables in the model. Exposure to mass media is also found to be important determinant
variable. The odds of couples practice of a method compared to not practice a FP method are
decreased by a factor of 0.649 for being not exposed to any kind of mass media like radio, TV,
newspapers and magazines rather than those who are exposed to family planning messages on
mass media, while controlling for other variables in the model for urban areas. In urban areas,

38
when a husband approves a FP method increase by one unit, the odds that a couple practice FP
increase by a factor of 10.718, when other variables are controlled.

Table 4.2 Selected Important Variables in the Model for Urban areas

Covariates
ˆ S.E.( ˆ ) Wald df p-value Exp( ˆ )
Region 241.880 10 .000*
Tigray -1.354 .181 55.998 1 .000* .258
Affar -.440 .257 2.921 1 .087 .644
Amhara -.602 .227 7.022 1 .008* .548
Oromiya -1.094 .201 29.562 1 .000* .335
Somali -2.814 .296 90.597 1 .000* .060
Ben-Gumuz -2.434 .241 102.143 1 .000* .088
SNNP -1.805 .219 67.673 1 .000* .164
Gambela -1.967 .271 52.561 1 .000* .140
Harari -1.270 .151 70.570 1 .000* .281
Dire Dawa -.484 .121 16.108 1 .000* .617
Addis Ababa(ref.) 1.000
Age of a woman 102.841 2 .000*
15-19 -.325 .197 2.708 1 .100 .723
20-34 .845 .107 62.255 1 .000* 2.327
35-49(ref.) 1.000
Number of living children 61.668 2 .000*
No children -.964 .145 43.925 1 .000* .382
1-2 children -.802 .115 48.558 1 .000* .449
Three or more(ref.) 1.000
Educational level of woman 37.447 2 .000*
No education -.698 .360 3.761 1 .052 .498
Primary .694 .251 7.643 1 .006* 2.002
Secondary and higher(ref.) 1.000
Women’s occupation 76.901 2 .000*
Not working -.828 .134 38.083 1 .000* .437
Non-agriculture -1.173 .134 76.733 1 .000* .309
Agric-employee(ref.) 1.000
Educational level of partner 8.126 2 .017*
No education -.647 .360 3.233 1 .072 .524
Primary -.717 .252 8.075 1 .004* .488
Secondary and higher(ref.) 1.000
Visited by FP worker in the last 12
months
No .916 .202 20.581 1 .000* 2.498
Yes(ref.) 1.000
Discuss FP with partner
Never -1.414 .130 117.776 1 .000* .243
Once and more (ref.) 1.000

39
 Husband/partner approval of FP 330.299 2 .000*
Disapproves 1.097 .153 51.542 1 .000* 2.996
Approves 2.372 .139 292.124 1 .000* 10.718
Don’t know(ref.) 1.000
Exposure to mass media
No -.432 .107 16.178 1 .000* .649
Yes(ref.) 1.000
Religion of a woman 147.932 3 .000*
Orthodox 3.756 .359 109.417 1 .000* 42.768
Protestant 5.066 .425 142.388 1 .000* 158.554
Muslim 3.638 .374 94.723 1 .000* 38.022
Other(ref.) 1.000
Constant -2.458 .376 42.625 1 .000* .086
* Statistically Significant at (p<0.05) ref. = reference category

4.2.3 Determinants of family planning practice in rural areas

Logistic regression analysis with forward variable selection procedure was also employed for
rural couples alone. The results showed region of residence, age of a woman, number of living
children, educational level and occupation of men and women, discussion about FP with
partner, visited by FP worker in the last 12 months, exposure to mass media, approval of male
partner’s on FP and economic status of household are found to be important predictors of
family planning practice status for rural couples (Table 4.4).

The second model on rural areas showed odds of couples practice FP compared to those who
do not practice FP are increased by a factor of 8.139 and 7.272 for being a couple in Tigray and
Ben-Gumuz regions, respectively rather than for a couple being in Dire Dawa region,
controlling other variables in the model for rural part of Ethiopia. Odds of couples practice FP
compared to who do not practice FP are decreased by the factor of 0.887 for a women being
completed primary education rather than those who completed secondary and higher level of
education, while controlling other variables in the model. In rural areas odds of practice a FP
method compared to not practice are increased by a factor of 3.072 for a husband employed in
non-agriculture rather than agricultural employee in a couple, when other variables in the
model are controlled. In rural areas, when a husband approves a FP method increase by one
unit, the odds that a couple practice FP increase by a factor of 4.826, when other variables are
controlled. The odds of couples practice a method compared to not to practice a FP method are
decreased by a factor of 0.958 for a couple with 1-2 living children rather than those with three

40
and more children, controlling for other variables in the model for urban areas. Results of all
variables in the logistic regression are presented in Appendix C using enter method.

Table 4.3 Important Selected Variables in the Model for Rural areas

Covariates
ˆ S.E.( ˆ ) Wald df p-value Exp( ˆ )
Region 271.182 9 .000*
Tigray 2.097 .258 66.226 1 .000* 8.139
Affar 1.347 .276 23.801 1 .000* 3.846
Amhara 1.683 .255 43.572 1 .000* 5.380
Oromiya 1.388 .256 29.311 1 .000* 4.008
Somali -.666 .345 3.734 1 .053 .514
Ben-Gumuz 1.984 .261 57.797 1 .000* 7.272
SNNP 1.462 .258 32.031 1 .000* 4.316
Gambela .939 .267 12.383 1 .000* 2.558
Harari 1.411 .290 23.678 1 .000* 4.099
Dire Dawa(ref.) 1.000
Age of a woman 6.495 2 .039*
15-19 -.047 .137 .120 1 .729 .954
20-34 .125 .057 4.752 1 .029* 1.133
35-49(ref.) 1.000
Number of living children 25.170 2 .000*
No children -.632 .128 24.598 1 .000* .531
1-2 children -.043 .056 .572 1 .449 .958
Three or more(ref.) 1.000
Educational level of woman 6.725 2 .035*
No education .243 .450 .291 1 .590 1.274
Primary -.120 .428 .079 1 .779 .887
Secondary and higher(ref.) 1.000
Women’s occupation 44.521 2 .000*
Not working .248 .057 18.929 1 .000* 1.282
Non-agriculture .525 .086 37.646 1 .000* 1.691
Agric-employee(ref.) 1.000
Educational level of partner 16.708 2 .000*
No education -.993 .449 4.886 1 .027* .370
Primary -.427 .428 .996 1 .318 .652
Secondary and higher(ref.) 1.000
Male Partner’s occupation
Non-agriculture 1.122 .076 215.581 1 .000* 3.072
Agric-employee(ref.) 1.000
Visited by FP worker in the last 12
months
No -1.896 .142 177.350 1 .000* .150
Yes(ref.) 1.000

41
 Discuss FP with partner
Never -1.362 .057 575.381 1 .000* .256
Once and more (ref.) 1.000
Husband/partner approval of FP 565.579 2 .000*
Disapproves .693 .076 82.848 1 .000* 1.999
Approves 1.574 .071 494.038 1 .000* 4.826
Don’t know(ref.) 1.000
Exposure to mass media
No -.472 .053 79.873 1 .000* .624
Yes(ref.) 1.000
Economic status of HH
Poor -.098 .049 4.009 1 .045* .907
Medium and higher (ref.) 1.000
Constant -.609 .289 4.444 1 .035* .544
* Statistically Significant at (p<0.05) ref. = reference category

4.2.3 Determinants of family planning practice at national level

The combined urban and rural (national) sample results showed that type of place of residence,
region, age of a woman, religion of a woman, educational level of women and men, occupation
of a female, visited by FP worker in the last 12 months, discussion about FP with partners,
husband’s/partner’s approval of a method, media exposure and number of living children are
found to be important variables (Table 4.5).

The third model on both rural and urban areas showed odds of couples practice FP compared to
who do not practice FP are decreased by a factor of 0.642 and 0.599 for being a couple in Ben-
Gumuz and Tigray regions, respectively rather than for a couple being in Addis Ababa,
controlling other variables in the model for all parts of Ethiopia. For a couple being reside in
urban areas rather than rural counterparts, odds of practice FP method compared with do not
practice a method is increased by a factor of 2.365, when other variables are controlled in the
model. When age of a woman increases one unit for age group 20-34 years rather than for age
group 35-49 years, the odds that couples practice of family planning increase by a factor of
1.405, when other variables are controlled in the model for Ethiopia. The odds of couples FP
practice are increased by a factor of 2.925 for a male partner being non-agricultural employee
rather than agricultural employee, controlling for other variables in the model. When a husband
approves a FP method increases by one unit, the odds that a couple practice FP increase by a
factor of 5.37, when other variables are controlled in the model for national level.

42
The odds of couples practice of a FP method compared to not to practice a method are
decreased by a factor of 0.632 for being not exposed to any kind of mass media like radio, TV,
newspapers and magazines rather than those who are exposed to family planning messages on
mass media, while controlling for other variables in the model. Results of all variables in the
logistic regression are presented in Appendix A using enter method. The estimated coefficients
( ˆ ’s) for the covariates in the final model, their standard error (S.E.( ˆ )) and the odds ratio
corresponding each estimated coefficients of not practicing family planning among urban
couples is presented in the following Table 4.5.

Table 4.4 Important Selected Variables in the model at National Level

Covariates
ˆ S.E.( ˆ ) Wald df p-value Exp( ˆ )
Place of residence
Rural .861 .082 109.705 1 .000* 2.365
Urban (ref.) 1.000
Region 319.145 10 .000*
Tigray -.512 .116 19.534 1 .000* .599
Affar -.830 .134 38.120 1 .000* .436
Amhara -.722 .110 42.927 1 .000* .486
Oromiya -.876 .107 67.464 1 .000* .417
Somali -2.499 .175 203.685 1 .000* .082
Ben-Gumuz -.443 .117 14.401 1 .000* .642
SNNP -.636 .113 31.545 1 .000* .529
Gambela -1.329 .125 113.069 1 .000* .265
Harari -1.025 .120 72.853 1 .000* .359
Dire Dawa -.905 .104 75.351 1 .000* .404
Addis Ababa(ref.) 1.000
Age of a woman 57.688 2 .000*
15-19 .010 .111 .009 1 .925 1.010
20-34 .340 .049 47.931 1 .000* 1.405
35-49(ref.) 1.000
Martial Status
Married -2.133 .216 97.774 1 .000* .118
Living together(ref.) 1.000
Number of living children 100.354 2 .000*
No children -.874 .088 99.017 1 .000* .417
1-2 children -.210 .048 18.769 1 .000* .811
Three or more(ref.) 1.000
Educational level of partner 141.528 2 .000*
No education -.817 .069 141.096 1 .000* .442
Primary -.480 .063 57.683 1 .000* .619
Secondary and higher(ref.) 1.000

43
 Male Partner’s occupation
Non-agriculture 1.073 .068 250.191 1 .000* 2.925
Agric-employee(ref) 1.000
Visited by FP worker in last 12 months
No -.814 .115 49.841 1 .000* .443
Yes(ref.) 1.000
Discuss FP with partner
Never -1.307 .050 673.682 1 .000* .271
Once and more (ref.) 1.000
Husband/partner approval of FP 881.877 2 .000*
Disapproves .765 .067 131.327 1 .000* 2.150
Approves 1.681 .061 770.039 1 .000* 5.370
Don’t know(ref.) 1.000
Exposure to mass media
No -.459 .046 98.870 1 .000* .632
Yes(ref.) 1.000
Religion of a woman 154.313 3 .000*
Orthodox 2.087 .172 147.345 1 .000* 8.060
Protestant 1.810 .177 104.031 1 .000* 6.109
Muslim 2.009 .174 132.585 1 .000* 7.456
Other(ref.) 1.000
Constant .699 .323 4.675 1 .031* 2.012
* Statistically Significant at (p<0.05) ref. = reference category

4.2.4 Goodness of Fit and Model Diagnostics

a. Goodness of Fit
In our logistic regression analysis of the data by SPSS package, results of several goodness of
fit test accompany the SPSS output. The classification table given below shows that of 25,240
couples included in the analysis 86.4% were correctly classified providing evidence that the
model will fit well.

Table 4.5 Classification Table

Predicted
Status of a couple about FP method Percentage
practice Correct
Observed Do not practice a Practice a FP
FP method method
Step1 Status of a couple Do not practice a FP
about FP method method 18245 1095 94.3
practice
Practice a FP method 2329 3571 60.5
Overall Percentage 86.4
The cut value is .500

44
The goodness of fit test for the multiple logistic regression model was assessed using the
Hosmer-Lemeshow test and the deviance-based chi-square test. For national level the Hosmer-
Lemeshow test the p-value 0.080 shows that the model adequately fits the data at 0.05 level of
significance. Similarly, the deviance-based chi-square test provided chi-square value of X2=
4455.702, d.f =10, P<0.01 imply good fit for the combined model.

b. Model Diagnostics

The adequacy of the fitted model was checked for possible presence and treatment of outliers
and influential values. The diagnostic test results for detection of outliers and influential values
are presented in Appendix C. The DFBETAs for model parameters including the constant term
and Cook’s influence statistic were both less than unity. DFBETAs less than unity imply no
specific impact of an observation on the coefficient of a particular predictor variable, while
Cook’s distance less than unity showed that an observation had no overall impact on the
estimated vector of regression coefficients β. The value of the leverage statistic less than one
shows that no subject has a substantial large impact on the predicted values of model. Thus,
from the above goodness of fit tests and diagnostic checking, we can say that our model is
adequate.

4.3 Determinants of family planning practice: A Multilevel Logistic

Regression Analysis
In the multilevel analysis, a two-level structure is used with regions as the second-level units
and couples as the first level units. This is basically with the expectation that there would be a
difference in the status of couples not to practice family planning methods among regions. The
nesting structure is couples within regions with a total of 25,240 couples.

A chi-square test statistic in equation (3.13) was applied to assess heterogeneity in the
proportion of couples who do not practice FP methods among the 11 regions. The test yield
X2= 4455.702, d.f=10, P<0.01. Thus, there is evidence for heterogeneity among regions with
respect to FP practice of couples.

The results presented in the subsequent sections are carried out using a student-version
software LISREL 8.8 for window. The program is restricted to analyze few number of
explanatory variables. As a result, only five explanatory variables (place of residence,

45
discussion about FP with partner, visited by FP worker in the last 12 months, male
partners/husbands approval of FP and exposure to mass media) are considered. These predictor
variables have two categories each.

4.3.1 The empty logistic regression model

The simplest non–trivial specification of the hierarchical linear model is a model in which only
the intercept varies between level two units and no predictor (explanatory) variables are
entered in the model. The empty model contains no explanatory variables and it can be
considered as a parametric version of assessing heterogeneity among regions with respect to
couples’s family planning practice status.

Table 4.6 Estimates for empty model

Fixed Part Coefficient S.E. Z-value P-value

 0 = Intercept -1.2182 0.0164 74.3669 0.0000*

Random Part Variance

S.E. Z-value P-value
Component
Level-two variance: 1.3480 0.4239 3.1800 0.0015*
2
 0  var(U 0 j )

Deviance-based chi-square
3960.7895 0.000*
* Statistically Significant at (p<0.05)

The deviance-based Chi-square (deviance = 3960.7895) indicated in Table 4.6 above is the
difference in deviance between an empty model without random effect (deviance = 27449.7626
see Appendix D) and an empty model with random effect (deviance =23488.9731 see
Appendix D). This implies that an empty model with random intercept is better than an empty
model without random intercept. The variance of the random factor in empty model is
significant which indicates that there are regional differences in the family planning practice
status of couples. The test indicates that the fitted model is good.

46
4.3.2 The random intercept model and fixed explanatory variables
In order to identify the effect of some selected explanatory variables a multilevel logistic
regression model with random intercept and fixed slope for explanatory variables were
estimated with the help of LISREL software. The variables considered were place of residence
(1= urban, 2= rural), discussion about FP with partner (0= never, 1= once and more), visited by
FP worker in the last 12 months (0= no, 1= yes), male partners/husbands approval of a FP
method (1= approves, 2= otherwise) and exposure to any mass media (0= no, 1= yes).The
results of two-level random intercept and fixed slope (coefficient) model are presented in the
following Table 4.7 below. The deviance based chi-square test for significance of random
effects (X2=349.8995, df=1, P<0.05) indicates that the random intercept model with the fixed
slope is found to give a better fit as compared to the empty model discussed in Section 4.3.1.
From Table 4.7, we observe that place of residence, husbands/partners approval of a FP
method, visited by FP worker in the last 12 months, exposure to mass media and discussion
about FP with male partner were found to be significant determinants of variation in couples
practice of family planning methods among regions. As we can see in Table 4.7, all predictors
are significant determinants of variation of contraceptive methods of couples among regions.

Table 4.7 Estimates for random intercept and fixed coefficient model

Estimate
Covariates S.E. Z-value P-value
coefficient
Intercept -3.5743 0.1080 -33.0809 0.0000*
Place of Residence
Urban 1.9089 0.0462 41.2895 0.0000*
Rural (Ref.)
Husband’s/partner’s approval
of a method
No 1.3792 0.0409 33.7433 0.0000*
Yes (Ref.)
Visited by FP worker in the last
12 months
No -0.7758 0.1157 -6.7034 0.0000*
Yes (Ref.)
Discuss FP with partner
-1.2715 0.0472 -26.9396 0.0000*
Never
Once and more (Ref.)
Exposure to mass media
No -0.7168 0.0410 -17.4637 0.0000*
Yes (Ref.)

47
 Estimate
Random Part Variance
S.E. Z-value P-value
component
Random Intercept: 0.2571 0.0816 3.1486 0.0016*
2
 0  var(U 0 j ) intercept
variance
Deviance-based chi-square 349.8995 0.0000*
* Statistically Significant at (p<0.05), (Ref.) = indicates the reference category

4.3.3 The random coefficient model

The random coefficient model is useful, because it shows how much variability exists at each
level. Result in Table 4.8 below is obtained by including level-2 random coefficients of Place
of residence, visited by FP worker in the last 12 months, discuss FP with partner, husband’s
approval of a FP method and exposure to mass media and an overall (level-2) or regional
2
variance constant term (  0 ) together with variance and covariance terms representing the
random effects of predictors.

Table 4.8 Results for Fixed and Random Effects of Random Coefficient Model

Estimate
Covariates S.E. Z-value P-value
coefficient
Intercept -4.2287 0.1101 -38.4217 0.0000*
Place of Residence
Urban 2.1290 0.0478 44.5611 0.0000*
Rural (Ref.)
Visited by FP worker in the last
12 months
No -2.0345 0.1198 -16.9824 0.0000*
Yes (Ref.)
Discuss FP with partner
-1.7434 0.0480 -36.2938 0.0000*
Never
Once and more (Ref.)
Husband’s/partner’s approval
of a method
No 1.6852 0.0409 33.7433 0.0000*
Yes (Ref.)
Exposure to mass media
No -0.6638 0.0421 -15.7743 0.0000*
Yes (Ref.)

48
 Random Part Estimate
Variance
S.E. Z-value P-value
component
Random Coefficient:
Level-2 variance covariance
5.5077 1.7535 3.1410 0.0017*
 0 2  var(U 0 j )
1.2381 0.3888 3.1844 0.0015*
 12  var(U 1 j )
 2 2  var(U 2 j ) 8.7196 2.8732 3.0348 0.0024*

0.9700 0.3075 3.1544 0.0016*

 3 2  var(U 3 j )
1.0199 0.3221 3.1662 0.0015*
 4 2  var(U 4 j )
 5 2  var(U 5 j ) 1.9157 0.6097 3.1420 0.0017*

-1.8754 0.7101 -2.6410 0.0083*

 01  cov(U 0 j ,U 1 j )
 02  cov(U 0 j ,U 2 j ) 0.7902 1.5658 0.5047 0.6138
1.4357 0.6085 2.3594 0.0183*
 03  cov(U 0 j ,U 3 j )
 04  cov(U 0 j ,U 4 j ) -1.2805 0.6130 -2.0888 0.0367*

 05  cov(U 0 j ,U 5 j ) -2.5071 0.9289 -2.6990 0.0070*

 12  cov(U 1 j ,U 2 j ) 1.2381 0.3888 3.1844 0.0015*

 13  cov(U 1 j ,U 3 j ) -0.1596 0.2503 -0.6376 0.5238

 14  cov(U 1 j ,U 4 j ) -0.1747 0.2572 -0.6794 0.4969

 15  cov(U 1 j ,U 5 j ) 0.8698 0.3887 2.2377 0.0252*

 23  cov(U 2 j ,U 3 j ) 1.8967 0.7899 2.4011 0.0163*

 24  cov(U 2 j ,U 4 j ) -2.3455 0.8541 -2.7462 0.0060*

 25  cov(U 2 j ,U 5 j ) 0.9318 0.9548 0.9759 0.3291

 34  cov(U 3 j ,U 4 j ) -0.8061 0.2852 -2.8262 0.0047*

 35  cov(U 3 j ,U 5 j ) -0.1298 0.3146 -0.4124 0.6800

 45  cov(U 4 j ,U 5 j ) 0.4159 0.3390 1.2268 0.2199

Deviance-based chi-square 1358.7496 0.0000*
*Statistically Significant at (P<0.05), (Ref.) = indicates the reference category

Table 4.9 Level-2 covariance matrix of the random coefficient

intercept RESIDE VISIT_FP DISSCPAR HUSAPPRO MEDIA
Intercept 5.507681
RESIDE -1.875390 1.238128
VISIT_FP 0.790197 1.427065 8.719584
DISSCPAR 1.435746 -0.159564 1.896726 0.970009
HUSAPPRO -1.280482 -0.174733 -2.345508 -0.80611 1.019946
MEDIA -2.507134 0.869833 0.931815 -0.129753 0.415866 1.915709

49
 Table 4.10 Level-2 correlation matrix of the random coefficient

intercept RESIDE VISIT_FP DISSCPAR HUSAPPRO MEDIA

Intercept 1.000000
RESIDE -0.718165 1.000000
VISIT_FP 0.114026 0.434323 1.000000
DISSCPAR 0.621163 -0.145601 0.652182 1.000000
HUSAPPRO -0.540257 -0.155490 -0.786503 -0.810438 1.000000
MEDIA -0.771842 0.564792 0.227991 -0.095184 0.297509 1.000000

Table 11. Test of the significance of Correlations

Visited by Male
FP worker partner/ Exposure to
Place of in the last Discussed FP Husband any mass
residence 12months with partner approves FP media
Place of residence Pearson Correlation 1 -.096** -.131** .326** -.419**
Sig. (2-tailed) .000 .000 .000 .000
Visited by FP worker in Pearson Correlation -.096** 1 .073** -.088** .073**
the last 12m Sig. (2-tailed)
.000 .000 .000 .000

Discussed FP with Pearson Correlation -.131** .073** 1 -.359** .132**

partner Sig. (2-tailed) .000 .000 .000 .000
Male partner/Husband Pearson Correlation .326** -.088** -.359** 1 -.283**
approves FP Sig. (2-tailed) .000 .000 .000 .000
Exposure to any mass Pearson Correlation -.419** .073** .132** -.283** 1
media Sig. (2-tailed) .000 .000 .000 .000
**. Correlation is significant at the 0.01 level (2-tailed).

In Table 4.8 above, the value of Var (U0j), Var (U1j), Var (U2j), Var (U3j), Var (U4j) and Var
(U5j) are the estimated variance of intercept, slope of place of residence, slope of visited by FP
worker in the last 12 months, slope of discussion about FP with partner, slope of husband’s
approval of a FP method and slope of exposure to mass media, respectively. All are regionwise
intercepts and the slopes vary significantly, that is, there is a significant variation in the effects
of these explanatory variables across the regions. Some of the variances of the interaction
terms between intercepts and slopes of explanatory variables are also found significant.
Interpretation of significant covariance terms can be easily made in terms of the correlation
coefficients displayed in Tables 4.9 and 4.10. The correlation matrix contains the estimated
correlation between random intercepts and slopes.

Positive correlation between intercepts and slopes implies that regions with higher intercepts
tend to have on average higher slopes (Table 4.10). For example, the correlation between the
intercept and random slope of “discuss FP with partner” is 0.621163, meaning that couples

50
who discuss family planning issues have better status of using family planning methods than
those who did not by a larger factor at regions with higher intercepts compared to regions with
lower intercepts.

The negative sign for the correlation between intercepts and slopes implies that regions with
higher intercepts tend to have on average lower slopes on the corresponding predictors (Table
4.10). Couples whose place of residence is in urban, who have access to family planning
lessons through mass media and whose husband/partner approve FP methods are less likely not
to practice the methods of contraception.

The results in Table 4.8 show that the listed predicted variables contribute significantly to the
status of couples practice of family planning. The deviance-based Chi-square (deviance =
1358.7496) with degrees of freedom d.f = 21 indicated in Table 4.8 is the difference between
the model without any random effects (deviance = 18029.8087 see Appendix F) and the model
with fixed and random effects (deviance = 16671.0590 see Appendix F). This difference
indicates that a model with a random coefficient is more appropriate to explain regional
variation than a model with fixed coefficients and the variables included are significant.

We note that the student version of the software LISREL does not provide model diagnostic
tests. We only checked a goodness of fit by using deviance-based chi-squared test.

In general, the results of the multilevel logistic regression suggest that there exist differences
in the status of couples practicing family planning methods among regions in Ethiopia. As a
result, the random coefficient model analysis generates separate estimates of logistic
regressions for each region (Table 4.11).

Table 4.12 Parameter estimates for regions

Region Coefficients
intercept RESIDE VISIT_FP DISSCPAR HUSAPPRO MEDIA
Tigray Estimate 2.5094 -.9560 .8689 1.1941 -.7224 -.5102
S.E .0810 .0225 .1441 .0138 .0096 .0139
Z-value 30.9839* -42.4489* 6.0306* 86.4316* -75.6215* -36.6273*
Affar Estimate -.0019 -1.3457 -6.7284 -1.9348 1.5856 -2.0228
S.E .3402 .1078 1.8726 .2133 .1138 .0569
Z-value -.0055 -12.4815* -3.5931* -9.0690* 13.9327* -35.5695*
Amhara Estimate .2430 .1432 .0589 .5277 -.5218 .1230
S.E .1123 .0294 .0730 .0094 .0072 .0173
Z-value 2.1645* 4.8644* .8065 56.1186* -72.7367* 7.0997*

51
 Oromiya Estimate .3254 -.0053 2.1321 .3749 -.3965 .1683
S.E .0834 .0230 .5861 .0081 .0074 .0088
Z-value 3.9019* -.2297* 3.6376* 46.5526* -53.2759* 19.1940*
Somali Estimate -5.4788 1.0841 -2.0021 -1.3816 2.1081 3.4080
S.E 1.2216 .0970 2.5712 .2485 .2567 .3688
Z-value -4.4850* 11.1783* -.7787 -5.5593* 8.2123* 9.2397*
Ben- Estimate 3.1336 -1.4993 -.4925 1.0877 -.3231 -.8331
Gumuz S.E .1421 .0313 1.3855 .0188 .0141 .0193
Z-value 22.0476* -47.9649* -.3554 57.7883* -22.9067* -43.2217*
SNNP Estimate -1.6432 .3282 .3180 -.3641 .3839 .9941
S.E .1324 .0316 .0701 .0122 .0087 .0163
Z-value -12.4066* 10.3996* 4.5342* -29.8301* 44.2797* 60.9245*
Gambela Estimate 1.9015 -.7762 2.4685 .8094 -.4804 -.0074
S.E .1301 .0354 .7164 .0163 .0116 .0186
Z-value 14.6133* -21.9434* 3.4456* 49.7556* -41.3870* -.3966
Harari Estimate 1.4568 -.3648 -1.9610 -.6476 .2986 -1.5979
S.E .1580 .0424 1.7065 .0480 .0283 .0533
Z-value 9.2195* -8.6002* -1.1492 -13.4804* 10.5520* -29.9607*
Dire Estimate -1.2409 1.4878 .7282 .3410 -.5319 .4163
Dawa S.E .1468 .0512 .0827 .0324 .0192 .0342
Z-value -8.4552* 29.0366* 8.8080* 10.5272* -27.7199* 12.1872*
Addis Estimate -1.2049 -.0068 4.6095 1.9041 -1.4000 -.1385
Ababa S.E .3087 .0458 .0620 .2967 .0186 .0271
Z-value -3.9025* -.1479 74.3407* 6.4164* -75.3997* -5.1127*
* Statistically significant at (p<0.05)
As we can see resuts in Table 4.11 above, all explanatory variables are statistically significant
in Tigray, Oromiya, SNNP and Dire Dawa regions and hence they are determinants of family
planning practice. For Addis Ababa region all predictor variables are statistically significant
except place of residence (because for Addis Ababa place of residence cannot be categorized
as urban and rural). “Visited by FP worker in the last 12 months” before the survey was
conducted is not significant factor in Amhara, Somali, Benishangul-Gumuz and Harari for
couples practice of FP.
Table 4.11 shows there prevails variation among regions with respect to the impact of the
predictor variables on the couples practice of family planning methods.

52
 CHAPTER FIVE
5 DISCUISSION, CONCLUSION AND RECOMMEDATION
5.1 DISCUISSION
This study is an attempt to identify some determinants of family planning practice of couples
based on Ethiopian Demographic and Health Survey (EDHS 2005) data. Accordingly
descriptive analysis, multiple logistic regression and multilevel logistic regression techniques
were used. Multiple logistic regression was applied separately for urban, rural and national
level. The results which are obtained are discussed as follows.

At first the study included sixteen predictor variables that were categorized under socio-
economic, demographic and proximate characteristics. The descriptive analysis of the study
revealed that only 23.4 percent of the sample couples were practicing family planning methods
and 76.6 percent did not practice family planning.

A finding that is consistent with past studies is that couples in urban areas practice FP method
more than their rural counterparts. Concerning the regional variation in couples FP practice
status, the results for both rural and urban parts of Ethiopia confirmed that couples in Somali
and Affar regions practice family planning methods less than couples who live in other regions.

Another important statistical method used in this study is multilevel logistic regression which
is an example of generalized hierarchal linear model. In the multilevel analysis couples are
nested within various regions in Ethiopia. Three multilevel models: empty model, random
intercept and slope or random coefficient model were applied in order to explain regional
differences in family planning practice among couples. The results obtained are discussed
briefly in the following paragraph.

Before the analysis of data using multilevel approach, first heterogeneity of the status of FP
practice of couples with regard to regions was checked. A nonparametric approach based on
the chi-square test and the parametric approach based on the empty model suggests that
couples status of FP practice differs among regions. Such heterogeneity is a requirement in the
multilevel analysis. In addition to a model without explanatory variables, two other models
were used in multilevel logistic regression for the national sample a whole. In both models

53
(random intercept and random coefficient) the overall variance constant term is found to be
statistically significant which may again imply the differences in couples status of FP practice.
Among the five variables considered due to restriction in the software used, the effect of the
random part of “place of residence”, “visited by FP worker in the last 12 months”, “discussed
FP with partner”, “husband’s approval of a FP method” and “exposure to mass media” differs
across regions. Similarly, the interaction of random parts of “place of residence” and “visited
by FP worker in the last 12 months”, “exposure to mass media” provided significant effect on
status of couples FP practice across regions. Also, the interaction of the random parts of
“visited by FP worker in the last 12 months” and “discussed FP with partner”, “husband’s
approval of a FP method” were found to be determinant factors for the practice of FP among
regions. The goodness-of-fit of the model is checked by deviance-based chi-square test in
multilevel logistic regression analysis by LISREL software and found comparatively from the
three models fitted the random coefficient model is better.

54
5.2 CONCLUSION
The study identified the following socio-economic, demographic and proximate variables as
determinants of family planning practice of couples in Ethiopia: Visited by FP worker in the
last 12 months, discuss FP with partner, husbands/partners approval of a method, exposure to
mass media, educational level of women, partners education level, number of living children,
age of a woman, occupation of male/female are the most important ones. The descriptive
results show that more than 50% of the respondents (couples) in the study do not practice
family planning methods.

Educational level of women and men do have a strong positive and significant association with
family planning practice status of couples. It is observed that couples with no education are
less likely to practice a FP than those who have completed primary, secondary and higher
levels. Educating individuals can significantly reduce the number of non users of family
planning methods. The study results also showed that couples engaged in agricultural activities
practiced FP less than those couples who are non-agricultural workers.
Even though couples have information about FP methods practice they do not practice it due to
religion of a woman and some other causes. Couples who had discussions about family
planning methods with their partners and whose husband/partner approved FP practice do
practice FP more. Involvement of men has contributions for couples to practice family
planning methods.

Although there is region-wise disparity in couples FP practice status, in all regions it is

observed that couples in rural parts of the country are more likely not to practice family
planning methods. Couples who are exposed to mass media like radio, TV, newspapers and
magazines have higher chance to practice FP methods than those who are not exposed to
family planning information from the media. When couples are visited by family planning
worker in the last 12 months before the survey they have more chance to practice FP.

55
5.3 RECOMMEDATIONS
Based on the findings of the current study we forward the following recommendations:

 In order to ensure wider expansion of contraceptive usages, government and non-

government organization involved in family planning should organize discussion groups
of couples on issues of family planning methods practice.
 Enhance information and communication activities regarding family planning services
using mass media, family planning workers and health centers.
 Policy makers should consider the adoption of family planning services for couples who
are living in rural areas of Ethiopia.
 All concerned bodies should conduct similar research to solve problems related to
family planning in all parts of the country so as to decrease the maternal mortality and
infant mortality and contribute to lower population growth.
 Family planning programs should be designed and implemented in accord with the
overall features of the regions to safeguard couples for practice of FP at each region.

56
 REFERENCES
Agresti, A. (1996). An Introduction to Categorical data Analysis. John Wiley and Sons, Inc.,
New York.
Agresti, A. (1990). Categorical Data Analysis. John Wiley and Sons, Inc., New York.
Almaz Terefe and C.O. Larsen (1993). “Modern contraception Use in Ethiopia: Does involving
Husbands make a difference?” America Journal of Public Health, Vol. 83, No.11, pp.15-
67. Washington, D.C.
Asefa, M., Drewett, R. and Tessema, F. (2000) A birth cohort study in southwest Ethiopia to
identify factors associated with infant mortality that is amenable for intervention. Ethiopian
Journal of Health Development 14(2), 161–168.
Bandarage, A. (1997). Women Population and Global Crisis: A Political Economic Analysis.
Zed Book Ltd London.

Bandura, A. (2002) Environmental sustainability by sociocognitive deceleration of population

growth. In The Psychology of Sustainable Development (Schmuch, P. & and Schultz, W.,
eds). Kluwer, Dordrecht, pp. 209–238.
Berhane, Y., Mekonnen E., Zerihun, M., and G. Assefa G (1999) Perception of fertility
regulation in remote community, Southern Ethiopia. Ethiopian Journal of Health
Development, 13 (3), 217–222.
Bongaarts, J. (1990). “The KAP-Gap and the Unmet Need for Contraception” Population and
Development Review Vol.17, No.2, pp.293-313.
Bongaarts, J. and S. W. Sinding (2009). “A Response to Critics of Family Planning
Programs” International Perspectives on Sexual and Reproductive Health Volume 35,
Number 1, March 2009.
Breslow, N. E. and D. G. Clayton ( 1993) Approximate inference in generalized linear mixed
models. J. Am. Statist. Assoc., 88:9–25.
Commenges, D. and Jacqmin, H. (1994). “The intra-class correlation coefficient distribution
free definition and test”. Biometrics, 50, 517-526.

Copas, J.B. (1988). “Binary Regression Models for Contaminated Data”. Journal of Royal
Statistical Association, B 50 (2): 225-265.

57
Central Statistical Authority (CSA), 2005. Ethiopia and ORC MACRO, Ethiopia demographic
and health survey, 2006. Addis Ababa Ethiopia and Calverton Maryland USA.
Daniel Sahle Yesus (1995). “Determinants of contraceptive non-use and unmet need among
married women in Ethiopia.” Unpublished Master Thesis, Addis Ababa University.

DHS (2000) Demographic and Health Survey. Central Statistical Authority, Addis Ababa,
Ethiopia.

DHS (2005) Demographic and Health Survey. Central Statistical Authority, Addis Ababa,
Ethiopia.
Ethiopia Trend Report (2007) “Analysis of the 2000 and 2005 Demographic and Health
Survey” Calverton, Maryland, USA: Macro International Inc.

Fitaw, Y., Berhane, Y. and Worku, A. (2003) Differentials of fertility in rural Butajira.
Ethiopian Journal of Health Development, 17 (1), 17–25.
Gaym, A. (2000) A review of maternal mortality at Jimma hospital, South-Western Ethiopia.
Ethiopian Journal of Health Development,14 (2), 215–223.
Gelfand A. E., S. E. Hills, A. Racine-Poon, and A. F. M. Smith. (1990) Illustration of
Bayesian inference in normal data models using Gibbs sampling. Journal of the American
Statistical Association, 85:972–985.
Goldstein, H., (1991) Nonlinear multilevel models with an application to discrete response data.
Biometrika, 78:45–51.
Goldstein, H. and J. Rasbash., (1996) Improved approximations for multilevel models with
binary responses. J. Roy. Statist. Soc. A, 159:505–513.
Goldstein, H., Multilevel Statistical Models. 3rd edition (2003) London: Arnold; New York:
Oxford University Press Inc.
Hailemariam A., Welsh M. and Nichols D. (2000). “Reproductive Health and Family
planning In Two Underserved community in Tigray Regional State”, Ethiopia:
Finding from a Baseline survey.

Hamilton, J., (1997), Contraceptive use increases in developing countries, New York.
Hardee-Cleaveland, K., and Judith Banister,(1988) Fertility Policy and Implementation in
china, 1986-88, Population and Development Review, Vol. 14, No. 2, , pp. 245-286.

58
Hogan, P.D., B. Berhanu and A. Hailemariam (1999). “Household Organization, Women’s
Autonomy and Contraceptive Behavior in Southern Ethiopia.” Studies in Family Planning
30[4]: 302–314.
Hosmer, D. and S. Lemeshow (2000, 2nd Ed). Applied Logistic Regression. John Wiley and
Sons Inc., New York.

Kubba, AK and Krasner N. (2000) An update in the palliative management of malignant

dysphagia. Eur J Surg Oncol; 26:116-29.
Laird, N. M. (1978) Empirical Bayes methods for two-way contingency tables. Biometrika,
65:581–590.
Levine, R. (2007). USAID’s Track Record in Family Planning. Available at
http://www.cgdev.org
Marston, C. and Cleland, J.C. (2003) Do unintended pregnancies carried to term lead to
adverse outcomes for mother and child? An assessment in five developing countries.
Population Studies, 57, 77–93.
McCauley, A.p., B. Robey, A.K. Blanc, and J.S. Geller, (1994). “Opportunities for Women
through Reproductive Choice.” Population Reports, Series M, No.12, Baltimore, Johns
Hopkins School of Public Health.
McCullagh, P. and Nelder, J.A. (1989). Generalized Linear Models. (2nd ed.), London:
Chapman and Hall.

McCulloch C. E., (1997) Maximum likelihood algorithms for generalized linear mixed models.
Journal of American Statistical Association, 92:162–170.
Merrick, T.W. (2002) Population and poverty: new views on an old controversy. International
Family Planning Perspectives, 28 (1), 41–28.
Ministry of Health (MOH), (2001). “Health and Health Related Indicators” Addis Ababa,
Ethiopia: Planning and Programming Department of MOH.
Ministry of Health (MOH), (1996). “Guidelines for Family Planning Services in Ethiopia.”
Addis Ababa, Ethiopia.
Mizanur R., Tilahun G., Mengistu A., and Jennifer W (2007) Enhanced Access to
Reproductive Health and Family Planning Pathfinder International in Ethiopia 2002-2007,
Addis Ababa, Ethiopia.

59
Nagase, T. (2003) Obstacles to modern contraceptive use among married women in southern
urban Maldives. Contraception, 68 (2), 125–134.
Population Reports. Why Family Planning Matters. Vol. XXVII, number 2. Series J. number
49, July 1999 (online). Available at: http:// www.Aol.com/astanar/smith.html (accessed 9
June 2004).
Population Reference Bureau (PRB). (2002). Family planning worldwide. 2002 Data sheet.
Data and estimates of contraceptive use and related reproductive health indicators for the
countries and regions of the world.
Potts, M. (2000) The Unmet Need for Family Planning. Scientific American Inc., 382 (1), 88–93.
Rhonda S., Lori A., Jay G., and Donna C., PRB, (2009) 4th edition. Family Planning Saves
Lives Washington DC, USA.
Robey, B., J. Ross, and I. Bhushan, (1996). “Meeting Unmet need: New strategies.” Population
Reports, Series J, No.43, Baltimore, Johns Hopkins School of Public Health, Population
Information Program.
Ross, J.A. and Winfrey, W.L. (2002) Unmet need for contraception in the developing world
and the former Soviet Union: an update estimate. International Family Planning
Perspectives, 28 (3), 138–143.
Sathara Z.A. and Chidambara, U.C. (1984). “Determinants of contraceptive
use”, Comparative Studies: Cross National Summaries. No. 36, 7-15.

Sayed H.A. El-Khorazaty, MN and Way, A.A. (1986). “Fertility and Family Planning in
Egypt”, 1984, Egypt, National Population Councel, Cairo.

Seltzer, J., (1998) - The origins and evolution of family planning programs in developing
countries/ Judith R. Seltzer., Published in 2002 by RAND, Santa Monica.

Shane, B. (1997). “Family Planning Saves Lives.” Washington DC: Population Reference
Bureau. Washington D.C January 1997.

Sinding S.W. (1994). “Seeking Common Ground: Unmeet need and Demographic Goals”
International Family planning Perspectives.

Snijders T. A.B. and Roel J. Bosker (1999). “An introduction to basic and advanced
multilevel modeling”. Department of Statistics, University of Poone.

60
Tularo. T., W. Deressa, A. Ali and G. Davey (2006). The role of men in contraceptive use and
fertility preference in Hosanna Town, South Ethiopia. Ethiopian J. Health Dev., 20(3) 152-
160.
UNFPA (1999). Advocating Change: “Population Empowerment Development”. United Nation
Family Planning Activities (UNFPA) 1999 Bulletin E/6000/1997 R/6000/1999.

USAID (2007). Making the care for U.S. International family Planning Assistance.
World Health Report (WHR) (2005). Make Every Mother and Child Count. World Health
Organization, Geneva, Switzerland.
WHO, Maternal Mortality in 2005. Estimates Developed by WHO, UNICEF, UNFPA and the
World Bank (Geneva: WHO, 2007).
WHO (1995) Progress report of maternal health and safe motherhood program 1993-1995:
World Health Organization, Geneva.
World Bank (WB) (2001). Women’s Empowering Women The Ethiopian Women’s
Development Initiatives Project., Geneva.

61
APPENDIX
Logistic Regression Output using Enter method
APPENDIX A National level

Case Processing Summary

Unweighted Cases(a) N Percent

Selected Cases Included in Analysis 25240 100.0
Missing Cases 0 .0
Total 25240 100.0
Unselected Cases 0 .0
Total 25240 100.0
a If weight is in effect, see classification table for the total number of cases.

Classification Table (a, b)

Predicted
Status of a couple about FP Percentage
method practice Correct
Observed Do not practice a Practice a
FP method FP method
Step 0 Status of a couple Do not practice a
about FP method FP method 19340 0 100.0
practice
Practice a FP
5900 0 .0
method
Overall Percentage 76.6
a Constant is included in the model.
b The cut value is .500

Model Summary

-2 Log Cox & Snell Nagelkerke R

Step likelihood R Square Square
1 15969.901(a) .365 .551
a Estimation terminated at iteration number 20 because maximum iterations has been reached.
Final solution cannot be found.

Omnibus Tests of Model Coefficients

Chi-square df Sig.
Step 1 Step 11479.862 33 .000
Block 11479.862 33 .000
Model 11479.862 33 .000

62
 Hosmer and Lemeshow Test

Step Chi-square df Sig.

1 162.012 8 .080

Contingency Table for Hosmer and Lemeshow Test

Status of a couple about Status of a couple about
FP method practice = Do FP method practice =
not practice a FP method Practice a FP method

Observed Expected Observed Expected Total

Step 1 1 2531 2531.000 0 .000 2531
2 2509 2520.518 20 8.482 2529
3 2396 2456.844 140 79.156 2536
4 2459 2401.698 71 128.302 2530
5 2378 2326.677 149 200.323 2527
6 2079 2189.425 443 332.575 2522
7 2055 1994.887 463 523.113 2518
8 1633 1658.244 894 868.756 2527
9 1000 949.650 1520 1570.350 2520
10 300 311.056 2200 2188.944 2500

Classification Table (a)

Predicted
Status of a couple about FP Percentage
method practice Correct
Do not
Observed practice a FP Practice a FP
method method
Step 1 Status of a couple Do not practice a FP
about FP method method 18245 1095 94.3
practice
Practice a FP method 2329 3571 60.5
Overall Percentage 86.4
a The cut value is .500

Variables in the Equation (National Level)

95.0% C.I.for
B S.E. Wald df Sig. Exp(B) EXP(B)
Lower Upper
Step RESIDENCE(1)
.816 .083 97.612 1 .000* 2.261 1.923 2.659
1(a)
AGEWOM 54.631 2 .000*
AGEWOM(1) .069 .113 .369 1 .543 1.071 .858 1.337
AGEWOM(2) .346 .050 47.744 1 .000 1.413 1.281 1.559
MARTIAL(1) -2.071 .227 83.411 1 .000* .126 .081 .197

63
 NUMCHILD 90.436 2 .000*
NUMCHILD(1) -.832 .089 87.213 1 .000 .435 .365 .518
NUMCHILD(2) -.236 .049 23.083 1 .000 .790 .718 .870
WOMEDU 3.395 2 .183
WOMEDU(1) .074 .234 .100 1 .751 1.077 .681 1.703
WOMEDU(2) .232 .206 1.261 1 .261 1.261 .841 1.889
PARTOCCUP(1) .998 .069 208.143 1 .000* 2.714 2.370 3.108
VISITEDFP(1) -.705 .113 38.964 1 .000* .494 .396 .616
DISSCPART(1) -1.202 .050 573.022 1 .000* .301 .273 .332
HUSAPPROV 773.518 2 .000*
HUSAPPROV(1) .800 .068 136.833 1 .000 2.225 1.946 2.544
HUSAPPROV(2) 1.611 .061 691.541 1 .000 5.009 4.442 5.648
MEDIA(1) -.357 .047 57.931 1 .000* .700 .638 .767
RELIGWOM 113.279 3 .000*
RELIGWOM(1) 1.782 .178 100.521 1 .000 5.941 4.193 8.416
RELIGWOM(2) 1.451 .183 62.686 1 .000 4.265 2.979 6.108
RELIGWOM(3) 1.705 .181 88.677 1 .000 5.499 3.856 7.840
REGION 307.352 10 .000*
REGION(1) -.565 .116 23.854 1 .000 .568 .453 .713
REGION(2) -.649 .138 21.970 1 .000 .523 .398 .686
REGION(3) -.722 .111 42.506 1 .000 .486 .391 .604
REGION(4) -.907 .106 73.124 1 .000 .404 .328 .497
REGION(5) -2.351 .175 179.702 1 .000 .095 .068 .134
REGION(6) -.269 .119 5.121 1 .024 .764 .605 .965
REGION(7) -.547 .114 22.958 1 .000 .579 .463 .724
REGION(8) -1.192 .127 88.610 1 .000 .304 .237 .389
REGION(9) -1.032 .120 73.402 1 .000 .356 .281 .451
REGION(10) -.885 .104 72.327 1 .000 .413 .336 .506
OCCUPWOM 4.801 2 .091
OCCUPWOM(1) .109 .050 4.796 1 .029 1.116 1.012 1.230
OCCUPWOM(2) .056 .066 .715 1 .398 1.057 .929 1.203
PARTEDU 14.073 2 .001*
PARTEDU(1) -.874 .233 14.027 1 .000 .417 .264 .659
PARTEDU(2) -.704 .207 11.549 1 .001 .494 .329 .742
KNOWLEDGE(1) 573.70
-18.668 .001 1 .974 .000 .000 .
9
WEALTH(1) -.034 .042 .646 1 .422 .967 .890 1.050
Constant .803 .332 5.843 1 .016* 2.233
-* Statistically significant at p < 0.05
a. Variable(s) entered on step 1: RESIDENCE, AGEWOM, MARTIAL, NUMCHILD, WOMEDU, PARTOCCUP, VISITEDFP,
DISSCPART, HUSAPPROV, MEDIA, RELIGWOM, REGION, OCCUPWOM, PARTEDU, KNOWLEDGE, WEALTH.

64
APPENDIX B: For Urban Areas
Case Processing Summary

Unweighted Cases(a) N Percent

Selected Cases Included in Analysis 4620 18.3
Missing Cases 0 .0
Total 4620 18.3
Unselected Cases 20620 81.7
Total 25240 100.0
a If weight is in effect, see classification table for the total number of cases.

Omnibus Tests of Model Coefficients

Chi-square df Sig.
Step 1 Step 2225.498 32 .000
Block 2225.498 32 .000
Model 2225.498 32 .000

Model Summary
-2 Log Cox & Snell Nagelkerke R
Step likelihood R Square Square
1 3507.412(a) .382 .538
a Estimation terminated at iteration number 20 because maximum iterations has been reached.
Final solution cannot be found.

Hosmer and Lemeshow Test

Step Chi-square df Sig.

1 166.417 8 .051

Contingency Table for Hosmer and Lemeshow Test

Status of a couple about Status of a couple about
FP method practice = Do FP method practice =
not practice a FP method Practice a FP method

Observed Expected Observed Expected Total

Step 1 1 420 439.779 40 20.221 460
2 380 340.181 80 119.819 460
3 180 227.779 283 235.221 463
4 233 154.937 237 315.063 470
5 86 105.128 380 360.872 466
6 41 72.781 453 421.219 494
7 71 49.451 410 431.549 481
8 29 32.207 438 434.793 467
9 0 15.457 458 442.543 458
10 0 2.298 401 398.702 401

65
 Variables in the Equation (Urban)

B S.E. Wald df Sig. Exp(B) 95.0% C.I.for EXP(B)

Lower Upper
Step1 REGION
253.786 10 .000*
(a)
REGION(1) -1.260 .185 46.385 1 .000 .284 .197 .408
REGION(2) -.370 .263 1.975 1 .160 .691 .412 1.157
REGION(3) -.616 .234 6.929 1 .008 .540 .342 .854
REGION(4) -1.117 .209 28.445 1 .000 .327 .217 .493
REGION(5) -2.830 .303 87.022 1 .000 .059 .033 .107
REGION(6) -3.471 .295 138.060 1 .000 .031 .017 .055
REGION(7) -1.794 .224 64.192 1 .000 .166 .107 .258
REGION(8) -2.411 .329 53.681 1 .000 .090 .047 .171
REGION(9) -1.223 .154 62.860 1 .000 .294 .217 .398
REGION(10) -.506 .125 16.255 1 .000 .603 .472 .771
AGEWOM 103.391 2 .000*
AGEWOM(1) -.237 .215 1.209 1 .272 .789 .518 1.203
AGEWOM(2) .941 .111 72.105 1 .000 2.562 2.062 3.183
MARTIAL(1) 2279.6
-22.456 .000 1 .992 .000 .000 .
19
NUMCHILD 76.113 2 .000*
NUMCHILD(1) -1.333 .160 69.469 1 .000 .264 .193 .361
NUMCHILD(2) -.710 .118 36.206 1 .000 .492 .390 .620
WOMEDU 39.186 2 .000*
WOMEDU(1) -.642 .367 3.057 1 .080 .526 .256 1.081
WOMEDU(2) .820 .258 10.121 1 .001 2.270 1.370 3.760
OCCUPWOM 80.477 2 .000*
OCCUPWOM(1) -.771 .137 31.540 1 .000 .462 .353 .605
OCCUPWOM(2) -1.224 .138 78.961 1 .000 .294 .224 .385
PARTEDU 13.274 2 .001*
PARTEDU(1) -.865 .367 5.555 1 .018 .421 .205 .864
PARTEDU(2) -.939 .258 13.249 1 .000 .391 .236 .648
PARTOCCUP(1) -.002 .206 .000 1 .993 .998 .667 1.495
KNOWLEDGE(1) 3569.5
-18.365 .000 1 .996 .000 .000 .
02
VISITEDFP(1) .828 .210 15.561 1 .000* 2.289 1.517 3.453
DISSCPART(1) -1.639 .139 138.854 1 .000* .194 .148 .255
HUSAPPROV 278.535 2 .000*
HUSAPPROV(1) 1.008 .164 37.682 1 .000 2.739 1.986 3.779
HUSAPPROV(2) 2.331 .153 233.282 1 .000 10.287 7.627 13.873
MEDIA(1) -.557 .116 23.102 1 .000* .573 .456 .719
RELIGWOM 133.782 3 .000*
RELIGWOM(1) 3.765 .375 100.651 1 .000 43.147 20.680 90.024
RELIGWOM(2) 5.107 .442 133.721 1 .000 165.246 69.531 392.724

66
 RELIGWOM(3) 3.875 .392 97.909 1 .000 48.200 22.370 103.854
WEALTH(1) .028 .094 .089 1 .765 1.029 .855 1.237
Constant 2279.6 604394
20.220 .000 1 .993
19 683.070
-* Statistically significant at p < 0.05
a Variable(s) entered on step 1: REGION, AGEWOM, MARTIAL, NUMCHILD, WOMEDU, OCCUPWOM, PARTEDU,
PARTOCCUP, KNOWLEDGE, VISITEDFP, DISSCPART, HUSAPPROV, MEDIA, RELIGWOM, WEALTH.

APPENDIX C: For Rural Areas

Case Processing Summary

Unweighted Cases(a) N Percent

Selected Cases Included in Analysis 20620 81.7
Missing Cases 0 .0
Total 20620 81.7
Unselected Cases 4620 18.3
Total 25240 100.0
a If weight is in effect, see classification table for the total number of cases.

Omnibus Tests of Model Coefficients

Chi-square df Sig.
Step 1 Step 4429.873 31 .000
Block 4429.873 31 .000
Model 4429.873 31 .000

Model Summary

-2 Log Cox & Snell Nagelkerke R

Step likelihood R Square Square
1 11653.848(a) .193 .357
a Estimation terminated at iteration number 20 because maximum iterations has been reached.
Final solution cannot be found.

Hosmer and Lemeshow Test

Step Chi-square df Sig.
1 105.473 8 .100

Contingency Table for Hosmer and Lemeshow Test

Status of a couple about Status of a couple about
FP method practice = Do FP method practice =
not practice a FP method Practice a FP method

Observed Expected Observed Expected Total

Step 1 1 2056 2056.000 0 .000 2056

67
 2 2065 2065.000 0 .000 2065
3 2034 2025.426 20 28.574 2054
4 1961 1978.176 100 82.824 2061
5 1968 1933.980 83 117.020 2051
6 1961 1897.595 100 163.405 2061
7 1752 1807.428 314 258.572 2066
8 1567 1679.909 494 381.091 2061
9 1531 1483.890 538 585.110 2069
10 1005 972.595 1071 1103.405 2076

Variables in the Equation (Rural)

95.0% C.I.for
B S.E. Wald df Sig. Exp(B) EXP(B)
Lower Upper
Step REGION 262.73
9 .000*
1(a) 7
REGION(1) 2.048 .264 60.341 1 .000 7.750 4.623 12.993
REGION(2) 1.605 .277 33.669 1 .000 4.979 2.895 8.563
REGION(3) 1.693 .259 42.649 1 .000 5.438 3.271 9.040
REGION(4) 1.533 .257 35.548 1 .000 4.632 2.799 7.668
REGION(5) -.382 .346 1.218 1 .270 .683 .347 1.344
REGION(6) 2.411 .264 83.492 1 .000 11.148 6.646 18.699
REGION(7) 1.921 .266 52.308 1 .000 6.829 4.058 11.494
REGION(8) 1.289 .272 22.524 1 .000 3.630 2.131 6.182
REGION(9) 1.459 .290 25.352 1 .000 4.301 2.437 7.588
AGEWOM 9.079 2 .011*
AGEWOM(1) -.037 .139 .069 1 .792 .964 .735 1.265
AGEWOM(2) .155 .059 7.014 1 .008 1.168 1.041 1.310
MARTIAL(1) 1554228
18.862 4570.759 .000 1 .997 .000 .
07.580
NUMCHILD 18.028 2 .000*
NUMCHILD(1) -.550 .130 17.892 1 .000 .577 .447 .745
NUMCHILD(2) -.092 .058 2.561 1 .110 .912 .815 1.021
WOMEDU 14.803 2 .001*
WOMEDU(1) 1483836.
14.210 489.366 .001 1 .977 .000 .
840
WOMEDU(2) 847709.4
13.650 489.366 .001 1 .978 .000 .
33
OCCUPWOM 35.318 2 .000*
OCCUPWOM(1) .177 .059 8.971 1 .003 1.193 1.063 1.340
OCCUPWOM(2) .509 .088 33.704 1 .000 1.663 1.401 1.974
PARTEDU 26.934 2 .000*
PARTEDU(1) -14.922 489.366 .001 1 .976 .000 .000 .
PARTEDU(2) -14.152 489.366 .001 1 .977 .000 .000 .
PARTOCCUP(1) 188.45
1.073 .078 1 .000* 2.923 2.508 3.407
1
KNOWLEDGE(1) -27.228 692.281 .002 1 .969 .000 .000 .
VISITEDFP(1) 145.49
-1.710 .142 1 .000* .181 .137 .239
4
DISSCPART(1) 490.36
-1.257 .057 1 .000* .285 .255 .318
4

68
 HUSAPPROV 484.19
2 .000*
8
HUSAPPROV(1) .718 .078 84.572 1 .000 2.051 1.760 2.390
HUSAPPROV(2) 436.31
1.483 .071 1 .000 4.407 3.834 5.065
4
MEDIA(1) -.363 .054 45.688 1 .000* .695 .626 .773
RELIGWOM 75.936 3 .000*
RELIGWOM(1) 1.346 .184 53.277 1 .000 3.840 2.676 5.512
RELIGWOM(2) .936 .188 24.886 1 .000 2.549 1.765 3.682
RELIGWOM(3) 1.394 .188 54.746 1 .000 4.033 2.787 5.835
WEALTH(1) -.081 .050 2.657 1 .103 .922 .836 1.017
Constant -21.062 4570.759 .000 1 .996 .000
-* Statistically significant at p < 0.05
a Variable(s) entered on step 1: REGION, AGEWOM, MARTIAL, NUMCHILD, WOMEDU, OCCUPWOM,
PARTEDU, PARTOCCUP, KNOWLEDGE, VISITEDFP, DISSCPART, HUSAPPROV, MEDIA, RELIGWOM,
WEALTH.

Result of diagnostic tests for outliers and influential value

Descriptive Statistics

N Minimum Maximum
Analog of Cook's influence statistics 25240 .00000 .04333
Leverage value 25240 .00001 .01319
Normalized residual 25240 -5.93748 7.77937
DFBETA for constant 25240 -.03855 .02499
DFBETA for REGION(1) 25240 -.00586 .00469
DFBETA for REGION(2) 25240 -.00553 .00993
DFBETA for REGION(3) 25240 -.00457 .00518
DFBETA for REGION(4) 25240 -.00422 .00465
DFBETA for REGION(5) 25240 -.00619 .02333
DFBETA for REGION(6) 25240 -.00643 .00502
DFBETA for REGION(7) 25240 -.00447 .00468
DFBETA for REGION(8) 25240 -.00838 .01008
DFBETA for REGION(9) 25240 -.00497 .00675
DFBETA for REGION(10) 25240 -.00456 .00593
DFBETA for AGEWOM(1) 25240 -.00638 .00840
DFBETA for AGEWOM(2) 25240 -.00178 .00188
DFBETA for MARTIAL(1) 25240 -.01656 .03639
DFBETA for NUMCHILD(1) 25240 -.00557 .00603
DFBETA for NUMCHILD(2) 25240 -.00188 .00160
DFBETA for WOMEDU(1) 25240 -.02167 .01321
DFBETA for WOMEDU(2) 25240 -.02113 .01331
DFBETA for OCCUPWOM(1) 25240 -.00150 .00175
DFBETA for OCCUPWOM(2) 25240 -.00240 .00362
DFBETA for PARTEDU(1) 25240 -.01323 .02125
DFBETA for PARTEDU(2) 25240 -.01320 .02094

69
DFBETA for PARTOCCUP(1) 25240 -.00408 .00373
DFBETA for KNOWLEDGE(1) 25240 -.00381 .00064
DFBETA for VISITEDFP(1) 25240 -.01078 .01143
DFBETA for DISSCPART(1) 25240 -.00181 .00166
DFBETA for HUSAPPROV(1) 25240 -.00391 .00183
DFBETA for HUSAPPROV(2) 25240 -.00308 .00201
DFBETA for MEDIA(1) 25240 -.00205 .00157
DFBETA for RELIGWOM(1) 25240 -.02798 .01911
DFBETA for RELIGWOM(2) 25240 -.02969 .01864
DFBETA for RELIGWOM(3) 25240 -.02776 .01902
DFBETA for WEALTH(1) 25240 -.00129 .00126
DFBETA for RESIDENCE(1) 25240 -.00452 .00741
Valid N (listwise) 25240

Change in Deviance against Predicted probabilities

70
 o
o

..
,•
,
-
>
•"< ,-
,-
•
;;
•
c

0
",- •
o

, ,. ,.
Expos ure to iIIny media

,- 0

I
I
0
,- •
•8
..
,•
>
,-
I
0
0
0

•"< ,-
•
"'•
C
_1 .00000

_2.00000
I
0
8
0
I
0

0
_3.00000

,
"M31e p3rtner/Husb3nd 3pproves FP
"

71
 ,- 0
0

,- 0

•
..,•
•
>

.,•
0
0
,-
,- I
•
c
.,-
.,- ft
I
0
,-
u
" "
,
"
Type of place of resldenu

..,•
>
•o
u

.,•
0 .00000 I
c•
_1 .00000

_2.00000
I
i
o
_3.00000

Discussed FP with p3rtner

72
OUTPUT FOR MULTILEVEL LOGISTIC REGRESSION
APPENDIX D: EMPTY MODEL

Model and Data Descriptions

Sampling Distribution = Bernoulli

Link Function = Logistic
Weight Variable = Weights
Number of Level-2 Units = 11
Number of Level-1 Units = 25240
Number of Level-1 Units per Level-2 Unit =
1960 1400 3760 4660 1820 2140 3720 1620 1000 1600 1560

o==================================================o
| Results for the model without any random effects |
o==================================================o

Goodness of fit statistics

Statistic Value DF Ratio

--------- ----- -- -----
Likelihood Ratio Chi-square 27449.7626 25238 1.0876
Pearson Chi-square 25240.0000 25238 1.0001
Deviance 27449.7626
Akaike Information Criterion 27453.7626
Schwarz Criterion 27470.0350

Estimated regression weights

Standard
Parameter Estimate Error z- Value P- Value
--------- -------- -------- ------- -------
intercept 0.0000 0.0000
intercept1 1.1872 0.0149 79.8255 0.0000

o=====================================================o
| Results for the model with fixed and random effects |
o=====================================================o

Total number of (macro) iterations 4

Goodness of fit statistics

Statistic Value DF Ratio

--------- ----- -- -----
Likelihood Ratio Chi-square 23488.9731 25237 0.9307
Pearson Chi-square 25177.2204 25237 0.9976
Deviance 23488.9731
Akaike Information Criterion 23492.9731
Schwarz Criterion 23509.2455

73
 Estimated regression weights

Standard
Parameter Estimate Error z-Value P-Value
--------- ------ -------- ------- -------
intercept 0.0000 0.0000
intercept1 1.2182 0.0164 74.3669 0.0000

Deviance-based Chi-Square test for significance of random effects

NDF Chi-Square P-Value

--- ---------- -------
1 3960.7895 0.0000*

Estimated level 2 variances and covariance’s

Standard
Parameter Estimate Error z-Value P-Value
--------- -------- -------- ------- -------
intercept/intercept 1.3480 0.4239 3.1800 0.0015

Level-2 covariance matrix

intercept
intercept 1.348047

Level-2 correlation matrix

intercept
intercept 1.000000

APPENDIX E: RANDOM INTERCEPT AND FIXED SLOPES VARIABLES

o==================================================o
| Results for the model without any random effects |
o==================================================o
Goodness of fit statistics

Statistic Value DF Ratio

--------- ----- -- -----
Likelihood Ratio Chi-square 18029.8087 25234 0.7145
Pearson Chi-square 25199.2362 25234 0.9986
Deviance 18029.8087
Akaike Information Criterion 18041.8087
Schwarz Criterion 18090.6258

Estimated regression weights

Standard
Parameter Estimate Error z-Value P-Value
--------- -------- -------- ------- -------
Intercept -4.1258 0.1058 -38.9919 0.0000
VISITE_FP -0.8943 0.1154 -7.7495 0.0000
DISSCPAR -1.2047 0.0473 -25.4787 0.0000
HUSAPPRO 1.4786 0.0404 36.5703 0.0000
MEDIA -0.7236 0.0405 -17.8460 0.0000
RESIDE 2.1014 0.0451 46.6303 0.0000

74
o=====================================================o
| Results for the model with fixed and random effects |
o=====================================================o
Total number of (macro) iterations 12

Goodness of fit statistics

Statistic Value DF Ratio

--------- ----- -- -----
Likelihood Ratio Chi-square 17679.9091 25233 0.7007
Pearson Chi-square 25325.5563 25233 1.0037
Deviance 17679.9092
Akaike Information Criterion 17691.9092
Schwarz Criterion 17740.7263

Estimated regression weights (interpretation)

Standard
Parameter Estimate Error z-Value P-Value
--------- -------- -------- ------- -------
Intercept -3.5743 0.1080 -33.0809 0.0000
VISIT_FP -0.7758 0.1157 -6.7034 0.0000
DISSCPAR -1.2715 0.0472 -26.9396 0.0000
HUSAPPRO 1.3792 0.0409 33.7433 0.0000
MEDIA -0.7168 0.0410 -17.4637 0.0000
RESIDE 1.9089 0.0462 41.2895 0.0000

Deviance-based Chi-Square test for significance of random effects

NDF Chi-Square P -Value

--- ---------- -------
1 349.8995 0.0000*

Estimated level 2 variances and covariances

Standard
Parameter Estimate Error z-Value P-Value
--------- -------- -------- ------- -------
Intercept/intercept 0.2571 0.0816 3.1486 0.0016

Level-2 covariance matrix

intercept
intercept 0.257050

Level-2 correlation matrix

intercept
intercept 1.000000

75
APPENDIX F: RANDOM INTERCEPT AND RANDOM SLOPES VARIABLES

o==================================================o
| Results for the model without any random effects |
o==================================================o
Goodness of fit statistics

Statistic Value DF Ratio

--------- ----- -- -----
Likelihood Ratio Chi-square 18029.8087 25234 0.7145
Pearson Chi-square 25199.2362 25234 0.9986
Deviance 18029.8087
Akaike Information Criterion 18041.8087
Schwarz Criterion 18090.6258

Estimated regression weights

Standard
Parameter Estimate Error z -Value P -Value
--------- -------- -------- ------- -------
intercept -4.1258 0.1058 -38.9919 0.0000
RESIDE 2.1014 0.0451 46.6303 0.0000
VISIT_FP -0.8943 0.1154 -7.7495 0.0000
DISSCPAR -1.2047 0.0473 -25.4787 0.0000
HUSAPPRO 1.4786 0.0404 36.5703 0.0000
MEDIA -0.7236 0.0405 -17.8460 0.0000
o=====================================================o
| Results for the model with fixed and random effects |
o=====================================================o

Total number of (macro) iterations 101

Goodness of fit statistics

Statistic Value DF Ratio

--------- ----- -- -----
Likelihood Ratio Chi-square 16671.0867 25228 0.6608
Pearson Chi-square 24141.5301 25228 0.9569
Deviance 16671.0590
Akaike Information Criterion 16683.0590
Schwarz Criterion 16731.8762

Estimated regression weights (interpretation)

Standard
Parameter Estimate Error z-Value P-Value
--------- -------- -------- ------- -------
Intercept -4.2287 0.1101 -38.4217 0.0000
RESIDE 2.1290 0.0478 44.5611 0.0000
VISIT_FP -2.0345 0.1198 -16.9824 0.0000
DISSCPAR -1.7434 0.0480 -36.2938 0.0000
HUSAPPRO 1.6852 0.0420 40.0884 0.0000
MEDIA -0.6638 0.0421 -15.7743 0.0000

76
Deviance-based Chi-Square test for significance of random effects

NDF Chi-Square P-Value

--- ---------- -------
21 1358.7496 0.0000*

Estimated level 2 variances and covariances

Standard
Parameter Estimate Error z-Value P-Value
--------- -------- -------- ------- -------
intercept/intercept 5.5077 1.7535 3.1410 0.0017
RESIDE/intercept -1.8754 0.7101 -2.6410 0.0083
RESIDE/RESIDE 1.2381 0.3888 3.1844 0.0015
VISIT_FP/intercept 0.7902 1.5658 0.5047 0.6138
VISIT_FP/RESIDE 1.4271 0.8153 1.7503 0.0801
VISIT_FP/VISIT_FP 8.7196 2.8732 3.0348 0.0024
DISSCPAR/intercept 1.4357 0.6085 2.3594 0.0183
DISSCPAR/RESIDE -0.1596 0.2503 -0.6376 0.5238
DISSCPAR/VISIT_FP 1.8967 0.7899 2.4011 0.0163
DISSCPAR/DISSCPAR 0.9700 0.3075 3.1544 0.0016
HUSAPPRO/intercept -1.2805 0.6130 -2.0888 0.0367
HUSAPPRO/RESIDE -0.1747 0.2572 -0.6794 0.4969
HUSAPPRO/VISIT_FP -2.3455 0.8541 -2.7462 0.0060
HUSAPPRO/DISSCPAR -0.8061 0.2852 -2.8262 0.0047
HUSAPPRO/HUSAPPRO 1.0199 0.3221 3.1662 0.0015
MEDIA/intercept -2.5071 0.9289 -2.6990 0.0070
MEDIA/RESIDE 0.8698 0.3887 2.2377 0.0252
MEDIA/VISIT_FP 0.9318 0.9548 0.9759 0.3291
MEDIA/DISSCPAR -0.1298 0.3146 -0.4124 0.6800
MEDIA/HUSAPPRO 0.4159 0.3390 1.2268 0.2199
MEDIA/MEDIA 1.9157 0.6097 3.1420 0.0017

Level-2 covariance matrix

intercept RESIDE VISIT_FP DISSCPAR HUSAPPRO MEDIA

intercept 5.507681
RESID -1.875390 1.238128
VISIT_FP 0.790197 1.427065 8.719584
DISSCPAR 1.435746 -0.159564 1.896726 0.970009
HUSAPPRO -1.280482 -0.174733 -2.345508 -0.806114 1.019946
MEDIA -2.507134 0.869833 0.931815 -0.129753 0.415866 1.915709

Level-2 correlation matrix

intercept RESIDE VISIT_FP DISSCPAR HUSAPPRO MEDIA

intercept 1.000000
RESID -0.718165 1.000000
VISIT_FP 0.114026 0.434323 1.000000
DISSCPAR 0.621163 -0.145601 0.652182 1.000000
HUSAPPRO -0.540257 -0.155490 -0.786503 -0.810438 1.000000
MEDIA -0.771842 0.564792 0.227991 -0.095184 0.297509 1.000000

77