Effect of Poverty on Healthcare Utilization, Choice

EFFECT OF POVERTY ON HEALTHCARE UTILIZATION, CHOICE

OF HEALTHCARE PROVIDERS AND HEALTH STATUS IN KENYA

PETER KATUNDU MUSYOKA

K96/CTY/24439/2013

A THESIS SUBMITTED TO THE SCHOOL OF ECONOMICS IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE AWARD OF THE DOCTOR OF PHILOSOPHY DEGREE IN ECONOMICS OF KENYATTA UNIVERSITY

APRIL, 2019

DECLARATION This thesis is my original work and has not been presented for a degree in any other University or any other award.

Signature……………………………………..Date……………………………… Peter Katundu Musyoka B.Ed (Hons), University of Nairobi M.A (Economics), University of Malawi Reg. No. K96/CTY/24439/2013

We confirm that this thesis was developed by the candidate under our supervision as University Supervisors

Signature……………………………………..Date………………………………

Dr. Julius Korir Department of Economic Theory Kenyatta University Nairobi, Kenya.

Signature……………………………………..Date………………………………

Dr. Jacob Omolo Department of Applied Economics Kenyatta University Nairobi, Kenya

Signature……………………………………..Date……………………………… Dr. Charles C. Nzai Department of Applied Economics Kenyatta University Nairobi, Kenya

DEDICATION To my wife Caroline Katundu and our children Gabriel Mutuku, Grace Ndinda and Geoffrey Musyoka.

iii

ACKNOWLEDGEMENTS I am very grateful and thankful to God for according me good state of health and energy during the entire PhD study period. Generous financial assistance from

Kenyatta University during the PhD study period is acknowledged and highly appreciated. I am grateful for invaluable comments and guidance received from my supervisors Dr. Julius Korir, Dr. Jacob Omolo and Dr. Charles Nzai. I would like also to recognize and appreciate all the lecturers from School of Economics,

Kenyatta University, who taught me in the coursework component of this PhD study. I would also like to thank other members of staff whose advice and word of encouragement impacted this work greatly. Special thanks go to Dr. Diana Ngui for her comments throughout the research portion of this PhD study. I would also like to appreciate resource persons especially Dr. James Ciera and Dr. Maurice

Mutisya who played various roles during writing of this thesis. I would also like to appreciate my classmates with whom we held discussions and encouraged each other during our PhD studies. I am also grateful to all participants at the School of

Economics Postgraduate Seminar for their comments during presentation of this

PhD work. Last but not least, I am greatly indebted to my family members for their prayers, support and understanding during the entire period of my PhD studies. In a special way I would like to acknowledge and appreciate my uncle,

Mr. Stephen Muthiani for support and encouragement throughout the study period. My wife Caroline and our children Gabriel, Grace and Geoffrey, I have no enough words to say thank you. I appreciate everything you did for me during my studies. I also appreciate my parents, Mr. and Mrs. A. Musyoka Katundu, brothers and sisters for the support and encouragement during my studies.

TABLE OF CONTENTS DECLARATION ...... ii DEDICATION ...... iii ACKNOWLEDGEMENTS ...... iv TABLE OF CONTENTS ...... v LIST OF TABLES ...... vii ABBREVIATIONS AND ACRONYMS ...... viii OPERATIONAL DEFINITION OF TERMS ...... x ABSTRACT ...... xi CHAPTER ONE: INTRODUCTION ...... 1 1.1 Background ...... 1 1.1.1 Policy Landscape Relating to Health and Poverty Linkages ...... 1 1.1.2 Trends in Poverty and Poverty Reduction Strategies in Kenya ...... 11 1.1.3 Health Policies and Health Indicators in Kenya...... 23 1.1.4 Health Care Utilization in Kenya ...... 36 1.2 Statement of the problem ...... 38 1.3 Research questions ...... 41 1.4 Objectives of the Study ...... 41 1.5 Significance of the Study ...... 42 1.6 Scope of the Study...... 42 1.7 Organization of the Study ...... 43 CHAPTER TWO: LITERATURE REVIEW ...... 44 2.1 Theoretical Literature ...... 44 2.1.1 Neo-materialist Hypothesis ...... 44 2.1.2 Behavioral Model of Health Care Utilization ...... 45 2.1.3 Grossman‟s Model of Human Capital ...... 48 2.1.4 Acton‟s Utility Maximization Model of Health Care Demand ...... 51 2.2 Empirical Literature ...... 54 2.3 Overview of Literature ...... 81 CHAPTER THREE: METHODOLOGY ...... 86 3.1 Research Design ...... 86 3.2 Theoretical Framework ...... 86 3.2.1 A Theoretical Framework for Production of Health ...... 86 3.2.2 A Theoretical Framework for Choice of Healthcare Provider ...... 89 3.3 Model Specification ...... 92 3.3.1 Effect of poverty on health care utilization in Kenya ...... 92 3.3.2 Effect of poverty on choice of health care providers in Kenya ...... 96 3.3.3 Effect of poverty on health status in Kenya ...... 102 3.4 Definition and Measurement of Variables ...... 106 3.5 Diagnostic Tests ...... 109 3.6 Data type and source ...... 109 3.7 Data Analysis ...... 110 CHAPTER FOUR: EMPIRICAL FINDINGS AND DISCUSSION ...... 112 4.1 Descriptive Statistics ...... 112 4.2 Empirical Findings ...... 120 4.2.1 Effect of poverty on health care utilization in Kenya ...... 120

4.2.2 Effect of poverty on choice of health care providers in Kenya ...... 137 4.2.3 Effect of poverty on health status in Kenya ...... 150 CHAPTER FIVE: SUMMARY, CONCLUSIONS AND POLICY IMPLICATIONS ...... 171 5.1 Summary ...... 171 5.2 Conclusions ...... 174 5.3 Policy Implications ...... 174 5.4 Contribution to knowledge ...... 179 5.5 Areas for further research ...... 180 REFERENCES ...... 181 Appendix ...... 191

LIST OF TABLES Table 1.1: Prevalence of Absolute Poverty Overtime in Kenya ...... 16 Table 1.2: Selected Health Targets as Per NHSSP I (1999-2004) ...... 27 Table 1.3: Kenya‟s Key Health Indicators and Global Development Targets for Selected Years ...... 34 Table 1.4: Comparison of Kenya‟s Key Health Indicators with Selected Regions ..35 Table 1.5: Proportion of People Reporting Illness and Total Number of Outpatient Visits and Utilization Rates (2003-2013) ...... 37 Table 4.1: Descriptive Statistics for Continuous and Count Variables ...... 112 Table 4.2: Summary Statistics for Discrete/Categorical Variables ...... 115 Table 4.3: Poisson, NBRM and ZIP Models Selection based on Vuong and LR tests ...... 121 Table 4.4: Regression Results of Poverty Status Model measured at household level ...... 125 Table 4.5: Regression results of NBRM, 2SRI and CFA ...... 127 Table 4.6: Average Marginal Effects for the Healthcare Provider Choice Model .139 Table 4.7: Average Marginal Effects of Probability of Reporting Own Health as Poor ...... 152 Table 4.8: Average Marginal Effects of Probability of Reporting Own Health as Very Good ...... 161 Table A1: Validity test of instrumental variable in health care utilization model ..191 Table A2: Validity test of instrumental variable in health care provider choice model...... 192 Table A3: Validity test of instrumental variable in health status model ...... 193 Table A4: Average Marginal Effects for the Healthcare Provider Choice Model (Government) ...... 194 Table A5: Average Marginal Effects for the Healthcare Provider Choice Model (Private)...... 195 Table A6: Average Marginal Effects for the Healthcare Provider Choice Model (Mission) ...... 196 Table A7: Average Marginal Effects for the Healthcare Provider Choice Model (Others) ...... 197 Table A8: Average Marginal Effects of Probability of Reporting Own Health as Satisfactory ...... 198 Table A9: Average Marginal Effects of Probability of Reporting Own Health as Very Good ...... 199

vii

ABBREVIATIONS AND ACRONYMS

2SRI Two-Stage Residual Inclusion

AIDs Acquired Immune Deficiency Syndrome

APHRC African Population Health Research Center

BMI Body Mass Index

CFA Control Function Approach

HIV Human Immunodeficiency Virus

KDHS Kenya Demographic Health Survey

KHPF Kenya Health Policy Framework

KIHBS Kenya Integrated Household Budget Survey

KIPPRA Kenya Institute for Public Policy Research and Analysis

KNBS Kenya National Bureau of Statistics

MDGs Millennium Development Goals

MTP Medium Term Plan

NCAPD National Coordinating Agency for Population and

Development

NGO Non-Governmental Organization

NHSSP National Health Sector Strategic Plan

OAU Organization of African Unity

ODA Official Development Assistance

OLS Ordinary Least Squares

PRSP Poverty Reduction Strategic Paper

SDGs Sustainable Development Goals

viii

VECM Vector Error Correction Model

WHO World Health Organization

WMS Welfare Monitoring Survey

OPERATIONAL DEFINITION OF TERMS Poverty: Inability to attain a predetermined minimum level of consumption at

which basic needs of a society or country are assumed to be satisfied.

Absolute poverty: Minimal requirements necessary to afford basic standards of

food, clothing, healthcare and shelter.

Health: State of physical, mental and social wellbeing and not merely absence of

disease or infirmity.

Health status: Is individual‟s perception of his/her relative level of wellness and

illness, taking into consideration the presence of biological or

physiological dysfunction, symptoms and functional impairment.

General health: This is health estimated at the population level as opposed to

specific groups such as infants, children, or mothers.

Health care: Prevention, treatment, and management of illness and the

preservation of mental and physical wellbeing through the services offered

by medical and allied health professions.

Health care utilization: Usage of health services either for prevention, treatment,

or management of illness or injury.

Access to health care: This is the ease with which individuals can obtain needed

health services

Health care providers: Individuals or institutions involved in the provision of

health services.

ABSTRACT The importance of good health cannot be under estimated. However, presence of high poverty rates can lead to under utilization or lack of utilization of health care thus hindering achievement of good health. Thus, poverty reduction and improvement of health care utilization are important in ensuring enjoyment of good health. Despite Kenya‟s commitment to reduce poverty and improve health status of her citizens, between 1982 and 2014, poverty remained high above 40 per cent. However, in 2015/2016, poverty was estimated to have reduced to 36.1 per cent. This was against the Millennium Development Goals target of halving poverty by 2015 and the Sustainable Development Goals target of eradicating poverty by 2030. Kenya‟s health indicators have also not been impressive. Infant mortality rate, for instance, stood at 39 deaths per 1,000 live births in 2014 against Millennium Development Goals target of 22 by 2015. Maternal mortality rate remained high at 362 deaths per 100,000 live births in 2014 against Millennium Development Goals target of 147 by 2015. The Sustainable Development Goals target is to have less than 70 deaths per 100,000 live births by 2030. This poor performance in health indicates that the country needs to address the health challenges otherwise it will miss on the development goals by 2030. Health care utilization has also been low. Household members who reported illness and never sought health care stood at 22.8 per cent in 2003, before dropping to 16.7 per cent and 12.7 per cent in 2007 and 2013, respectively. Those who fell sick and reported lack of finances as the main reason for not seeking medical attention constituted 44 per cent in 2003, 38 per cent in 2007 and 21.4 per cent in 2013. These statistics point to poor health care utilization due to poverty. The aim of this study, therefore, was to investigate the effect of poverty on healthcare utilization, choice of healthcare providers and health status in Kenya. The study employed a non-experimental cross-sectional research design. The study used the Kenya Household Health Expenditure and Utilization Survey dataset of 2013. To achieve objective one, the study used Negative Binomial Regression Model, while Multinomial probit was used to address objective two. Objective three was addressed using Ordered probit model. In all the three objectives, Two Stage Residual Inclusion and Control Function models were used to control for possible endogeneity and unobserved heterogeneity. Study findings showed that increase in wealth increases health care utilization. Further, the results revealed that as wealth increased, the probability of visiting private hospitals increased while those of visiting government, mission and other health facilities declined. The results also revealed that, those with higher wealth index were more likely to report better health status compared to those with lower wealth index. The results, therefore, indicates that increase in wealth increases healthcare utilization, motivates individuals to seek healthcare from providers considered to offer high quality health services and improves health status. Thus, although Kenya missed some health related Millennium Development Goals, if poverty is addressed, the country can do better in its efforts to achieving the Sustainable Development Goals and the country‟s development plan, Kenyan Vision 2030.

CHAPTER ONE

INTRODUCTION 1.1 Background

1.1.1 Policy Landscape Relating to Health and Poverty Linkages

Health is a fundamental human right, a valued asset and a prerequisite for improved productivity. Attainment of the highest possible level of health is, therefore, one of the most important social goals world-wide. Health status of an individual or a population refers to all forms of the individual‟s or population‟s health (Awiti, 2014). The main indicators of general health status at individual level includes self reported general health status, self reported morbidity, illness and normal activity. Others include self reported physical functioning, nutrition based indicators, haemoglobin levels and whether one is suffering from an acute or chronic disease (Mwabu, 2008; Strauss & Thomas, 1998, 2007).

Individual‟s health is important as it affects all aspects of his/her life. The effects of good health accrues not only to an individual, but are also extended to the family, community and the nation. Poor health reduces working hours; lowers production and productivity; reduces Gross Domestic Product (GDP) and savings, and increases health care expenses. Health expenditures, are borne by the individuals and also by the society at large (Bourne, 2009; Lawson, 2004).

Because of increased health care expenditures, ill-health leads to re-allocation of expenditure from social development sectors such as education to health care.

This switching of costs due to ill health can lead to or increase poverty for an individual or his/her family (Awiti, 2014; Bourne, 2009; Lawson, 2004). Thus,

health is a key determinant for social and economic development. Therefore, pursuit for long life, need to be supported by a healthy population. It is the interrelations among health, social and economic development that account for a demand in health care services (Bourne, 2009).

Health of an individual can be influenced by many factors among them unobservable biological factors, health-related behaviours, non-medical market inputs, market medical inputs and various socio-economic factors (Awiti, 2014).

In case of poor health, there is a wide variety of actions that an individual could take to improve it. The individual may decide to choose self medication, consult a traditional healer, or seek treatment from a private or public health facility or a pharmacist. The particular action taken depends on individual characteristics, health care provider characteristics, societal factors, and geographical factors.

However, a major individual factor is the affordability of the required health care

(Asfaw, 2003; Awiti, 2014). Hence, poor people may opt not to seek any treatment or if they do, they may visit cheap health care facilities that may not offer the best health care. Alternatively, they may not utilize fully the minimum required health services leading to more complications.

Thus, health care seeking behavior of individuals is an indicator of their willingness to improve their health and preserve life (Bourne, 2009). Access to health care is, therefore, a crucial determinant of health worldwide. However, there exist potential barriers to accessing health care. These barriers include distance and travel costs to health facilities, socio-cultural factors, poverty and cost of service especially in resource challenged countries like Kenya. Other

barriers may include functional accessibility to health service and inequalities in health services and infrastructural distribution across the country (Yawson, Malm,

Adu, Wontumi, & Biritwum, 2012).

Poverty, which manifests itself in various forms such as high mortality rate, lack of access to basic education, lack of safe drinking water, lack of main health facilities and shelter can lead to ill health (Nafula, Onsomu, Mwabu, & Muiruri,

2005; Nkpoyen, Eteng, & Abul, 2014; Salihi, Madeline, & Norlaila, 2012). On one hand, poverty can create barriers to accessing and utilizing health services.

This is because poverty may lead people not to utilize or under utilize health care.

Non-utilization or under utilization of health care may in turn lead to poor health.

On the other hand, poor health can lead to poverty due to higher health costs and inability or reduced ability to work hence losing income (Buddelmeyer & Cai,

2009; Gupta, Wit, & McKeown, 2007; Peters et al., 2008). Ill health can also lead to increased school absenteeism and reduced ability by children to learn when they do attend school. Reduced ability to learn and increased absenteeism by children may later in life, negatively impact their productivity levels. In turn, reduced productivity levels will affect children‟s health outcomes (Peters et al.,

2008; Salihi et al., 2012). Thus, the vicious cycle of ill-health and poverty continues to propagate.

Despite the realization of the importance of health, that requires the actions of many other social and economic sectors in addition to the health sector, there exists gross inequality in the health status of individuals. The inequality applies to people across regions and countries. This is particularly between developed and

developing countries as well as within countries. The inequality in health status may not be acceptable socially, economically and politically (Edeme, Ifelunini, &

Obinna, 2014; WHO, 1978). Edeme et al. (2014) observes that of the 30 countries with the world‟s highest child mortality rates, 27 are in sub-Saharan Africa (SSA).

Wagstaff, Claeson, Hecht, Gottret, and Fang (2006) makes a similar observation by indicating that, of the 11 million under five mortality rates that occurred in the world in the year 2000, less than one per cent occurred in high-income countries.

This is compared to 42 per cent in SSA, 35 per cent in South Asia, and 13 per cent in East Asia (Wagstaff et al., 2006). The authors further indicated that of the 3.1 million people who died from Human Immunodeficiency Virus and Acquired

Immune Deficiency Syndrome (HIV/AIDs) in 2003, almost all (99 per cent) were in the developing world with SSA contributing 74 per cent.

The international community has over time recognized the important linkage between health and poverty. This has seen the world community put effort towards addressing the twin challenges of health and poverty (Lawson, 2004;

Wagstaff et al. 2006). It is part of these efforts that saw the World Health

Organization (WHO) and the United Nations Children‟s Fund (UNICEF) convene the Alma Ata Conference in 1978 (WHO, 1978). The Alma-Ata Conference participants undertook to tackle “politically, socially and economically unacceptable” health inequalities in all countries (WHO, 1978:1). The declaration outlined social justice and the right to better health for all as values that needed to be pursued. It is this declaration that entrenched health as a human right (Gillam,

2008; WHO, 1978). Primary health care, which include primary medical services

and activities tackling determinants of ill health was considered the key to delivering health for all by the year 2000 (Gillam, 2008; WHO, 1978).

Early efforts to expand primary health care as per the declaration in the 1970s and early 1980s were, however, undermined in many parts of the developing world by sharp reductions in public spending, emerging diseases, economic crisis and political instability (Gillam, 2008). Further, the Alma Ata political and social goals aggravated early ideological antagonism. Therefore, in the market oriented capitalist countries, the goals were never fully embraced (Gillam, 2008). The opposition and lack of commitment to the goals made most countries fail to provide even the limited packages of primary health care (Gillam, 2008). Lack of accessibility due to geographical and financial constraints, scarcity of resources, and shortage of drugs, equipment and human capital left health care services in many countries limited in terms of coverage, range and impact (Gillam, 2008).

Thus, primary health care, which was the driving force build upon Alma-Ata declaration was hardly achieved.

In the 1990s, there was renewed enthusiasm of the importance of health in development. During this period, there was increased development assistance in health in real terms. The development assistance for health rose from 4 per cent in

1992 to 7 per cent in 1998 (Faure, 2001). About a third of the development assistance for health was meant for population and reproductive health while another third was for basic health (Faure, 2001). The World Bank increased its global lending to health from 8 per cent in 1992 to about 14 per cent in 1998

(Faure, 2001; Wagstaff et al., 2006). Due to increased lending, global total debt

stock rose from USD 1,712 billion in 1992 to USD 2,567 billion in 1998. Nearly half of the total debt stock was held by developing countries (Faure, 2001). This raised global concern over the rising debt in many developing countries. The concern was especially due to the belief that debt repayments were limiting government spending in health. To free resources into health and other social sectors, debt relief initiatives such as Highly Indebted Poor Country Initiative

(HIPC) were initiated in 1996 by World Bank and International Monetary Fund

(IMF) (Wagstaff et al., 2006).

The 1990s period saw growth of key global health initiatives and partnerships aimed at addressing diseases of the poor such as malaria, tuberculosis, and cholera

(Faure, 2001; Wagstaff et al., 2006). The initiatives included the Global Alliance for Vaccines and Immunization; the Joint United Nations Programme on

HIV/AIDS (UNAIDS); and the Stop TB Partnership. Other initiatives included the Global Fund to Fight AIDS, Tuberculosis, and Malaria; the Global Alliance for Improved Nutrition; and the Roll Back Malaria Partnership. A number of non- governmental organizations (NGOs) were also set up. The setting up of the NGOs was to hasten discovery and uptake of low-cost health technologies, especially in developing countries (Gillam, 2008). The NGOs included the Global Alliance for

Tuberculosis, the International Trachoma Initiative, the International AIDS

Vaccine Initiative, and the Medicines for Malaria Venture. The global initiatives together with new partnerships availed more resources for health. In spite of the increased resources, new challenges emerged. These included harmonization, coordination and link of local actions with global goals in the fight against

disease, malnutrition and death in the developing countries (Gillam, 2008;

Wagstaff et al., 2006).

In 2000, heads of 147 states at the United Nations Millennium Summit endorsed the Millennium Development Goals (MDGs). The MDGs placed premium on poverty reduction and improving health (Lawson, 2004). Nearly half of the

MDGs were concerned with different aspects of health and poverty (WHO,

2015a). Three of the eight MDGs directly focused on improving health status. The health related goals aimed at reducing under-five mortality rates by two-thirds; reducing maternal mortality rates by three quarters; achieving universal access to reproductive health; and halting the spread of major diseases such as HIV/AIDS, malaria and tuberculosis (TB). All these were to be achieved by 2015 (Lawson,

2004; Wagstaff et al., 2006). In relation to poverty, the first MDG aimed at eradicating extreme poverty and hunger (Lawson, 2004; WHO, 2015a). The target of the first MDG focused on halving between 1990 and 2015 the proportion of people living with extreme poverty and hunger.

In 2001, the Heads of States of African Union (AU) committed through the Abuja

Declaration to allocate at least 15 per cent of their annual budget to improve health sector (Mburu, Folayan, & Akanni, 2014; OAU, 2001). They further urged donor countries to fulfill their promise of committing 0.7 per cent of their Gross

National Product (GNP) as Official Development Assistance (ODA) to developing countries (Mburu et al., 2014; OAU, 2001). The commitment was an acknowledgement of the key role played by public funding in ensuring

sustainable and equitable health coverage for all (WHO, 2016) and also the need to achieve MDGs .

Since the Abuja Declaration in 2001, health funding in Africa has risen. Health budgets in AU Member States increased from 9 per cent to 11 per cent of public expenditures in 2001 and 2011, respectively (UNAIDS, 2013). By the end of

2011, only six AU Member States namely Liberia, Madagascar, Malawi,

Rwanda, Togo and Zambia had fulfilled the Abuja Declaration of committing at least 15 per cent of national budget to health (UNAIDS, 2013). In 2014, the average annual public expenditure on health in Africa was 10 per cent of total public spending. This ranged from 4 per cent in Cameroon to 17 per cent in

Swaziland. Unlike in 2011 when six countries had fulfilled the Abuja declaration of 15 per cent, only four countries, namely Malawi, Ethiopia, Gambia and

Swaziland were allocating 15 per cent of their total public expenditure to health in

2014 (WHO, 2016). Thus, Kenya was among countries that had not fulfilled the

Abuja Declaration.

Kenya‟s health expenditure as a percentage of total government expenditure stood at 8 per cent in 2001/02. In 2005/06 and 2009/10 total health expenditure as a percentage of total government expenditure reduced to 5.2 per cent and 4.6 per cent, respectively. However, the government expenditure on health as a percentage of total government expenditure increased to 6.1 per cent in

2012/2013. This was, however, below the 15 per cent agreed through the Abuja

Declaration (Republic of Kenya, 2015b). Therefore, majority of African countries, Kenya included have not fulfilled the commitment of allocating 15 per

cent of their total government spending to health (Mburu et al., 2014; UNAIDS,

2013).

Achievements of the MDGs were mixed. In 2015, 12 per cent of the world‟s population lived below the international poverty line of USD 1.90 a day, down from 36 per cent in 1990. In East Asia and Pacific, extreme poverty rate fell from

61 per cent in 1990 to 7 per cent in 2012, while in South Asia it fell from 51 per cent to 19 per cent (United Nations, 2015a; World Bank, 2016). In contrast,

SSA‟s extreme poverty rate did not fall below its 1990 level of 57 per cent until

2002 (World Bank, 2016). In 2015, the poverty rate in SSA stood at 41 per cent

(United Nations, 2015a).

The global under-five mortality rate declined by more than half, dropping from 90 deaths per 1,000 live births in 1990 to 43 deaths per 1,000 live births in 2015. In

SSA, the under-five mortality rate dropped from 179 deaths per 1,000 live births in 1990 to 86 deaths per 1,000 live births in 2015. This rate fell short of the MDG target of reducing the under-five mortality rate by two thirds by 2015 (United

Nations, 2015a; World Bank, 2016). Global maternal mortality declined from 385 deaths per 100,000 live births in 1990 to 216 deaths per 100,000 live births in

2015. In SSA, maternal deaths declined from a high of 990 deaths per 100,000 live births in 1990 to 510 deaths per 100,000 live births in 2013 again missing the target of reducing the deaths by three quarters.

Haemorrhage is one of the largest contributors of maternal deaths. Haemorrhage and other child birth related complications can be prevented through health care

interventions. Such health care interventions include antenatal care during pregnancy, skilled care during childbirth and care and support after childbirth

(United Nations, 2015a). The proportion of deliveries attended by skilled health professionals, globally, increased from 59 per cent in 1990 to 71 per cent in 2014.

In SSA, the proportion of deliveries attended by skilled health professionals stood at 52 per cent in 2014, an increase from 43 per cent in 1990. However, despite the improvement in proportion of child births attended by skilled health professionals, the MDG target of reducing maternal mortality ratio by three quarters was not met

(World Bank, 2016).

Building on the achievements of the MDGs, the United Nations General

Assembly adopted the 2030 Agenda for Sustainable Development in 2015 (World

Bank, 2016). This agenda was built upon the Rio+20 United Nations Conference on Sustainable Development of 2012. The conference had recognized the link between poverty, population health, inequality, creating inclusive economic growth and preserving the planet (United Nations, 2012). According to United

Nations (2012), the relationship between each of these elements are dynamic and reciprocal. This symbiotic relationship could explain why Goals 1 and 3 of the

Sustainable Development Goals (SDGs) focuses on poverty eradication and promotion of healthy lives and wellbeing for all (United Nations, 2015b; WHO,

2015a).

Kenya has not been left out in poverty reduction efforts and improvement of health for the betterment of her citizens‟ lives. Since attaining political independence in 1963, Kenya has put in place policies and programmes aimed at

ensuring good health, improved access and utilization of health care and eradication of poverty in the country. However, little has been achieved as the country still face high poverty rates and poor health outcomes as elaborated in the sections that follow.

1.1.2 Trends in Poverty and Poverty Reduction Strategies in Kenya Following Kenya‟s political independence in 1963, the government committed itself to addressing challenges of diseases, poverty and illiteracy. The country‟s development strategy at the time was informed by the doctrine of African

Socialism (Republic of Kenya, 1965). Kenya‟s development strategy emphasized rapid economic growth with the assumption that poverty, unemployment and income disparities would improve due to the trickle-down effect of a robust economy (Kimalu, Nafula, Manda, Mwabu, & Kimenyi, 2002; Misati & Mngoda,

2012; Nafula et al., 2005). Thus, access to health services, education and political participation were envisioned from an economic perspective.

Kenya‟s development strategies were outlined in various development plans. In order to attain republican status, Kenya hurriedly drafted the First Development

Plan (1964-1970) also called the Red Plan (1964-1970). Due to the speed of drafting the plan, it was revised several times. Therefore, the Red Plan (1964-

1970) was replaced with the Green Plan (1966-1970) (Mureithi, 1988). The

Green Plan (1966-1970), which was an extension of the Red Plan (1964-1970), was wider in scope, more systematic in analysis and growth targets. However, the plan had little in terms of project content. During the Red Plan (1964-1970)

period, the set economic growth target was 6.3 per cent per year. The target was surpassed as the economy grew at 6.9 per cent per annum (Mureithi, 1988).

The Second Development Plan (1970-1974), also referred to as the Blue Plan

(1970-1974), aimed at achieving more prosperity and better living conditions.

With an assumption of trickle-down effect of economic growth, the government set a target of growing the economy at 6.7 per cent per annum. However, the target was missed as the economy grew at 5.5 per cent per annum (Mureithi,

1988; World Bank, 1982).

Analysis of the first decade (1964-1974) shows that there was rapid economic growth, but manifest failure of the trickle-down effect (Mureithi, 1988). Available poverty estimates show that in 1972 and 1974, poverty stood at 30 per cent and 29 per cent, respectively (Manda, Kimenyi, & Mwabu, 2001). This indicates that majority of the population did not enjoy the benefits of the rapid economic growth.

The Third Development Plan (1974-1978) placed high emphasis on job creation, redistribution of income, family welfare, science and appropriate technology.

Through the plan, the government targeted to grow the economy by 7.4 per cent per annum. This was to translate to an employment growth of 3.2 per cent per year and eventual reduction in poverty. Manufacturing was targeted to grow at

10.2 per cent per year. However, the set target for economic growth was missed, as the economy grew at 4.6 per cent per year. This was attributed to the oil crisis of 1973/74 and drought in 1974 (Mureithi, 1988; World Bank, 1982). During the

plan period, poverty is estimated to have increased to 32 per cent by 1976 (Manda et al., 2001).

With the increasing poverty in the country, the Development Plan (1979-1983) focused more on poverty alleviation and better income distribution. The plan was largely inspired by the basic needs strategy proposed by the World Employment

Conference in Geneva in 1976 (Mureithi, 1988). The strategy focused on the provision of basic services such as food, clothing, shelter, water, education and health care for the poor. Also poverty alleviation was to be pursued by creating income-generating opportunities. The plan recognized that about 85 per cent of the population lived in rural areas and that‟s where majority of the poor are located. Thus, the plan placed strong emphasis on policies, programmes and projects for rural development (Manda et al., 2001; Mureithi, 1988; Nafula et al.,

2005).

The strategy was designed to improve the economic and wellbeing status of the rural poor. However, the rural dimension of poverty alleviation was not combined explicitly to the political, cultural, social and environmental concerns, which were either completely ignored or mentioned in passing as by-products of development programmes (Manda et al., 2001; Nafula et al., 2005). The economy was targeted to grow at 6.3 per cent per year. However, the target was missed, as the economy grew at an average of 4.2 per cent. This performance was attributed to the global economic recession of 1979. The poor performance worsened the poverty situation in the country, which stood at 46.8 per cent in 1981/82 (Manda et al.,

2001; Mureithi, 1988).

Against the background of the global recession, the Development Plan (1984-

1988) focused on mobilizing domestic resources for equitable development. The objectives of the plan included economic growth, job creation, equality of opportunities and equitable income distribution. The plan also targeted to grow the economy at 4.9 per cent per year (Mureithi, 1988). The plan also ushered in the District Focus for Rural Development (DFRD) strategy, which aimed at allocating resources in a more geographically equitable basis. The goal of the equitable geographical distribution of resources was expected to offer the possibility of social and economic equity and poverty alleviation (Manda et al.,

2001; Nafula et al., 2005).

Through DFRD, planning responsibility and implementation of projects was shifted to the rural areas. This was expected to improve the participation of the local people in the projects funded and help in poverty alleviation. However, because of poor preparation, lack of skills among staff in methods of participatory planning and lack of monitoring and evaluation, some decentralized projects were ill conceived and designed leading to their failure. The strategy did little to alleviate poverty, as the targeted poor were largely left out during project design and implementation (Manda et al., 2001).

Furthermore, in the 1980s Structural Adjustment Programmes (SAPs) were adopted. This strategy was expected to create a new pace in economic growth

(Manda et al., 2001). Therefore, to improve on economic performance, SAPs aimed at increasing economic efficiency through strengthening the role of private sector, trade liberalization and reduction in government expenditure (Manda et

al., 2001; Misati & Mngoda, 2012). The adjustment policies were also expected to have direct effects on social welfare through changes in the level of income and its distribution. To achieve reduction in government expenditure, cost-sharing was introduced in basic public social services such as health. There was also retrenchment of workers in the public sector and removal of government subsidies in a range of basic production inputs. These reforms of retrenchment and liberalization of the economy compounded with corruption and nepotism, though necessary, in the short term, worsened the unemployment situation in the country.

Worsened unemployment implied lack of income necessary to meet basic needs such as food, clothing, shelter, education and medical services especially among the majority poor and the vulnerable in the society (Misati & Mngoda, 2012;

Nafula et al., 2005).

Even though, the poverty indicators were worsening in Kenya, it was not until 1992 that the Kenyan government intensified poverty monitoring and analysis activities through Welfare Monitoring Surveys (WMSs) (Manda et al., 2001). In total, three

WMSs were carried out with comprehensive analysis on poverty trends. The

WMSs were done in 1992, 1994 and 1997 (Manda et al., 2001). The main goal of the WMSs was to determine the current and the future net socio-economic consequences of structural adjustment in Kenya. This is because internal structural problems had been identified as hindrances to Kenya‟s economic growth in the 1970s and 1980s and poverty reduction efforts (Manda et al., 2001;

World Bank, 1982). In addition to WMSs, poverty was also monitored using

Small Area Estimation (SAE). The small area poverty estimates are normally not

available even for census years. To solve the problem, a method of SAE is used.

The method is used to generate reliable estimates of quantities of interest for geographical regions, when the regional sample sizes are small in the survey data set (KNBS, 2015). The method of SAE was applied in Kenya‟s census data for

1999 and 2009. In addition to WMSs and SAE, poverty was monitored using the

Kenya Integrated Household Budget Survey (KIHBS) of 2005/06 and 2015/16.

KIHBS was designed to provide various indicators and data needed to measure living standards and poverty in Kenya.

Table 1.1 shows the national and regional prevalence of absolute poverty over time in Kenya.

Table 1.1: Prevalence of Absolute Poverty Over time in Kenya WMS WMS WMS Small KIHBS Small KIHBS I 1992 II III Area 2005/06 Area 2015/2016 1994 1997 Estimation Estimation 1999 2009 Kenya 45.0 40.3 52.6 52.6 46.6 45.2 36.1 Rural 48.0 46.8 53.1 52.8 49.7 50.5 40.1 Urban 29.0 29.0 50.1 49.5 34.4 33.5 29.4 Source of data: Gakuru and Mathenge (2012); Republic of Kenya (2015), Republic of Kenya (2018)

Table 1.1 indicates that poverty has remained above 40 per cent between 1992 and 2009 before declining slightly in 2015/2016 to 36.1 per cent. This percentage implied that about 16.4 million Kenyans were living in poverty. During the period, substantial regional differences in the prevalence of poverty have existed as indicated in Table 1.1. Poverty is largely concentrated in the rural areas. In

1990s, about half of the Kenya‟s rural population were poor and between 29 and

50 per cent of the urban population were poor. In 2000s, the situation remained

the same with rural poverty at 49.7 per cent and 50.5 per cent in 2005/6 and 2009, respectively. Urban poverty was 34.4 per cent and 33.5 per cent in 2005/6 and

2009, respectively. In 2015/2016, little had changed as rural poverty stood at 40.1 per cent while that of urban areas remained at 2.91 per cent.

Rural poverty is linked to agricultural activities and land, whereas urban poverty is linked to how incomes are generated (Manda et al., 2001). The poor people in the rural areas depend on agriculture more than the non-poor people (Manda et al., 2001). In addition, prosperity of non-farm activities in the rural areas are linked to forward and backward production linkages with agriculture (Manda et al., 2001; Nafula et al., 2005). Therefore, low access to physical assets such as land, low agricultural productivity, non-farm job opportunities and health care and schooling tend to explain rural poverty more than the labour market distortions experienced in urban sector (Manda et al., 2001; Nafula et al., 2005). According to KIPPRA (2013), those living in poverty in the country were 49.8 per cent and

49.5 per cent in 2012 and 2013, respectively, an increase from 45.2 per cent in

2009. The gap between rural and urban absolute poverty also remained high, a reflection of regional inequalities. This was attributed to the 2007/2008 post- election violence and other external shocks (KIPPRA, 2015).

In her efforts to reduce poverty, the Kenyan government has over time adopted various policies. Even though the government intensified its poverty monitoring and analysis efforts in 1992, the first major poverty reduction policy was adopted in June, 2000. The government adopted the Interim Poverty Reduction Strategy

Paper (IPRSP, 2000-2003), which preceded the Poverty Reduction Strategy Paper

(PRSP, 2001-2004) (Republic of Kenya, 2001). The IPRSP (2000-2003) outlined measures aimed at revamping the economic growth and poverty reduction. The strategy had five basic policy objectives. These were to facilitate sustained and rapid economic growth; to improve governance and security; to increase the ability of the poor to raise their incomes; to improve the quality of life of the poor; and to improve equity and participation. The elements of the strategy aimed at addressing the underlying causes of poverty (Republic of Kenya, 2000). The

(IPRSP, 2000-2003) identified quantitative targets of various dimensions of poverty such as educational attainment and health status. Economic growth was targeted to average 3.2 per cent over the period 2002-2004. The government believed that the benefits of economic growth will trickle down and reduce unemployment from 25 per cent in 1998 to 20 per cent in 2002 leading to poverty reduction (Republic of Kenya, 2001).

Analysis of the outcomes of the IPRSP (2000-2003) and PRSP (2001-2004) show that the policy objectives were not achieved. Those living in poverty in the year

2001 were 55.4 per cent, an increase from 52.6 per cent in 1999 (Republic of

Kenya, 2004). Odima (2014) argued that, the policy objectives were not achieved largely because of non-implementation, inadequate resource allocation, lack of prioritization, lack of participation and involvement of the poor, poor planning and budgeting for key social sectors.

In 2003, the government embarked on the process of preparing an economic recovery strategy that focused on reviving the economy and creation of job opportunities. This was built on previous government policy documents such as

the IPRSP (2000-2003) and the PRSP (2001-2004). Thus, the government introduced Economic Recovery Strategy for Employment and Wealth Creation

(ERS, 2003-2007). The ERS (2003-2007) pursued goals that closely focused on economic growth, poverty reduction, employment creation, and improving the wellbeing of the people (Republic of Kenya, 2004).

The strategy targeted to increase economic growth rate from 1.2 per cent in

2002/03 to 1.9 per cent by 2003/04. This was to be increased further to 3.1 per cent and to 4.5 per cent in 2004 and 2006/07, respectively. Also, the government targeted to reduce the proportion of people below the absolute poverty line by 10 per cent by the year 2006 from 57 per cent in 1997. The implementation of the

ERS (2003-2007) saw the country‟s economy back on track to rapid growth since

2002. In 2002, GDP grew from a low of 0.6 per cent and gradually rose to 6.1 per cent in 2006 and 7 per cent in 2007 (Republic of Kenya, 2007, 2008). Resulting from strong economic performance, real per capita income increased to an annual average rate of 3 per cent. This resulted to poverty reduction from 55.4 per cent in

2001 to 46 per cent in 2006 (Republic of Kenya, 2008).

In 2008, the government of Kenya launched the country‟s long term development plan: Kenya Vision 2030 (2008-2030) (Republic of Kenya, 2008). The Kenya

Vision 2030 was motivated by a desire to have a development strategy that built on the ERS (2003-2007). It was also motivated by a desire to transform Kenya into a newly industrializing middle income country with high quality of life in a clean and secure environment by the year 2030 (Republic of Kenya, 2008). The

Kenya Vision 2030 also sought to ensure that Kenya provides equitable and

affordable health care for all. In addition, the Kenya Vision 2030 also aims at establishing a socially just and equitable society without extreme poverty

(Republic of Kenya, 2008). The Kenya Vision 2030 is implemented through five- year rolling Medium Term Plans (MTPs).

The objective of the First Medium Term Plan (MTP I, 2008-2012) was to realize a higher and sustainable economic growth in a more equitable atmosphere accompanied by increased employment opportunities (Republic of Kenya, 2008).

The MTP I (2008-2012) aimed at increasing real GDP growth from 7 per cent in

2007 to 7.9 and 8.7 per cent in 2009 and 2010, respectively. This was to be increased further to 10 per cent by the year 2012. Further, the government set a target of reducing poverty from 46 per cent in 2006 to 28 per cent by 2012.

During the implementation of the MTP I (2008-2012), Kenya adopted a new

Constitution. The new Constitution changed the country‟s governance structure by creating a two-tier government: a national government and 47 county governments (Republic of Kenya, 2013a). The Constitution introduced devolution, which is playing a major part in service delivery and gives all

Kenyans key economic and social rights. The Constitution through the Bill of

Rights places a heavy duty on the health sector to ensure realization of the right to health (Republic of Kenya, 2013a). The goal of the health sector under the

Constitution is to provide equitable, affordable and quality health care to all citizens. To achieve this goal, the Constitution devolved health services with aim of improving service delivery (Republic of Kenya, 2013a).

The MTP I (2008-2012)‟s growth targets were however missed. Kenya‟s economy grew by 2.1 per cent in 2008, 5.1 per cent in 2010, and 4.1 per cent in

2012 (Republic of Kenya, 2013a). This poor performance of the economy was attributed to adverse multiple shocks among them the post-election crisis, drought, the global financial and economic crisis, high international oil and food prices. These shocks affected efforts to reduce poverty and address other social problems like health (Republic of Kenya, 2013a). People living in poverty were estimated to be 49.8 per cent in 2012 (KIPPRA, 2013).

The Second Medium Term Plan (MTP II, 2013-2017) built on MTP I (2008-

2012) and taking into account the aspirations in the constitution, especially devolution (Republic of Kenya, 2013a). The MTP II (2013-2017) aimed at implementing devolution, accelerating economic growth, reducing poverty, transforming the structure of the economy and creating more quality jobs, and increasing access to health care. The MTP II (2013-2017) also aimed at addressing the unmet MDG targets (Republic of Kenya, 2013a). In 2013, people living in poverty were estimated to be 49.5 per cent (KIPPRA, 2013).

The MTP II (2013-2017) targeted to grow the economy from 6.1 per cent in 2013 to 7.2 and 8.7 per cent in 2014 and 2015, respectively. In 2016 and 2017, the government targeted to grow the economy to 9.1 per cent and 10.1 per cent, respectively. The higher economic growth was premised on the increased investment, which was targeted to reach 30.9 per cent of GDP by 2017 from 24.7 per cent in 2013 (Republic of Kenya, 2013a). The benefits of the growth were

expected to trickle down and reduce absolute poverty. However, the MTP II

(2013-2017) was silent on the targeted level of poverty.

Review of the MPT II (2013-2017) shows that the targets were missed. The economy grew at 5.9 per cent and 5.4 per cent in 2013 and 2014, respectively

(Republic of Kenya, 2018c). In 2015, the economy grew at 5.7 per cent missing the set target of 8.7 per cent. In 2016 and 2017 the economy grew at 5.9 per cent and 4.9 per cent, respectively missing targets of 9.1 per cent in 2016 and 10.1 per cent in 2017 (Republic of Kenya, 2018c). During the period under review, poverty declined slightly in 2015/2016 to 36.1 per cent from 49.8 per cent in 2012

(KIPPRA, 2013; Republic of Kenya, 2018a) an indication that little was achieved regarding poverty reduction.

The country is implementing the Third Medium Term Plan (MTP III, 2018-2022).

The MTP III (2018-2022) builds on MTP II (2013-2017) and prioritizes policies, programmes and projects aimed at supporting the implementation of the “Big

Four” agenda launched in 2017(Republic of Kenya, 2018d). The “Big Four” focuses on key basic aspects that are critical in improving the living standards of

Kenyans towards becoming an upper middle-income country in line with Kenya

Vision 2030 (KIPPRA, 2018). The plan prioritizes affordable and decent housing, achievement of universal health coverage, food and nutritional security, and creation of job opportunities through manufacturing. The government targets to increase the contribution of the manufacturing sector to GDP to 15 per cent by

2022 (Republic of Kenya, 2018b, 2018d). The vibrant manufacturing sector is expected to increase job opportunities in the manufacturing sector by more than

800,000. The increase is anticipated to accelerate economic growth and lead to poverty reduction (Republic of Kenya, 2018b, 2018d). Thus, the government assumes that there will be a trickle-down effect of the benefits coming from improved economy. However, the government has no specific target of the poverty levels that it intends to achieve by 2022.

1.1.3 Health Policies and Health Indicators in Kenya Kenya‟s Health Policy has been based on the country‟s post-colonial philosophy of nation building and socio-economic development anchored in the Sessional

Paper No. 10 of 1965 (Wamai, 2009). As evidenced in the Sessional Paper No. 10 of 1965 and subsequent government policy documents, the government considers good health as a fundamental right for all her citizens (Republic of Kenya, 1965,

2010b; Wamai, 2009). Since attaining political independence in 1963, the government committed itself to the provision of health services that meets the basic needs of the population. The policy aspiration was geared at providing health services within easy reach by all Kenyans. The evidence that investing in health leads to human capital outcomes that positively impact in the social and economic development of a country, motivated the policy. The government emphasized on preventive, promotive, rehabilitative services and curative services

(Oyaya & Rifkin, 2003; Republic of Kenya, 1994). To achieve this goal, the government has over time pursued various health policies aimed at addressing health challenges. These policies have been informed by national priorities, regional and international commitments (Oyaya & Rifkin, 2003; Republic of

Kenya, 1994).

From the beginning, the Kenyan government health policy focused on four main aspects for the development of the country‟s health care. The four components includes developing an insurance scheme, expanding the health care system coverage, preventive health care and family planning (Wamai, 2009). Before

1980s, health policies were outlined in the National Development Plans, indicating the government‟s intentions and strategies for socio-economic development. In the First Development Plan (1966-1970), the government introduced the free access health policy, which abolished the Kshs 5 user-fees that existed until 1965 (Wamai, 2008, 2009). The free access health policy aimed at expanding health care coverage. The expansion was to be achieved through centralization of delivery tasks from the devolved units to the Ministry of Health

(Oyaya & Rifkin, 2003).

Again, in 1966, the government introduced National Health Insurance Fund

(N.H.I.F) for persons formally employed. This was part of the government‟s development of health infrastructure. The N.H.I.F was expected to increase insurance coverage in the population. The N.H.I.F covered the contributor‟s spouse and children under the age of 18 years regardless of the type of illness one suffered or the number of children (Wamai, 2009). The N.H.I.F paid for health care received from private health care providers. However, introduction of user fees in government health facilities created problems. Thus, N.H.I.F was restructured in 1972 in order to cater for services offered in government health facilities and also allow participation of self-employed persons (Mwabu, 1995).

In the Development plan (1984-1988), due to inadequate resources, poor economic performance, international donor pressures and declining budget allocations, the government introduced major reforms in the health sector (Oyaya

& Rifkin, 2003). Through the Fifth Development Plan (1984-1988), the government published the National Guidelines for the Implementation of Primary

Healthcare in Kenya, following WHO policy direction for primary health care for all by the year 2000. The guidelines re-emphasized decentralization, community participation and intersectoral collaboration. The decentralization was in line with the DFRD concept ushered in 1983, which saw the management of health programmes taken to the rural areas. Through the guidelines, the government re- introduced user fees in accessing health care to supplement health budget at health facilities, which was implemented in December,1989 (Chuma, Musimbi, Okungu,

Goodman, & Molyneux, 2009; Kimani, 2014). Introduction of user fees complicated government‟s pledge to provide free health services to all. This led to widespread protests by Kenyans. As a result, the user fees was suspended in

August, 1990 before its re-introduction in April, 1992 (Kimani, 2014).

With an aim of accelerating health care coverage, another major policy in the

1980s was to integrate traditional medicine in to modern medicine (Mwabu, 1995;

Wamai, 2009). In 1980, a research unit was established in the Ministry of Health.

This unit was aimed at professionalizing the knowledge and practices of the traditional practitioners. The traditional health care providers were also licensed by the government to operate outside the government health system. Further, traditional mid-wives were engaged by the government to work in its health

facilities, mostly in the rural areas where there was huge need. The traditional mid-wives were involved in maternal and child health activities (Mwabu, 1995;

Wamai, 2009). The policy to involve traditional practitioners was in line with international debates and directions such as the Alma Ata Declaration of 1978 on

Health for All by Year 2000 (Wamai, 2009). However, despite recognition of traditional medicine as an integral part of health system nationally and internationally, there remained integration challenges such as lack of information for the health system and mistrust from conventional medicine. The use of traditional medicine while linked with specific cultural practices was also amplified by poverty, which made health care unaffordable for the over 50 per cent poor Kenyans (Wamai, 2009).

It is not until 1994 when the Kenyan government introduced a more comprehensive Kenya Health Policy Framework Paper (KHPF, 1994-2010). The policy was developed to guide the health sector reforms. It explicitly outlined the vision for development of the health sector and reforms aimed at sustainable, accessible and affordable quality health care; resource mobilization; participation and collaboration with other health stakeholders; and the government‟s regulatory role (Republic of Kenya, 1994; Wamai, 2009). The Implementation and Action

Plans (IAP) of 1996 laid out the implementation framework for the KHPF (1994-

2010). However, the IAP ignored the priorities and views shared by all concerned stakeholders, hence it did not have the necessary commitment for effective implementation. It was therefore abandoned and in its place was developed the

First National Health Sector Strategic Plan (NHSSP I, 1999-2004), and later the

Second National Health Sector Strategic Plan (NHSSP II, 2005-2010). The objectives contained in KHPF (1994-2010) and NHSSP I (1999-2004) were linked to the IPRSP (2000-2003) and PRSP (2001-2004) (Republic of Kenya,

2000, 2005a).

The NHSSP I (1999-2004) articulated the health care targets. The government set a target of reducing malaria morbidity and mortality by 30 per cent by the year

2006. It also targeted to reduce child malnutrition by 30 per cent over the period

1999-2004 (Republic of Kenya, 2005a). Table 1.2 shows some of the selected health targets as per the NHSSP I (1999-2004).

Table 1.2: Selected Health Targets as Per NHSSP I (1999-2004) Indicator 2000 2001 2002 Infant mortality rate (per 1000 live births) 70 68 65 Under five mortality rate (per 1000 live births) 105 100 98 Maternal mortality ratio (per 100,000 live births) 550 520 500 Incidence of stunting in children under 5 years (%) 36.5 36 35 Wasting among children under 5 years (%) 6.2 6.1 6.0

Life expectancy at birth 53 54 55

Source of Data: Republic of Kenya (2001)

Table 1.2 illustrates the trends in infant mortality rate, which declined from 70 deaths per 1,000 live births in 2000 to 65 deaths per 1,000 live births in 2002. The government also set out to reduce under-five mortality rate to 98 deaths per 1,000 live births in 2002. In regard to maternal health, the government set out to reduce maternal mortality ratio to 550 deaths per 100,000 live births in 2000 and further reduce it to 500 deaths per 100,000 live births by 2002.

To achieve the objectives outlined in KHPF (1994-2010) and NHSSP I (1999-

2004), and the targets set therein, in 2004 the government reversed the user fees policy of 1992 (Chuma et al., 2009). The government announced that health care at dispensary and health centres should be free for all. However, there was a requirement of minimal registration fee of Kshs 10 and Kshs 20 in dispensaries and health centres, respectively (Chuma et al., 2009; Kimani, 2014). Under the new policy, commonly referred to as the 10/20 policy, exemptions for the registration fees was made for children aged five years and below. Exemptions were also made for specific health services including maternity services in government health centres and dispensaries, treatment of tuberculosis and vaccinations. Registration fee for patients was also waived for the poor (Chuma et al., 2009; Kimani, 2014). The introduction of 10/20 policy in 2004 led to an immediate increase in health care utilization of 70 per cent. Using data collected in 2004, an evaluation done in 2005 reported that the increase was not sustained, although utilization remained 30 per cent higher than before the introduction of

10/20 policy (Chuma et al., 2009). Lack of sustainability was attributed to the declining finances for purchase of non-medical items and supplementary drugs, payment for staff allowance during outreach activities and support staff (Chuma et al., 2009; Kimani, 2014).

Evaluation of the outcomes of the NHSSP I (1999-2004) show a downward trend of some health indicators. Infant mortality rate increased from 74 deaths per 1,000 live births in 1998 to 78 deaths per 1,000 live births in 2003. Thus, the set target under the KHPF (1994-2010) of 65 deaths per 1,000 live births by 2002 was

missed. The under-five mortality rate target of 98 deaths per 1,000 live births was also missed. The under-five mortality rate rose from 112 deaths per 1,000 live births in 1998 to 114 deaths per 1,000 live births in 2003. Life expectancy at birth declined from 60 years in 1998 to 47 years in 2000. This indicates that the target of increasing life expectancy to 53 years by 2000 was missed (Republic of Kenya,

2004). This implies that NHSSP I (1999-2004) did not contribute to the improvement of health status of Kenyans (Republic of Kenya, 2005b). The poor performance was attributed to absence of legislative framework to support decentralization, weak management systems, and lack of accountability. Other factors that led to the poor performance were inadequate funding and lack of well- articulated, prioritized and costed strategies (Republic of Kenya, 2005a).

The NHSSP II (2005-2010) was put in place in 2005 to reverse the downward spiral of Kenya‟s health status (Republic of Kenya, 2005a). The NHSSP II (2005-

2010) was an integral part of the ERS (2003-2007) and its objectives linked to

MDGs. It aimed at reducing health inequalities and to reverse the decline in the health outcome indicators. The objectives of the NHSSP II (2005-2010) were to increase equitable access to health services, improve the efficiency and effectiveness of service delivery, and financing of the health sector (Republic of

Kenya, 2005b).

The government under the NHSSP II (2005-2010) targeted to reduce maternal mortality rate from 414 deaths per 100,000 live births in 2004/05 to 170 deaths per 100,000 live births by 2010. Further, the government projected to reduce

HIV/AIDs prevalence from 10.6 per cent in 2004/05 to 6 per cent in 2010. The

government also set a target of reducing malaria morbidity from 30 per cent in

2004/05 to 10 per cent in 2008, a target, which had been set in ERS (2003-2007).

Immunization coverage was to be increased from 57 per cent in 2004/05 to 90 per cent in 2010. In addition, the government planned to increase the proportion of deliveries by skilled health staff from 42 per cent in 2004/05 to 90 per cent by

2010. The government planned to reduce infant and under-five mortality rates from 77 and 114 deaths per 1,000 live births, respectively in 2004/05. However, the NHSSP II (2005-2010) did not set targets for both infant and under-five mortality rates. The targets were only projected to meet the MGD targets of 25 deaths per 1,000 live births and 33 deaths per 1,000 live births for infant and under-five mortality rates, respectively, by 2015 (Republic of Kenya, 2005a).

The implementation of the NHSSP II (2005-2010) coincided with the MTP I

(2008-2012), which had outlined a number of health sector targets. The MTP I

(2008-2012) health sector goal was to provide affordable and quality health care to all citizens through restructuring of the health service delivery system

(Republic of Kenya, 2008). This was aimed at shifting focus from curative to preventive health care in order to lower the nation‟s disease burden. The government under the MTP I (2008-2012) projected to reduce under-five mortality rate from 92 deaths per 1,000 live births in 2007 to 45 deaths per 1,000 live births in 2012. Maternal mortality rate was targeted to reduce from 414 deaths per 100, 000 live births in 2007 to 200 deaths per 100,000 live births in

2012. The government set a target of increasing the proportion of births attended by skilled health staff from 51 per cent in 2007 to 64 per cent in 2012.

Immunization coverage was targeted to increase from 71 per cent in 2007 to 90 per cent in 2012. Further, the government set a target of reducing HIV/AIDs prevalence from 7.4 per cent in 2007 to 6.4 per cent in 2012. The government also focused on reducing the proportion of in-patient malaria fatality from 19 per cent in 2007 to 17 per cent in 2012 (Republic of Kenya, 2012).

To achieve the targets, the government introduced Reproductive Health Output-

Based Aid (OBA) voucher programme in 2005. The programme targeted the poor in the society. Thus, it was first introduced in three rural districts, namely

Kisumu, Kiambu and Kitui; and two urban informal settlements, Korogocho and

Viwandani in Nairobi (NCAPD, 2008). The programme targeted three main areas, namely clinical family planning, safe motherhood and gender violence recovery services (NCAPD, 2008). The programme aimed at offering quality reproductive health care services for less fortunate populations (Kimani, 2014).

In 2007, the government abolished all fees for deliveries in public health facilities.

This was aimed at reducing maternal mortality (Chuma et al., 2009). In line with the objectives of NHSSP II (2005-2010) of increasing service delivery and improve financing, the government established a Health Sector Services Fund

(HSSF) in 2010 (Kimani, 2014). The scheme was established to disburse funds directly to government dispensaries and health centres. This was aimed at improving health services at the local levels. The establishment of the scheme was to enable local communities to participate in health care delivery and to bestow local facilities with the autonomy to manage their resources (Kimani, 2014).

Analysis of health outcomes in MTP I (2008-2012) show mixed results. The target of reducing maternal mortality from 414 deaths per 100,000 live births in

2007 to 200 deaths per 100,000 live births by 2012 was missed. It instead rose to

488 deaths per 100,000 live births. The target of reducing HIV/AIDs prevalence rate from 7.4 per cent in 2007 to 6.4 per cent in 2012 was also missed. The achieved target stood at 6.3 per cent in 2012. In addition, the government missed the projection of increasing the proportion of birth deliveries attended by skilled health staff from 51 per cent in 2007 to 64 per cent in 2012. The realized target stood at 43 per cent. Further, the government failed to meet the target of increasing immunization coverage from 71 per cent in 2007 to 90 per cent by

2012. The achieved target stood at 84.5 per cent in 2012. The target of reducing under-five mortality rate from 92 deaths per 1,000 live births to 45 deaths per

1,000 live births was also not met. The realized target stood at 74 deaths per 1,000 live births in 2012. However, the government realized gains in reducing in-patient malaria fatality from 19 per cent in 2007 to 15 per cent in 2012, surpassing the target of 17 per cent by 2012 (Republic of Kenya, 2012).

A new Kenya Health Policy (KHP, 2012-2030) is in place. The new policy will guide the implementation of health reforms in Kenya. The intended health reforms are aimed at improving health status of Kenyan people. According to the policy, reforming of the health sector will be done in line with the Constitution of

Kenya 2010, the Kenya Vision 2030, and other global commitments such as SDGs

(Republic of Kenya, 2014c). The key objectives of the KPH (2012-2030) are to eliminate communicable conditions, halt and reverse the rising burden of non-

communicable conditions, provide essential health care, and minimize exposure to health risk factors (Republic of Kenya, 2014c).

The policy targets to increase life expectancy from 60 years in 2010 to 72 years by 2030. It also aims to improve neonatal mortality and infant mortality rates to

13 deaths and 20 deaths per 1,000 live births by 2030 from 31 deaths and 52 deaths per 1,000 live births in 2010, respectively. Under-five mortality rate is targeted to reach a low of 24 deaths per 1,000 live births by 2030 from 74 deaths per 1,000 live births in 2010. Also, the government aims at reducing maternal mortality rate from 488 deaths per 100,000 live births in 2010 to 113 deaths per

100,000 live births by the year 2030 (Republic of Kenya, 2014c).

To achieve the targets set in the KHP (2012-2030), the government has put in place some policies. In 2013, the government reversed the 10/20 policy, which was introduced in 2004 by abolishing user fees charged in government dispensaries and health centres (KIPPRA, 2014). The government also launched the free delivery policy, which led to the waiver of fees charged for delivery in all public health facilities (Kimani, 2014; KIPPRA, 2014). Introduction of free delivery policy was aimed at improving maternal health and child health. In 2017, the government through its “Big Four” agenda announced its intention to achieve universal health care by the year 2022. This is aimed at contributing towards reduction of out-of-pocket health expenditure from 26 per cent to 22 per cent by

2022. Reduction of out-of-pocket health expenditure will guarantee Kenyans access to health care and at minimal cost (KIPPRA, 2018).

The development of the country‟s health care system is not isolated to the country but follows international debates and directions. Thus, it is important to compare the country‟s health indicators with the international targets. Table 1.3 gives a comparison of Kenya‟s health indicators with international development targets.

Table 1.3: Kenya’s Key Health Indicators and Global Development Targets

for Selected Years

3 4

1 2

Indicator

2013 2014 MDGs Target SDGs Target Life expectancy at birth (years) 61 58 - -

Neonatal mortality rate (per 1000 live births) 26.3 22 10 12

Infant mortality rate (per 1000 live births) 47.5 39 22 -

Under-five mortality rate (per 1000 live births) 70.7 52 32 25

Maternal mortality ratio (per 100, 000 live births) 400 362 147 <70 Source of Data: 1WHO (2015b); 2Republic of Kenya (2014b);3Republic of Kenya

(2013b); 4WHO (2015a)

Table 1.3 shows that life expectancy at birth declined from 61 years in 2013, to 58 years in 2014. This was an indication of deteriorating health status. Maternal mortality rate declined from 400 deaths per 100,000 live births in 2013 to 362 deaths per 100,000 live births in 2014. However, the rate was high compared to the MDG target of 147 deaths per 100,000 live births by 2015 and SDG target of less than 70 deaths per 100,000 live births by 2030. Under-five mortality rate improved from 70.7 deaths per 1,000 live births in 2013 to 52 deaths per 1,000 live births in 2014. However, rate was above the global targets of 32 deaths per

1,000 live births for MDGs by 2015 and 25 deaths per 1,000 live births for SDGs

by 2030. Infant mortality rate also improved from 47.5 deaths per 1,000 live births in 2013 to 39 deaths per 1,000 live births in 2014. However, this rate was high compared to MDG target of 22 deaths per 1,000 live births. Improvement was also reported in neonatal mortality rate, which stood at 22 deaths per 1,000 live births in 2014 from 26.3 deaths per 1,000 live births in 2013. However, despite the improvement, the mortality rate was higher than the MDG and SDG targets of 10 deaths per 1,000 live births and 12 deaths per 1,000 live births, respectively.

Kenya is not isolated from the rest of the world. Table 1.4 gives a comparison of the Kenya‟s key health indicators and those of selected regions, namely Africa,

Europe and the World as at 2013.

Table 1.4: Comparison of Kenya's Key Health Indicators with Selected

Regions

Indicator

Kenya Africa Europe World Life expectancy at birth (years) 61 58 76 71 Neonatal mortality rate (per 1000 live births) 26.3 30.5 6.1 20 Infant mortality rate (per 1000 live births) 47.5 59.9 10.5 33.6 Under-five mortality rate (per 1000 live births) 70.7 90.1 12.2 45.6 Maternal mortality ratio (per 100, 000 live births) 400 500 17 210

Source of Data: WHO (2015b).

Table 1.4 shows selected indicators in 2013, for Kenya and averages for Africa,

Europe and the world. It shows that for all the selected indicators, Kenya was performing poorly in comparison to averages of other selected regions except,

Africa. Kenya‟s maternal mortality ratio was 400 deaths per 100,000 live births, which was almost double that of the world average. Further, Kenya‟s under-five

mortality rate remained at 70.7 deaths per 1,000 live births. This was higher than the average rates for Europe and the world, which stood at 12.2 deaths per 1,000 live births and 45.6 deaths per 1,000 live births, respectively. However, on average, Africa, which had a mortality rate of 90.1 per 1,000 live births performed poorly than Kenya. In terms of the other indicators such as life expectancy at birth, neonatal mortality rate and infant mortality rate, Kenya only performed better than Africa on average. However, in comparison to averages for Europe and the rest of the world, it performed poorly in those health indicators.

1.1.4 Health Care Utilization in Kenya

Over time, the Kenyan government has implemented various policies and initiatives aimed at addressing the challenge of health care utilization (Kimani,

Mugo, & Kioko, 2016). These broad initiatives are contained in various papers such as KHPF (1994-2010) and KHPF (2012-2030). Through the Kenya

Household Health Expenditure and Utilization Surveys (KHHEUS) of 2003, 2007 and 2013, the government examined how utilization of health care has changed overtime. Demand for health care services by individuals and households depends on their perception on the need for a health care (Kimani et al., 2016). Table 1.5 shows the proportion of people who reported illness in the 4 weeks preceding survey and total number of visits to a health facility and utilization rate. Table 1.5 indicates that household members who reported illness but did not seek health care stood at 22.8 per cent in 2003.

Table 1.5: Proportion of People Reporting Illness and Total Number of Outpatient Visits and Utilization Rates, 2003-2013 Description Year 2003 2007 2013 Total number of visits made in 4-week recall period to all health 4.8 7.4 9.1 care providers(Millions) People with some sickness reported (%) 17.4 15.1 19.3 People with some sickness reported but did not seek health care 22.8 16.7 12.7 (%) Average number of visits (in 4 weeks) a) Per 100 people 15 20 24 b) Per 100 sick 85 132 122 people Average number of visits (utilization rate) per person per year 1.9 2.6 3.1 Source of Data: Republic of Kenya (2014d)

In 2007 and 2013, those who reported some sickness and failed to seek treatment stood at 16.7 per cent and 12.7 per cent, respectively, despite the government‟s efforts. The utilization rate has also been increasing from 1.9 visits per person in

2003 to 2.6 visits per person in 2007. Among the reasons given by those who reported sickness and failed to seek treatment were self medication, poor quality of service and distance to health care provider. Other reasons included cultural and religious reasons, fear of discovering serious diseases and not taking illness seriously (Republic of Kenya, 2014d). Individuals who felt sick and reported lack of finances as the main reason for not seeking health care stood at 21.4 per cent in

2013, a drop from 37.7 per cent in 2007. In 2013, the most reported reason for not seeking health care was that illness was not considered to be serious enough to warrant medical attention (Republic of Kenya, 2014d).

Accessibility and utilization of health care depends on the availability of the services, finances and distance to the nearest health facilities. In Kenya, delivery of health services is done through various health care providers. The providers

include government, private-for-profit and voluntary agencies such as faith based organizations (FBOs), missions and NGOs (Gakii, 2013; Kosimbei, 2005). In

2013, utilization of outpatient services showed a high dependency of public facilities. Total out-patient visits to public health facilities stood at 59 per cent in

2013, while for private health facilities stood at 31 per cent (Republic of Kenya,

2014d). A major concern is those people who reported illness and never sought treatment due to lack of finances. Also, it is important to understand how individuals choose where to seek medical services, considering majority chose to visit government hospitals.

1.2 Statement of the problem Health is a highly valued asset, a fundamental human right and a prerequisite for improved productivity. This right and enjoyment of good health may not be realized in the presence of high incidences of poverty. To guarantee good health to her citizens, the Kenyan government has instituted various policies. These policies include policy of rapid economic growth, 10/20 policy, waiver and exemptions, Reproductive Health Output Based Aid (OBA), Health Sector

Services Fund (HSSF) scheme, devolution of healthcare management, public health insurance schemes, and abolition of all fees for maternity services at public health facilities (Chuma et al., 2009; Kimani, 2014; Nyambura, 2016). The policies are aimed at increasing health care accessibility, utilization and improved welfare. Despite the operationalization of these policies, poverty still remains high. Poverty headcount ratio remained above 40 per cent since 1982 until 2009 before dropping slightly to 36.1 per cent in 2015/16. This was against a target of

28 per cent by 2015 (KIPPRA, 2013; Manda et al., 2001; Republic of Kenya,

1999, 2004, 2015a, 2018a). If the situation continues, it is unlikely the country will meet the SDG target of eradicating poverty in all its forms by 2030. The delay in traction is also likely to hinder achievement of Kenya Vision 2030.

Compared with global goals such as MDGs and SDGs, Kenya‟s health indicators have not been impressive either. Under-five mortality rate was 71 deaths per

1,000 live births in 2015 compared to MDG target of 32 deaths per 1,000 live births. The rate is still way above the SDG target of 25 deaths per 1,000 live births by the year 2030. Maternal mortality rate remained at 362 deaths per 100,000 live births in 2014 against MDG target of 147 deaths per 100,000 live births by 2015

(Republic of Kenya, 2014b). This is high compared to the set SDG target of less than 70 deaths per 100,000 live births by 2030. In relation to health care utilization, children who were fully immunized in 2003 stood at 59 per cent. The rates for 2007 and 2012 were 73 per cent and 84.7 per cent, respectively

(Republic of Kenya, 2008, 2012, 2013a). The realized immunization levels were below the MDG target of 100 per cent by 2015. Further, the proportion of women who delivered assisted by skilled health care staff stood at 46 per cent in 2012 compared to a global target of 90 per cent by 2015 (Republic of Kenya, 2013a). In comparison to international directions, the country has low population densities of doctors and nurse, which may compromise its capacity to provide adequate health care and in turn health status. The country has an average of 22 doctors and 173 nurses per 100,000 populations, against the WHO recommended minimum staffing of 36 doctors and 356 nurses, respectively per 100,000 populations

(KIPPRA, 2013; Republic of Kenya, 2016). Lack of enough health care providers may hinder people from accessing and utilizing essential health care. High poverty levels experienced in the country, could explain the low levels of health care utilization as evidenced from the few births attended by skilled health workers and low immunization levels. Low health care utilization in turn could be the reason why the country has been experiencing poor health outcomes as indicated by high maternal and under-five mortality rates.

Attempts to explain the extent to which poverty affects health care utilization and health status have been made on specific segment of the population such as maternal health, child health and infants health in Kenya but not on the general health (Akunga, Menya, & Kabue, 2014; Kabubo-Mariara , Karienyeh, &

Kabubo, 2012; Mutunga, 2011). Existing studies also concentrated on parts of the country such as rural or slum areas and not the entire country (Muriithi, 2013;

Mutua, Kimani-Murage, & Ettarh, 2011; Ochako, Fotso, Ikamari, & Khasakhala,

2011). The studies used datasets collected before health care was devolved.

However, despite the limitations, the studies showed that individuals from wealthier families have better health and are more likely to utilize health care than those from poor families (Kabubo-Mariara et al., 2012; Mutunga, 2011; Ochako et al., 2011). Further, the studies showed that poverty reduces probability of seeking health care, and also influences individual choices on where to seek treatment (Awiti, 2014; Mwabu, Ainsworth, & Nyamete, 1993; Nyambura, 2016).

However, the findings may not be representative of the country, and hence cannot be generalized. This study filled the gap by firstly, focusing on effects of poverty

on general health as opposed to a specific segment of a population in Kenya. This gave a better picture of how the health status in the country is affected by poverty, hence on how best to address the problem. Secondly, the study took a national outlook, unlike other studies, which concentrated on small areas like slums in

Nairobi or rural parts of the country. Lastly, the study used the most current cross- sectional dataset, collected after devolution of health services, which was expected to give robust results and most recent trends in regard to health and poverty for policy purposes.

1.3 Research questions This study sought to answer the following questions:

i. What are the effects of poverty on health care utilization in Kenya?

ii. What are the effects of poverty on choice of health care providers in

Kenya?

iii. What are the effects of poverty on health status in Kenya?

1.4 Objectives of the Study The broad objective of the study was to investigate the effect of poverty on health care utilization, choice of health care providers and health status in Kenya.

The specific objectives were to:

i. Determine the effect of poverty on health care utilization in Kenya

ii. Investigate the effect of poverty on choice of health care providers in

Kenya

iii. Establish the effect of poverty on health status in Kenya

1.5 Significance of the Study The findings of this study will benefit key government ministries, departments, and agencies both at the national and county governments, and other health sector stakeholders involved in promotion and improvement of health and fight against poverty. The findings will inform them of the effects of poverty on health care utilization, choice of health care providers and health status. This information is intended to inform policy and planning in the health sector with an aim of ensuring better health for all Kenyans. The study has also contributed to the existing knowledge on the effect of poverty on health care utilization, choice of health care providers and health status in Kenya. The study has demonstrated that failure to control for potential endogeneity and unobserved heterogeneity may lead to understatement of effect of poverty on health care utilization, choice of health care providers and health status.

1.6 Scope of the Study The study sought to investigate the effect of poverty on health care utilization, choice of health care providers and health status among households in Kenya using Household Health Expenditure and Utilization survey data of 2013. The data was collected from a total of 33,675 households drawn from 1,347 clusters divided into 814 (60 per cent) rural and 533 (40 per cent) urban clusters. The survey covered 44 counties in Kenya. The counties not included in the survey were Garissa, Mandera and Wajir. The 2013 Household Health Expenditure and

Utilization Survey provides the most recent and relevant data set available.

1.7 Organization of the Study The study is organized as follows: Chapter one presents the background of the study, research questions, the objectives and the significance of the study. Chapter two provides a review of theoretical and empirical literature relevant to the study.

Chapter three presents the research design, theoretical framework, empirical models that were estimated, and data analysis and estimation procedures. Chapter four presents the findings of the study with respective discussions while chapter five presents conclusion, summary, policy implications, contribution to knowledge and areas of further research.

CHAPTER TWO

LITERATURE REVIEW

2.1 Theoretical Literature

A number of theories have been advanced to explain the relationship between poverty, health care utilization, choice of health care providers and health status.

These include neo-materialist hypothesis, behavioural model of health care utilization, Grossman‟s model of human capital, and the Acton‟s model of health care utilization. These theories have been reviewed in the sections that follow.

2.1.1 Neo-materialist Hypothesis

The Neo-materialist hypothesis was advanced by Lynch, Smith, Kaplan, and

House (2000). The hypothesis contends that differences in health among nations, regions, cities and individuals are as a result of the level and distribution of material resources within the population (Lynch et al., 2000). The proponents of the hypothesis argued that how a society decides to distribute resources among its citizens is an important contributor to the quality of various social determinants of health. The neo-materialist hypothesis suggests that the effect of income inequality on health is a combination of negative exposures and lack of resources held by individuals together with under-investment in human, physical, health and social infrastructure. This argument avers that bad health could be due to increased income inequalities that reduces state spending on health care, goods and services for the poor (Drabo, 2011).

The hypothesis brings out an understanding on how income and resource distribution may affect health of individuals. It shows the importance of the government in the provision of social services such as health care and how government‟s actions may affect health of her citizens. This is an important policy issue, which can be of much help to policy makers in deciding on how to allocate resources if they need to influence health status of the population. However, the hypothesis does not inform the amount of resources that can influence health positively. It also ignores other channels through which income may influence health such as environmental quality (Drabo, 2011).

Therefore, though distribution of material resources and society decisions to distribute resources, negative exposures and reduced government spending could be important determinants of health and health care utilization, this study did not incorporate them in the modeling framework. This is because their inclusion was beyond the scope of this study and there was possibility of data limitations especially on the negative exposures and the society‟s decisions on how to distribute resources among individuals.

2.1.2 Behavioral Model of Health Care Utilization

The behavioural model of health care utilization was formulated by Andersen

(1968). The model aimed at demonstrating the factors that lead to the use of health services. It was developed to assist in understanding why families use health services; to define and measure equitable access to health care; and to assist in developing policies to promote equitable access to health care services

(Awiti, 2014; Kimalu 2013; Kimani, 2014). The model considers an individual‟s 45

use of health services to be a function of three types of factors; predisposing characteristics, enabling and need factors (Andersen, 1995).

The predisposing factors are those conditions that influence people to use or not to use health services even though the conditions are not directly responsible for use. The predisposing factors include the socio-cultural characteristics of individuals that exist prior to their illness (Andersen & Newman, 1995; 2005).

The factors are based on the argument that a family‟s propensity to utilize health services can be predicted from a set of individual characteristics such as demographic factors, social structure, and health beliefs, which predate illness

(Andersen, 1995).

Demographic factors such as sex, past illness and gender represent biological imperatives suggesting the probability that individuals will need health services.

Social structure is a representative of a range of factors that determine an individual‟s status in the society, his/her ability to deal with challenges and the resources he/she commands to deal with the challenges. Social structure includes marital status, education, religion, household size, occupation, culture, ethnicity, social networks and social interactions (Awiti, 2014; Babitsch, Gohl, & Lengerke,

2012; Andersen, 2005).

Health beliefs are values concerning health and illness, attitudes towards health services and knowledge that people have about disease or illness and health services (Andersen, 1995). The health beliefs might influence individual‟s subsequent perceptions of need and use of health services (Andersen, 1995,

Kimani, 2014; Rebhan, 2011). Individuals who believe on the importance of health services are more likely to use them.

Enabling factors are those conditions that facilitate or impede use of health services. The factors are based on the argument that even if individuals have a predisposition to use health services, certain resources must be in place to enable them access health services (Andersen & Newman, 2005; Andersen, 1995).

Enabling factors include poverty status of an individual/household, illness level, availability of health facilities, access to health care, health insurance, health policies and other individual, family and community resources (Awiti, 2014;

Kimani, 2014).

Need factors draw from a premise that for individuals to utilize health services, there must be felt need from the individuals to use those services (Andersen,

1995). Need can be classified in two categories namely, perceived need and evaluated need for health services (Andersen, 1995; Kimani, 2014). Perceived need relate to how individuals perceive their general health. It includes whether or not people judge their problems to be of sufficient magnitude to warrant professional help. Evaluated need represents professional assessment and objective measurement of patient‟s health status and need for medical attention

(Andersen, 1995; Andersen & Newman, 2005; Rebhan, 2011).

The behavioral model gives a conceptual basis for understanding human behavior especially in regard to health care utilization. The model incorporates concepts and constructs that are representative of individual behavior as well as public

health and health care such as resources as discussed under enabling factors.

Thus, the model gives a basis for studying relationship between poverty and health care utilization since it captures well the issues of individual behavior and resource availability (Riegle & Stewart, 2013). The model also allows the unit of analysis to be either at family or individual level.

However, the model ignores an important environmental channel through, which income may affect health. Even though the model could either explain or predict use of health service, predisposing factors might be exogenous and enabling factors may be necessary but not sufficient. Therefore, assuming the presence of predisposing factors and enabling conditions, individuals must perceive sickness as a need for utilization of health services. This study sought to examine the effect of poverty on health care utilization and also on health status in Kenya. The study incorporated aspects of the behavioural model for health care utilization such as enabling factors like poverty status, and predisposing factors in order to understand how poverty affects health care utilization.

2.1.3 Grossman’s Model of Human Capital

Much of the economic theory of health analysis is based on the Grossman‟s human capital model (Grossman, 1972; 2000; 2004). The model explains individual‟s health status using the human capital theory. In the Grossman‟s model, individuals are assumed to maximize utility through consumption of health and non-health related goods subject to given income and wealth levels.

Grossman also held that individuals produce and consume their own health. In his approach, the individual chooses his level of health and, therefore, his life span. 48

Grossman (1972) showed that every individual inherits an initial stock of health, which depreciates with age. However, the health stock can be replenished by investments like health care, diet, and exercise. Thus, health care services are demanded in order to improve health status (Grossman, 1972; 2000). Other inputs individuals use to produce their own health include education, nutrition and lifestyle choices such as physical exercises, smoking and consumption of alcohol

(Mwabu, 2007; Kimani, 2014; Namubiru, 2014). Therefore, the level of health is not treated as exogenous but depends on the amount of resources the individual allocates to the production of health.

Grossman (1972; 2000) argued that health care demand differs from other goods and services because it is a derived demand. Thus demand for health services is derived from demand for good health. Good health increases individual‟s productivity and the total amount of time allocated on market and non-market activities. Therefore, health demanded is a consumption good, which enters directly into the individual‟s utility function. It is also an investment good, which increases the number of healthy days. The increased number of healthy days allows an individual to participate in both market and non-market activities, which in turn increase their earnings.

In the theory of consumer behaviour each individual has a utility function by, which various combinations of goods and services that can be purchased are ranked. The theory assumes that individuals are rational. Therefore, individuals will choose a most preferred bundle of goods and services from the feasible set of consumption bundle allowed by their budget. Thus, individuals will buy goods

and services that will generally increase their utility level (Grossman, 1972). The theory of human capital explains the motives for an individual to invest in human capital to raise productivity in both market and non-market sectors. The theory, therefore, highlights the role of human capital in producing earnings and commodities, which in turn feeds into the individual‟s utility function (Becker,

1962, Grossman, 1972, 2000).

Grossman (2000) also incorporated a household production function to explain the gap between health outcomes as an output and health care as an input.

Grossman stressed that some output of household production function enters directly into the utility function. Further, Grossman (2000) distinguished goods and services from commodities, by presenting commodities as a function of goods and services, and consumer time. Grossman (2000), indicated that individuals buy health services and other goods to produce health, which is a commodity. Health enters the utility function directly rather than health care being an input that enters directly into the utility function.

Despite the great theoretical and intuitive insights of the Grossman (1972) model in understanding demand for health and health care utilization it has been criticized by several researchers. For instance, the model is said to be deterministic, since it does not take into account random occurrences of illness or stochastic shocks (Hren, 2012). The model also assumes that health deterioration is due to age only. However, empirical research has shown that other factors such as lifestyle influences the deterioration of health. In the model, an individual is also assumed to have complete and perfect information about their health capital,

marginal benefits of current and future investment into health, current and future depreciation rate and interest rate, and complete insight into the health production process (Hren, 2012). The assumption of perfect knowledge is an abstraction in reality as it evades uncertainty linked to the stochastic nature of disease occurrence and the unpredictability of future health care expenses.

Further, the model simplifies the complexity of individual‟s health status into a binary state, which assumes the individual is either sick or healthy (Galama,

Hullegie, Meijer, & Outcault, 2012). However, as argued by Liljas (1998), a more realistic approximation can be introduced by a continuum of health states.

Although the model has received some criticism since its inception, it is unique in its approach within the context of health economics to both theoretically and empirically conceptualise a complex demand for health. Also, theoretical extensions and competing models are relatively low and so, the Grossman (1972) model has remained influential in understanding demand for health and health care services.

2.1.4 Acton’s Utility Maximization Model of Health Care Demand

The theory was advanced by Acton (1975). The model extended the Grossman

(1972) model by embracing the argument that time costs and other demographic factors are involved in the consumption of health care. Acton‟s model starts from a behavioural model of utility maximization, where utility depends on health care and the consumption of other goods. In the model, on experiencing an illness or an injury, an individual is assumed to choose among various treatment alternatives including no treatment so as to maximize utility subject to budget 51

constraint. The individual is constrained by both monetary and non-monetary costs such as travelling and waiting time while seeking health care (Mwabu,

1993). The impact of these monetary and non-monetary costs in accessing health facilities are seen as defining the quality of a particular facility or a certain healthcare provider (Mwabu, 1993).

The model concentrated on the role of monetary and non-monetary costs in determining demand for health care. The model explains that, demand for treatment in response to a particular episode of illness or injury can be modeled in terms of the provider choice. The choice of health care providers involves selecting among various available alternatives such as public, private or no care.

Thus, Acton (1975) recognized the discrete nature of health care decisions in health care demand. Therefore, estimation of demand for health care calls for use of discrete choice models (Mugwila, 2005).

In the model, Acton (1975) assumed that individuals consider health care as a normal good. The implication of the assumption is that, the effect of unearned income has a positive effect on demand for health care. Further, the assumption implies that, the earned income has a negative impact on the demand for health care. This is because, in the case of unearned income, people with higher incomes buy more of normal goods. However, in the case of earned income, the increase in wages raise income and the opportunity cost of time, which increases the time cost component of consumption activities. Consequently, goods and services that require relatively large commitments of time for them to be consumed become

more expensive and, therefore, substituted by other goods and services, which require less time.

The major advantage of Acton‟s model is its simplicity in application and appreciation of other demographic variables in determining demand for health care. The model also gives a theoretical basis for analysis of discrete choice models. However, the model ignores the role of health need. It also fails to acknowledge that, health care has a derived demand for good health, which should enter the utility function directly. Further, the direct inclusion of time is not logical, since it violates the household production theory, where time enters the budget constraint. This is because, in the household production theory, a household is seen as a production unit, which combines its own time with market purchased goods to produce pleasure giving commodities such as health.

However, despite, the model‟s limitations, it gives the theoretical foundation for analyzing how individuals choose health care providers, which requires use of discrete choice models (Mugwila, 2005). The model, therefore, gives a starting point for this study. The current study did not include health care services directly into the utility function but through the health care production function. This is because, the study appreciates that health care has a derived demand for good health (Grossman, 2000). Thus, health care can only enter the utility function through the health demand function. This study also did not incorporate time used in production of health, although it is an important variable in production of health. This is because incorporation of time costs was beyond the scope of this study, and also data on time costs was not readily available.

2.2 Empirical Literature

A number of empirical studies have been done looking at the relationship between poverty, health care utilization and health status. Among the studies includes:

Mwabu et al. (1993) conducted an empirical analysis of the relationship between medical care quality and medical care demand in rural Kenya. The study used data from a randomized household survey, enriched with exogenous information on health facility attributes collected in 1980 and 1981. The data was collected from 1,721 individuals spread across 315 households. At facility level, data was collected from 8 health facilities. The authors estimated multinomial and nested multinomial logit models. Among the variables included in the study were type of healthcare provider, distance to the nearest health facility, sex, education level, user fee, availability of drugs, number of health workers in a health facility and household income.

Study findings indicated that availability of drugs in a health facility is positively related to medical use. Also lack of prescription drugs was found to be positively related to medical care demand. Further, health care demand was found to decrease with user fees and with greater distance to the provider, but increased with income. Variables and estimation techniques used were appropriate. The dependent variable was discrete and so applying discrete choice model is recommended (Mugwila, 2002). However, the study findings cannot be generalized as data was collected from a small rural area in Kenya. The data was also collected in 1980 and 1981 and since then a lot has changed in Kenya. The current study used a recent and richer data in terms of required information, which 54

was collected in 2013 and is nationally representative and hence the findings can be generalized.

Asenso-Okyere, Dzator, and Osei-Akoto (1996) estimated a disease specific demand function to study the determinants of utilization of the services of a health care provider or a treatment regiment for malaria. The study used survey data collected in two districts in Ghana. The data was collected from 1,389 households consisting of 5,323 individuals. The authors used a multinomial logit framework.

Variables studied included type of health care provider, price, travel time, waiting time, distance, drugs provision, and age. Other variables were ethnicity, religion, education, severity of illness, gender and household expenditure.

The results confirmed the popular use of self-medication as a first choice of action in treating malaria. The choice of malaria care providers was found to be influenced by facility price, travel time, waiting time for treatment, education, age, sex and quality of care measured in terms of drugs availability. The results further showed that, as income increases, the odds are in favour of self-medication when people get malaria. The study used a rich data type collected from a large representative sample. Since the dependent variable was polychotomous in nature, the authors used appropriate methodology. Also variables included in the study are supported by recent studies such as Awiti (2014). However, the study only looked at health service utilization in regard to only one disease yet there are so many other ailments that affect people. Thus, people who did not suffer malaria and were sick were left out even if were using health services. This may not give the true picture on usage of health services.

Hallman (1999) studied child health care demand in Philippines which is a developing country. The author examined how quality, price and access to curative health care influence use of modern public, modern private and traditional health care providers. The study involved a sample of 3,000 children aged 0-2 years in Cebu, Philippines. The study relied on household, community, and health facility data from the Cebu Longitudinal Health and Nutrition Study collected in 33 rural and urban communities during 1983-1986. This period coincided with the country‟s severe economic downturn and also introduction of structural adjustment programs. For analysis, the author made use of multinomial logit model. Child age, parental education, mother‟s height, value of household assets and number of health personnel were among the variables studied. Others included availability of drugs, user fees, type of health care providers, distance to the nearest hospital and number of hospital visits.

Study results indicated that distance to the nearest health facility reduced demand for utilization of modern and private health services. Further, the results showed that maternal human capital and household income increased utilization of private health service providers. It was also found that non modern health care practitioners were used more if they had recently attended health training session, regardless of whether it was sponsored by government or non-government agencies. Children who were male and younger than 6 months of age were found to be more likely to be taken to private and traditional health care providers, than to public health facilities. The study used a rich data set and hence included

relevant variables as used in similar empirical studies (Mwabu et al., 1993). It also used appropriate methodology for estimation making the results robust.

Gelberg, Andersen, and Leake (2000) examined determinants of health care use and health status of 363 homeless individuals from a longitudinal study: RAND

Course of Homelessness Study, conducted in 1990 and 1991 in Los Angeles,

United States of America. Variables included in the study were age, gender, ethnicity, education, employment status, length of residence in Los Angeles, mental health status, criminal history, substance use, homeless history and victimization.

The study examined four health conditions namely high blood pressure, functional vision impairment, skin/leg/foot problems and tuberculosis skin test positivity. Multivariate techniques used in the study included least squares linear regression and multiple logistic regressions. Use of health care services was found to improve self-reported health status. Results further indicated that use of health services between baseline survey and follow-up surveys was statistically and significantly associated with better far, but not better near vision.

Lichtenberg (2002) examined the effects of medical care on health care utilization and outcomes by single year of age, for ages close to age 65 years in America.

The study used data from various sources namely, National Hospital Discharge

Survey, 1979-1992, National Ambulatory Medical Care Surveys, 1973-1998,

Medical Expenditure Panel Survey, 1996 and National Health Interview Survey,

1987-1991. Among the variables used in the study were survival rates, number of

physician visits, mortality rates, morbidity incidences, age, gender and use of ambulatory care. The study used both descriptive and econometric regression techniques to achieve its objectives.

Findings of the study indicated that insurance induced increase in health care utilization led to a reduction in days spent in bed by about 13 per cent and reduced growth in the probability of death after age 65. Physician visits had a negative effect on the male death rate, conditional on age and the death rate in the previous year. The short-run elasticity of the death rate with respect to the number of physician visits was -0.95, and the long-run elasticity was -0.497. A sustained 10 per cent increase in the number of physician visits led to a 5 per cent reduction in the death rate.

Wang (2003), using Demographic and Health Survey (DHS) data for the period between 1990 and 1999 from over 60 low-income countries, investigated determinants of child mortality both at the national level and for rural and urban areas separately. Among the studied variables were mortality rates, asset index, use of antenatal care, use of vaccination, education level, access to electricity, access to safe water, access to sanitation, gender and share of health expenditure to GDP. Other variables included region, wealth index, level of urbanization, quality of government and cultural effects. Various regression techniques such as

OLS and Weighted Least Squares (WLS) were applied to achieve objectives of the study.

Results of the study showed that, at the national level, access to electricity, vaccination in the first year of life and public expenditure statistically and significantly reduced child mortality. In urban areas, only access to electricity had a significant health impact while in rural areas, increased vaccination use was important for mortality reduction. Female education, access to safe water, access to sanitation and vaccination coverage were found to be jointly statistically and significantly reduce infant mortality rates in rural areas. This was a cross-country study which used rich data for various DHS making the results more robust.

However, the study results are highly aggregated making it difficult to deduce any relevant country-specific policy. Thus, it is important to have a country specific study that investigates how health care use affects health status.

Borah (2006) investigated the determinants of outpatient health care provider choice in rural India using data from National Sample Survey Organization of

India of 1995. The author used a mixed multinomial logit framework to model health care utilization. The variables used in the study included type of health care provider, household‟s expenditure, residence, age, gender, education level, severity of illness, distance to nearest hospital, price of services and household size.

The results of the study showed that price and distance to a health facility play a significant role in health care provider choice decision. However, when health status is poor, the results showed that distance plays a less significant role in an adult‟s provider choice decision. Further, the results indicated that price elasticity of demand for outpatient care varies with income. People in lower-income groups

were found to be more price sensitive in their choice of given health care provider than those in higher-income groups.

Results also showed that price elasticities were generally higher for children than for adults except for those in top income quartile. Further, results for both boys and girls showed that, boys‟ price elasticities in lower-income groups were generally higher for private hospital care in all price ranges while girls‟ price elasticities were higher for private doctors. Increase in the price of health care by private doctors would reduce girls‟ demand for care from private doctors more than that of boys for all price ranges and for all income groups. The study used appropriate methodology to achieve its objectives.

Amaghionyeodiwe (2008) investigated the determinants of households‟ choice of health care provider in Nigeria. The study used primary data collected from a sample of 7,920 households from both rural and urban areas. The study used a nested multinomial logit to analyse the data. Variables used in the study included distance to hospital, consultation fee, drugs availability, number of medical staff and availability of medical services. Other variables included age, location, household income, type of illness, household health care expenditure, gender, tribe, education, household head and type of health care provider.

Study findings revealed that both distance and money prices are significant factors in discouraging individuals from seeking modern health care services but money prices were less important as a determinant of the choice of health care provider. Income was found not to be a statistically significant determinant of

health care provider‟s choice. However, for the middle income group, income was found to be a statistically significant determinant of health care provider‟s choice for private clinics and private hospitals. In examining how households choose health care providers, the study did well in linking the demand and supply factors, and this is important.

Thus, the study appropriately used the demand and supply factors. In line with other studies, the study used nested multinomial logit, which is less restrictive because patient‟s alternative choices were more than two and were assumed to be distinct. Such choices can be assumed to have different attributes and therefore could be considered to be mutually exclusive (Mwabu et al., 1993).

Horta, Gigante, Candiota, Barros, and Victora (2008) carried out a study in

Pelotas, Southern Brazil to analyze the causes of death from 1982 to 2006. The study aimed at assessing mortality in a birth cohort followed between 1982 and

2006 and its associated factors. The study used data collected from hospitals, cemeteries, and death registries through the Mortality Information System. The variables analysed included gender, color of mother, mother's schooling rate, family income, weight at birth, weight and height per age. The authors used a

Poisson regression technique to estimate the relative mortality risk.

The study established that low-income children had higher mortality rates in all age brackets. The study also indicated that low socioeconomic level at infancy was associated to higher mortality rates during the whole period of the study. Due to infectious diseases, low income was found to be associated to a higher risk of

death. However, the study omitted some key variables such as occupation of the mother, marital status of the mother and household size. The omitted variables are important as they are expected to affect health outcomes through altering the household preference function (Kabubo-Mariara, Mwabu, & Ndeng‟e, 2009).

Buddelmeyer and Cai (2009) carried out a study to examine the joint dynamics of health and poverty in Australian families. In particular, the study was interested in establishing the causal relationship between health and poverty. The study used a balanced panel data consisting of 1,769 families, obtained from the Household,

Income and Labour Dynamics (HILDA) Survey (2001-2006). The study used a joint dynamic model for estimation in order to control for the correlation between the unobserved determinants of health and poverty. Poverty in this study was defined at the family level whereas health was defined at individual level. The study used a standard self reported health status as a measure of general health.

Poverty was measured using equivalised family disposable income. The study adopted family as the unit of analysis implying that health was represented by the family head as the reference person of a family.

The findings indicated that the relationship between health and poverty could be confounded by unobserved heterogeneity. Particularly, it was found that families headed by a person in ill-health were more likely to be in poverty compared to families headed by a person with good health. On the other hand, a head of a family whose family was in poverty in the current year was more likely to be in ill-health in the following year compared to a family head whose family was not in poverty. The type of data and the methodology applied was appropriate. This

study though not looking at causality borrowed some ideas such as measurement of general health. However, instead of measuring poverty using equivalised family disposable income, the current study used asset ownership due to data limitations. Previous studies have shown that the asset index is an important measure of wealth just like expenditures or incomes, whether instrumented or not in explaining health outcomes (Kabubo-Mariara et al., 2012).

Halasa and Nandakumar (2009) examined factors influencing a patient‟s choice of provider for outpatient care services in Jordan. The authors used data from a sample of 1,031 outpatients from the Jordan health care utilization and expenditure survey of 2000. The study used a multinomial logit model to achieve its objectives. Variables used in the study included type of health care provider, sex, geographical location, age, education, employment, marital status, wealth index, household size, out-of-pocket expenditure and insurance. Other variables used included health status, waiting time, privacy of medical examination, staff treatment and sufficient treatment time.

Results of the study indicated that place of residence, insurance status, poverty level, marital status, cost of treatment, quality of care, and health status were statistically significant and influenced the patients‟ choice of healthcare provider.

The study findings further indicated that education, age and family size had no significant influence on the choice of health care provider. Due to the multi- category nature of the dependant variable, the study used multinomial logit for estimation, which was appropriate as suggested in other empirical studies

(Amaghionyeodiwe, 2008; Asfaw, 2003; Awiti, 2014).

Kabubo-Mariara et al. (2009) analyzed the evolution and determinants of children‟s nutritional status in Kenya using KDHS data for 1998 and 2003. The study used a sample of 2,914 and 2,956 children aged less than 36 months in 1998 and 2003 respectively. Variables included in the study were age of the child, weight of a child, height of a child, ethnicity, availability of information, household size, gender, age of household head, education of household head, mother‟s height, mother‟s education, asset index, religion, region and fertility preferences. Other variables included health care variables such as use of vaccination, prenatal and delivery care by a health care professional and use of modern contraceptives. The study employed descriptive and regression techniques to achieve its objectives.

Results of the study showed that in 1998, all vaccination and prenatal care by a professional health care staff had negative signs instead of the expected positive signs. In 2003, the vaccination had unexpected sign but was statistically significant. In both periods, access to modern contraception methods and prenatal care by a health professional in 2003 were highly correlated with children‟s height. Further, the results showed that fertility preferences were statistically insignificant implying that they are not important determinants of children‟s nutritional status in Kenya.

In measuring children‟s nutritional status, the study used height for age and weight for height, which are indicators of children health status. The study used relevant variables as indicated in reviewed literature and the methods used were appropriate (Abuya, Ciera, & Kimani-Murage, 2012; Namubiru, 2014). However,

the study was limited in its scope as it only looked at the children health status, and ignored the general health of the rest of the population. It also used data, which is almost 10 years since it was collected and since then a lot has changed in the health system in Kenya including devolution of health.

Cisse (2011) analysed health care utilitzation in Cote D‟Ivoire using data from the survey of the National Institute of Statistics for 1993. The study was anchored on the theory of utility maximization to produce health. A multinomial logit model estimated. Variables used in the study included choice of health care provider, household income, household size, religion, education, age, ethnicity, gender, religion, price for medication and time taken to reach a health facility.

The results of the study showed that the education level of the household head, the household‟s income, the price of medication and the time taken to reach the health care provider, determine the choice for a specific health care provider.

Further, it was found that the level of education and income positively affect choice of health care provider. In addition, it was established that the cost of medication and the time taken to reach to health care provider negatively affect the choice of health care provider. Methodology used and variables chosen were appropriate for the study.

Klaauw and Wang (2011) examined the determinants of mortality at different ages of a child in rural India. The study used data from the National Family

Health Survey of 1998/1999 of India. The variables analysed included religion, wealth, immunization rates, gender, parental education, availability of sanitation,

source of drinking water, type of cooking fuels and availability of medical services. The data was analysed using flexible parametric framework based on hazard rate model.

The study found that infant and child mortality are negatively and statistically correlated with the household wealth index constructed based on asset ownership.

The study findings further showed that increased schooling expenditures, higher female immunization rates and lower poverty levels significantly and statistically reduce child mortality. A higher head-count poverty rate within a state was found to increase child mortality rate. Attainment of parental education was found to be statistically significant in reducing child mortality especially when parents had completed primary education. The choice of variables and the model used in the study was appropriate. The current study borrowed some explanatory variables to use such as parental education and the measure of poverty using asset index.

Mutua et al. (2011) carried out a study to determine the vaccination status of children aged between 12-23 months living in two slums of Nairobi and to identify the risk factors associated with incomplete vaccination. The study used data from the Maternal and Child Health component of a longitudinal study done in Korogocho and Viwandani slums of Nairobi by African Population and Health

Research Center (APHRC) since 2006. The study used data collected from children aged 12-23 months totaling to 1,848. These were children who were expected to have received all vaccinations recommended by the WHO. All vaccination data was collected during household visits from vaccination cards and also from reported information by the mothers. In order to identify the risk factors

associated with incomplete vaccination, the study used multivariate models.

Variables used in the study included vaccination status, sex, maternal education, parity, household asset index, ethnicity and birth weight. Others included antenatal care, postnatal care, marital status, maternal age at index child‟s birth and place of delivery.

Results showed that at 12 months, up-to-date coverage with all vaccinations was

41.3 per cent and 51.8 per cent with and without the birth dose of Oral Polio

Vaccine (OPV), respectively. Multivariate results indicated that poverty measured with household assets index and household monthly expenditure, place of delivery, ethnicity, level of mother‟s education, parity and age were predictors of full vaccination among children living in the slums. Marital status of the mother was found not to be statistically significant associated with full child vaccination.

To achieve the intended objectives, the methods, variables and data type used were appropriate as evidenced in other similar studies (Kamau & Esamai, 2001).

However, even though the study contributed in knowledge of risk factors associated with incomplete vaccination, the findings were of a limited geographical area and population. Thus, the findings may not be a good representation of the country‟s situation on health care utilization. The current study used a nationally representative dataset with large sample size. Hence the findings therein are expected to be more robust and can be generalized.

Mutunga (2011) examined how infant and child mortality is related to the household‟s environmental and socio-economic characteristics in Kenya. The

study used the Kenya Demographic Health Survey (KDHS) data for 2003. Among the studied variables included mother‟s education, source of drinking water, sanitation facility, type of cooking fuels, access to electricity, wealth index, household size and sex. Data was analysed using the Weibull and Cox models.

Study findings indicated that children born in wealthier families had better survival prospects. Further, child mortality was found to be negatively related to household size an indication that lower child survival prospects are experienced in smaller households. The choice of variables and data in the study are appropriate.

However, the study by only considering infant and child health status missed an important segment of the society. The study did not incorporate the youth and the adult health status, which is critical in an economy as they form a large part of the productive population of a country. The current study addressed this by incorporating the health status of the youth and adult population.

Ochako et al. (2011) studied utilization of maternal health services among women in Kenya. The study used data from the 2003 Kenya Demographic and Health

Survey (KDHS), with a focus on young women aged 15-24. Variables used in the study included timing of first antenatal clinic (ANC) visit, education, household wealth, urban-rural residence, ethnicity, parity, age at birth of the last child and marital status. Timing of the first antenatal clinic was used as the dependent variable. To achieve its objectives, the study used multivariate ordered logistic regression.

The study found that household‟s wealth, ethnicity, education, parity, place of residence, marital status and age at birth of the last child had strong influence on timing of first ANC visit and the type of delivery assistance received. Women from rich households were found to be more likely to seek antenatal care early than those from poor households. The study also found a strong association between early timing of the first ANC visit and the use of skilled professionals at delivery.

The results further indicated that a large percentage of young pregnant women do not seek ANC during the first trimester as is recommended by WHO. The sample size used in the study was large enough to make inference, and variables and methods of analysis were appropriate to achieve the intended objectives.

However, the study was only concerned with maternal health and, therefore, it could not say anything about the health of the rest of the population. The current study addressed the limitation by considering the general health of the population so that it can inform policy formulation capturing the larger society. General health at individual levels, which can be measured using self reported health status is important as it can capture all segments of the population as opposed to a small segment (Awiti, 2013).

Skordis-Worrall, Hanson, and Mills (2011) estimated demand for health services among residents of the poorest communities in four districts of Cape Town, South

Africa. Data was collected from 250 individuals aged above 18 years and spread across 144 households. Two models of health care were presented: one estimating the probability of using any service, and the other modeling the number of visits

among users. The authors estimated a random effects probit model to predict likelihood of using any provider in a given time and period.

They also used a negative binomial model to analyse the number of visits to any provider amongst the users. The study used variables such as number of visits to health care provider, travel time, waiting time, availability of drugs, age, gender, wealth, education level, insurance and type of illness. Other variables included severity of illness, extra household funds, membership to a savings scheme, self reported health status, household expenditure, household income, and price of drugs.

Study findings indicated that health care use is predicted by perceived financial situation, gender, mental and physical health, extra-household resource and the price of a private consultation. The study further found that the number of visits to a health care provider was predicted by age, physical and mental health, extra- household resources and quality of the private provider. The findings showed that extra financial resources by a household enable access to health care service. Men were also found to be less likely to seek care at the same level of illness as women. Estimation techniques, variables and data type employed in the study were appropriate. The current study borrowed from the estimation of health care utilization using number of visits to health care provider as used in this study.

Number of visits to a health facility is a good measure of health care utilization

(Kimani, 2014).

Salihi et al. (2012) investigated the relationship between poverty and health in

Nigeria using HIV prevalence as the case study. The study used data from the

Nigeria National Bureau of Statistics for the years 1990-2009. The study used

Granger Causality and co-integration techniques to investigate the causal relationship between poverty and HIV in Nigeria. It used Vector Error Correction

Model (VECM) to estimate the long-run relationship between poverty and HIV prevalence.

The authors found that there were no direct and statistically significant linkages between poverty and HIV prevalence. This was an indication that poverty is not the driving force behind increased rates of HIV prevalence in Nigeria. Choice of model for the study to address the intended objectives was appropriate. However, the number of observations for a time series study was inadequate and this could lead to inappropriate results. The study also made an assumption that other variables such as employment status, residence, marital status and education cannot influence poverty and HIV prevalence. Failure to include such variables in the model may lead to misspecification of the model. The current study was not interested in causal relationship but rather on one direction effect. This study used cross-sectional data and included more variables to investigate the effects of poverty on health status.

Kabubo-Mariara et al. (2012) analysed child poverty in Kenya using two measures of child well-being, namely child survival and asset index. The study further analysed determinants of child survival. The study used 1993, 1998 and

2003 rounds of KDHS data. The authors employed the Weilbull parametric survival-time model for analysis.

The study findings indicated that asset-poor children were more likely to experience death than asset-richer ones. This suggested a strong responsiveness of mortality to change in assets. In addition, the results showed boys, first born and children of multiple births have lower survival time than the respective reference groups. Maternal education was found to significantly lower the risk of mortality, while age variables suggested the importance of reducing teenage births. The variables, data and methodology used in the study were appropriate. The current study borrowed from this study methods of measuring poverty using asset ownership.

Wang, Chen, Hsu, and Wang (2012) examined the association between reduced health care utilization and health outcomes measured by cause-specific mortality rates in Taiwan. The study applied a natural experiment design to the Taiwan population between 2000 and 2004, which included the 2003 severe acute respiratory syndrome (SARS) epidemic. The study applied poisson regression models to estimate mortality rates since it used count data. Among the variables included were sex, age, 5 year dummy variables, 12 indicator variables of month, number of mortality cases, intervention variable-SARS and monthly disease- specific utilization.

Results of the study showed that compared with the non-SARS period, mortality caused by diabetes mellitus and cerebrovascular diseases increased statistically

and significantly when health care utilization decreased by an average of 20 per cent in 2003. The study concluded that the statistically and significantly negative association of health care utilization and the mortality suggested that a dramatic decrease in health care utilization within a short period increased the risk of mortality from specific disease, indicating the value of health care.

The study employed rightly the poisson regression techniques which are required when using count data (Cameron & Trivedi, 2005). Although the study contributes on the evidence of how health care utilization affects health status, the study was done in Taiwan, which is different from Kenya. The study also concentrated on disease-specific and not the general health status, and so it may not be possible to conclude the effects in relation to other diseases not studied.

Ali and Noman (2013) carried out a study on the determinants of demand for health care in Bogra, Bangladesh. The study used cross-sectional data collected in

2007 from a sample of 276 randomly selected patients. Among the variables included in the study were income, cost of drug, age, level of education, duration of illness, distance, waiting time, quality and price of health care. A binary logistic regression model was employed to identify the determinants of the demand for health care.

The study results showed that the estimated coefficient of price variable is negatively related to demand for health care and it is statistically significant.

Further the results indicated that the level of education and income has positive effect on demand for health care. Moreover, the study found a positive relation

between waiting time and demand for health care. The study used relevant variables and methodology. However, the sample was small to make conclusion for a cross-sectional study and this may affect the robustness of the results.

Muriithi (2013) carried out a study on the determinants of health seeking behaviour in Kibera slum in Kenya. The study aimed at explaining the underlying determinants of the demand for health care services in the slum. The study used survey data collected in 2008 in Kibera slum. Variables used in the study included self treatment, user fees, distance to health facility, sex, age, trust index, health information and quality of health facility. Others included household size, occupation, acreage of land holding and health information score. The size and magnitude of acreage of land holding was used as a proxy for asset base indicating household wealth level. The study tested the hypothesis that the information available about service quality in a health facility affects demand for health care. Multinomial logit model was used in the study.

The study results indicated that service quality, information about the service quality, wealth, user fees, and gender are the main determinants of patient‟s choice among alternative medical treatments. The results further showed that people with more resources are less likely to seek medical care from a public facility. However, since the study did not specify the sample size used, it is not possible to determine the appropriateness of the methodology used. In addition, slums are heterogeneous and form a small portion of the country‟s population

(Abuya et al., 2012). Thus, the results may not be used to explain determinants of demand for health care in a conclusive manner. The current study differed with

this one in that it used a nationally collected data, which is richer in information regarding health care seeking behavior, utilization and health status in Kenya.

Adeoti and Awoniyi (2014) examined the determinants of health status and the demand for health care in Nigeria. The study used Nigeria‟s Demographic Health

Survey data for 2008. Estimation methods used included Ordinary Least Squares

(OLS), Two stage least square (2SLS) and Control Function Approach. Study findings showed that immunization was not significant in reducing underweight among children born in poor households. However, in the non-poor households, immunization was found to significantly reduce underweight.

Further, the study found that children in poor households born by educated parents and living in urban areas were better off than those in rural areas in terms of children health status measured by anthropometric measures of children

(underweight). Results further showed that age of the child, mother‟s educational status, household size and area of residence significantly affect child‟s health status. The study used appropriate variables, type of data and methodology making the results robust. The methodology was appropriate since Body Mass

Index (BMI) used as a proxy for maternal health status is continuous. However, the study concentrated on the health status of children. The current study focused on the general health status of all population groups.

Akunga et al. (2014) examined determinants of Postnatal Care (PNC) use in

Kenya using KDHS (2008-2009) data for a sample of 3,970 women. Data was analyzed using bivariate and multivariate techniques. Variables used in the study

included PNC use, skill of antenatal care (ANC) provider, skill of delivery provider, birth interval, age of the mother at the time of her last birth, religion, marital status, residence, region and wealth index. Other variables were level of education of the mother, size of the baby, weight of the baby, whether pregnancy for the last child was wanted or not, birth interval of the preceding child, number of ANC visits and place of delivery.

Results of the study indicated that lack of education and unskilled delivery are associated with low use of PNC services. Factors, which were found to be associated with PNC use are mothers‟ age at delivery of the last child, four and above ANC) visits, urban residence, and skilled delivery. However, the study found that, birth interval, number of children ever born, marital status, religion and wealth index do not determine use of PNC. The study used a large sample size, enough to produce robust results. However, findings of the study may not be used to inform policy makers on utilization of health care by other segments of the population such as infants and children. The current study addressed this by considering the general utilization of healthcare by using number of hospital visits by those who reported to have been sick.

Awiti (2014) examined the effect of poverty on an individual‟s choice of a health care provider in the event of sickness or injury in Kenya. The author used data from the Kenya Integrated Household Budget Survey which was carried out between 2005 and 2006. The data was analysed using a multinomial probit model.

The study found that poverty reduces the probability of visiting a modern health

care provider amongst all age groups (infants, children aged 1 to 5 years, children aged 6 to 14 years, and adults).

The results showed that among infants, poverty increases the probability of not visiting any health care provider when ill but reduces the probability of visiting a modern health care provider, holding other factors constant. The results specifically showed that among infants, poverty increases the probability of not visiting any health care provider by 0.375 but reduces the probability of visiting a modern health care provider by 0.449, holding other factors constant.

Further, the results showed that living in the rural areas as opposed to living in urban areas increased the probability of visiting non-modern health care provider by 0.070 but decreased the probability of visiting a modern health care provider by 0.080, holding all other factors constant. Also an increase in the average distance to the nearest health facility by one kilometer was found to reduce the probability of visiting a non-modern health care provider by 0.005 but increased the probability of visiting a modern health provider by 0.005, holding other factors constant.

The study used appropriate methodology, and variables. However, the data type used was collected more than 10 years ago. Since then there has been change of political landscape and adoption of new constitution, which led to devolution of health services. Thus, the findings of the study may not help much in policy direction. The current study used a more recent rich data collected nationally in

2013. This gave more updated information and may help much in policy direction.

Edeme et al. (2014) used a survey data from Multiple Indicator Cluster Survey

(2012) and General Household Survey (2012) to examine the relationship between household income and child mortality in Nigeria. The study used both random and fixed effects models. The study modeled neonatal, infant, and under- five mortality rate against household income in Nigeria. The study controlled for access to safe water and sanitation, maternal education, household size, and access to antenatal care. Study findings showed that household income had negative effect on neonatal, infant and child mortality in Nigeria. Further, the study found that increase in household size worsened neonatal and infant mortality rates given low household incomes. The approach used in the study in terms of methodology, data type and variables was appropriate and the results are robust.

Fink, Günther, and Hill (2014) assessed child morbidity and mortality differences across residential areas focusing on health disparities between poor urban slum areas and rural and better-off urban areas. The study used data from 191

Demographic and Health Surveys (DHS) across 73 developing countries. The study used Cox proportional hazard model to assess the associations between residence and mortality outcomes. Further, a logistic model was used in the study to analyse morbidity outcomes. Study findings indicated that children in slums have better health outcomes than children living in rural areas but worse off than children in better-off neighborhoods of the same urban settlements. Much of the

observed health differences were explained by large differences in maternal education, household wealth and access to health services across residential areas.

The study results further indicated that maternal education and household wealth are important in protecting children from health risks. The study concluded that children in cities have relatively good health outcomes regardless of the nature of their neighborhood. It indicated that children in smaller urban areas and especially in slums of such towns fare less well and better than rural children on most outcomes. This indicated that urban advantage still persists. The study chose appropriate variables and type of data. Although the study used Cox Proportional

Hazard for assessing associations of residence and mortality outcomes which is appropriate, it consistently used fixed effects ignoring the fact that countries differ and hence there was need for use of random effects. The study also used a pooled data for 73 countries. This means the results are not representative of a specific country and, therefore, cannot be used to give a clear policy guideline for a country.

Hamad and Rehkopf (2015) examined the effects of income on perinatal health, which is health pertaining to the period immediately before and after birth, in

United States. The study used a sample of 2,985 women surveyed in 1979

National Longitudinal Survey of Youth and their 4,683 children born during

1986-2000. The data collected annually in 1979-1994, and biennially thereafter up to 2000. The authors used multivariate linear regressions and linear probability models in their study. Some of the variables included in the study were breast- feeding status, use of alcohol and tobacco during pregnancy, birth weight,

household income, child‟s gender, mother‟s age, marital status and educational attainment. Other variables included utilization of prenatal and postnatal care.

The study findings showed that higher income had a statistically significant association with an earlier first prenatal care visit and higher birth weight. Further, the results indicated that higher income was statistically and significantly associated with reduced likelihood of tobacco use during pregnancy, and increased likelihood of attending a well-child check in the first month after birth.

The study used multivariate and linear probability models suitable for binary outcomes, controlling for covariates. However, the study only looked at effect of income on use of health services immediately before and after birth.

The study did not look at how income affects use of health services during the other periods such as during early stages of pregnancy, and later stages after delivery. Health care utilization in early stages of pregnancy through ANC visits can help detect pregnancy complications early enough and hence can be controlled (Ochako et al., 2011). Health care utilization in later periods after delivery may improve maternal health outcomes for both the mother and the child

(Akunga et al., 2014). Such health outcomes can easily be captured using self reported health status and number of visits to a hospital.

Teerawichitchainan and Knodel (2015) examined association between poverty, economic inequality, and health among elderly in Myanmar using a national

Aging Survey data for 2012. The authors used ordinary least square (OLS) and binary logistic regression models to examine how economic status (proxied by

wealth index) is associated with health statuses (measured by self reported health status, functional limitations, sensory impairment and disability status).

The authors found a significant association between wealth quintiles and most health indicators included in the study. The results revealed that self-assessed health, sensory impairment and for most part, functional limitations improve with higher levels of economic status. The improvement as per the authors is consistent and statistically significant with increasing wealth quintiles. The study used appropriate variables and methodology to address the issues at hand. However, despite the study giving important information on how poverty is associated with health status of the elderly, it left out the majority of the population, which is less than 60 years old. The current study factored in this segment of the population to understand how poverty affects health status.

2.3 Overview of Literature

The theoretical literature reviewed neo-materialist hypothesis, the behavioural model of health care utilization, the Grossman model of human capital and the

Acton‟s utility maximization model of health care demand. The analysis of the theoretical literature showed theories explaining linkage of poverty and health exists. Theoretically, there are underlying economic, social and demographic foundations that determine health and consequently utilization of health care.

From the theories, it was clear that socio-economic status is important in determining the health status of a population. Further, the theories suggest that,

health status improves with improvements in socio-economic status. Low socio- economic status pushes households into poor health status.

The theories that were most applicable to this study were the Grossman‟s model of human capital, the behavioural model of health care utilization and the Acton‟s utility maximization model of health care demand. This was because the

Grossman‟s model helps in understanding how individuals who have limited resources would behave in case their health status deteriorates. This model was important in achieving the objective seeking to understand how poverty affects health care utilization and health status.

The behavioral model of health care utilization helps in understanding what determines health care utilization and how individuals decide where to seek health services once they fall sick. This assisted in modeling effects of poverty on health care utilization. On the other hand Acton‟s utility maximization model of health care demand gives a theoretical foundation for analyzing individual‟s behaviour in choosing health care providers, which needs application of discrete choice models. This assisted in understanding how poverty affects choice of health care providers.

Neo-materialist hypothesis is mainly concerned with the level and distribution of material resources by the society among its citizens. It specifically considers the role of government in distribution of resources and how the distribution affects health outcomes. This study did not look at how the society decides to distribute resources among its citizens. The study was also not interested on how the

government spends on health care and other goods. Thus, the hypothesis was not applicable as it was beyond the scope of this study.

From the reviewed empirical literature, it was evident that most studies used cross-sectional and panel data sets in examining the link between poverty and health. Different methodologies have been applied to estimate effects of poverty on health status, health care utilization and choice of health care providers.

Among the estimation techniques used include ordinary least squares (OLS), two stage least squares (2SLS), Weibull and Cox models, Poisson regression models,

Granger Causality Models, Vector Error Correction Models, probit, and logit models. Others included multinomial probit, multinomial and nested multinomial logit, multivariate ordered logistic, and negative binomial logit. Other studies considered the problems associated with cross-sectional data such as endogeneity and used instrumental variables approach to solve the problem. Further, some studies used Control Function Approach to take care of unobservable effects of welfare measures on health status.

To investigate the link between poverty and health, most studies used health indicators such as self-assessed health, functional limitations, sensory impairment, disability status, anthropometric measures like underweight, mortality rate and morbidity on one hand and on the other hand welfare indicators such as household asset index, household expenditure, and income. In estimating health care utilization, and following behavioral model of health care utilization, some studies have used number of visits to health care providers while others used information

on the usage of some health care services such as post natal care, immunization and ante natal care.

Various studies investigating the link between poverty and health used a number of factors such age, sex, education level/attainment, marital status, household size and distance to health facility. Other variables used by various studies were place of residence, employment status, health status, household size, religion, race, access to health care services, utilization of health care services and household head level of education. In addition, wealth index, sex of household head, parental age and education and insurance status were used.

This study used the same but with more emphasis on poverty measures to establish its effects on health care utilization, choice of health care providers and health status in Kenya. In measuring poverty, different methods have been adopted by different studies. Some used income as a proxy for poverty while others used expenditure as a proxy for poverty. Others used subjective measures where individuals were requested to rank themselves whether they were poor or rich. Others constructed a wealth index based on household ownership of assets and then classified households as rich or poor. This study used a wealth index based on asset ownership as a proxy for poverty.

Studies done in Kenya in the area of interest for this study are few. This study differed from the previous studies in various ways. Firstly, this study focused on effect of poverty on general health as opposed to health of specific population groups such as maternal health, child health among others in Kenya. This study

differed from studies by Mutunga (2011), which only was concerned with infant and child health, and Kabubo-Mariara et al (2012), which tried to address child health only. Looking at the general health for the whole population gave a better picture of how the health status in the country is affected by poverty. Hence it is possible to come up with better ways of addressing health challenges in the country.

Secondly, the study identified how poverty affects health care utilization in the country. The study took a national outlook, unlike studies by Muriithi (2013) and

Mutua et al. (2011), which concentrated only on small slum areas of Nairobi and the findings cannot be generalized to the entire country. Using a nationwide dataset rich in formation can help in estimating with precision the effects of poverty on various groups and areas by disaggregating the data. This is important for policy makers in order to be sure of the areas and populations to target with health programs intended for the poor.

Lastly, the study used the most current cross-sectional dataset which was expected to give robust results and most recent trends in regard to health and poverty for policy purposes. This is unlike studies by Awiti (2014), Kabubo-Mariara et al.

(2012), Mutua et al. (2011), Mutunga (2011), Muriithi (2013) and Ochako et al.

(2011), which used datasets collected about ten years ago and before the introduction of new governance structure in 2010. The new governance structure led to devolution of health services in the country, leading to change in management of health and distribution of resources towards addressing health challenges in the country.

CHAPTER THREE

METHODOLOGY

3.1 Research Design

The broad objective of this study was to investigate the effect of poverty on health care utilization, choice of health care providers and health status in Kenya. To achieve the objective, this study used non-experimental cross-sectional research design. The non-experimental research design was best suited for this study, because, the study did not involve variables that could be manipulated by the researcher.

3.2 Theoretical Framework

Different theoretical frameworks have been used in addressing the relationship between poverty, health care utilization, choice of health care providers and health status. Among them includes:

3.2.1 A Theoretical Framework for Production of Health

To respond to the first and third objectives, this study was anchored on the

Grossman‟s human capital model (Grossman, 1972, 1982, 2000). In the model, individuals are assumed to maximize utility through consumption of health and non-health related goods subject to income. To operationalize the model, this study borrowed from Adeoti and Awoniyi (2014), Ajakaiye and Mwabu (2007),

Mwabu (2007), and Mwabu (2008). The study used a standard economic model of individual behavior in, which utility function is maximized subject to health production and income constraints. The individual utility (U) depends on

consumption of health related goods (X ) , consumption of health neutral goods

(Y), and health status (H) given as:

U  u(X,Y, H)………...…………..………………………………………… (3.1)

Where X is health related goods that have direct influence on the health status and also yields utility. Some of the health related goods include exercising, smoking, and engaging in risky behavior such as unprotected sex; Y denotes the health neutral goods that have no direct effect on health status of the members such as clothing; and H is health status of an individual.

According to Grossman (1972), health is dependent on investment in health, which is a function of health care and individual characteristics that influence the efficacy of health services. Thus, from equation (3.1), the production of health

( H ) by an individual can be described by the function given as:

H  h(X,Z, P,G,) ………………...………………………………………. (3.2) where Z is the purchased market inputs (health investment goods) such as medical services that affect individual health directly; X is health related goods;

P are control variables such as insurance coverage and employment status of individuals; G are household characteristics and geographical characteristics such as marital status, age, residence, education, religion, and household size, and  represents component of health due to genetic traits or environmental factors known to but not influenced by individuals or households (Ajakaiye & Mwabu,

2007; Mwabu, 2007, 2008).

An individual maximizes equation (3.1) subject to health production function

(3.2) and a budget constraint given by equation (3.3)

M  XPx  YPy  ZPz ……………………………………………………...…. (3.3)

where M is exogenous money income, and Px , Py , Pz are prices of health related goods (X ) , health neutral goods (Y), and health investments goods (Z) , respectively. From equations (3.1) and (3.2), health investment good (Z) is assumed to be purchased only to improve individual‟s health so that it only enters the utility function (equation 3.1) through health production function H given by equation (3.2).

The utility maximization problem can therefore be expressed in Lagrangian function as:

LX ,Y ,Z ,  UX ,Y,h(X ,Z, P,G, (M  XPx YPy  ZPz )…………...….. (3.4)

From equation (3.4), the first order necessary condition (FONC) for utility maximization can be given as:

LX U X X,Y,h(X,Z,P,G,)*hX (X,Z,P,G,)  Px  0…………...…. (3.5)

LZ  U Z X ,Y,h(X , Z, P,G, *hZ (X , Z, P,G, )  Pz  0.……..……...…. (3.6)

LY  UY X,Y,h(X,Z, P,G,) Py  0 ………………………………….... (3.7)

L  M  XPx YPy  ZPz  0 ……………………………………………..... (3.8)

Following Mwabu (2008) and Ajakaiye and Mwabu (2007), solving the FONCs simultaneously yields the health input demand functions of the optimal solutions to the individuals problem expressed as:

* X  DX (Px , Py , Pz , M , P,G, ) …….………………………….……...…….. (3.9)

 Y  Dy (Px , Py , Pz , M , P,G, ) ……………………………....…………...… (3.10)

* Z  Dz (Px , Py , Pz , M , P,G, ) …………………….……………...……...… (3.11)

The resulting reduced form demand function for health status may be written as:

H  (Px , Py , Pz ,M, P,G,) ………….……...…………………………….. (3.12)

where H and Px , Py , Pz are as explained earlier. M is wealth index which is a proxy for poverty status.

3.2.2 A Theoretical Framework for Choice of Healthcare Provider

The second objective of this study sought to understand how poverty affects choice of health care providers. The study used the framework advanced by Acton

(1975) on health care utilization, which recognized the discrete nature of health care decisions in health care demand. In case of illness or injury, individuals are faced with different alternatives in which they have to make a decision.

Individuals make a decision on whether to seek treatment or not. In the event they decide to seek treatment, they have to decide where to visit for treatment. This is a case of a discrete choice decision making, which is founded on choice theories.

A choice theory is a collection of procedures that defines four elements namely: the decision maker; the alternatives; the attributes; and the decision rule (Ben-

Akiva & Lerman, 1985). In a discrete choice model, an individual is faced with a challenge of choosing between two or more alternatives that have different combinations of attribute levels. An individual is expected to act rationally in

evaluating the available alternatives in each choice set and choose the alternative which gives the greatest relative utility, by making trade-offs across the different alternatives (de Bekker-Grob, 2009). Thus, an individual will choose alternative

A over B if U(X A , Z)  U(X B , Z) where U is indirect utility function of an

individual from certain alternatives, X A are attributes of alternative ; X B are attributes of alternative ; and Z is socioeconomic characteristics of the individual that influence his/her utility.

In a discrete choice decision making framework, choices are modeled within a

Random Utility Theory (RUT) (de Bekker-Grob, 2009). In the random utility models, the decision maker is assumed to have a capability to perfectly discriminate. However, a researcher is assumed to not have complete information and, therefore, there is need to take into account uncertainty (Ben-Akiva &

Bierlaire, 1999). The uncertainty emanates from four sources including: alternative attributes that are unobserved; individual‟s unobserved tastes; measurement errors; and instrumental variables or proxy (Ben-Akiva & Bierlaire,

1999). In the case of choosing who to consult in case of illness/injury, an individual is faced with a number of health care providers each of which yields indirect utility. An assumption is made that individual n chooses alternative j that maximizes his/her indirect utility amongst all alternatives in the choice set

Cn .

Thus, in order to reflect the uncertainty, the latent utility of an alternative j in a

choice set Cn (as perceived by individual n ) is written as:

U jn  V jn   jn …………………………..……………...…………………... (3.13)

where U jn is the indirect utility of individual n for choosing health care provider

j ; V jn is the deterministic (systematic) part of utility, and  jn is the random component. Therefore, the likelihood that alternative is selected by decision

maker n from a choice set Cn is expressed as:

Pr( j | Cn )  Pr(U jn  U kn ,k Cn )  Pr(U jn  maxU kn ) …………....….… (3.14) kCn

Substituting equation 3.13 into equation 3.14, it becomes:

Pr( j | Cn )  Pr{[V jn  jn]  [Vkn   kn ]} ………………………………….… (3.15)

Let the deterministic utility conditional on receiving treatment from provider j be given as:

V jn  V (h jn , y jn ) ………………………………………………..………...… (3.16)

Where h jn is the expected health improvement after individual n receives

treatment from health care provider j , and y jn is the consumption of non-health care goods, the amount which depends on the choice of j due to monetary and

non-monetary costs of treatment from provider j . Let h0n be the expected health from a reference health care provider such as self treatment. Thus, the change in

the expected health from choosing health care provider j is h jn  h0n . If the change is positive, then health care provider j is supposed to have a positive

impact on health of individual . Let E jn denote the change in expected health improvement from choosing health care provider j . Thus, the expected health production function can be expressed as:

h jn  h0n  E jn …………………………………………………..…………... (3.17)

where E jn depends on X jn which includes education level, health status, age, household income and other individual n characteristics, and the attributes of

health care provider j chosen by individual ( Z jn ). That is,

E jn  E(X jn ,Z jn ) ………………………………………………..……….… (3.18)

Thus, the conditional utility function can be expressed as:

U jn  V[E(X jn ,Z jn )  h0n ]   jn …………………………………..……...… (3.19)

Therefore, the unconditional utility maximization problem for individual can be expressed as:

* * U n  max(U kn ) where U n is the highest level of utility that individual can get. kCn

The solution to the utility maximization problem of individual yields a system of demand functions for alternative. To solve the problem therefore requires use of choice model.

3.3 Model Specification

3.3.1 Effect of poverty on health care utilization in Kenya To achieve the first objective of this study, equation (3.11) was used. Equation

(3.11) is a reduced form of health investment good equation. In this study, health investment good is medical services. Utilization of the medical services was captured by number of visits to a health facility. Thus, the dependent variable was a count variable and was not expected to take any negative figure. In practice, the distribution of health care utilization in terms of hospital visits, may display over- dispersion. This over-dispersion can be caused by presence of a large number of

zero counts in the data. This is because in health care data, there will be some individuals who never utilize health care services during the survey period. In other words, the mean of the variable is smaller than the variance of the variable

(Jones, 2007). To estimate counts of an event when the event has over-dispersion

(extra-Poisson variation), Negative Binomial regression model is normally used.

This is because, Negative binomial model is not sensitive to event dependency and variable event probabilities; hence it is considered a more flexible model as it relaxes the assumption of equi-dispersion property of Poisson model. Thus, it is more attractive in the modeling of health care utilization than Poisson model.

Following Kimani, Mugo, and Kioko (2016), and Fabbri and Monfardini (2003), this study estimated equation (3.11) using the Negative Binomial Regression

Model. The negative binomial model is an extension of the standard Poisson model. In estimating models with count variables, the starting point is the standard Poisson regression model where the variable is assumed to have a

Poisson distribution. Specifically for this study, the probability that health care

utilization (Y ) takes a specific value ( yi ) is given by (see Cameron & Trivedi,

2005).

i yi e i PrY  yi | X i   , yi  0,1,..., ……..……………………………...... (3.20) yi !

Where yi is observed number of health facility visits (health care utilization), i

is the mean parameter, X i are the covariates of health care utilization, and Pr represents probability. In most cases, the mean parameter is expressed in log- linear model (Greene, 2002) such that:

' lni  X i    i or

' i  exp(X i ) , i  0…………………………………………………… (3.21)

where  i is individual heterogeneity in a cross-sectional data.

The Poisson distribution implies the property of equi-dispersion:

Eyi | xi  Vyi | xi   i which is restrictive in empirical applications (Fabbri &

Monfardini, 2003), where, the conditional variance exceeds the conditional mean.

The distribution of yi conditioned on X i and ui (i.e  i ) remains Poisson with

conditional mean and variance i :

iui yi e iui  Pryi | X i ,ui   …………………………………...………….. (3.22) yi !

Integrating ui out of the expression (3.22) produces the unconditional distribution

of yi . The formulation of this distribution is given by

   y   i  yi  i Pryi | X i   i 1 i  , where i  ……………...... (3.23) yi 1  i 

Which, is a form of the negative binomial model; where, is number of hospital visits made by individual i , and are covariates of health care utilization.

The equivalent empirical equation model may be expressed as:

Visits  0  1 Age   2 Education 3householdsize   4 Re ligion

5 Maritalstatus  6 Sex 7 Re sidence 8 Povertystatus 

9 Dis tan ce to facility  ...... (3.24)

However, poverty is potentially endogenous in health care utilization model. To address the endogeneity issue, this study used Two Stage Residual Inclusion

(2SRI) approach to consistently estimate health care utilization. The approach involves two steps in which the first step is to estimate the reduced form equation.

Thus, in this study, the first step was to estimate the reduced form poverty status equation. The poverty status of household n was determined as follows:

PovSn  1PFn 2 HS n 3PC n 4 Kn  2n …………...………...... (3.25) where PovS is poverty status of a household; PF is predisposing factors such as age, sex, religion, household size, education level; HS is health status (level of illness); PC is price of care; K is instrument variables; and  is the error term capturing unobservable factors influencing poverty status.

The second stage involves estimation of the health care utilization model, where both the residuals from first stage model and the endogenous variable are included as additional regressors. The addition of the residuals from the first stage is to control for unobservable variables that are correlated with the endogenous variable. This act of including the residuals in the primary model allows the endogenous variable to be treated as if it is an exogenous covariate during estimation (Kimani, 2014). The model estimated in second stage is expressed as follows:

Visits   0  1 Age   2 Education  3householdsize   4 Re ligion

 5 Maritalstatus   6 Sex   7 Re sidence 8 Povertystatus    9 Dis tan ce to facility  10  2n  1n ...... (3.26)

 where  is the first stage residuals. If poverty status is exogenous in the health

care utilization model, then 10 will equal to zero.  1 is a stochastic disturbance

term. If poverty status is exogenous, the study will estimate equation (3.24) using maximum likelihood estimation.

If in this study, there is a non-linear interaction between unobserved factors and poverty status that cause the effect of poverty status on utilization of health care differ amongst the population subjects, then there could be a problem of unobserved heterogeneity (Awiti, 2014; Cameron & Trivedi, 2005). To solve the problem of unobserved heterogeneity, this study used the Control Function

Approach (CFA) as proposed by Awiti (2014), and Ajakaiye and Mwabu (2007).

The approach involves inclusion of interactions between the generalized residuals from reduced from poverty status model, and the poverty status variable in the health care utilization model. The estimated model was expressed as follows:

Visits   0  1 Age   2 Education 3householdsize   4 Re ligion

5 Maritalstatus   6 Sex   7 Re sidence 8 Povertystatus    9 Dis tan ce to facility  10  2n  11Povertystatus* 2n  1n ...... (3.27)

3.3.2 Effect of poverty on choice of health care providers in Kenya To achieve the second objective of this study, equation (3.19) was used. The study borrowed from Awiti (2014), Kosimbei (2005), Asfaw (2003), Mwabu et al.

(1993) and Ellis and Mwabu (1991), in specifying the model. In the event an individual falls sick, it is assumed the individual or his/her relative would seek help from a health care system characterized by many providers. The sick individual or a relative is further assumed to choose the health care alternative that yields the maximum expected utility. Thus, conditional on seeking treatment,

the direct utility derived by individual n from treatment alternative j can be expressed as follows:

U jn  1h jn   2 y jn   jn …………………………………………………… (3.28)

where U jn is the conditional utility function; h jn is the expected health improvement by individual after receiving treatment from health care provider

; and y jn is the consumption of non-health care goods.

As individual maximizes utility, he/she faces a budget constraint such that consumption plus the cost of health care must be less or equal to income, implying that:

y jn  M  e jn ……………………………...…...…………………………… (3.29)

where e jn is the value of resources devoted by individual n in receiving health

care from provider j . The level of e jn is determined by such factors as treatment fees, waiting time, and access variables such as distance and travel time. Thus, the choice of health care provider is constrained by the health production function which is a function of both quality and set of individual and household characteristics. Given the role of prices in consumption of health related and non- health goods, and the assumption that consumer preferences over entire range of consumption of goods are well defined, it can be shown that ill individuals

maximize an indirect utility function V jn , expressed as follows:

V jn  V jn (xn , z j ,rjn ,M,an ) …………..……..…...………………………..… (3.30)

Where xn is vector of individual and household characteristics; z j is vector of

medical and physical attributes of provider j ; rjn is price of health care received

by individual n from provider j ; M is household income; an is price of non- health related goods consumed by individual . Equation (3.30) shows the maximum utility that individual n can achieve conditional on seeking treatment for an illness/injury controlling for income, health care providers, prices of other goods, personal attributes, and provider specific characteristics.

To operationalise equation (3.30), this study borrowed from Mwabu et al. (1993).

Assuming that the utility function in equation (3.30) is stochastic, then, it can be expressed as follows:

* V jn  V jn   j …………………………………………..………………….... (3.31)

* where V jn is the deterministic component of utility function and  j is an additive disturbance term.

In a semi-log linear functional form, the deterministic component of the utility function may be expressed as follows:

* ' ' V jn   Q jn   j S n ……………………………..…………………...……… (3.32)

where Q jn is a vector of generic attributes (in log form) that individual n faces in

facility j ; S n is a vector of characteristics specific (in log form) to individual ;

 and  are vector of parameters to be estimated. Assuming that  j is distributed independently with an extreme value distribution, then the probability

Pjn that individual n will seek treatment from provider j can be expressed as follows:

' ' ' ' Pjn  exp( Q jn   jSn ) / exp( Qkn   k S n ) ………………………..... (3.33)

Equation (3.33) is a multinomial logit specification. However, a multinomial logit is faced with a problem of red bus-blue bus or the independence of irrelevant alternatives (IIA) assumption. This emanates from the assumption of the model that the random terms across alternatives are independent and thus, addition of any other alternatives will not change choices of the decision maker. Therefore, the IIA assumption may lead to idealistic predictions that may be impractical. To avoid such a problem, a multinomial probit (MNP) model can be used to estimate the health care provider choice model (Awiti, 2014). The MNP model is a m - choice multinomial model, with utility of individual i from jth choice (Cameron

& Trivedi, 2005) given by

U ij  Vi (xij ;)   j ,i  1,2,..., N; j  1,2,..., m………….………………...… (3.34)

where, U ij is utility derived by individual from choosing health care provider

j ; xij is the observed characteristics of individual and alternative provider

chosen; V j (xij ;) is deterministic component of the utility, and  j is the error term denoting the random component of the utility function. In a MNP model, the error terms are assumed to be correlated across choices and they jointly follow a normal distribution such that:

 ~ MND(0,) , with   I m   and

 11 . . .  1m     . . . . .  '   E( i ki )   . . . . .     . . . . .     m1 . . .  mm 

Where  is a Kronecker product; and m is the number of alternatives.

The probability that individual n will choose treatment from health care provider

j is, therefore, given by the deterministic component (V jn ) and the random effect

( jn ) of the utility, expressed as:

P  Pr[V    V   ;k  1,..., I,k  j | x ; ] ……….………..... (3.35) jn j n j n k n kn jn

where V jn is a function of the characteristics ( x jn ) of both the individual n and provider j , and a constant, with an assumption that option j yields a higher utility as opposed to option k .

Thus, to estimate the effect of poverty on the choice of health care providers, this study estimates the probability that an individual n , chooses health care provider j as follows:

Pr(yn  j)  0  1PF  2 PovS  3 HS  4 PC   j ……...…………… (3.36) where PF are predisposing factors such as age, sex, religion, household size, education level; PovS is poverty status which falls under enabling factors; HS is health status (level of illness); PC is price of care, and  is the error term capturing the unobservable characteristics/factors that influences choice of health care provider.

100

Due to potential reverse causality between choice of health care provider and poverty status, individual‟s poverty status is potentially endogenous in the health care provider choice model (Awiti, 2014). Problem of endogeneity can lead to inconsistency of estimated coefficients and a risk of inability to infer causality between poverty status and choice of health care provider. To address the problem of endogeneity, this study used a Two Stage Residual Inclusion (2SRI) method to consistently estimate the health care provider choice model. The 2SRI involves two stages. The first stage is a consistent estimation of the reduced form model for the endogenous variable. The poverty status of individual n will be determined as follows:

PovSn  1PFn 2 HS n 3PC n 4 Kn  2n ………...………………… (3.37) where PovS is poverty status of an individual; PF is predisposing factors such as age, sex, religion, household size, education level; HS is health status (level of illness); PC is price of care; K is instrument variables; and  is the error term capturing unobservable factors influencing poverty status.

The second stage involves estimation of the health care provider choice model, where both the residuals from first stage model and the endogenous variable are included as additional regressors. The rationale for including the residuals from the first stage model is to serve as a control for unobservable variables that are correlated with the endogenous variable. This act of including the residuals in the primary model allows the endogenous variable to be treated as if it is an exogenous covariate during estimation (Kimani, 2014). The model estimated in second stage was expressed as follows:

101

 Pr(yn  j)  1PFn  2 PovSn  3HS n  4 PC n  5  2n  1n ………...... (3.38)

 The notations are similar to those in equation (3.37).  is the first stage residuals.

If poverty status is exogenous in the health care provider choice model, then 5

will equal to zero.  1 is a stochastic disturbance term. If poverty status is exogenous, the study will estimate equation (3.36) using maximum likelihood estimation.

There is a possibility that there is a non-linear interaction between unobserved factors and poverty status in this study. Such an interaction cause the effect of poverty status on choice of health care providers to differ amongst the population subjects, an indication of possible presence of a problem of unobserved heterogeneity (Awiti, 2014; Cameron & Trivedi, 2005). To address the unobserved heterogeneity problem, this study used the Control Function

Approach (CFA) as proposed by Awiti (2014), and Ajakaiye and Mwabu (2007).

The approach involves inclusion of interactions between the generalized residuals from poverty status model, and the poverty status variable in the health care provider choice model. The model that was estimated is as follows:

  Pr(yn  j)  1PF  2 PovS  3HS  4 PC  5  2n  6 PovS * 2n  1n . (3.39) where the notations are as those used in (3.36).

3.3.3 Effect of poverty on health status in Kenya

In order to understand how poverty affects health status, this study used equation

(3.12). The indicator used to capture the health status was self reported

102

assessment of individual‟s health. The health status was in four categories namely very good, good, satisfactory, and poor. Thus, supposing that there is a natural ordering of alternatives, a more parsimonious model that takes account of ordering will be necessary. The starting point is, therefore, a model of the form:

* ' H i  X i   i ………………………………………………..…………….. (3.40)

* where H i is a latent (unobservable) variable related to observed health status of individual i , and X are covariates. Thus, model (3.40) cannot be estimated as

is not observed. Instead, what is observed can be constructed as follows:

1 if individuali rateshis/ herhealthas poor  2 if individuali rateshis/ herhealthassatisfactory H i   ………….. (3.41) 3 if individuali rateshis/ herhealthas good 4 if individuali rateshis/ herhealthasverygood

This indictor of capturing health status is ordinal. Thus, the appropriate model for estimation is an ordered choice regression model (Cameron & Trivedi, 2005).

Following Awiti (2013) and Greene and Hensher (2010), this study assumed that there was a continuous latent (unobservable) variable, , that was related to the observed health status of individual i , through the following equation:

* 1 if  0    H i  1  * 2 if 1  H i   2 H  …………………………………...…... (3.42) i  * 3 if  2  H i   3  * 4 if  3  H i   4  

where 1 ,  2 , and 3 are the thresholds to be estimated. The latent variable is in turn related to the various covariates through the following equation:

103

* H i  1  2 PovS  3W1  1i …………………………………………..… (3.43)

where PovS is the poverty status of individual i , W1 is a vector of controls, and

 1 is a stochastic disturbance term.

The probability of observing a given outcome for a given value of the independent variable can be expressed as:

* Pr(H i  j | PovS,W1 )  Pr( j1  H i   j ) ……………………………..… (3.44) where j 1,2,3,4.

Substituting equation (3.43) in to equation (3.44) and simplifying gives the equation of the predicted probabilities of the observed outcomes as follows:

Pr(H i  j | PovS,W1 )  ( j  1  2 PovS  3W1 )  ( j1  1  2 PovS  3W1 ) ...… (3.45)

where j 1,2,3,4 , and  is the cumulative distribution function for  1 .

Assuming that  1 has a standard normal distribution, then the model becomes an ordered probit model, which is estimated in this study.

According to Adeoti and Awoniyi (2014), Awiti (2013), Awiti (2014), Fuchs

(2004), Mwabu (2007) and Mwabu (2008), poverty status of an individual, PovS , is potentially endogenous. In this study, endogeneity could be due to reverse causality of health status and poverty (Namubiru, 2014). To address the problem of endogeneity, this study adopted a 2SRI method. The approach involves including generalized residuals from poverty status model into health status model as additional regressor. Poverty variable is said to be endogenous if the coefficient of the poverty variable residuals is statistically significant in the health status

104

model (Awiti, 2013). In order to get the generalized residuals, a poverty status model to be estimated is of the form:

PovS  W  2i …………………………………………………….…….... (3.46)

where W is exogenous set of covariates comprising of W1 variable that also

belong to the health status equation plus a vector of instrumental variables W2 that affect poverty status, but have no direct influence on health status.  is a

parameter to be estimated, and  2 is a stochastic disturbance term. The obtained generalized residuals from equation (3.46) are then included as an additional regressor in the structural equation of interest (equation 3.43) to control for endogeneity of poverty status. The resulting equation can therefore be written as:

 * Hi  1  2 PovSi  3W1  4  2i   2i ………………..………………..... (3.47)

Apart from endogeneity, there could be a problem of unobserved heterogeneity due to unobserved factors of health that are interacted with variable of interest

(Mwabu, 2008). Unobserved heterogeneity could arise from unobserved preferences and health endowment of individuals that influence their choice of health inputs, but are also correlated with health outcomes. In the health production function, heterogeneity may arise from the presence of exogenous health factors that can be known to the individual household but are unobserved by the researcher (Kabubo-Mariara et al., 2009). In this study unobserved heterogeneity could arise from unobserved individual and household characteristics that are correlated with poverty as well as health status. To address the problem of unobserved heterogeneity in this study, CFA was used. The

105

approach involves inclusion of the interaction of poverty status variable with its generalized residuals, as an additional regressor in the health status model. Thus, to control for potential endogeneity of poverty status and potential unobserved heterogeneity equation (3.43) can be extended and expressed as follows:

  * Hi  1  2 PovSi  3W1  3  2i  4 PovSi * 2i  1i ………..………..... (3.48)

where PovSi is the poverty status of individual i , W1 is vector of controls,  2 are

generalized residuals from the poverty status model, and  1 is a stochastic disturbance term. Substitution of equation (3.48) into equation (3.45), and assuming a standard normal distribution for , the estimated model can be expressed as:

  Pr(H i  j | PovS,W1 )  ( j  1   2 PovS  3W1   4  2i  5 PovS * 2i )   ( j1  1   2 PovS  3W1   4  2i  5 PovS * 2i )...... (3.49)

where j 1,2,3,4 and  is the cumulative distribution function for  1 (Cameron

& Trivedi, 2005).

3.4 Definition and Measurement of Variables

The variables used in this study are defined and measured as follows:

Healthcare provider choice (HP)-these are persons consulted by individuals who reported to have been sick or injured. It is classified as government health facilities, private health facilities, mission health facilities and other health facilities. The other category consists of traditional healers, kiosks, faith healers,

106

and herbalists. It was coded 1=Government health facilities, 2=Private health facilities, 3=Mission health facilities, 4=Others.

Health status (HS)-it is the self rated health status by individuals. It was categorized as 1= Poor, 2=Satisfactory, 3=Good, 4=Very Good.

Health care utilization (HCU)-it is the use of health services by those who reported to have been sick. It is measured by the number of visits made to a health facility by an individual who reported to have been sick.

Age of individual (A)-it is the number of years of an individual at the time of survey. It is measured in years.

Sex (S)-it is the biological characteristics of the individual who reports to have been sick. It was coded 1=Male and 2=Female.

Marital status (MS)-This captures whether an individual is married or not, categorized as 1=Never married, 2=Married, 3=Divorced/Separated/Widowed.

Education level (EL)-is the highest level of education completed by an individual and head of household. It was measured using 1=No education,

2=Primary, 3=Secondary and 4=College/University.

Household wealth index (HWI)-it is the index capturing the standards of living of a family where an individual belongs based on asset ownership. The index was used as proxy for poverty status in this study. The index was constructed based on survey responses from households in the survey. The wealth index was assigned to each household based on a weighted average of variables in the dataset related to asset ownership. The index was generated using the Multiple Correspondence

Analysis (MCA). The wealth index scores are continuous. Those with more assets have a higher score than those with less assets and are considered less poor. 107

Household size (HS)-it is the number of members in a household measured using the actual number.

Religion (R)-it is the religion of individuals categorized as

1=Traditionalists/Atheists/Others, 2=Catholic, 3=Protestant, 4=Muslim.

Employment- It is the employment status of an individual. Dummy=1 if employed and 0, otherwise.

Distance to facility (DF)-it is the distance from the home of an individual to the nearest health facility measured in kilometers.

Residence (RS)-it is place of residence where the individual resides. It was

1=Rural, 2=Urban.

Waiting time- it is the number of hours that individuals spent in a health facility before they can be attended to.

Insurance cover-it is number of individuals with health insurance cover.

Dummy=1 if yes, 0 otherwise.

County average access to electricity-It is the average number of households that have access to electricity. The variable is continuous. Counties with more households with access to electricity have a higher mean score.

County average access to piped water-It is the average number of households that have access to piped water. The variable is continuous. Those counties with more households accessing piped water have a higher mean score.

Illness-It is the type of illness an individual is suffering from. There are three types used in the study. If suffering from malaria/fever=1, 0 otherwise. If

108

suffering from respiratory disease/pneumonia=1, 0 otherwise. If suffering from diarrhea=1, 0 otherwise.

3.5 Diagnostic Tests

To avoid biased and inconsistent estimates, diagnostic tests were carried out. The tests included Wald Chi-Square test and Pseudo R2 for goodness of fit; and Link test for model specification. Also tested was multicollinearity using Variance

Inflation Factor. Likelihood ratio (LR) test was also done to test whether models fitted the data well. Also principles of 2SRI and CFA were used to test existence of endogeneity and unobserved heterogeneity. Results of these tests are reported together with empirical results in every table in chapter four as appropriate.

3.6 Data type and source

In order to achieve objectives of this study, micro dataset obtained from Ministry of Health for Kenya was used. This is a secondary source. The dataset was from the 2013 Kenya Household Expenditure and Utilization Survey (KHHEUS). The dataset was used due to availability of variables of interest for this study. The dataset is reach in information related to health care utilization and health status in

Kenya. It was the most recent dataset, which captured this vital information and, therefore, it enabled this study to achieve its objectives as specified in chapter one. KHHEUS is a nationally representative cross-sectional data. The dataset was collected from a total of 33,675 households drawn from 1,347 clusters divided into 814 (60%) rural and 533 (40%) urban clusters. The survey covered 44 counties. Garissa, Mandera, and Wajir counties were not covered by the survey.

109

3.7 Data Analysis

This study used regression analysis to achieve its objectives specified in chapter one. Each of the objectives of this study was analysed as follows:

The first objective sought to determine the effects of poverty on health care utilization in Kenya. To achieve this objective, equation 3.24 was estimated using

Negative Binomial regression. This is because health care utilization was measured by number of visits to a health facility by those individuals who reported to have been sick. Thus, the dependant variable was a count variable, which was not expected to take any negative figure. Again due to possible presence of large number of zeros in the data due to those who do not utilize health care, there was evidence of over-dispersion problem. Negative binomial model is more suitable in cases where there is over-dispersion.

The second objective sought to investigate the effects of poverty on the choice of health care providers in Kenya. Seeking to achieve the objective, there is a possibility of having problems of endogeneity and heterogeneity. To avoid the problems, a 2SRI and CFA methods were used. In applying 2SRI, equation 3.34 was estimated first. Then, the residuals from equation 3.34 were used as an additional regressor in equation 3.35. Finally, equation 3.36 was estimated using multinomial probit with residuals from equation 3.34 and interaction between endogenous variable and residuals from equation 3.34 included as additional regressors.

110

The third objective of the study sought to establish the effect of poverty on health status in Kenya. The dependant variable was a self reported health status, which was ordered. In trying to achieve the objective, there is possibility of encountering problems of endogeneity and unobserved heterogeneity. Thus, to control the problems, two different procedures were used. A 2SRI method was applied to control for endogeneity. In the first step, equation 3.43 was estimated and obtained the generalized residuals. In the second step, the residuals from equation

3.43 were used as additional regressor in the primary equation 3.40, which was then estimated using ordered probit model. Lastly, equation 3.46 was estimated using ordered probit model. Equation 3.46 contained residuals from equation

3.43, interaction between endogenous variable and residuals from equation 3.45.

All the models were estimated using Stata software.

111

CHAPTER FOUR

EMPIRICAL FINDINGS AND DISCUSSION

4.1 Descriptive Statistics

The study used the KHHUES survey data of 2013, which covered 44 counties in

Kenya. Three counties namely Garissa, Mandera, and Wajir were not included in the survey since they had not been mapped due to insecurity in the areas. Table

4.1 shows the descriptive statistics derived from the data for the continuous and count variables that were used. The statistics include the mean, standard deviation, number of observations and range of variables.

Table 4.1: Descriptive Statistics for Continuous and Count Variables Variable Mean Standard Range Number of Deviation Observations Minimum Maximum Number of - - - - 29,114 Households Age in years 34.84 16.76 15 99 80,774 Waiting time 0.79 1.18 0 50 19,627 in hours Household 5.42 2.62 1 22 80,774 size Number of 1.73 1.38 0 12 17,903 Hospital visits Wealth Index -0.13 0.55 -0.96 1.82 80,774 Average 0.25 0.16 0.08 0.88 80,774 county access to electricity Average 0.33 0.20 0.04 0.81 80,774 county access to piped water Source: Author computation, Study Data, 2013.

Table 4.1 shows there were 80,774 household members spread across 29,114 households sampled for this study. Table 4.1 shows that the household size ranged

112

from 1 to 22 members with a mean of about six members and a deviation of three members. Though not in Table 4.1, the data also showed that 3,872 of the 29,114 households had only one member, and only one household had 22 members.

These statistics about household size corresponds with findings of Kimani (2014) on Kenya who reported an average household size of 5.18 with a deviation of 2.35 and a maximum number of household members of 15.

The period individuals waited before they could be attended to in a health facility ranged between zero and 50 hours with an average of 0.8 hours and a deviation of

1.2 hours. This indicates that in most cases, people were attended to within one hour of arrival in a health facility. However, there are cases where individuals waited for long hours before they could be attended to in a health facility. The long waiting hours could reflect the shortage of health care staff in some health facilities. This observation is consistent with findings of Daniels et al. (2017) in

Kenya who found that on average, patients waited for 1.35 hours before they could be treated. Kimani (2014) also found that in Kenya, patients waited on average for 0.92 hours before they could be attended to in a health facility and a maximum of 15 hours.

Age of individuals covered in the survey ranged between 15 and 99 years old with a mean of 34.8 years and a deviation of 16.8 years. This implies on average, those in the sample were the youth and those in their most productive age. The statistics relating to age when countered against the waiting time in health facilities points to possible wastages in productivity at individual, organizational and national levels.

113

In terms of health care utilization, the number of hospital visits ranged between zero and 12 hospital visits with a mean of 1.7 visits and a deviation of 1.4 visits.

Thus, the sampled individuals made on average two hospital visits. This could mean that after the two hospital visits individuals got well. Alternatively, it could imply that people stop going for health care after two visits due to other engagements even if they are unwell. Individuals who for example, are in the informal sector or other atypical forms of employment are most likely to find it difficult to miss work or income generating opportunity to seek medical care.

Wealth index, which is a measure of the value of household assets ranged between -0.96 and 1.82 scores with a mean of -0.13 scores and a deviation of 0.55 score. The wealth index was calculated using multiple correspondence analysis

(MCA). The scores were then rotated for easier interpretation of results.

Households with more assets were clustered together and those with less were also clustered together. Individuals with more household assets were ranked higher than those with less household assets. The low mean for wealth index is an indication that poverty level amongst households was high, consistent with findings of other studies in Kenya (Gakuru & Mathenge, 2012; KIPPRA, 2015).

The county index capturing average access to electricity ranged between 0.08 and

0.88, with a mean of 0.25 and deviation of 0.16. This indicates low penetration of electricity in most counties in the country. This observation is supported by findings of Kagiri and Wainaina (2017) on Kenya who found that the national average access to electricity was 15 per cent. The low electricity penetration may

114

have a negative impact on health sector especially due to equipments that require electricity for them to function.

The county index for access to piped water ranged between 0.04 and 0.81 with a mean of 0.33 and a deviation 0.20. The low mean is an indication that access to clean water remains a challenge in Kenya. Inadequate access to piped clean water may increase the risk of illness. The situation may also lead to increased spread of waterborne diseases and burden of health service delivery due to increased demand for health care. The observation on inadequate access to piped clean water corresponds with that of Kenya Demographic Health Survey in 2014. It reported that on average only 22.8 per cent of Kenyan population have access to piped water (Republic of Kenya, 2014b).

Table 4.2 presents the summary statistics of the discrete and categorical variables used in the study.

Table 4.2: Summary Statistics for Discrete and Categorical Variables Variable Number of Percentage Observations Reported illness Yes 17,903 22.16 No 62,871 77.84 Total 80,774 100 Health Status Poor 4,588 5.68 Satisfactory 10,848 13.44 Good 44,708 55.37 Very Good 20,598 25.51 Total 80,742 100 Religion Traditionalist/Atheist/Others 2,828 3.50 Catholic 19,814 24.53 Protestant 49,963 61.86

115

Muslim 8,169 10.11 Total 80,774 100 Area of residence Rural 51,728 64.04 Urban 29,046 35.96 Total 80,774 100 Sex Female 43,281 53.58 Male 37,493 46.42 Total 80,774 100 Distance to nearest health facility <1KM 3,529 17.17 1-3 KM 8,304 40.41 4-5 KM 3,061 14.90 6-9 KM 2,468 12.01 10+ KM 3,188 15.51 Total 20,550 100 Marital Status Never Married 29,618 36.67 Married 43,394 53.72 Divorced/separated/widowed 7,762 9.61 Total 80,774 100 Education level No education 13,190 16.33 Primary education 36,684 45.42 Secondary education 23,722 29.37 College/university education 7,178 8.89 Total 80,774 100 Having malaria/fever Yes 7,216 8.93 No 73,558 91.07 Total 80,774 100 Having respiratory disease/pneumonia Yes 2,400 2.97 No 78,374 97.03 Total 80,774 100 Having diarrhea Yes 300 0.37 No 80,474 99.63 Total 80,774 100 Healthcare providers Government 12,101 58.76 Private 6,504 31.58 Mission 1,659 8.06 Others 329 1.60

116

Total 20,593 100 Having Health Insurance Yes 13,425 16.62 No 67,349 83.38 Total 80,774 100 Employment status Employed 45,755 56.65 Not employed 35,019 43.35 Total 80,774 100 Source: Author computation, Study Data, 2013.

Table 4.2 shows that, of those surveyed, 22.16 per cent reported to have been sick while 77.84 reported not to have been sick four weeks before the survey. This study was more interested in those who reported illness. Those who reported illness that constituted 22.16 per cent of the sampled individuals possibly never went for work. This could have a huge impact on the productivity of the country if not addressed on time. It also has implications on longevity of working life.

Concerning health status, 5.68 per cent of the household members surveyed rated own health as poor, and 13.44 per cent rated own health as being satisfactory. The results also indicated that 55.37 per cent of individuals rated own health as good while 25.51 per cent rated their own health as being very good. This statistics imply that among those sampled, almost 20 per cent rated their own health as being poor or just satisfactory. Considering the importance good health both to individuals and to the Kenyan society, the percentage is high and should worry policy makers. Those with poor health may spend more time seeking treatment and loosing time, which they could have otherwise used in productive activities. If the individuals are suffering from communicable diseases the situation could even be worse since they can spread to others very easily. If individuals are suffering

117

from terminal diseases, the situation could lead to higher dependency ratio, which could in turn push or drag families deeper in to poverty.

In the sample surveyed, majority (61.9%) of the household members were

Protestants followed by Catholics at 24.5 per cent. Muslims were 10.11 per cent while the traditionalists, atheists and other groups constituted 3.5 per cent of the sampled population. In terms of area of residence, the results revealed that 64.04 per cent of the household members were residing in rural areas while the rest

(35.95%) were residing in urban areas. The results are consistent with the Kenya

Population and Housing Census report of 2009, which found that 32 per cent of

Kenyans were residing in urban areas and the rest were in rural areas. The results further showed that of all the household members, 53.58 per cent were females and 46.42 were males. The Population and Housing Census report of 2009 report also indicated that about 49 per cent of the country‟s population were males while

51 per cent were females (Republic of Kenya, 2010a).

Table 4.2 shows that 53.7 per cent of the household members were married, 36.7 per cent had never married and 9.6 per cent were divorced, separated or widowed.

Regarding education, 16.33 per cent of the household members had no education,

45.4 per cent had primary level education, 29.4 per cent had secondary level education and 8.89 per cent had either college or university level education. This indicates a higher literacy level in the country, which is important when making decisions relating to health issues. According to summary statistics presented in

Table 4.2, of those who reported to have been sick, 8.93 per cent indicated that they were suffering from malaria, 2.97 per cent reported to suffer from respiratory

118

disease or pneumonia and a small number (0.37%) reported suffering from diarrhea. These statistics are consistent with those documented by the government of Kenya in her Health Sector Human Resources Strategy (2014-2018). These statistics are worrying considering that one of the leading causes of deaths in

Kenya is malaria, which contributes about 5.8 per cent of total deaths. Diarrheal diseases on the other hand contributes about six per cent of the total deaths in

Kenya (Republic of Kenya, 2014a).

Table 4.2 also indicates that 56.65 per cent of the household members were employed while 43.44 were not employed. The results further indicate that only a handful (16.6%) of the household members had health insurance cover. This statistic implies that about 80 per cent of Kenyan population does not have health insurance cover. Thus, in case of illness or injury, the uncovered population relies on out of pocket expenditure to cater for their health care. This can be catastrophic in case of prolonged illness or in the event no available money at the time of illness.

Regarding distribution of health care providers, Table 4.2 shows that government health care providers were the majority (58.8%), followed by private health care providers (31.6%), Mission health care providers (8.03%) and lastly, other health care providers (1.6%). This is not surprising as the government has a responsibility of ensuring that her citizens enjoy good health as a right as enshrined in the constitution (Republic of Kenya, 2010b). Thus, the government has to invest in health. Also, investing in health requires high capital, which many private individuals and organizations may not be able to afford. The summary

119

results given in Table 4.2 also show that individuals residing within 1 kilometer from nearest health facility were 17 per cent while those living within 10 kilometers and above were 15.5 per cent. This is worrying considering that health is a human right and every individual should have unlimited access to it. Thus, with such long distances, people may be denied that right of accessing and enjoying good health.

4.2 Empirical Findings

The broad objective of the study was to investigate the effect of poverty on health care utilization, choice of health care providers and health status in Kenya. The empirical findings are reported and discussed based on themes as per the specific objectives of the study stated earlier in chapter one.

4.2.1 Effect of poverty on health care utilization in Kenya

The first objective of this study was to determine the effect of poverty on health care utilization in Kenya. In order to achieve this objective, equations 3.25 and

3.27 were estimated using data from the 2013 KHHEUS. Several model selection tests were carried out before final estimation. Firstly, the test between the model of equi-dispersion and over-dispersion, which refers to Poisson regression and negative binomial regression, respectively was carried out. The test was done using Likelihood Ratio (LR) test. Secondly, in order to determine whether ZIP was preferred to standard Poisson regression, a Vuong test was carried out (Baum,

2010).

120

Table 4.3 presents the results of model selection based on the Vuong and

Likelihood Ratio (LR) tests.

Table 4.3: Poisson, NBRM and ZIP Models Selection based on Vuong and LR tests Poisson Model NBRM Model ZIP Model Dependant Variable=Number of visits Variable Coeff. P>z Coeff. P>z Coeff. P>z

Wealth Index 0.037** 0.022 0.037** 0.022 0.037** 0.015 Age 0.0005 0.802 0.0005 0.811 0.0004 0.804 Age Squared 7.42e-06 0.691 7.63e-06 0.682 7.42e-06 0.711 Insurance Cover (Not insured=Reference) Insured -0.009 0.620 -0.09 0.618 -0.009 0.603 Log of waiting time -0.018*** 0.001 -0.018*** 0.001 -0.018*** 0.000 Gender (Male=Reference) Female 0.200*** 0.000 0.200*** 0.000 0.200*** 0.000 Religion (Traditionalist/Atheist/Others=Reference) Catholic 0.003 0.922 0.003 0.922 0.003 0.932 Protestant 0.070** 0.042 0.070* 0.042 0.070* 0.076 Muslim 0.075* 0.051 0.075* 0.051 0.075* 0.088 Log of household size 0.115*** 0.000 0.115*** 0.000 0.115*** 0.000 Distance to nearest health facility:<1KM (Reference) 1-3KM 0.114*** 0.000 0.113*** 0.000 0.114*** 0.000 4-5KM 0.096*** 0.000 0.095*** 0.000 0.096*** 0.000 6-9KM 0.177*** 0.000 0.177*** 0.000 0.177*** 0.000 10+ 0.121*** 0.000 0.120*** 0.000 0.121*** 0.000 Marital status (Never Married=Reference) Married -0.177*** 0.000 -0.177*** 0.000 -0.177*** 0.000 Divorced/separated/ Widowed -0.327*** 0.000 -0.326*** 0.000 -0.327 0.000 Education Level (No education=Reference) Primary Education 0.052*** 0.006 0.052*** 0.006 0.052*** 0.007 Secondary Education 0.039* 0.082 0.039* 0.082 0.039* 0.091 College/university education 0.004 0.903 0.004 0.900 0.004 0.898 Employment status (1=employed; 0 otherwise) 0.001 0.936 0.001 0.939 0.001 0.935 Area of residence (Rural=Reference) Urban -0.007 0.649 -0.007 0.650 -0.007 0.625 Constant 0.142*** 0.010 0.143*** 0.010 0.142** 0.014

121

Number of Observations 16,619 16,619 16,619 Zero Observations - - 89 Non-Zero Observations - - 16,530 Pseudo R2 0.0142 0.0138 - Wald χ2 (21) 859.2*** 0.000 859.78*** 0.000 - - LR χ2 (21) - - - - 728.04*** 0.000 Likelihood Ratio Test (Poisson Vs. NBRM) - - 6.49*** 0.005 - - Vuong Test of ZIP Vs. Standard Poisson - - - - Z= -0.03 0.514 Linktest: hat 0.940*** 0.000 0.940*** 0.000 - - hat squared 0.057 0.724 0.057 0.729 - - Note: ***, **, and * denote statistical significance at 1 per cent, 5 per cent and 10 per cent, levels of significance, respectively. Source: Author computation, Study Data, 2013.

Results presented in Table 4.3 indicate that the LR statistic of 6.49 had a probability value of 0.005. The LR test of α = 0 strongly rejected the null hypothesis that the errors did not exhibit overdispersion. Thus, the study rejected

Poisson regression model in favor of its generalized version, the NBRM. The results of the Vuong test indicate that the Z-value was -0.03 with a probability value of 0.514. Conclusion from the test was that Standard Poisson regression was preferred over ZIP. However, since the LR test had preferred NBRM over standard Poisson due to presence of over-dispersion, the study opted for the

NBRM. The NBRM is also favored as it relaxes the assumption of equi- dispersion (Kimani, 2014). The NBRM was used since the dependant variable

(health care utilization proxied by number of hospital visits) is a count variable, which is not expected to take any negative figure. The displayed over-dispersion could be due to large number of zero counts, which is common with most health care utilization data (Jones, 2007; Kimani et al., 2016).

122

In addition to using NBRM, 2SRI and CFA methods were estimated. Since poverty status is potentially endogenous in health care utilization, a 2SRI model was used to ascertain and control for endogeneity (Terza, Basu, & Rathouz,

2008). Further, there was possibility of unobserved heterogeneity in the health care utilization model. This is due to a non-linear interaction between unobserved factors and poverty status that make effect of poverty status on health care utilization to vary amongst population subjects. Therefore, to test presence and to control for unobserved heterogeneity, CFA was used (Ajakaiye & Mwabu, 2007;

Awiti, 2013; Kabubo-Mariara et al., 2009; Namubiru, 2014).

The 2SRI method involves two steps. The first stage involves estimation of the model of endogenous regressor. In this study, poverty status is potentially endogenous. Thus, a poverty status model was estimated first using OLS and its generalized residuals were obtained and used in the second stage as additional regressors (Awiti, 2013; Kimani et al., 2016). The poverty status model contained among other explanatory variables, instrument variables that were considered to affect health care utilization through poverty. The second stage involved estimation of the health care utilization model where both the poverty status variable and its generalized residuals were included as explanatory variables. If the coefficient of the generalized residuals is statistically different from zero, then poverty status is considered to be endogenous (Awiti, 2013).

Use of 2SRI, however, has one challenge of getting appropriate instrument variables. Since only poverty status (proxied by wealth index) was potentially endogenous in this study, only one instrumental variable was needed. Thus, it was

123

important to instrument wealth index by looking for a valid instrumental variable for it. According to Awiti (2013), Kabubo-Mariara et al. (2009) and Kimani et al.

(2016), validity entails relevance, strength and exogeneity of an instrument variable. Relevance means that the instrumental variable should be strongly correlated to the endogenous variable. The strength of an instrumental variable implies that the magnitude of its coefficient should be large, while exogeneity means that the instrumental variable should be uncorrelated with the structural disturbance term.

Common variables that have been used as instruments for poverty status includes distance to the nearest Non-Governmental Organization (NGO) health unit, distance to the nearest market, time to get to water source (Namubiru, 2014), and proportion of children who are severely underweight in a region (Awiti, 2014).

Dataset used in this study lacks such information, hence, called for innovation regarding variables that can serve as instruments. This study used the average number of households at the county level that have access to electricity. The choice of this variable was motivated by the fact that the average number of households at the county level that have access to electricity is not expected to influence how households utilize health care. However, access to electricity and poverty are deemed to be highly correlated. It is expected that households found in counties with lower electricity access should have a higher probability of being poor and the reverse should be true.

Table A1 in the appendix shows results for the test of validity, strength and relevance of the instrument. The results indicate that the instrument was highly

124

correlated with the endogenous variable with a P-value of 0.000 and was uncorrelated with the structural error term. Thus, average number of households at county level with access to electricity was a valid, strong and relevant instrument variable. In case of instrumental variables being uncorrelated with the structural error term, identification tests are needed. In case of over-identified model, where instruments are more than the endogenous variables, tests of whether the instrumental variables are uncorrelated with the error term are performed. However, if a model is just identified, then test of identification restrictions is not needed (Kimani, 2014). In this study, there was only one endogenous variable and one instrument. Thus, there was no need of performing identification test.

The results for the first stage of 2SRI method are shown in Table 4.4. The purpose of the first stage of 2SRI was to get generalized residuals, which were then used as additional regressor in the second stage of 2SRI estimation. Inclusion of the generalized residuals in the second stage was aimed at controlling potential endogenity in the health care utilization model (Kabubo-Mariara et al., 2009).

Table 4.4: Regression Results of Poverty Status Model measured at household level Variable Dependant Variable=Wealth index (proxy for poverty) Coefficient Robust Std. P>|t| Err Age 0.004*** 0.001 0.000 Age Squared -0.00002*** 7.91e-06 0.048 Sex: Male(Reference) Female 0.073*** 0.006 0.000 Religion: Traditionalist/Atheist/Others(Reference) Catholic 0.138*** 0.011 0.000

125

Protestant 0.172*** 0.011 0.000 Muslim 0.202*** 0.013 0.000 Marital: Never married(Reference) Married 0.023*** 0.009 0.010 Divorced/separated/Widowed -0.085*** 0.010 0.000 Log of household size -0.067*** 0.004 0.000 Education Level: No education(Reference) Primary Education 0.252*** 0.006 0.000 Secondary Education 0.491*** 0.008 0.000 College/university education 0.845*** 0.010 0.000 Employment Status: No (Reference) Yes 0.068*** 0.007 0.000 Area of residence: Rural (Reference) Urban 0.348*** 0.005 0.000 County average access to electricity: No (Reference) Yes 0.633*** 0.022 0.000 County average access to piped water Yes 0.174*** 0.018 0.000 Constant -0.955*** 0.022 0.000 Number of observations 28968 R-Squared 0.5373 F(16, 28951) 2413.96*** - 0.000 Linktest: hat 0.9987*** - 0.000 hat squared 0.0054 - 0.620 Mean VIF 6.68 Note: ***, **, and * denote statistical significance at 1 per cent, 5 per cent and 10 per cent, levels of significance, respectively. Source: Author computation, Study Data, 2013.

Table 4.4 presents results of the poverty status model estimated at household level. However, since the sole purpose for the first stage of the 2SRI was to obtain the generalized residuals for use in the second stage, the results are not discussed here for brevity. Instead, the study presents and discusses results for the second stage of the 2SRI.

The second stage of 2SRI approach entails estimation of the health care utilization model. In this second stage, both wealth index variable and its generalized residuals were included in the health care utilization model as additional

126

explanatory variables. Table 4.5 shows the baseline model (NBRM), the model controlling for endogeneity of poverty status (2SRI) and the model controlling endogeneity of poverty status and unobserved heterogeneity (CFA).

Table 4.5: Regression results of NBRM, 2SRI and CFA Variable Dependant variable= Number of hospital visits NBR Model 2SRI Model CFA Model Wealth index 0.035** 0.072** 0.076** (0.015) (0.030) (0.030) Age 0.0004 0.0004 0.001 (0.002) (0.002) (0.002) Age Squared 7.97e-06 7.60e-06 6.58e-06 (1.85e-05) (1.86e-05) (1.85e-05) Sex: Male(Reference) Female 0.200*** 0.198*** 0.198*** (0.013) (0.013) (0.013) Religion: Traditionalist/Atheist/Others(Reference) Catholic 0.003 -0.003 -0.005 (0.035) (0.036) (0.036) Protestant 0.070** 0.062* 0.060* (0.034) (0.035) (0.035) Muslim 0.075** 0.070* 0.069* (0.038) (0.039) (0.039) Marital Status: Never married(Reference) -0.177*** -0.180*** -0.179*** Married (0.019) (0.020) (0.020) Divorced/separated/Widow -0.326*** -0.325*** -0.325*** ed (0.027) (0.027) (0.270) Log of household size 0.115*** 0.120*** 0.120*** (0.010) (0.011) (0.011) Education Level: No education(Reference) 0.052*** 0.043** 0.043** Primary Education (0.019) (0.020) (0.020) 0.038* 0.024 0.024 Secondary Education (0.022) (0.026) (0.026) College/university 0.001 -0.023 -0.022 education (0.034) (0.039) (0.039) Employment Status: No (Reference) Yes 0.001 0.0004 -0.0001 (0.015) (0.015) (0.015) Area of residence: Rural (Reference) Urban -0.007 -0.023 -0.021

127

(0.015) (0.018) (0.018) -0.018*** -0.018*** -0.018*** Log of waiting time (0.005) (0.005) (0.005) Distance to nearest health facility: <1KM (Reference) 0.113*** 0.114*** 0.115*** 1-3 KM (0.017) (0.017) (0.017) 0.095*** 0.095*** 0.095*** 4-5 KM (0.021) (0.021) (0.021) 0.177*** 0.177*** 0.178*** 6-9 KM (0.024) (0.024) (0.025) 0.120*** 0.118*** 0.119*** 10+ KM (0.022) (0.022) (0.022) -0.048 -0.036 Poverty residual (0.034) (0.034) Interaction of wealth index and -0.062* poverty residuals (0.035) 0.143*** 0.163*** 0.171*** Constant (0.055) (0.057) (0.058) Number of observations 16,619 16,560 16,560 Pseudo R2 0.0138 0.0140 0.0140 Wald χ2 855.61*** 857.65*** 868.83*** 707.79 711.39(0.000 717.05(0.000 LR χ2 (2) (0.000)a )a )a 0.935 0.883(0.00)a 0.777(0.00)a Linktest: hat (0.00)a 0.062(0.704 0.111(0.491)a 0.213(0.184)a hat squared )a Mean VIF 6.10 Note: ***, ** and * denote statistical significance at 1 per cent, 5 per cent and 10 per cent levels of significance, respectively. (.)=Robust Standard Errors; (.)a=P- value Source: Author computation, Study Data, 2013.

The estimation results of the second model, 2SRI, indicated that generalized residuals of poverty status were -0.048 and not statistically significant. This suggested that poverty status was not endogenous in the health care utilization model. The third model, (CFA), indicated that the interaction of poverty status and its generalized residuals were -0.062 and statistically significant at 10 per cent level of significance. This showed presence of unobserved heterogeneity. Thus,

128

the appropriate model for this study was the CFA regression since there was evidence of unobserved heterogeneity. The results of the CFA showed that the

Wald Chi-square and LR Chi-square tests of goodness of fit rejected the null hypothesis that all parameters were equal to zero. This means that the model fitted the data well. Further, the link test for model specification showed that the probability values of hat and hat squared were 0.000 and 0.184, respectively. This implied that the model was well specified and, therefore, all the regressors were sufficient to explain the changes in the dependant variable. Test of multicollinearity indicated presence of low multicollinearity as the mean VIF was 6.10. However, multicollinearity is only considered a problem if the VIF is greater than 10.

In this study, wealth index was used as a proxy for poverty. Estimation results presented in Table 4.5 showed that the coefficient of wealth index was statistically significant at 5 per cent level of significance with a magnitude of

0.076. This indicated that an increase in wealth leads to increased use of health care. The finding was not surprising since wealth is considered an important enabling factor that influences demand for health care. As argued by Kyegombe

(2003), poor households in most cases have poor quality shelter, water and sanitation, which lead to increased mortality rates. The higher mortality rates are due to problems associated with respiratory diseases linked to cooking fires and lack of ventilation, and waterborne disease such as diarrheal due to dirty water.

Wealthier individuals may not suffer from health problems associated with poverty. However, in the event they suffer from such problems or any other type of illness or injury, wealthy people generally have higher levels of education and

129

their social networks can provide them with high levels of information on where to get quality health care (Kyegombe, 2003). With such information on where to get quality health services and ability to pay for the services, wealthier people tend to use more health services.

Thus, wealthier individuals are expected to seek health care for their families whenever necessary. In addition, majority poor depend more on casual jobs and have to take work in hazardous environment without any protective gear.

However, wealthier individuals are most probably employed and salaried and working in safe environments unlike their poor counterparts. This may imply that, wealthy people can easily miss work by a day or two to seek medication and still earn a living as they can seek a day off. The employed also enjoy medical insurance covers from their employers. However, the poor who may be causal labourers will find it difficult to miss work even for a day to seek medication due to high opportunity cost involved.

Further, the wealthy individuals do possess assets, which they can dispose off in times of distress due to illness. They can as well use such assets as security to seek medical care or loan facility to pay for medical care. This is not possible by the poor who may not possess assets that can be used as security for medication or, which can fetch enough cash even if they were sold off. The finding that wealth increases use of health care was consistent with earlier studies in Kenya

(Kimani, 2014; Kimani et al., 2016; Ochako et al., 2011), which found that increase in wealth increases number of visits to hospitals. It was also consistent

130

with those of Hamad and Rehkopf (2015) who showed that higher income had a positive and statistically significant association with prenatal care visits.

Results of estimation also showed that the coefficient of females was positive and statistically significant at 1 per cent significance level with a magnitude of 0.198.

This implies that females were more likely to use health care compared to their male counterparts other factors being constant. The difference between females and males in the utilization of health care could be associated with reproductive and conditions specific to gender such as monthly periods associated with females only. In Kenya, maternal health services are free in government hospitals. This may partly explain the finding that females utilize health care more than males due to reproductive related services they use mostly related to sexual and reproductive health, prenatal care and maternal and child health. Anectodal evidence also shows that males are slow in seeking health care unless the illness is serious. The finding is consistent with those of Dias, Gama, Cortes, and de Sousa

(2011) on Portugal, Skordis-Worrall et al. (2011) on South Africa, and Zyaambo,

Siziya, and Fylkesnes (2012) on Zambia who found that men are less likely to seek health care when they fall sick leading to less hospital visits.

Concerning religion, which was categorized as traditionalists/atheists/others,

Catholics, Protestants and Muslims, the estimation results showed that the coefficients for Protestants and Muslims were 0.060 and 0.069, respectively. The coefficients were statistically significant at 10 per cent significance level. This implied that Protestants and Muslims were more likely to utilize health care than traditionalists/atheists/others, other factors being constant. This is an indication

131

that Protestants and Muslims may believe in modern medicine compared to traditionalists who are conservative and will shun use of modern medicine even when seriously ill. This finding was consistent with those of Stephenson,

Baschieri, Clements, Hennink, and Madise (2006) on Kenya, who found that protestants were more likely to visit a hospital for maternal health care compared to those who adhere to other beliefs.

Estimation results given in Table 4.5 further showed that individuals who were divorced/separated/widowed and those currently married were less likely to utilize health care compared to those individuals who never married ceteris paribus. The coefficients for the divorced/separated/widowed and the currently married were -0.179 and -0.325, respectively. All the coefficients were statistically significant at one per cent level of significance. This finding indicates that individuals who are divorced/separated/widowed and the currently married probably have better health status compared to those never married. The finding could as well suggest a higher opportunity cost of seeking health care for currently married and divorced/separated/widowed individuals who may be working hard to cater for their dependants. This finding is similar to those of

Awiti (2014) and Mwabu, Wang‟ombe, and Nganda (2003) in Kenya. The authors found a negative relationship between marital status and health care utilization in

Kenya.

The estimation results also indicated that the larger the household, the more the use of health care. The coefficient of the log of household size was positive with a magnitude of 0.120. The coefficient was statistically significant at one per cent

132

significance level. This finding could be because in larger households, individuals are more likely to fall sick, especially from communicable diseases due to congestion. Also, individuals from large households may suffer from nutrition related diseases such as malnutrition, especially if the household poor. This high likelihood of individuals from large households falling sick may lead to more health care utilization. This study finding was consistent with those of Kimani et al. (2016) on Kenya. The author found that a ten per cent increase in household size led to 0.95 increase in the difference in logs of expected number of hospital visits.

According to estimation results, health care utilization increases with increase in education level. Compared to individuals with no education, those with primary level of education were more likely to use health care other factors being constant.

The effect of education was positive for primary level of education with a magnitude of 0.043. The coefficient was statistically significant at 5 per cent significance level. This could be because educated individuals may understand better the benefits of good health and hence demand more health care. The educated individuals are also likely to have better jobs and earn income, which enables them to afford health care.

The results also showed that, those with the college/university level of education had a negative effect on health care utilization though statistically insignificant.

This finding was surprising and unexpected. The explanation for this inconsistency could be that, individuals with higher levels of education tend to have less exposure to stress related to economic deprivation. They may, therefore,

133

be less prone to illness than those with lower levels of education who are likely to adopt unhealthy coping behaviors for stress such as drug abuse and smoking

(Zimmerman & Woolf, 2014).

In addition, individuals with more education tend to have greater socioeconomic resources for a healthy lifestyle and a greater relative ability to live and work in environments with resources and built designs for healthy living (Zimmerman &

Woolf, 2014). As such, more educated individuals rarely fall sick hence do not visit hospitals so often for health care utilization. This study finding is consistent with Mwabu et al. (2003) on Kenya. The authors found that education increases health care utilization. However, the finding contradicts those of Kimani (2014) who found a positive and statistically insignificant effect of education on health care utilization in Kenya.

The estimation results reported in Table 4.5 further showed that the coefficient of log of waiting time was negative with a magnitude of -0.018. The coefficient was statistically significant at one per cent significance level. This implied that long waiting time may discourage individuals from visiting hospitals for health care.

This is mainly due to high opportunity costs associated with waiting time while seeking health care. In this case, individuals who are in informal sector and those with unstable source of income are more likely to opt to go to work rather than spending many hours in hospitals seeking health care and lose their daily income.

This finding contradicts those of Ali and Noman (2013) on Bangladesh and

Kimani (2014) on Kenya. The authors found a positive relationship between waiting time and health care utilization.

134

Results on employment status showed that being employed reduces health care utilization. The coefficient of employment status was -0.0001 and it was statistically insignificant. Although the finding was not expected, the negative coefficient could be because majority of employed Kenyans (83%) work in the informal sector (Republic of Kenya, 2016). Thus, a visit to a health facility would mean lost valuable time and lost earnings. This, therefore, would mean that those working in the informal sector would rather fail to go to seek health care rather than lose their earnings. This finding is consistent with those of Kimani et al.

(2016) on Kenya who found that working status significantly reduced health care use.

According to estimation results, individuals in urban areas were less likely to seek health care compared to their rural counterparts ceteris paribus. The coefficient on urban residence was negative with a magnitude of -0.021 and statistically insignificant. This finding is surprising and was not expected. The negative relationship, however, could be due to the urban advantage, which shows that urban residents are more likely to have better health than their rural counterparts.

Thus, urban residents are less likely to seek health due to their good health. This study finding contradicts those of Kimani et al. (2016) and Mwabu et al. (2003).

The authors found a positive and statistically significant relationship between urban residence and health care utilization in Kenya.

Estimation results given in Table 4.5 further showed that distance to the nearest health facility had a positive effect on health care utilization. Distance to the nearest health facility was categorized in to four: 1) 1-3 Kilometers, 2) 4-5

135

Kilometers, 3) 6-9 Kilometers, and 4) more than 10 Kilometers. The coefficients for categories 1 to 4 were 0.115, 0.095, 0.178 and 0.119, respectively. All the coefficients were statistically significant at 1 per cent significance level. Although this finding was not expected, the positive relationship between distance and health care utilization may suggest that distance is not a hindrance to health care utilization. This could be so especially if individuals are more concerned with quality of services offered or the cost of seeking health care at any given health facility.

The finding on the relationship between distance and health care utilization is consistent with those of Awiti (2014) and Kimani et al. (2016) on Kenya who found a positive and statistically significant relationship between distance and health care utilization. However, the study finding contradicts those of Mwabu et al. (2003) on Kenya who found a positive and statistically insignificant effect between distance and health care utilization in Kenya. The finding also contradicts those of Awoyemi, Obayelu, and Opaluwa (2011) on Nigeria who found an inverse relationship between distance to nearest health facility and health care utilization.

Overall, the results presented and discussed revealed that increasing wealth increases health care utilization. In this study, wealth was a proxy for poverty status. Thus, it could be argued that, decrease in poverty increased health care utilization and vice versa. Other factors that were found to have a statistical significance in increasing health care utilization were being a female, a protestant and being a Muslim. Further, increase in household size, having primary

136

education level and increase in distance to the nearest health facility increased utilization of healthcare. Factors that were found to have a statistical significance in influencing health care utilization negatively included being married, being divorced/separated/widowed, and waiting time in a health facility before one can be attended to.

4.2.2 Effect of poverty on choice of health care providers in Kenya

The second objective of this study was to investigate the effect of poverty on choice of health care providers in Kenya. To achieve this objective, a multinomial probit model was estimated using the 2013 KHHEUS dataset. Equations 3.37 and

3.39 were estimated. Further, 2SRI model was estimated to test and control for endogeneity since poverty is potentially endogenous in the choice of health care providers. In addition, CFA was used to ascertain and control for unobserved heterogeneity. In this study, average number of households at the county level that have access to piped water was used as instrumental variable in the model. The variable was used since the average number of households at the county level that have access to piped water is not expected to influence choice of health providers directly. However, access to piped water is expected to have a high correlation with poverty. Also, households in areas with low access to piped water are more likely to be poor.

Table A2 presents results for the test of validity, strength and relevance of the instrumental variable. The results presented in Table A2 indicated that the instrument was highly correlated with the endogenous variable with a P-value of

0.000 and a magnitude of 0.174. However, the instrument was not correlated with 137

the structural error term as it was not statistically significant in the health care provider choice model. The results implied that the average number of households that had access to piped water at the county level was a valid, strong and relevant instrument variable for wealth index. Wealth index was the only endogenous variable and so the model was just identified implying that there was no need for identification tests.

Results of the first stage of 2SRI were same as those presented under section

4.2.1. The second step of 2SRI approach involved estimation of the choice of health care provider‟s model. In the model of choice of health care providers, wealth index and its generalized residuals were included as additional explanatory regressors. Health care provider variable had four categories namely, government health facilities, private health facilities, mission health facilities and others. The

„others‟ category includes traditional/religious/cultural healers, village health works, and shops. The „others‟ category was the reference category in all the health care provider choice models. Thus, results were interpreted using other category as the reference.

Three models were estimated in each category of health care provider, a baseline model that does not control for endogeneity and unobserved heterogeneity, a model that controls for endogeneity (2SRI) and a model that controls for unobserved heterogeneity (CFA). Estimation results presented in Table A4, A5,

A6 and A7 showed that even though poverty was endogenous in health care provider choice model, there was no evidence of unobserved heterogeneity in the models. Therefore, the most appropriate model was that, which controls for

138

endogenity to avoid biased results. Table 4.6 presents results of the average marginal effects for the health provider choice for government, private, mission and other health facilities.

Table 4.6: Average Marginal Effects for the Healthcare Provider Choice Model Dependent variable=Health care provider Variable Government Private Mission Others -0.167(- 0.175(11.28)** -0.004(- -0.004(- Wealth Index 10.09)*** * 0.41) 1.06) -0.0004(-1.58) -0.0005(- 0.001(5.50) 0.0001(2.1 Age 1.96)** *** 5)** Sex: Male(Reference) -0.060(- 0.020(4.67) -0.004(- Female 0.044(5.60)*** 7.98)*** *** 2.09)** Malaria/Fever: No (Reference) 0.046(6.04)*** -0.020(- -0.006(- Yes -0.020(-2.52)** 4.44)*** 3.74)*** Respiratory Disease/Pneumonia: No (Reference) -0.002(-0.17) 0.025(2.31)** -0.016(- -0.007(- Yes 2.72)*** 4.23)*** Diarrhea: No (Reference) 0.031(1.11) 0.046(2.33) -0.004(- Yes -0.073(-2.51)** ** 0.74) Marital Status: Never married(Reference) -0.010(-1.06) 0.025(2.72)*** -0.013(- -0.001(- Married 2.17)** 0.62) Divorced/Separated/ -0.039(- 0.064(4.52)*** -0.025(- -2.5e-05(- Widowed 2.64)*** 3.12)*** 0.01) 0.007(1.07) -0.007(-1.13) -0.002(- 0.002(1.26) Log of Household size 0.51) Distance to nearest health facility: <1 KM (Reference) 0.192(18.10)** -0.180(- 0.010(1.75) -0.022(- 1-3 KM * 17.12)*** * 6.26)*** 0.210(16.25)** -0.199(- 0.013(1.94) -0.025(- 4-5 KM * 15.77)*** * 6.64)*** 0.221(15.87)** -0.241(- 0.038(4.70) -0.018(- 6-9 KM * 18.17)*** *** 4.17)*** 0.165(12.63)** -0.198(- 0.054(7.07) -0.021(- 10+ KM * 15.72)*** *** 5.31)*** Education Level: No education (Reference) 0.058(5.08)*** -0.051(- -0.005(- -0.003(- Primary Education 4.52)*** 0.79) 0.99) Secondary Education 0.054(3.84)*** -0.049(- 1.49e- -0.005(-

139

3.57)*** 05(0.00) 1.60) College/University 0.0005(0.02) -0.016(-0.78) 0.019(1.48) -0.003(- education 0.661) Area of residence: Rural (Reference) -0.048(- 0.034(3.32)*** 0.013(2.07) 0.0002(0.0 Urban 4.36)*** ** 6) Poverty Residuals 0.036(2.00)** -0.056(- 0.021(2.14) -0.001(- 3.31)*** ** 0.32) Wald χ2 (54) 1816.98*** Number of 17,797 observations Mean VIF 1.29 Note: ***, **, and * denote statistical significance at 1 per cent, 5 per cent and 10 per cent levels of significance, respectively. (.)=Robust Z Statistics Source: Author computation, Study Data, 2013.

Estimation results presented in Table 4.6 showed that Wald Chi-square was

1816.98 and was statistically significant at one per cent significance level. This showed that the model was well fitted. Hence, all the model parameters were jointly different from zero and the model fitted the data well. Results also indicated presence of low multicollinearity problem since the mean VIF was 1.28.

Multicollinearity is considered a problem if the mean VIF is 10 or more (Verbeek,

2012). Thus, the detected level of multicollinearity cannot affect the stability and consistence of results.

Estimation results presented in Table 4.6 showed that the coefficients for wealth index for government, private, mission and other health facilities were -0.167,

0.175, -0.004, and -0.004, respectively. Coefficients for government and private health facilities were statistically significant at one per cent level of significance while those of mission and „others‟ were statistically insignificant. The results imply that a unit increase in wealth index increases the probability of visiting a

140

private health facility by 0.175 holding other factors constant. The results further suggest that the probability of visiting government declines by 0.167 with a unit increase in wealth index holding other factors constant.

It may be argued based on the foregoing findings that wealthy people prefer private hospitals because they consider them to offer better clinical services and standard health care (Prasad, 2013). Thus, private hospitals are preferred by those who can afford their services because they are considered to offer high quality services compared to government and other health facilities. Due to high costs charged in private hospitals, they attract few people compared to government health facilities. Thus, there are shorter queues in private health facilities as compared to government hospitals.

Wealthy people in most cases would prefer to visit high cost private hospitals, and spend less time rather than visit government hospitals associated with inefficiencies and long queues. This minimizes time wastage by the wealthy people. Therefore, reduction in poverty through accumulation of assets increases the probability of visiting private health facilities compared to any other health facility, holding all other factors constant. This finding is consistent with those of

Gakii (2013) on Kenya who found that increase in income increased the probability of visiting private hospitals and reduced probability of visiting government hospitals. The finding is also consistent with those of Hallman (1999) on Philippines who found that increase in household income increased use of private health care providers.

141

The estimation results given in Table 4.6 also indicated that the coefficients for age in relation to government, private, mission and other health facilities were -

0.0004, -0.0005, 0.001 and 0.0001, respectively. The coefficient for age in relation to government health facility was not statistically significant. The coefficient for age in relation to mission hospitals was statistically significant at one per cent significance level. The remaining coefficients for age in relation to private and other health facilities were all statistically significant at five per cent significance level. The results imply that if the age of an individual increases by one year, the probability of visiting a private hospital reduces by 0.0005 and that of visiting mission hospital increases by 0.001 holding other factors constant.

The finding that as individual‟s age increases, people are more likely to visit mission hospitals could be because many mission hospitals provide specialized treatment and are relatively cheaper compared to private hospitals. This could as well explain the negative relationship between age and the probability of visiting private hospitals. This finding is inconsistent with those of Amaghionyeodiwe

(2008) on Nigeria who found that older health care seekers tend to patronize public and private hospitals. The finding further contradicts those of Gakii (2013) on Kenya who found that older people were less likely to visit mission hospitals.

Regarding sex, estimation results revealed that the probability of females visiting private hospitals was lower by 0.060 compared to their male counterparts ceteris paribus. The results also indicated that the probability of visiting health facilities under „other‟ category was lower by 0.004 compared to males other factors being constant. However, the results showed that the probability of females visiting

142

government and mission health facilities was higher by 0.044 and 0.020, respectively compared to that of males other factors held constant. The coefficients of females visiting various health facilities were all statistically significant at one per cent significance level for all categories of health facilities.

The high probability of females visiting government health facilities in Kenya could be explained by the fact that maternal health care is free in those facilities.

Thus, females are more likely to visit those government facilities for health care related to reproductive health, prenatal and ante natal care, maternal and child health. Such health care services are offered at a cost in private hospitals and the cost may hinder females from seeking them in those facilities.

Females may also visit mission hospitals, which offer health care services at relatively low cost compared to other private health facilities. The finding could also be explained by empowerment differences that exist between males and females. Males are more empowered economically and socially. Thus, males can afford to visit private hospitals more than females. This finding agrees with those of Gakii (2013) on Kenya who found that males were less likely to visit government and mission hospitals but were more likely to visit private and other health facilities compared to females.

The estimation results presented in Table 4.6 further revealed that those suffering from malaria/fever and respiratory/pneumonia diseases were more likely to visit private health facilities and less likely to visit mission and government health facilities ceteris paribus. On the other hand, the results indicated that those

143

suffering from diarrhea were more likely to visit mission hospitals and less likely to visit government hospitals. Specifically, the coefficients for the variable representing those suffering from malaria/fever and those suffering from respiratory/pneumonia diseases and sought health care from private were 0.046 and 0.025, respectively. The coefficients were statistically significant at one and five per cent significance levels, respectively. The results imply that ceteris paribus, the probabilities of visiting private hospital by those suffering from malaria/fever and respiratory/pneumonia diseases were high by 0.046 and 0.025, respectively compared to those not suffering from such diseases.

The results further suggest that the probability of visiting mission hospital by those suffering from diarrhea was high by 0.046 compared to those not suffering from diarrhea, holding other factors constant. This could be explained by the perceived severity and seriousness of a disease. Malaria/fever and respiratory diseases are considered fatal and, may, therefore, prompt people to seek medical care in private hospitals, which are considered efficient and offer high quality health services. Hence, the aspect of high cost may be overlooked in relation to such diseases. The finding contradicts those of Asenso-Okyere et al. (1996) on

Ghana who found that individuals suffering from malaria were more likely to favor self medication.

The estimation results reported in Table 4.6 also showed that the coefficients for the married were positive and statistically significant at one per cent significance level in relation to choosing private hospitals. The results further revealed that the coefficient for the married was negative and statistically significant at five per

144

cent level of significance in relation to choosing mission hospitals. However, coefficients for the married in relation to visiting government and other health facilities were negative and not statistically significant. The coefficients for the married in relation to visiting government, private, mission and other health facilities were -0.010, 0.025, -0.013, and -0.001, respectively.

In relation to the divorced/separated/widowed, the coefficients were -0.039,

0.064, -0.025, and -0.000025 for those visiting government, private, mission and other health facilities, respectively. All the coefficients were statistically significant at one per cent significant level except that of the „other‟ category, which was not statistically significant.

The reported results imply that the probability of those married visiting private hospital was higher by 0.025 compared to that of the never married ceteris paribus. The results also suggested that the probability of those married visiting mission hospital was lower by 0.013 compared to those never married holding other factors constant. This could be explained by the marriage benefits associated with resource model, which suggests that the married have more resources at their disposal compared to those never married. The married could be having more resources than the „never‟ married especially if the couple is engaged in income generating activities. Thus, the married are able to afford health care in private hospitals, which charge higher fees compared to mission hospitals and are considered to offer high quality health care.

145

The results further indicate that the probability of the divorced/separated/widowed visiting private hospital is higher by 0.064 compared to the never married other factors being constant. Also the results revealed the divorced/separated/widowed compared to those never married had a lower probability of visiting government and mission hospitals by 0.039 and 0.025, respectively holding other factors constant. This could be because, even though the marriage is dissolved, the concerned parties may share assets they owned during marriage. In addition, even if those involved are divorced, they may continue to support each other, especially where children are involved. Thus, the divorced/separated/widowed will still be better off than those never married and hence they can still afford health care in private hospitals.

The estimation results presented in Table 4.6 further showed that there exists a positive relationship between distance to the nearest health facility and seeking health services in government hospitals. The coefficient of the variable was statistically significant at one per cent significance level. However, there was a negative relationship between distance to nearest health facility and visiting private health facility. The coefficient was statistically significant at one per cent level of significance. According to the estimation results, the probability of visiting government hospital was higher by 0.192, 0.210, 0.221, and 0.165 for those residing in distances of 1-3, 4-5, 6-9, and 10 and above kilometers, respectively from health facilities. This was compared to those residing in less than a kilometer ceteris paribus.

146

The results further suggests that the probability of visiting mission hospitals increased by 0.010, 0.013, 0.038 and 0.054 for people residing in 1-3, 4-5, 6-9, and 10 and above kilometers from the nearest hospital, respectively. This was compared to those residing in less than a kilometer from a health facility, holding other factors constant. In addition, the results showed that as distance from a health facility increased, the probability of visiting a private health facility decreased other factors being constant.

The positive relationship between distance to the nearest health facility and health care providers is contrary to expectation. This is because, longer distance is associated with direct and indirect costs in terms of money and time spent in seeking health care. However, the finding provides an evidence of bypassing nearer health facilities in search of health care in far away facilities. This could be motivated mainly by price and quality concerns. Government hospitals are generally cheaper compared to private hospitals in terms of direct costs. Thus, individuals may prefer to travel long distances to seek medical care from government hospitals, which are deemed to be cheaper.

In addition, individuals may not travel long distance to seek health care from private hospitals, which offer health care at relatively high rates. This finding is consistent with Awiti (2013) on Kenya who found that an increase in distance to the nearest health facility by one kilometer increased the probability of visiting a modern health care provider by 0.005. The finding is also consistent with those of

Amaghionyeodiwe (2008) on Nigeria. The author found that the coefficient for distance to nearest health facility was negative and significantly different from

147

zero in relation to use of health care from public health facilities. The author also found that the coefficient for distance to nearest health facility was positive and statistically significant in relation to choice of private health facilities. This finding contradicts those of the current study, which found that distance to the nearest health facility was negative and statistically significant in relation to choice of private health care providers. The current study finding also contradicted those of Gakii (2013) on Kenya who found that an increase in distance to health facilities decreased the probability of people visiting government health facilities and increased that of visiting private health facilities.

Regarding education, estimation results showed that, compared to individuals with no education, those with primary and secondary levels of education were more likely to visit government hospitals and less likely to visit private hospitals, holding other factors constant. Specifically, the probability of visiting government hospitals by those individuals with primary and secondary level of education was higher by 0.058 and 0.054, respectively compared to those with no education ceteris paribus. In addition, the probability of visiting private hospitals by individuals with primary and secondary level of education was lower by 0.051 and 0.049, respectively compared to those with no education, holding other factors constant. The coefficients for visiting government and private hospitals for those with primary and secondary levels of education were all statistically significant at one per cent level of significance. This could imply that as people become more educated, they are able to make informed decisions regarding their health. Thus, they are able to decide which health care would improve their health

148

best, especially in case of serious illness/injury. In most cases, it is relatively cheaper in government health facilities as compared to private health facilities.

This study finding is consistent with that of Gakii (2013) on Kenya who found that a year of schooling reduced the probability of visiting a private health facility by 0.511 per cent.

The estimation results reported in Table 4.6 further indicated that compared to rural residents, those living in urban area, had a lower probability of visiting government hospital by about 0.048. These urban residents had a higher probability of visiting private, mission and other health facilities by 0.034, 0.013 and 0.0002, respectively other factors being constant. The coefficients for government and private were statistically significant at one per cent level of significance, while that of mission health facilities was statistically significant at five per cent significance level. The coefficient of „other‟ health facilities was statistically insignificant. The results imply that urban residents are less likely to visit government hospitals compared to rural residents holding other factors constant. However, urbanites are more likely to visit private and mission hospitals for health care compared to rural residents ceteris paribus.

Compared to rural residents, most urban residents are in gainful employment and, therefore, they have a relatively higher and or stable source of income. This could explain why urban residents prefer private hospitals than government hospitals. In addition, most employed urban residents have medical insurance covers provided mostly by their employers. Thus, they are able to afford medical care in private hospitals, which are considered to be expensive but offer quality health care.

149

The study findings are consistent with those of Habtom and Ruys (2005) on

Eritrea who found that choice of private hospitals and mission hospitals by those living in urban areas was positive and statistically significant. However, the study finding on Eritrea that choice of government hospitals by those living in urban areas was positive and statistically significant contradicts the current study finding, which found a negative and statistically significant relationship. This contradiction could be due to differences in place of the studies and also the timing. The finding of the current study, however, agrees with those of Gakii

(2013) on Kenya who found that living in an urban area decreased the probability of visiting government health care provider, but increased the probability of visiting a private health care provider.

In summary, the results presented and discussed indicated that increase in wealth index increased the probability of choosing a private health provider and reduced that of choosing a government, mission and other health care providers ceteris paribus. Other factors that were found in this study to significantly influence how patients choose health care providers were age, sex, marital status, distance to the nearest health facility, education level, residence, and the nature of the illness.

4.2.3 Effect of poverty on health status in Kenya

The third objective of this study was to establish the effect of poverty on health status in Kenya. To achieve this objective Ordered Probit model and the 2013

KHHEUS dataset were used. Equations 3.46 and 3.49 were estimated. Ordered

Probit was used because health status had four categories in the survey namely, very good, good, satisfactory and poor. Thus, considering that there is a natural 150

ordering of the various alternatives, it was better to use ordered probit model that accounts for such ordering. Further, the study used 2SRI model in order to ascertain and control for potential endogeneity, and a CFA model to account for unobserved heterogeneity (Kabubo-Mariara et al., 2012; Kabubo-Mariara et al.,

2009).

This study used average number of households at the county level that have access to electricity as instrument variable. The variable was used because the average number of households at the county level that have access to electricity is not expected to have a direct influence on health status of individuals. However, there is a high correlation between access to electricity and poverty. Households found in areas with low electricity access are expected to have a higher probability of being poor.

Table A3 presents results for the test of validity, strength and relevance of the instrumental variable. The results show that the instrumental variable is highly correlated with the endogenous variable with a P-value of 0.000 and a magnitude of 0.633. However, it is uncorrelated with the structural error term. Therefore, it can be concluded that the average number of households at county level with access to electricity is a valid, strong and relevant instrument variable for wealth index. Since, there is only one endogenous variable and one instrumental variable, the model is just identified and hence there is no need for identification tests.

Results of the first stage of 2SRI are similar to those presented under section

4.2.1. The second step of 2SRI approach involved estimation of the self-rated

151

health status model. Wealth index and its generalized residuals were included in the health status model as additional explanatory regressors. Since health status had four categories namely poor, satisfactory, good and very good, only results of the two extremes (poor and very good) are presented and discussed here for brevity. Results of health status rated as satisfactory and good are presented in the appendix in Table A8 and Table A9. Results of health status rated as poor and very good are presented in Table 4.7 and Table 4.8. Each table presents results of the baseline model, which does not control for endogeneity and unobserved heterogeneity, the model controlling for endogenity (2SRI) and the model controlling both endogenity and unobserved heterogeneity (CFA).

Table 4.7 presents results of three different models. The first model is the baseline model, which presents results before controlling for endogeneity and unobserved heterogeneity. The second model is 2SRI model, which presents results after controlling for endogeneity. The last model is the CFA model, which presents results after controlling for endogeneity and unobserved heterogeneity.

Table 4.7: Average Marginal Effects of Probability of Reporting Own Health as Poor Self Rated Health Status=Poor Variable Baseline Model(1) 2SRI Model(2) CFA Model(3) Wealth Index -0.013(-12.40)*** -0.013(-6.00)*** -0.012(-5.46)*** Age 0.001(4.43)*** 0.001(4.39)*** 0.001(4.51)*** 1.37e-05(8.73)*** 1.34e- Age Squared 1.37e-05(8.72)*** 05(8.57)*** Sex: Male(Reference) Female 0.012(13.64)*** 0.012(13.72)*** 0.012(13.64)*** Marital Status: Never married(Reference) Married -0.005(-3.93)*** -0.005(-3.83)*** -0.005(-3.77)*** Divorced/Separated/Wid 0.012(5.91)*** 0.012(5.90)*** owed 0.012(6.01)***

152

Religion: Traditionalist/Atheist/Others(Reference) Catholic 0.003(1.43) 0.004(1.47) 0.003(1.40) Protestant 0.006(2.76)*** 0.006(2.74)*** 0.006(2.63)*** Muslim -0.007(-2.84)*** -0.007(-2.75)*** -0.007(-2.73)*** Household size 0.0003(1.84)* 0.0003(1.74)* 0.0003(1.66)* Education Level: No education (Reference Primary Education 0.002(1.71)* 0.002(1.71)* 0.002(1.27) Secondary Education -0.008(-5.34)*** -0.008(-4.73)*** -0.009(-5.12)*** College/University -0.019(-7.70)*** -0.019(-7.97)*** education -0.019(-9.04)*** Employment Status: No (Reference) -0.011(-10.09)*** -0.011(- Yes -0.011(-10.20)*** 10.19)*** Access to piped water: No (Reference) -0.026(-11.06)*** -0.027(- Yes -0.026(-12.28)*** 11.18)*** Area of residence: Rural (Reference) Urban 0.004(4.26)*** 0.004(3.37)*** 0.005(3.77)*** Poverty Residuals 0.0004(0.19) 0.002(1.04) Interaction of wealth -0.013(-6.58)*** index and poverty residuals Number of observations 80742 80450 80450 Note: ***, ** and * denote statistical significance at 1 per cent, 5 per cent and 10 per cent levels of significance, respectively. (.)=Robust Z Statistics Source: Author computation, Study Data, 2013.

Results of 2SRI model presented in Table 4.7 indicated that there is no evidence of endogeneity since the generalized residuals of poverty are statistically insignificant. Results of the CFA model on the other hand indicated presence of unobserved heterogeneity, which biases the results in models that do not control for it. Thus, the preferred model is that which control for unobserved heterogeneity. In this case, the model of interest was the CFA model. The results of the CFA model are, therefore, the ones to be interpreted.

According to the CFA model results, the coefficient of wealth index was negative with a magnitude of 0.012. The coefficient of the variable was statistically 153

significant at one per cent significance level. This implies that the probability of reporting own health as being poor falls by 0.012 with a unit increase in wealth index ceteris paribus. Thus, a decrease in poverty minimizes the probability of reporting poor health. The reason could be that poor people are likely to report poor health because they cannot afford to buy things that are needed for good health including sufficient amounts of quality food and health care. This could be because poor individuals do not have the capability and ability to purchase things needed for good health in case they fall sick. Thus, they are likely to report poor health. The poor are also not able to afford balanced meals and clean drinking water and, therefore, they are most likely to suffer from nutrition related diseases and waterborne diseases. Thus, they are likely to report poor health.

The poor are also more likely to have less education compared to wealthy individuals. Thus, they may lack information that is appropriate for health- promoting practices or they may lack voice that is needed to make social services work for them. This finding is consistent with Kodzi, Gyimah, Emina, and Ezeh

(2011), and Awiti (2013) on Kenya. The finding is also consistent with those of

Ahmad, Jafar, and Chaturvedi (2005) on Pakistan; Chin (2010) on Malawi; and

Bora and Saikia (2015) and Corsi and Subramanian (2012) on India. The authors found that increase in wealth status reduces the probability of rating own self health as poor. Thus, as wealth increases, self rated health status improves.

The estimation results reported in Table 4.7 further indicated that, the coefficient for age of an individual was 0.001 and was statistically significant at one per cent significance level. This means that a one year increase in individual‟s age

154

increases the probability of an individual reporting own health as being poor by

0.001 other factors being constant. This could be because with advancing age, individuals are more likely to experience life threatening health events. As children grow, they start exploring outside life that may expose them to diseases.

Adolescents may start experimenting with their bodies trying to discover themselves and hence engage in risky behaviours thus affecting their health.

Adults on the other hand my engage in risky income generating activities thus endangering their health. This finding is consistent with Hu et al. (2016) on

Taiwan. The authors found that as people‟s age increase, they are likely to experience poor health. The findings also agree with early studies in Kenya such as Awiti (2013), Gakii (2013) and Kodzi et al. (2011) who found that as individuals advance in their ages, they are more likely to report poor health.

The estimation results presented in Table 4.7 show that in relation to sex, the coefficient of female was positive with a magnitude of 0.012. The coefficient of the variable was statistically significant at one per cent significance level. This suggests that females have a higher probability of reporting their own health status as being poor compared to their male counterparts. Specifically, holding other factors constant, the probability of females reporting their own health as being poor was higher than that of males by 0.012. Existing literature indicates that females are more likely to report poor health than males due to social, cultural, economic and biological factors that all impact negatively on the health of females compared to their male counterparts (Hosseinpoor et al., 2012; Sen &

Östlin, 2008). These may include factors associated with pregnancy and childbirth

155

complications, contraception, Female Genital Mutilation (FGM), and also lack of autonomy in seeking and realizing health care opportunities. This result is consistent with other studies that show that women are more likely to report their own health as being poor compared to men (Ahmad et al., 2005; Awiti, 2013;

Bora & Saikia, 2015; Gakii, 2013; Kodzi et al., 2011).

Regarding marital status, which was categorized as never married, married, and divorced/separated/widowed, results indicated that married individuals were

0.005 less likely to report poor health ceteris paribus. This was compared to those never married. On the other hand, those who are divorced/separated/widowed were 0.012 more likely to report poor health compared to their counterparts who were never married, holding other factors constant. The coefficients for the variable representing the married and divorced/separated/widowed were all statistically significant at one per cent significance level. The finding that married individuals were less likely to report poor health while those divorced/separated/widowed were more likely to report poor health compared to those never married could be because married people may have more resources, compared to the unmarried and the divorced. This finding could be true, especially if the couple is engaged in income generating activities.

Married people, thus, are in a better position to seek medical care in case of illness and they are also able to afford good meals, which are good for their health. Further, as Liu and Umberson (2008) argues, marriage has benefits in that it enhances mental health, which in turn positively affects physical health. On the other hand, people who are divorced/separated/widowed may suffer from

156

psychosocial problems, which may affect their health. They may also not be able to afford health care services in case of illness due to limited resources compared to the married people. This finding agrees with those of Awiti (2013) and Kodzi et al. (2011) on Kenya. The authors found that married people were more likely to report better health than the unmarried. However, the finding contradicts Bora and

Saikia (2015) on India who found that the probability of reporting poor health was lower among the unmarried compared to their married counterparts.

Regarding religion, which was categorized as Catholics, Protestants, Muslims and traditionalists/atheists/others, the estimation results presented in Table 4.7, show that Catholics and Protestants were more likely to report poor health compared to the traditionalists/atheists/others. On the other hand, the results show that

Muslims were less likely to report poor health compared to traditionalists/atheists/others. The magnitudes of the coefficients of Catholics,

Protestants and Muslims were 0.003, 0.006 and -0.007, respectively. Apart from the coefficient of Catholics, which was statistically insignificant, all the other coefficients were statistically significant at one per cent significance level. This implies that, the probability of Protestants reporting their health as being poor was higher by 0.006 compared to traditionalists/atheists/others, other factors being constant. On the other hand, compared to traditionalists/atheists/others, the probability of Muslims reporting their own health as being poor was less by

0.007. This finding is inconsistent with Kodzi et al. (2011) on Kenya. The author found that Protestants were more likely to report better health and Muslims were less likely to report better health. The finding also contradicts Bora and Saikia

157

(2015) on India who found that Muslims were more likely to rate their own health as being poor.

The estimation results reported in Table 4.7 further show that the coefficient of household size was positive and statistically significant at 10 per cent significance level. The magnitude of the coefficient of household size was 0.003. Specifically, the results imply that the probability of reporting own health as being poor increases by about 0.003 with a one member increase in household size, ceteris paribus. Thus, larger households could mean increased competition for the limited household resources and basic necessities such as shelter, food and clothing. The increased competition may negatively impact on health of individuals in the household. In congested households, for example, members may suffer from airborne and other communicable diseases. Hence, they are more likely to report poor health than the non-congested households. This finding is consistent with Gakii (2013) on Kenya who found a negative relationship between household size and the probability of reporting good health.

Regarding education level, which was categorized as no education, primary education level, secondary education level and college/university education level, estimation results indicated that apart from the coefficient for primary education level, coefficients for all other levels of education were negative. The results also indicated that the coefficient for primary education level category was statistically insignificant. However, coefficients for secondary and college/university education level categories were statistically significant at one per cent significance level. This suggests that compared to individuals with no education,

158

those with secondary and college/university level of education are less likely to report their own health as being poor by 0.008 and 0.019, respectively holding other factors constant.

Though the coefficient of variable representing individuals with primary level of education was found to be statistically insignificant, it indicates a higher probability of rating own health as being poor by about 0.002 compared to those with no education, other factors being constant. This could be because, as individuals attain higher levels of education, they become more conscious of their health and they are more likely to take better care of themselves. The results are consistent with those Awiti (2013) and Gakii (2013) on Kenya, Bora and Saikia

(2015) on India, and Teerawichitchainan and Knodel (2015) on Myanmar. The authors established that individuals with no education or have lower education levels report worse health than those with higher levels of education.

Concerning employment status, the estimation results presented in Table 4.7 indicate that the coefficient of being employed was negative and statistically significant at one per cent significance level. Thus, employed individuals are less likely to report their own health as being poor by about 0.011 compared to their unemployed counterparts other factors being constant. Being employed is important in producing better health outcomes. Being in gainful employment brings greater access to health and welfare benefits. Thus, employed individuals are able to afford health care and other health promoting services and goods, which the unemployed may not afford. This finding is consistent with those of

Hosseinpoor et al. (2012) on 18 African countries and 19 European countries. The

159

authors found that being in a paid employment compared to being unemployed, had a much larger positive effect on good health.

The estimation results in Table 4.7 also showed that the coefficient of residence was positive and statistically significant at one per cent significance level.

Specifically, holding other factors constant, the probability of urban residents reporting own health as being poor was higher than their rural counterparts by about 0.005. This could be because majority of urban residents are well educated compared to rural residents. Thus, urban residents could have a conscious perception of their health compared to the rural residents. This result is consistent with Gakii (2013) on Kenya and Teerawichitchainan and Knodel (2015) on

Myanmar. The authors established that urban residents reported poor health than their rural counterparts. However, the finding contradicts those of Ahmad et al.

(2005) on Pakistan and Bora and Saikia (2015) on India who found that people residing in urban areas were more likely to report better health than those residing in rural areas.

As mentioned earlier, only results on poor health status and very good health status are presented and discussed in this section. Table 4.8 shows the results of the models for the probability of rating own health as very good.

160

Table 4.8: Average Marginal Effects of Probability of Reporting Own Health as Very Good Self Rated Health Status=Very Good Baseline Model 2SRI Model(2) CFA Model(3) Variable (1) Wealth Index 0.037(12.56)*** 0.038(6.00)*** 0.035(5.47)*** Age -0.002(-4.46)*** -0.002(-4.42)*** -0.002(-4.55)*** -4.1e-05(- -4e-05(-8.50)*** Age Squared -4.1e-05(-8.66) 8.66)*** Sex: Male(Reference) -0.035(- -0.035(- Female -0.035(-13.83)*** 13.92)*** 13.83)*** Marital Status: Never married(Reference) Married 0.015(3.95)*** 0.014(3.84)*** 0.014(3.79)*** Divorced/Separated/ -0.035(-5.91)*** -0.035(-5.90)*** Widowed -0.035(-6.00) Religion: Traditionalist/Atheist/Others(Reference) Catholic -0.010(-1.43) -0.011(-1.47) -0.010(-1.41) Protestant -0.019(-2.76)*** -0.019(-2.74)*** -0.018(-2.63)*** Muslim 0.022(2.84)*** 0.022(2.75)*** 0.022(2.73)*** Household size -0.001(-1.84)* -0.001(-1.74)* -0.001(-1.66)* Education Level: No education (Reference Primary Education -0.007(-1.71)* -0.007(-1.71)* -0.005(-1.27) Secondary 0.025(5.35)*** 0.025(4.74)*** 0.027(5.13)*** College/University 0.055(7.75)*** 0.057(8.03)*** education 0.056(9.11)*** Employment Status: No (Reference) Yes 0.032(10.30)*** 0.032(10.19)*** 0.032(10.29)*** Access to piped water: No (Reference) Yes 0.078(12.41)*** 0.078(11.18)*** 0.079(11.30)*** Area of residence: Rural (Reference) Urban -0.012(-4.26)*** -0.013(-3.37)*** -0.014(-3.77)*** Poverty residuals -0.001(-0.19) -0.007(-1.04) Interaction of wealth 0.037(6.59)*** index and poverty residuals Number of 80450 80450 observations 80742 Note: ***, ** and * indicates statistical significance at 1 per cent, 5 per cent and 10 per cent levels of significance, respectively. (.)=Robust Z Statistics, (.)a=P- value Source: Author computation, Study Data, 2013.

161

Table 4.8 shows the average marginal effects. Three different models are presented. Baseline model is the first one. The model does not control for endogeneity and unobserved heterogeneity. The second model is the 2SRI, which controls for endogeneiy. The last model is the CFA, which controls for endogeneity and unobserved heterogeneity.

Results presented in Table 4.8 on endogeneity test indicate no evidence of endogenity. This is evident from 2SRI results, which show that generalized residuals of poverty are statistically insignificant. Further, CFA results show evidence of unobserved heterogeneity. This is because the interaction of poverty status and its residuals is statistically significant. Thus, the preferred model is the

CFA since it controls for the unobserved heterogeneity and so the results are not biased. Interpretation of results is therefore based on the CFA model only.

The estimation results given in Table 4.8 indicate a positive relationship between wealth index and reporting very good health status. The coefficient of wealth index was positive with a magnitude of 0.035 and was statistically significant at one per cent significance level. This implies that a unit increase in wealth index increases the probability of rating own health as very good by 0.035 holding other factors constant. It may be argued that based on the foregoing findings, ownership of assets is crucial in influencing health status of an individual. Members of households with more assets are, therefore, associated with better health status than those with fewer assets. This is because household assets have a direct protective effect on health. Moreover, ownership of assets is considered as a sign of one‟s status in the society. Thus, those with more household assets are

162

accorded a higher status in the society. As such, ownership of household assets may affect the psychosocial well-being by psychosocial processes (Ahmad et al.,

2005).

Individuals with higher wealth status, therefore, enjoy the direct protective effect on health by items they possess. In addition, wealthy people enjoy a higher social status in the society, which positively affects their mental health, which in turn has a positive impact on physical health. Due to the assets they own, wealthy individuals are able to afford health care in case of illness or injury and other health promoting goods and services such as balanced meals, good shelter, cleaning drinking water, and exercises. This is unlike the poor who may be struggling to afford meals and even good shelter.

Besides, wealthy individuals are likely to live in good environment, which is not congested and is well supplied with basic amenities. As such, wealthy people are less likely to suffer from communicable diseases like the poor individuals who in most cases live in highly congested areas with no basic amenities such as toilets.

Thus, individuals with higher wealth index are more likely to report own health as being very good compared to their poor counterparts, ceteris paribus. This finding is consistent with earlier studies by Awiti (2013), Bora and Saikia (2015),

Teerawichitchainan and Knodel (2015) who found that the probability of reporting own health as being very good increases with increase in wealth index.

The estimation results given in Table 4.8 also indicate that females were less likely to report very good health compared to male counterparts, holding other

163

factors constant. The coefficient of a female rating own health as being very good was about -0.035. The coefficient was statistically significant at one per cent significance level. This suggests that compared to male counterparts, females are less likely to report their own health as being very good by about 0.035. This could be due to biological, social, cultural and economic differences that exist between males and females. Females are less empowered economically and in most societies are looked down upon. Hence, since they do not possess economic power, they have no much influence on economic decisions. This may have a negative influence on their health especially in situation where they have to ask for financial assistance for them to seek medical care. Females also experience pregnancy and childbirth complications, which may sometimes affect their health negatively compared to their male counterparts. Thus, females are more likely to experience and report poor health status than males. This finding is consistent with McGee, Liao, Cao, and Cooper (1999) on US and Bora and Saikia (2015) on

India who found women were more likely to report poorer health status compared to men.

Further, the estimation results presented in Table 4.8 indicate that as one‟s age advanced, he/she was less likely to report very good health status, other factors held constant. The coefficients for age and age squared were -0.002 and -

0.000041, respectively. The coefficients were statistically significant at one per cent level of significance. This suggests that as one became aged, he/she was less likely to report very good health status. This could be because as people become older, they are more likely to suffer from old age diseases such as cardiovascular

164

diseases, diabetes and other chronic diseases, which require prolonged medication periods and higher treatment costs. Also as individuals advance in age, their income generating opportunities diminishes after their productive age, and they become economically challenged. The economic challenge may negatively affect health care utilization and other health promoting activities by the aged, thus leading to poor health status. This finding is consistent with those of Ahmad et al.

(2005) on Pakistan; Ghosh and Husain (2010) on India; and Singh, Arokiasamy,

Singh, and Rai (2013) on India. The authors found that as people became old, they were less likely to report very good health.

Regarding marital status, which was categorized as being never married, married, divorced/separated/widowed, the results showed that married individuals were more likely to rate own health as very good than those never married ceteris paribus. On the other hand the divorced/separated/widowed were less likely to rate own health as being very good compared to the unmarried counterparts, holding other factors constant. The coefficients for the married and divorced/separated/widowed were 0.014 and -0.035, respectively. The coefficients for the variables were also statistically significant at one per cent significance level. The results imply that the probability of the married reporting own health as very good was higher by 0.014 compared to those never married holding other factors constant.

The results further indicated that the probability of reporting own health as very good by the divorced/separated/widowed was lower by 0.035 compared to those never married holding other factors constant. This could be because of benefits of

165

marriage, which according to the resource model shows that the married are more likely to command considerable resources especially if the spouses are engaged in income generating activities (Liu & Umberson, 2008). The married may also be psychosocially fit and hence may not suffer from depression and anxiety due to family support. Thus, they are more likely to report very good health. On the other hand, as explained under the stress model, dissolution of marriage undermines the health of the divorced, the separated and the widowed (Liu &

Umberson, 2008).

Dissolution of marriage may lead to stress and reduced availability of resources.

Those involved in marriage dissolution may also suffer psychologically depending on their status in the society. Thus, the divorced/separated/widowed are less likely to report very good health compared to their unmarried counterparts who may not suffer from problems of marriage dissolution, other factors being constant. The finding of the current study is consistent with earlier studies done in

Kenya by Awiti (2013) and Kodzi et al. (2011) who found that those currently married reported very good health status.

Concerning religion, which was categorized as Catholics, Protestants, Muslims, and Traditionalists/atheists/others, the estimation results indicated that coefficients for Catholics, Protestants, and Muslims were -0.010, -0.018 and

0.022, respectively. All the coefficients were statistically significant at one per cent significance level except for the Catholics, which was statistically insignificant. The results imply that the probability of individuals who are

Muslims reporting own health as very good is higher by 0.022 compared to that of

166

traditionalists/atheist/others holding other factors constant. The results further suggest that the probability of Protestants and Catholics reporting own health as very good is lower be 0.010 and 0.018, respectively compared to traditionalists/atheist/others, other factors held constant.

The finding that Muslims were more likely to rate their own health as very good could be because, Muslims have many opportunities to better their lives. For instance, there are many banks in Kenya such as First Community Bank and

Kenya Commercial Bank targeting Muslims by introducing Sharia compliant products like loans. Thus, Muslims can have access to loan facilities, which they may use to invest in business and accumulate wealth. They may use their earnings to access health care thereafter. Muslims are also becoming educated as evidenced by many institutions of learning targeting them such as Umma University.

Education has enabled them to realize importance of better health and how to achieve it. This finding contradicts Kodzi et al. (2011) on Kenya; Singh et al.

(2013) on India; and Bora and Saikia (2015) on India, who established that

Muslims were less likely to report better health.

The estimation results given in Table 4.8 also indicate that the coefficient of household size was negative and statistically significant at ten per cent significance level. Specifically, if household size increases by one member, the probability of an individual rating own health as very good falls by 0.001. This could be due to increased competition by household members on the limited household resources. Thus, increased household size may imply that some members may not get adequate resources especially those that impact on health

167

positively such as food. This finding is consistent with Gakii (2013) on Kenya who found a negative relationship between household size and probability of reporting very good health.

Regarding education level, which had four categories namely no education, primary education level, secondary education level, and college/university level, estimation results showed that the coefficients for primary, secondary and college/university levels of education were -0.005, 0.027 and 0.057, respectively.

The coefficients for secondary and college/university education levels were statistically significant at one per cent significance level. However, the coefficient for primary education level was statistically insignificant. The results imply that the probability of people with secondary and college/university education levels reporting own health as very good is higher by 0.027 and 0.057, respectively, compared to that of those with no education, holding other factors constant. This may be so because those with secondary and college/university levels of education are more informed about health and its importance. Hence, those with higher levels of education may engage more on preventive and promotive health care as compared to curative health care. The highly educated individuals are also likely to be employed and earning income, which they can in turn use to purchase health care and other health promoting goods and services. This finding agree with those of Bora and Saikia (2015) on India who found that as education increased, the percentage of people reporting good health increased.

Estimation results also showed that the coefficient of employment status was positive and statistically significant at one per cent significance level. The

168

magnitude of the coefficient was 0.032. The results imply that the probability of those employed reporting their own health as being very good was higher by

0.032 compared to their unemployed counterparts holding other factors constant.

Through employment, individuals are able to earn income, which they can in turn use to invest in their health. Thus, employment is an enabling factor for accessing goods and services necessary for health promoting. This finding contradicts those of Gakii (2013) on Kenya who found that those employed were more likely to report poor health.

Regarding area of residence, estimation results showed that the coefficient of being in urban area had a magnitude of -0.014 and it was statistically significant at one per cent significance level. Specifically, the results suggest that the probability of individuals residing in urban areas reporting own health as being very good is lower by 0.014 compared to those individuals residing in rural areas ceteris paribus. This may be because about 60 per cent of Kenya‟s urban households reside in informal settlements where basic amenities are lacking and chances of contracting communicable disease are high (Cira, Kamunyori, &

Babijes, 2016). This finding is inconsistent with those of Teerawichitchainan and

Knodel (2015) on Myanmar who found that urban residence was positively associated with self-assessed health. The finding also contradicts those of Gakii

(2013) on Kenya who found that people residing in urban areas reported better health. Further, the finding is inconsistent with those of Bora and Saikia (2015) on

India who found that urbanites enjoy better health than their rural counterparts.

169

Results on how poverty affects health status has revealed that increase in wealth, implying decrease in poverty, reduced the probability of reporting poor health status, and increased the probability of reporting very good health status ceteris paribus. The results have also revealed that increase in age, being a female, being divorced/separated/widowed, being a protestant, residing in urban area and increase in household size increased the probability of reporting poor health status and decreased that of reporting very good health status, ceteris paribus.

Estimation results further indicated that being married, being a Muslim, having secondary education, having college/university education, being employed, having access to piped water reduced the probability of reporting poor health status and increased that of reporting very good health status other factors held constant.

170

CHAPTER FIVE

SUMMARY, CONCLUSIONS AND POLICY IMPLICATIONS 5.1 Summary

The Kenyan government has overtime recognized that good health is necessary in improvement of productivity, but poverty adversely affects it. In order to mitigate the negative effects of poverty on health, the government has since her political independence in 1963 instituted various policies aimed at reducing poverty and at the same time improving health. Despite the government‟s efforts, poverty rates have remained high and health outcomes have not been impressive. This study explored the effect of poverty on health care utilization, choice of health care providers and health status in Kenya. The specific objectives of the study were: (i) to determine the effect of poverty on health care utilization in Kenya; (ii) to investigate the effect of poverty on choice of health care providers in Kenya; and

(iii) to establish the effect of poverty on health status in Kenya.

The first objective was accomplished by use of control function approach. The estimation results showed that increase in wealth index, which implied reduction in poverty led to increase in health care utilization. Other factors from the estimation results that were found to have a positive influence on health care utilization included being a female, being a protestant and being a Muslim.

Additionally, having primary education level, increase in distance to the nearest health facility and increase in household size, led to increased healthcare utilization. Estimation results also showed that, being married, being

171

divorced/separated/widowed and waiting time in a health facility had a negative effect in influencing health care utilization.

In the second objective, there was no evidence of unobserved heterogeneity, but there was evidence of endogeneity. Thus, the two-stage residual inclusion model was estimated. The results showed that increase in wealth index increased the probability of visiting a private health facility and reduced that of visiting a government, mission or any other health facility, other factors held constant.

Thus, increase in wealth leading to poverty reduction, increased the probability of individuals choosing private health facilities compared to government, mission and other health facilities ceteris paribus. This was expected since as one becomes wealthier, his/her preference for private health facilities increases since he/she can afford the services, which are considered to be of high quality compared to government and other health facilities. Factors that positively influenced individuals to choose private health care facilities included being married, divorced/separated/widowed, residing in urban areas, suffering from malaria and respiratory diseases. However, distance to the nearest health facility, education level, age and being a female reduced the probability of choosing private health facilities other factors held constant.

The results also showed that distance to the nearest health facility, having primary and secondary levels of education, and being a female increased the probability of choosing government health facility other factors held constant. However, suffering from malaria, suffering from diarrhea, being divorced/separated/widowed and residing in urban areas, reduced the probability

172

of individuals visiting government health facilities, ceteris paribus. Further, the results showed that increase in age, being a female, distance to the nearest health facility, suffering from diarrhea and residing in urban areas increased the probability of choosing mission health facilities other factors held constant.

However, being married, being divorced/separated/divorced, suffering from respiratory diseases and suffering from malaria, decreased the probability of choosing mission health facilities other factors being constant.

In the analysis of the data to accomplish objective three, CFA was used given evidence of unobserved heterogeneity and absence of endogeneity. The estimation results showed that increase in wealth index, implying decrease in poverty reduced the probability of reporting own health as being poor and increased that of reporting own health as being very good ceteris paribus. The estimation results further indicated that being a female, being a protestant, being divorced/separated/widowed, increase in age, residing in urban area and increase in household size, increased the probability of reporting poor health status and decreased that of reporting very good health status, other factors held constant.

Further, the results showed that being a Muslim, having secondary education level, having college/university level of education, being married, being employed and having access to piped water, increased the probability of reporting own health as being very good and reduced that of reporting own health as being poor holding other factors constant.

173

5.2 Conclusions

The results of the study led to the conclusion that poverty level is important in influencing health care utilization, choice of health care provider and health status. This notwithstanding, it can be concluded that other individual and household variables also play important role in influencing utilization of health services, choice of health providers as well as perceived health status.

Thus, from the foregoing, it could be concluded that increase in wealth, which implies reduction in poverty, could increase health care utilization. Due to increased wealth, individuals are able to afford health care and hence they are able to utilize the services more. The study findings showed that increase in wealth and subsequent reduction in poverty motivates individuals to seek healthcare from private health facilities, which are considered to provide high quality health services. Further, since increase in wealth increases use of health care and motivates individuals to seek health care from providers who give quality services, leads to improved health status. Therefore, although Kenya missed some health related MDGs, it can do much better in her efforts towards achieving SDGs and the Kenya Vision 2030 if poverty is addressed.

5.3 Policy Implications

The following policy implications can be drawn from the estimation results of this study.

This study found that increase in wealth index led to increased health care utilization. The study also established that as wealth increased, implying reduction in poverty, individuals were more likely to report very good health status. This

174

study, therefore, recommends that both the county and the national governments should pay attention to programs aimed at reducing and or eradicating poverty in

Kenya. The government should support and put more resources in the education sector especially in the Technical and Vocational Education and Training

(TVET), which provides knowledge and skills for employment. This will ensure those individuals who are not academically gifted or who do not proceed with their studies past primary level can still get skills necessary for employment. This will increase individual chances of getting employed and earn a living. The governments should also support entrepreneurship practices by encouraging and giving technical and financial support to those who are starting businesses. This will create more income generating opportunities and hence reduce poverty.

The study further established that wealthier individuals tend to choose private hospitals as compared to government, mission and any other health facility. The wealthier individuals consider private health facilities to be well equipped and to offer high quality health services. The private hospitals are costly and may hinder individuals from accessing and utilizing health care especially the poor. The study also established that there exists a positive effect of distance to the nearest health facility on demand for health care services from government health facilities.

Thus, national and county governments should introduce mobile clinics in every county as a mechanism of reducing distance to the nearest health facility in the short term. In the long term, the government should invest more in health care promotion and provision through allocation of more resources to the public sector.

Improved allocation of resources will help the sector in equipping public health

175

facilities and recruitment of additional qualified health personnel. The health personnel who are well equipped and motivated will help in improving the quality of health care provided in the public health facilities. This will ensure that people do not avoid seeking health care from public health facilities due to quality concerns.

The estimation results showed that a one year increase in age of individuals increased the probability of reporting poor health status. The study also found that the poor were more likely to report poor health status and were also less likely to utilize health care. Thus, the national and county governments should invest in social safety nets for the aged and the poor who cannot be insured by private insurance companies due to advanced age or those who cannot afford to pay the insurance premiums. The government through the National Hospital Insurance

Fund (NHIF) and other players should come up with an insurance cover for the aged and the poor. In the short term, governments at the national and county level should introduce health based voucher programmes targeting the aged and the poor. This will ensure that the elderly and the poor in the society have access to and are able to utilize health services whenever they fall sick. In the long term, the government should endeavor to introduce and maintain universal health care.

This study found that increase in household size increased the probability of reporting poor health status and reduced that of reporting very good health status.

Therefore, there is need for community education by the government and other health stakeholders on the importance of small household sizes and involvement in family planning programmes especially by men. Political and religious leaders

176

should also stop sending conflicting messages to the public that may hinder or undermine uptake of family planning. Advocacy groups should also put more pressure to the government to allocate more resources towards family planning programmes.

Apart from increase in wealth, which was found to have a positive effect on health care utilization and health status, education also had a positive and statistical significant influence on health care utilization and health status. Thus, as the government puts more effort in reducing poverty, it should also ensure that people have access to education by promoting access to quality education. The government has over the years tried to ensure that individuals have access to education by introducing free primary education and subsidized secondary education. However, that is not enough. The government should construct more schools and equip them especially in the regions considered to have been marginalized for long. The government together with other education stakeholders and partners should also improve the teacher-student/pupil ratio, and ensure appropriate training and retraining of teachers.

Employment status was found to positively affect health status. The government should, therefore, enact policies that ensure expansion of the economy so that many people can secure job opportunities. The national government and the county governments responsible for policy formulation and regulation should design programs aimed at formalizing the informal sector, which employs about

83 per cent of all working Kenyans. The policies should include mechanisms for improving productivity and competitiveness, and the quality of working life. This

177

will ensure its sustainability and promote creation of job opportunities. The youth should also not be fixated to only seeking white collar jobs, but they should have attitudinal change and embrace entrepreneurship including agribusiness.

Attitudinal change and embracing of entrepreneurship will assure youths of having source of income, which they can in turn invest in their health. Higher earnings will allow individuals to have balanced meals, good shelter, clothing, and good health care, which are all important in improving health status.

Creation of wealth to reduce poverty without minimizing time individuals spend waiting in health facilities before they could be attended to, may not achieve the desired outcome. This study found that on average, individuals waited for 0.8 hours before they could be attended to in a health facility. The long waiting time negatively affected health care utilization. Thus, various health care providers should adopt technology and introduce queue management system to minimize time spent while waiting to be attended to in a health facility. Minimization of waiting time in a health facility will encourage many people to seek treatment whenever they are sick and hence improve their health status.

Access to piped clean water was also found to have a positive and statistically significant effect on health status. Access to clean water may reduce waterborne diseases and, therefore, improve the health of individuals. Thus, the national and county government together with other development partners should invest in provision of clean water. It is appreciated that the government has been investing in construction of dams and drilling of water. However, the government should do much more. The government should formulate and implement policies that make

178

it punitive to anyone found interfering with water sources. Households should be educated and encouraged to harvest rain water and also drink treated water.

5.4 Contribution to knowledge

This study contributes to the existing literature on role of poverty in influencing health care utilization, choice of healthcare providers and health status. It is the first study to examine the effect of poverty on healthcare utilization, choice of health care providers and health status using national dataset that was collected after devolution of health care in Kenya. This enabled the study to have large observations and it was able to estimate results with precision, which could be generalized. Previous studies used datasets from small regions like slum areas, certain villages or rural areas, hence limiting generalization of the results.

In addition to demonstrating how poverty affects health care utilization, choice of health care providers and health status, the study showed that the effects vary by sex. On sex, the study indicated that females have a higher probability of utilizing health care than their male counterparts. The study also showed that although females utilize more health care, they are more likely to report poor health status.

Further, the study showed that compared to their male counterparts, females were less likely to visit private health facilities, which are assumed to offer quality health care. The scenario points to inherent socio-economic and cultural obstacles, which may be affecting the health of females. Therefore, this study reveals that the design of policies and programmes should be aligned to the need of different

179

segments of the populace so as to maximize their effect towards use of health services, choice of health care providers and health status.

The study also used general health as opposed to a specific health issue. This enabled the study to come up with policy implications that can cut across in all areas of the health. Previous studies were concerned with particular health issues such as maternal health, child health, and malaria among others.

5.5 Areas for further research

This study has presented an analysis of how poverty affects health care utilization, choice of health care providers and health status in Kenya. To broaden the understanding of these relationships, future research may focus on the following areas:

i) An empirical investigation on the effect of poverty on health care since the

devolution of health care was introduced.

ii) An investigation on how poverty has affected health status overtime taking

in to account regional differences.

180

REFERENCES

Abuya, B. A., Ciera, J., & Kimani-Murage, E. (2012). Effect of mother‟s education on child‟s nutritional status in the slums of Nairobi. BMC pediatrics, 12(1), 80. Acton, J. P. (1975). Nonmonetary factors in the demand for medical services: some empirical evidence. Journal of Political Economy, 83(3), 595-614. Adeoti, A. I., & Awoniyi, O. A. (2014). Demand for Health Care Services and Child Health Status in Nigeria- A Control Function Approach. African Research Review, 8(1), 273. doi: 10.4314/afrrev.v8i1.18 Ahmad, K., Jafar, T. H., & Chaturvedi, N. (2005). Self-rated health in Pakistan: results of a national health survey. BMC Public Health, 5(1), 51. doi: 10.1186/1471-2458-5-51 Ajakaiye, O., & Mwabu, G. (2007). The demand for reproductive health services: Frameworks of analysis. AERC, Nairobi. Akunga, D., Menya, D., & Kabue, M. (2014). Determinants of Postnatal Care Use in Kenya. African Population Studies, 28(3). Ali, K. J., & Noman, A. N. K. (2013). Determinants of demand for health care in Bangladesh: An Econometric Analysis. World Journal of Social Sciences, 3(6), 153-163. Amaghionyeodiwe, L. A. (2008). Determinants of the choice of health care provider in Nigeria. Health Care Manage Sci, 11, 215-227. doi: 10.1007/s10729-007-9038-3 Andersen, R. (1995). Revisiting the behavioral model and access to medical care: does it matter? Journal of health and social behavior, 1-10. Andersen, R., & Newman, J. F. (2005). Societal and individual determinants of medical care utilization in the United States. Milbank Quarterly, 83(4). Asenso-Okyere, W. K., Dzator, J. A., & Osei-Akoto, I. (1996). The behaviour towards malaria care - A Multinomial Logit Approach. Social Indicators Research, 39(2), 167-186. Asfaw, A. (2003). How poverty affects the health status and the health care demand of households? The case of rural Ethiopia. Paper presented at the Staying Poor: Chronic Poverty and Development Policy, Institute for Development Policy and Management, University of Manchester, U.K. http://r4d.dfid.gov.uk/PDF/Outputs/ChronicPoverty_RC/Asfaw.pdf Awiti, J. O. (2013). Essays on Health Determinants in Kenya. Unpublished PhD Thesis. School of Economics, University of Nairobi. Nairobi, Kenya. Awiti, J. O. (2014). Poverty and health care demand in Kenya. BMC Health Services Research, 14(560). doi: 10.1186/s12913-014-0560-y Awoyemi, T., Obayelu, O., & Opaluwa, H. (2011). Effect of distance on utilization of health care services in rural Kogi State, Nigeria. Journal of human Ecology, 35(1), 1-9. Baum, C. (2010). Models for Count Data and Categorical Response Data. Boston College and DIW, Berlin.

181

Ben-Akiva, M. E., & Bierlaire, M. (1999). Discrete choice methods and their applications to short-term travel methods. Handbook of transportaion science, 23, 5-34. Ben-Akiva, M. E., & Lerman, S. R. (1985). Discrete choice analysis: Theory and application to travel demand (Vol. 9): MIT Press. Bora, J. K., & Saikia, N. (2015). Gender differentials in self-rated health and self- reported disability among adults in India. PloS one, 10(11), e0141953. Borah, B. J. (2006). A mixed logit model of health care provider choice: analysis of NSS data for rural India. Health Econ, 15(9), 915-932. doi: 10.1002/hec.1166 Bourne, P. A. (2009). Impact of poverty, not seeking medical care, unemployment, inflation, self-reported illness, health insurance on mortality in Jamaica. North Am J Med Sci, 1(3), 99-109. Buddelmeyer, H., & Cai, L. (2009) Interrelated Dynamics of Health and Poverty in Australia. IZA Discussion Paper Vol. 4602. Bonn, Germany: Institute for the Study of Labor (IZA). Cameron, A. C., & Trivedi, P. K. (2005). Microeconomics: Methods and Applications. United States of America: Cambridge University Press. Chin, B. (2010). Income, health, and well-being in rural Malawi. Demographic research, 23(35), 997. Chuma, J., Musimbi, J., Okungu, V., Goodman, C., & Molyneux, C. (2009). Reducing user fees for primary health care in Kenya: Policy on paper or policy in practice? International Journal for Equity in Health, 8(1), 1. Cira, D. A., Kamunyori, S. W., & Babijes, R. M. (2016). Kenya urbanization review. Washington, DC: World Bank. Cisse, A. (2011). Analysis of health care utilization in Côte d'Ivoire. Nairobi, Kenya: The African Economic Research Consortium. Corsi, D. J., & Subramanian, S. V. (2012). Association between socioeconomic status and self-reported diabetes in India: a cross-sectional multilevel analysis. BMJ open, 2(4), e000895. Daniels, B., Dolinger, A., Bedoya, G., Rogo, K., Goicoechea, A., Coarasa, J., . . . Das, J. (2017). Use of standardised patients to assess quality of healthcare in Nairobi, Kenya: a pilot, cross-sectional study with international comparisons. BMJ Global Health, 2(2), e000333. de Bekker-Grob, E. W. (2009). Discrete choice experiments in health care. BMJ (Clinical research ed.). Erasmus Universiteit Rotterdam. Dias, S., Gama, A., Cortes, M., & de Sousa, B. (2011). Healthcare‐seeking patterns among immigrants in Portugal. Health & social care in the community, 19(5), 514-521. Drabo, A. (2011). Impact of income inequality on health: does environment quality matter? Environment and Planning A, 43(1), 146-165. Edeme, R. K., Ifelunini, I. A., & Obinna, O. S. (2014). Relationship between household income and child mortality in Nigeria. American Journal of Life Sciences, 2(6-4), 1-12. doi: 10.11648/j.ajls.s.2014020604.11 Ellis, R. P., & Mwabu, G. M. (1991). The demand for outpatient medical care in rural Kenya: Institute for Economic Development, Boston University.

182

Fabbri, D., & Monfardini, C. (2003). Public vs. private health care services demand in Italy. Giornale degli economisti e annali di economia, 93-123. Faure, J.-C. (2001). Development Co-Operation Report 2000: Efforts and Policies of the Member of the Development Assistance Committee (Vol. 2): OCDE. Fink, G., Günther, I., & Hill, K. (2014). Slum Residence and Child Health in Developing Countries. Demography. doi: 10.1007/s13524-014-0302-0 Fuchs, V. R. (2004). Reflections on the socio-economic correlates of health. Journal of Health Economics, 23(4), 653-661. doi: http://dx.doi.org/10.1016/j.jhealeco.2004.04.004 Gakii, J. (2013) Demand for Health Care in Kenya: The Effect of Health Insurance. KIPPRA Discussion Paper: Vol. 155. Nairobi, Kenya: KIPPRA. Gakuru, R., & Mathenge, N. (2012). Poverty, Growth, and Income distribution in Kenya: A SAM Perspective. AGRODEP Working Paper, 0001. Galama, T., Hullegie, P., Meijer, E., & Outcault, S. (2012). Empirical evidence for decreasing returns to scale in a health capital model: HEDG, Department of Economics, University of York. Gelberg, L., Andersen, R. M., & Leake, B. D. (2000). The Behavioral Model for Vulnerable Populations: Application to Medical Care Use and Outcomes for Homeless People. Health Services Research, 34(6). Ghosh, S., & Husain, Z. (2010). Economic independence, family support and perceived health status of elderly: recent evidences from India. Asia Pacific Population Journal, 25(1), 47-77. Gillam, S. (2008). Is the declaration of Alma Ata still relevant to primary health care? Bmj, 336(7643), 536-538. Greene, W. H. (2002). Econometric Analysis (5th ed.). New Jersey, Prentice Hall: Pearson Education, Inc. Greene, W. H., & Hensher, D. A. (2010). Modeling ordered choices: A primer: Cambridge University Press. Grossman, M. (1972). The demand for health; a Theoretical and Empirical Investigation. New York: National Bureau of Economic Research. Grossman, M. (1982). The demand for health care after a decade. Journal of Health Economics, 1, 1-3. Grossman, M. (2000). The human capital model. Handbook of health economics, 1, 347-408. Gupta, R. P. S., Wit, M. L. D., & McKeown, D. (2007). The impact of poverty on the current and future health status of children. Paediatr Child Health, 12(8), 667-672. Habtom, G. K., & Ruys, P. H. M. (2005). The choice of a health care Provider: The Central and Northern highlands of Eritrea. Tilburg University (http://www. eeaesem. com/papers/EEA/2005/709/article. pdf). Halasa, Y., & Nandakumar, A. K. (2009). Factors determining choice of health care provider in Jordan. Eastern Mediterranean Health Journal, 15(4), 959-968.

183

Hallman, K. (1999). Child health care demand in a developing country: Unconditional estimates from the Philippines. IFPRI Food Consumption and Nutrition Division. Discussion Paper NO. 70. Hamad, R., & Rehkopf, D. H. (2015). Poverty, Pregnancy, and Birth Outcomes: A Study of the Earned Income Tax Credit. Paediatr Perinat Epidemiol, 29(5), 444-452. doi: 10.1111/ppe.12211 Horta, B. L., Gigante, D. P., Candiota, J. S., Barros, F. C., & Victora, C. G. (2008). Monitoring mortality in Pelotas birth cohort from 1982 to 2006, Southern Brazil. Rev Saude Publica, 42(Suppl 2), 108-114. Hosseinpoor, A. R., Williams, J. S., Amin, A., De Carvalho, I. A., Beard, J., Boerma, T., . . . Chatterji, S. (2012). Social determinants of self-reported health in women and men: understanding the role of gender in population health. PloS one, 7(4), e34799. Hren, R. (2012). Theoretical shortcomings of the Grossman model. Bulletin: Economics, Organisation and Informatics in Healthcare, 28(1), 63-75. Hu, Y.-N., Chen, P.-C., Hsu, C.-C., Yu, H.-K., Chien, K.-L., Li, C.-C., & Hu, G.- C. (2016). Age and Gender Differences in the Relationship Between Self- rated Health and Mortality Among Middle-aged and Elderly People in Taiwan—Results of a National Cohort Study. International Journal of Gerontology, 10(2), 91-95. Jones, A. M. (2007). Applied Econometrics for Health Economists A Practical Guide (2 ed.). Oxford, Seattle: Radcliffe Publishing. Kabubo-Mariara , J., Karienyeh, M. M., & Kabubo, F. M. (2012). Child Survival, Poverty and Inequality in Kenya: Does Physical Environment Matter? African Journal of Social Sciences, 2(1), 65-84. Kabubo-Mariara, J., Mwabu, D., & Ndeng‟e, G. (2009). The Consequences of Fertility for Child Health in Kenya: Endogeneity, Heterogeneity and Application of the Control Function approach. Paper presented at the Center for the study of African Economies (CSAE) Conference 2009 on Economic Development in Africa, St Catherine's College, Oxford. http://www.csae.ox.ac.uk/conferences/2009-edia/papers/280-mariara.pdf Kagiri, D., & Wainaina, G. (2017). Time and Cost Overruns in Power Projects in Kenya: ACase Study of Kenya Electricity Generating Company Limited. ORSEA JOURNAL, 3(2). Kamau, N., & Esamai, F. (2001). Determinants of immunisation coverage among children in Mathare Valley, Nairobi. East African Medical Journal, 78(11), 590-594. Kimalu, P., Nafula, N., Manda, D. K., Mwabu, G., & Kimenyi, M. S. (2002) A Situational Analysis of Poverty in Kenya. KIPPRA Working Paper: Vol. 6. Nairobi, Kenya: KIPPRA. Kimani, D. N. (2014). Out-of-Pocket Health Expenditures and Household Poverty: Evidence from Kenya Unpublished PhD Thesis. School of Economic, University of Nairobi. Nairobi, Kenya. Kimani, D. N., Mugo, M. G., & Kioko, U. M. (2016). An Econometric Analysis Of Health Care Utilization In Kenya. European Scientific Journal, 12(16).

184

KIPPRA. (2013) Kenya Economic Report 2013. Nairobi, Kenya: Kenya Institute for Public Policy Research and Analysis. KIPPRA. (2014) Kenya Economic Report 2014. Nairobi, Kenya: Kenya Institute for Public Policy Research and Analysis. KIPPRA. (2015) Kenya Economic Report 2015. Nairobi, Kenya: Kenya Institute for Public Policy Research and Analysis. KIPPRA. (2018). Realizing the "Big Four" Agenda through Energy as an Enabler. Policy Monitor. Nairobi, Kenya: Kenya Institute for Public Policy Research and Analysis. Klaauw, B. v. d., & Wang, L. (2011). Child mortality in rural India. Journal of Population Economics, 24(2), 601-628. Kodzi, I. A., Gyimah, S. O., Emina, J., & Ezeh, A. C. (2011). Religious involvement, social engagement, and subjective health status of older residents of informal neighborhoods of Nairobi. Journal of Urban Health, 88(2), 370-380. Kosimbei, G. (2005) Child Healthcare Seeking Behaviour in Kenya. KIPPRA Discussion Paper: Vol. 50. Nairobi, Kenya: KIPPRA. Kyegombe, N. (2003). Health and chronic poverty, Background Paper to the Chronic Poverty Report 2004-5. Available at www.chronicpoverty.org Lawson, D. (2004). Health, poverty and poverty dynamics in Africa. Paper presented at the IV Mediterranean Seminar on International Development. http://www.uib.cat/depart/deaweb/smed/pdf/lawson.pdf Lichtenberg, F. R. (2002). The effects of Medicare on health care utilization and outcomes. Frontiers in health policy research, 5(3). Liljas, B. (1998). The demand for health with uncertainty and insurance. Journal of Health Economics, 17(2), 153-170. Liu, H., & Umberson, D. J. (2008). The times they are a changin': Marital status and health differentials from 1972 to 2003. Journal of health and social behavior, 49(3), 239-253. Lynch, J. W., Smith, G. D., Kaplan, G. A., & House, J. S. (2000). Income inequality and mortality: importance to health of individual income, psychosocial environment, or material conditions. British Medical Journal, 320(7243), 1200. Manda, K. D., Kimenyi, M. S., & Mwabu, G. (2001) A Review of Poverty and Antipoverty Initiatives in Kenya. KIPPRA Working Paper: Vol. 3. Nairobi, Kenya: KIPPRA. Mburu, R. W., Folayan, M. O., & Akanni, O. (2014). The Abuja+ 12 declaration: implications for HIV response in Africa. African journal of reproductive health, 18(3), 34-46. McGee, D. L., Liao, Y., Cao, G., & Cooper, R. S. (1999). Self-reported health status and mortality in a multiethnic US cohort. American journal of epidemiology, 149(1), 41-46. Misati, J., & Mngoda, D. (2012). Re-Visioning Kenya’s Social Development Agenda for Industrialization and Sustainable Development. Paper presented at the Scientific Conference Proceedings.

185

Mugwila, L. O. (2002). Demand for health care in Kenya: The case of Vihiga district. Unpublished M.A Thesis. School of Economics, University of Nairobi. Nairobi, Kenya. Mureithi, P. L. (1988). Development Planning in Kenya. In M. Urrutia & S. Yukawa (Eds.), Development Planning in Mixed Economies (pp. 198- 219). Tokyo, Japan: The United Nations University. Muriithi, M. K. (2013). The determinants of health seeking behaviour in a Nairobi slum, Kenya. European Scientific Journal, 9(8), 151-164. Mutua, M. K., Kimani-Murage, E., & Ettarh, R. R. (2011). Childhood vaccination in informal urban settlements in Nairobi, Kenya: Who gets vaccinated? BMC Public Health, 11(6). Mutunga, C. J. (2011). Environmental determinants of child mortality in Kenya Health Inequality and Development (pp. 89-110): Springer. Mwabu, G. (1995). Health care reform in Kenya: a review of the process. Health Policy, 32(1), 245-255. Mwabu, G. (2007). Health Economics for Low-Income Countries. Center Discussion Paper, 955. http://www.econ.yale.edu/growth_pdf/cdp955.pdf Mwabu, G. (2008). The Production of Child Health in Kenya: A Structural Model of Birth Weight. Journal of African Economies, 18(2), 212-260. Mwabu, G., Ainsworth, M., & Nyamete, A. (1993). Quality of Medical Care and Choice of Medical Treatment in Kenya: An Empirical Analysis. The Journal of Human Resources, 28(4), 838-862. Mwabu, G., Wang‟ombe, J., & Nganda, B. (2003). The demand for medical care in Kenya. African Development Review, 15(2‐3), 439-453. Nafula, N. N., Onsomu, E. N., Mwabu, G., & Muiruri, S. (2005) Review of Policy Options for Poverty Reduction in Kenya. KIPPRA Discussion Paper: Vol. 49. Nairobi, Kenya: KIPPRA. Namubiru, L. (2014). Effect of Household Poverty on Women's Fertility and Child Nutritional Status in Uganda. Unpublished PhD Thesis. School of Economics, University of Nairobi. NCAPD. (2008) An Output-Based Approach to Reproductive Health: Vouchers for Health in Kenya. NCAPD Policy Brief No. 2. Nkpoyen, F., Eteng, B. G., & Abul, U. F. (2014). Health Capital and Poverty Reduction in Rural Cross River State, Nigeria. International Journal of Education and Research, 2(5). Nyambura, W. A. (2016). Determinants of Demand For Healthcare Services in Private Hospitals in Kenya. Unpublished Msc Thesis. School of Economics, University of Nairobi. Nairobi, Kenya. OAU. (2001). Abuja Declaration on HIV/AIDS, tuberculosis and other Related Infectious Diseases. Addis Ababa, Ethiopia: Organization of African Unity. Ochako, R., Fotso, J.-C., Ikamari, L., & Khasakhala, A. (2011). Utilization of maternal health services among young women in Kenya: Insights from the Kenya Demographic and Health Survey, 2003. BMC Pregnancy and Childbirth, 11(1).

186

Odima, A. O. (2014). Exploring Poverty Health Linkages: A case of Child Mortality in Kenya. Unpublished M.A Research project. School of Economics, University of Nairobi. Oyaya , C. O., & Rifkin, S. B. (2003). Health sector reforms in Kenya: an examination of district level planning. Health Policy, 64, 113-127. Peters, D. H., Garg, A., Bloom, G., Walker, D. G., Brieger, W. R., & Rahman, H. M. (2008). Poverty and Access to Health Care in Developing Countries. Annals of the New York Academy of Sciences, 1136, 161-171. Prasad, S. (2013). Preference of hospital usage in India. Annals of Tropical Medicine and Public Health, 6(4), 472. Rebhan, D. P. (2011). Health Care Utilization: Understanding and applying theories and models of health care seeking behavior. Case Western Reserve University, 1-19. Republic of Kenya. (1965). African Socialism and its Application to Planning in Kenya. Nairobi, Kenya: Government Printer. Republic of Kenya. (1994). Kenya's Health Policy Framework, 1994-2010. Nairobi, Kenya: Ministry of Health. Republic of Kenya. (1999). National Poverty Eradication Plan 1999-2015. Nairobi, Kenya: Office of the President, Department of Development Co- ordination. Republic of Kenya. (2000). Interim Poverty Reduction Strategy Paper 2000 - 2003. Nairobi, Kenya: Ministry of Finance and Planning. Republic of Kenya. (2001). Poverty Reduction Strategy Paper for the Period 2001-2004. Nairobi, Kenya: Ministry of Finance and Planning. Republic of Kenya. (2004). Economic Recovery Strategy for Wealth and Employment Creation (2003 - 2007). Nairobi, Kenya: Government Printer. Republic of Kenya. (2005a). National Health Sector Strategic Plan of Kenya (NHSSP II), 2005-2010. Nairobi, Kenya: Ministry of Health. Republic of Kenya. (2005b). Reversing the Trends: The Second National Health Sector Strategic Plan of Kenya-NHSSP II (2005–2010). Nairobi, Kenya: Ministry of Health. Republic of Kenya. (2007). Vision 2030, A Globally Competitive and Prosperous Kenya. Nairobi: Government Printer. Republic of Kenya. (2008). Kenya Vision 2030: First Medium Term Plan, 2008- 2012. Nairobi, Kenya: Ministry of State for Planning, National Development and Vision 2030. Republic of Kenya. (2010a). 2009 Kenya Population and Housing Census. Nairobi, Kenya: Ministry of State for Planning, National Development and Vision 2030 and Kenya National Bureau of Statistics. Republic of Kenya. (2010b). Constitution of Kenya. Nairobi, Kenya: Government Printer. Republic of Kenya. (2012). Health Sector Working Group Report. Medium Term Expenditure Framework (MTEF) for the period 2013/14-2015/16. Nairobi, Kenya: Ministry of Health.

187

Republic of Kenya. (2013a). Kenya Vision 2030: Second Medium Term Plan, 2013-2017. Nairobi, Kenya: Ministry of Devolution and Planning. Republic of Kenya. (2013b). Millenium Development Goals Status Report for Kenya 2013. Nairobi, Kenya: Ministry of Devolution and Planning. Republic of Kenya. (2014a). Health Sector Human Resources Strategy 2014- 2018. Nairobi, Kenya: Ministry of Health. Republic of Kenya. (2014b). Kenya Demographic and Health Survey 2014. Nairobi, Kenya: Kenya National Bureau of Statistics (KNBS). Republic of Kenya. (2014c). Kenya Health Policy 2012-2030. Nairobi, Kenya: Ministry of Health. Republic of Kenya. (2014d). Kenya Household Health Expenditure and Utilisation Survey 2013. Nairobi, Kenya: Ministry of Health. Republic of Kenya. (2015a). Economic Survey 2015. Nairobi, Kenya: Kenya National Bureau of Statistics (KNBS). Republic of Kenya. (2015b). Kenya National Health Acounts 2012/13. Nairobi, Kenya: Ministry of Health. Republic of Kenya. (2016). Economic Survey 2016. Nairobi, Kenya: Kenya National Bureau of Statistics (KNBS). Republic of Kenya. (2018a). Basic Report on Well-Being in Kenya: Based on the 2015/16 Kenya Integrated Household Budget Survey (KIHBS). Nairobi, Kenya: Kenya National Bureau of Statistics (KNBS). Republic of Kenya. (2018b). Budget Statement Financial Year 2018/2019. Nairobi, Kenya: The National Treasury and Planning. Republic of Kenya. (2018c). Economic Survey 2018. Nairobi, Kenya: Kenya National Bureau of Statistics (KNBS). Republic of Kenya. (2018d). Kenya Vision 2030: Third Medium Term Plan, 2018- 2022. Nairobi, Kenya: Ministry of Devolution and Planning. Riegle, A. L., & Stewart, S. D. (2013). Health Care Access and Utilization among the Poor. Poverty and Health: A Crisis Among America's Most Vulnerable: A Crisis among America's Most Vulnerable, 183. Salihi, A. H., Madeline, B., & Norlaila, A. B. (2012). On the causal links between poverty and HIV in Nigeria, 1990-2009: An application of Granger causality and co-integration techniques. Malaysia Journal of Society and Space, 8(1), 50-59. Sen, G., & Östlin, P. (2008). Gender inequity in health: why it exists and how we can change it. Global Public Health, 3(sup1), 1-12. doi: 10.1080/17441690801900795 Singh, L., Arokiasamy, P., Singh, P. K., & Rai, R. K. (2013). Determinants of gender differences in self-rated health among older population: evidence from India. Sage Open, 3(2), 2158244013487914. Skordis-Worrall, J., Hanson, K., & Mills, A. (2011). Estimating the demand for health services in four poor districts of Cape Town, South Africa. International Health, 3, 44-49. Stephenson, R., Baschieri, A., Clements, S., Hennink, M., & Madise, N. (2006). Contextual influences on the use of health facilities for childbirth in Africa. American journal of public health, 96(1), 84-93.

188

Strauss, J., & Thomas, D. (1998). Health, Nutrition, and Economic Development. Journal of Economic Literature, 36(2), 766-817. Strauss, J., & Thomas, D. (2007) Health over the life course. California Center for Population Research On-Line Working Paper Series: Vol. CCPR-011- 07. Los Angeles: University of California. Teerawichitchainan, B., & Knodel, J. (2015). Economic status and old-age health in poverty-stricken Myanmar. Journal of aging and health, 27(8), 1462- 1484. doi: 10.1177/0898264315584577 Terza, J. V., Basu, A., & Rathouz, P. J. (2008). Two-stage residual inclusion estimation: addressing endogeneity in health econometric modeling. Journal of Health Economics, 27(3), 531-543. UNAIDS. (2013). Abuja +12 Shaping the future of health in Africa. (JC2524/1/E). UNAIDS Retrieved from http://www.unaids.org/sites/default/files/media_asset/JC2524_Abuja_repo rt_en_0.pdf. United Nations. (2012). The Future we want. (A/RES/66/288). United Nations Headquarters in New York. United Nations. (2015a). The Millenium Development Goals Report 2015. New York. United Nations. (2015b). Transforming our world: the 2030 Agenda for Sustainable Development. (A/RES/70/1). United Nations Headquarters in New York. Verbeek, M. (2012). A Guide to Modern Econometrics (4 ed.). USA: John Wiley & Sons, Ltd. Wagstaff, A., Claeson, M., Hecht, R. M., Gottret, P., & Fang, Q. (2006). Millennium Development Goals for Health: What Will It Take to Accelerate Progress? In D. T. Jamison, J. G. Breman, A. R. Measham, G. Alleyne, M. Claeson, D. B. Evans, P. Jha, A. Mills & P. Musgrove (Eds.), Disease Control Priorities in Developing Countries (2nd ed.). Washington (DC): World Bank. Wamai, R. G. (2008). Reforming health systems: the role of NGOs in decentralization: lessons from Kenya and Ethiopia: International society for third-sector research (ISTR). Wamai, R. G. (2009). Healthcare policy administration and reforms in post- colonial Kenya and challenges for the future. Local and Global Encounters: Norms, Identities and Representations in Formation. Helsinki: The Renvall Institute for Area and Cultural Studies. Wang, S., Chen, L., Hsu, S. H., & Wang, S. (2012). Health care utilization and health outcomes: a population study of Taiwan. Health Policy and Planning, 27, 590-599. WHO. (1978). Declaration of Alma-Ata International Conference on Primary Health Care, Alma-Ata, USSR, 6-12 September 1978. Geveva: WHO Retrieved from http://www.who.int/publications/almaata_declaration_en.pdf.

189

WHO. (2015a). Health in 2015: From MDGS, Millenium Development Goals to SDGs, Sustainable Development Goals. Geneva, Switzerland.: World Health Organization. WHO. (2015b). World Health Statistics 2015. Geneva: World Health Organization WHO. (2016). Public Financing for Health in Africa: from Abuja to the SDGs. Geneva, Switzerland: World Health Organization. World Bank. (1982). Growth and Structural Change in Kenya: A Basic Economic Report. (3350-K). Washington, DC: World Bank Retrieved from http://documents.worldbank.org/curated/en/844971468047746716/Main- report World Bank. (2016). World Development Indicators 2016. Washington, DC: World Bank. Yawson, A., Malm, K., Adu, A., Wontumi, G., & Biritwum, R. (2012). Patterns of health service utilization at a medical school clinic in Ghana. Ghana medical journal, 46(3), 128. Zimmerman, E., & Woolf, S. H. (2014). Understanding the relationship between education and health: Institute of Medicine of the National Academies. Zyaambo, C., Siziya, S., & Fylkesnes, K. (2012). Health status and socio- economic factors associated with health facility utilization in rural and urban areas in Zambia. BMC Health Services Research, 12(1), 389.

190

Appendix

Table A1: Validity test of instrumental variable in health care utilization model Variable Poverty status model Health care utilization model Wealth index 0.037(0.017)** Age 0.004(0.000)*** 0.0004(0.002) Age Squared -0.00002(7.91e- 7.63e-06(1.86e-05) 06)*** Sex: Male(Reference) Female 0.073(0.006)*** 0.200(0.013)*** Religion: Traditionalist/Atheist/Others(Reference) Catholic 0.138(0.011)*** 0.003(0.035) Protestant 0.172(0.011)*** 0.070(0.034)** Muslim 0.202(0.013)*** 0.075(0.038)* Marital Status: Not married(Reference) Married 0.023(0.087)** -0.177(0.019)*** Divorced/separated/Widowed -0.085(0.097)*** -0.326(0.027)*** Log of household size -0.067(0.004)*** 0.115(0.010)*** Education Level: No education(Reference) Primary Education 0.252(0.006)*** 0.052(0.019)*** Secondary Education 0.491(0.008)*** 0.039(0.022)* College/university education 0.845(0.010)*** 0.004(0.034) Employment Status: No (Reference) Yes 0.068(0.007)*** 0.001(0.015) Area of residence: Rural (Reference) Urban 0.348(0.005)*** -0.007(0.015) Insured: Not insured (Reference) Insured -0.009(0.019) Log of waiting time -0.018(0.005)*** Distance to nearest health facility: <1 KM (Reference) 1-3KM 0.113(0.017)*** 4-5KM 0.095(0.021)*** 6-9KM 0.177(0.024)*** 10+ KM 0.120(0.022)*** County average access to electricity: No (Reference) Yes 0.633(0.022)*** 0.001(0.042) County average access to piped water: No (Reference) Yes 0.174(0.018)*** Constant -0.955(0.022)*** 0.143(0.057)** Number of observations 28,968 16,619 R-Squared/Pseudo R2 R-Squared=0.5373 Pseudo R2=0.0138 F(16, 28951) 2413.96*** Wald χ2 (22) 859.85(0.000)a*** Linktest: hat 0.9987(0.000)a*** 0.9404(0.000)a*** hat squared 0.0054(0.620)a 0.0568(0.729)a

191

Mean VIF 6.68 5.45 Note: ***, ** and * indicates statistical significance at 1 per cent, 5 per cent and 10 per cent levels of significance, respectively. (.)=Robust Standard Errors, (.)a=P-value Source: Author computation, Study Data, 2013.

Table A2: Validity test of instrumental variable in health care provider choice model Variable Poverty status Healthcare provider choice model model Government Private Mission Wealth -0.115(0.093) 0.504(0.094)* 0.230(0.098)* index ** * Age 0.004(0.000)** - - 0.003(0.002) * 0.004(0.002)* 0.006(0.02)** Age -0.00002(7.91e- Squared 06)*** Sex: Male(Reference) Female 0.073(0.006)** 0.217(0.073)* -0.008(0.073) 0.293(0.078)* * ** ** Malaria/Fever: No(Reference) Yes 0.158(0.076)* 0.348(0.077)* 0.014(0.081) * ** Respiratory Disease/Pneumonia: No(Reference) Yes 0.357(0.118) 0.444(0.119)* 0.226(0.125)* ** Diarrhea: No(Reference) Yes 0.087(0.280) 0.258(0.281) 0.439(0.289) Religion: Traditionalist/Atheist/Others(Reference) Catholic 0.138(0.011)** * Protestant 0.172(0.011)** * Muslim 0.202(0.013)** * Marital Status: Not married(Reference) Married 0.023(0.087)** 0.032(0.097) 0.131(0.097) -0.049(0.102) - -0.062(0.133) 0.175(0.134) -0.183(0.141) Divorced/separated/ 0.085(0.097)** Widowed * Log of household - 0.001(0.059) -0.079(0.060) -0.019(0.063) size 0.067(0.004)** * Education Level: No education(Reference) Primary 0.252(0.006) 0.239(0.097)* 0.021(0.099) 0.070(0.104) Education *** * Secondary 0.491(0.008) 0.361(0.125)* 0.170(0.127) 0.224(0.132)* Education *** ** 192

College/unive 0.845(0.010) 0.114(0.178) 0.161(0.180) 0.196(0.187) rsity education *** Employment Status: No (Reference) Yes 0.068(0.007) *** Area of residence: Rural (Reference) 0.348(0.005) - 0.160(0.078)* -0.032(0.083) Urban *** 0.139(0.078)* * Distance to nearest -0.001(0.001) -0.001(0.001) 0.001(0.001) health facility County average access to electricity: No (Reference) Yes 0.633(0.022) *** County average access to piped water: No (Reference) Yes 0.174(0.018) -0.114(0.171) -0.033(0.173) -0.086(0.182) *** - 2.238(0.218)* 1.919(0.220)* 0.680(0.229)* 0.955(0.022) ** ** ** Constant *** Number of 28968 17860 17860 17860 observations R-Squared/Pseudo R2 R- Squared=0.53 73 F(16, 28951) 2413.96*** Wald χ2 (45) 1273.69*** 0.999(0.000)a 1.109(0.000)a 0.147(0.000)a 1.878(0.000)a Linktest: hat *** *** *** *** 0.005(0.620)a - 0.006(0.986)a -.344(0.372)a hat squared 0.027(0.940)a Mean VIF 6.68 1.28 Note: ***, ** and * indicates statistical significance at 1 per cent, 5 per cent and 10 per cent levels of significance, respectively. (.)=Robust Standard Errors, (.)a=P-value Source: Author computation, Study Data, 2013.

Table A3: Validity test of instrumental variable in health status model Variable Poverty status model Health Status model Wealth index 0.1221(0.010)*** Age 0.004(0.000)*** -0.006(0.001)*** Age Squared -0.00002(7.91e-06)*** -0.0001(1.52e-05)*** Sex: Male(Reference) Female 0.073(0.006)*** -0.113(0.008)*** Religion: Traditionalist/Atheist/Others(Reference) Catholic 0.138(0.011)*** -0.033(0.023) Protestant 0.172(0.011)*** -0.062(0.022)*** Muslim 0.202(0.013)*** 0.072(0.026)*** Marital Status: Not married(Reference)

193

Married 0.023(0.087)** 0.049(0.012)*** Divorced/separated/Widowed -0.085(0.097)*** -0.116(0.019)*** Log of household size -0.067(0.004)*** -0.018(0.008)** Education Level: No education(Reference) Primary Education 0.252(0.006)*** -0.022(0.013)* Secondary Education 0.491(0.008)*** 0.082(0.015)*** College/university education 0.845(0.010)*** 0.018(0.020)*** Employment Status: No (Reference) Yes 0.068(0.007)*** 0.104(0.010)*** Area of residence: Rural (Reference) Urban 0.348(0.005)*** -0.040(0.009)*** County average access to electricity: No (Reference) Yes 0.633(0.022)*** -0.027(0.039) County average access to piped water: No (Reference) Yes 0.174(0.018)*** 0.267(0.031)*** Constant -0.955(0.022)*** Number of observations 28968 80742 R-Squared/Pseudo R2 R-Squared=0.5373 Pseudo R2=0.0427 F(16, 28951) 2413.96*** Wald χ2 (17) 6810.75(0.000)*** Linktest: hat 0.9987(0.000)a*** 0.9953(0.000)a*** hat squared 0.0054(0.620)a -0.0038(0.854)a Mean VIF 6.68 6.14 Note: ***, ** and * indicates statistical significance at 1 per cent, 5 per cent and 10 per cent levels of significance, respectively. (.)=Robust Standard Errors; (.)a=P-value Source: Author computation, Study Data, 2013.

Table A4: Average Marginal Effects for the Healthcare Provider Choice Model (Government) Variable Dependent variable=Health care provider (Government) Baseline Model 2SRI Model CFA Model -0.139(- -0.167(- -0.167(- Wealth Index 16.34)*** 10.09)*** 10.07)*** Age -0.0004(-1.71)* -0.0004(-1.58) -0.0004(-1.58) Sex: Male(Reference) Female 0.043(5.40)*** 0.044(5.60)*** 0.044(5.59)*** Malaria/Fever: No (Reference) Yes -0.019(-2.38)*** -0.020(-2.52)** -0.020(-2.52)** Respiratory Disease/Pneumonia: No (Reference) Yes -0.002(-0.19) -0.002(-0.17) -0.002(-0.16) Diarrhea: No (Reference) Yes -0.068(-2.33)** -0.073(-2.51)** -0.073(-2.51)** Marital Status: Never married(Reference) Married -0.012(-1.28) -0.010(-1.06) -0.010(-1.06)

194

Divorced/Separated/Widowed -0.039(-2.64)*** -0.039(-2.64)*** -0.039(-2.63)*** Log of Household size 0.010(1.62) 0.007(1.07) 0.006(1.07) Distance to nearest health facility: <1KM (Reference) 1-3KM 0.194(18.33)*** 0.192(18.1)*** 0.192(18.10)*** 4-5KM 0.211(16.42)*** 0.210(16.25)*** 0.210(16.26)*** 6-9KM 0.233(16.06)*** 0.221(15.87)*** 0.221(15.87)*** 10+KM 0.165(12.68)*** 0.165(12.63)*** 0.165(12.62)*** Education Level: No education (Reference) Primary Education 0.051(4.62)*** 0.058(5.08)*** 0.058(5.08)*** Secondary 0.042(3.18)*** 0.054(3.84)*** 0.054(3.84)*** College/University education -0.020(-1.07) 0.0005(0.02) 0.0005(0.02) Area of residence: Rural (Reference) Urban -0.062(-7.18)*** -0.048(-4.36)*** -0.048(-4.34)*** Poverty Residuals 0.036(2.00)** 0.035(1.94)** Interaction of wealth index 0.002(0.01) and poverty residuals Wald χ2 (51) 1815.00*** 1816.98*** 1817.12*** Number of observations 17,860 17,797 17,797 Mean VIF 1.29 1.29 1.29 Note: ***, ** and * indicates statistical significance at 1 per cent, 5 per cent and 10 per cent levels of significance, respectively. (.)=Robust Z Statistics Source: Author computation, Study Data, 2013.

Table A5: Average Marginal Effects for the Healthcare Provider Choice Model (Private) Variable Dependent variable=Health care provider (Private) Baseline Model 2SRI Model CFA Model Wealth Index 0.131(16.58)*** 0.175(11.28)*** 0.175(11.26)*** -0.0004(-1.92)* -0.0005(-1.96)** -0.0005(- Age 1.96)*** Sex: Male(Reference) Female -0.069(-7.80)*** -0.060(-7.98)*** -0.061(-7.98)*** Malaria/Fever: No (Reference) Yes 0.044(5.80)*** 0.046(6.04)*** 0.046(6.04)*** Respiratory Disease/Pneumonia: No (Reference) Yes 0.024(2.28)** 0.025(2.31)** 0.025(2.31)** Diarrhea: No (Reference) Yes 0.027(0.96) 0.031(1.11) 0.031(1.11) Marital Status: Never married(Reference) Married 0.028(3.10)*** 0.025(2.72)*** 0.025(2.72)*** Divorced/Separated/ 0.064(4.55)*** 0.064(4.52)*** 0.064(4.52)*** Widowed Log of Household size -0.011(-1.83)* -0.007(-1.13) -0.007(-1.13)*** Distance to nearest health facility: <1KM (Reference) -0.182(-17.35)*** -0.180(- -0.178(- 1-3KM 17.12)*** 17.11)*** -0.200(-15.91)*** -0.199(- -0.199(- 4-5KM 15.77)*** 15.77)***

195

-0.242(-18.33)*** -0.241(- -0.241(- 6-9KM 18.17)*** 18.17)*** -0.199(-15.84)*** -0.198(- -0.198(- 10+KM 15.72)*** 15.71)*** Education Level: No education (Reference) Primary Education -0.041(-3.83)*** -0.051(-4.52)*** -0.051(-4.52)*** Secondary -0.031(-2.46)** -0.049(-3.57)*** -0.049(-3.57)*** College/University 0.016(0.91) -0.016(-0.78) 0.016(-0.78) education Area of residence: Rural (Reference) Urban 0.056(6.90)*** 0.034(3.32)*** 0.034(3.32)*** Poverty Residuals -0.056(-3.31)*** -0.056(-3.23)*** Interaction of wealth index -0.001(-0.06) and poverty residuals Wald χ2 (51) 1815.00*** 1816.98*** 1817.12*** Number of observations 17,860 17,797 17,797 Mean VIF 1.29 1.29 1.29 Note: ***, ** and * indicates statistical significance at 1 per cent, 5 per cent and 10 per cent levels of significance, respectively. (.)=Robust Z Statistics Source: Author computation, Study Data, 2013.

Table A6: Average Marginal Effects for the Healthcare Provider Choice Model (Mission) Variable Dependent variable=Health care provider (Mission) Baseline Model 2SRI Model CFA Model Wealth Index 0.012(2.56)*** -0.004(-0.41) -0.003(-0.36) Age 0.001(5.70)*** 0.001(5.50)*** 0.001(5.48)*** Sex: Male(Reference) Female 0.020(4.61)*** 0.020(4.67)*** 0.020(4.66)*** Malaria/Fever: No (Reference) Yes -0.019(- -0.020(- -0.020(- 4.40)*** 4.44)*** 4.43)*** Respiratory Disease/Pneumonia: No (Reference) Yes -0.015(- -0.016(- -0.016(- 2.60)*** 2.72)*** 2.70)*** Diarrhea: No (Reference) Yes 0.045(2.30)** 0.046(2.33)** 0.046(2.33)** Marital Status: Never married(Reference) Married -0.015(-2.46)** -0.013(-2.17)** -0.013(-2.13)** -0.025(- -0.025(- -0.025(- Divorced/Separated/Widowed 3.14)*** 3.12)*** 3.11)*** Log of Household size -0.0004(-0.14) -0.002(-0.51) -0.002(0.53) Distance to nearest health facility: <1KM (Reference) 1-3KM 0.009(1.70)* 0.010(1.75)* 0.010(1.77)* 4-5KM 0.012(1.83)* 0.013(1.94)* 0.013(1.95)*

196

6-9KM 0.037(4.58)*** 0.038(4.70)*** 0.038(4.71)*** 10+KM 0.053(6.98)*** 0.054(7.07)*** 0.054(7.09)*** Education Level: No education (Reference) Primary Education -0.009(-1.41) -0.005(-0.79) -0.005(-0.81) Secondary -0.006(-0.81) 1.49e-05(0.00) -5.1e-05(-0.01) College/University education 0.009(0.8) 0.019(1.48) 0.019(1.49) Area of residence: Rural (Reference) Urban -0.002(-0.41) 0.013(2.07)** 0.014(2.13)** Poverty Residuals 0.021(2.14)** 0.032(2.23)** Interaction of wealth index and -0.007(-0.68) poverty residuals Wald χ2(51) 1815.00*** 1816.98*** 1817.12*** Number of observations 17,860 17,797 17,797 Mean VIF 1.29 1.29 1.29 Note: ***, ** and * indicates statistical significance at 1 per cent, 5 per cent and 10 per cent levels of significance, respectively. (.)=Robust Z Statistics Source: Author computation, Study Data, 2013.

Table A7: Average Marginal Effects for the Healthcare Provider Choice Model (Others) Variable Dependent variable=Health care provider (Others) Baseline Model 2SRI Model CFA Model Wealth Index -0.005(-2.29) -0.004(-1.06) -0.005(-1.14) Age 0.0001(2.03)** 0.0001(2.15)** 0.0001(2.20)** Sex: Male(Reference) Female -0.004(-1.88)* -0.004(-2.09)** -0.004(-2.03)** Malaria/Fever: No (Reference) Yes -0.006(-3.29)*** -0.006(- -0.006(- 3.74)*** 3.75)*** Respiratory Disease/Pneumonia: No (Reference) Yes -0.007(-4.17)*** -0.007(- -0.007(- 4.23)*** 4.38)*** Diarrhea: No (Reference) Yes -0.004(-0.73) -0.004(-0.74) -0.004(-0.74) Marital Status: Never married(Reference) Married -0.001(-0.43) -0.001(-0.62) -0.002(-0.71) Divorced/Separated/Widowed -7.2e-05(-0.02) -2.5e-05(0.01) -0.0002(-0.07) Log of Household size 0.001(0.78) 0.002(1.26) 0.002(1.27) Distance to nearest health facility:<1KM (Reference) -0.022(-6.24)*** -0.022(- -0.022(- 1-3KM 6.26)*** 6.26)*** -0.024(-6.61)*** -0.025(- -0.025(- 4-5KM 6.64)*** 6.63)*** -0.017(-4.09)*** -0.018(- -0.018(- 6-9KM 4.17)*** 4.20)*** 10+KM -0.019(-4.87)*** -0.021(- -0.021(-

197

5.31)*** 5.35)*** Education Level: No education (Reference) Primary Education -0.002(-0.88) -0.003(-0.99) -0.003(-0.94) Secondary -0.005(-1.70)* -0.005(-1.60) -0.005(-1.57) College/University education -0.003(-0.69) -0.003(-0.66) -0.003(-0.68) Area of residence: Rural (Reference) Urban 0.0003(0.18) 0.0002(0.06) -0.0003(-0.13) Poverty Residuals -0.001(-0.32) -0.002(-0.54) Interaction of wealth index and 0.007(1.98) poverty residuals Wald χ2 1815.00*** 1816.98*** 1817.12*** Number of observations 17860 17797 17797 Mean VIF 1.29 1.29 1.29 Note: ***, ** and * indicates statistical significance at 1 per cent, 5 per cent and 10 per cent levels of significance, respectively. (.)=Robust Z Statistics Source: Author computation, Study Data, 2013.

Table A8: Average Marginal Effects of Probability of Reporting Own Health as Satisfactory Self Rated Health Status=Satisfactory Baseline 2SRI CFA Variable Model(1) Model(2) Model(3) -0.018(- -0.019(- -0.017(- Wealth Index 12.49)*** 5.99)*** 5.46)*** 0.001(4.42)* 0.001(4.54)* Age 0.001(4.46)*** ** ** 2.01e- 1.97e- Age Squared 2e-05(8.67)*** 05(8.67)*** 05(8.52)*** Sex: Male(Reference) 0.017(13.76) 0.017(13.84) 0.017(13.75) Female *** *** *** Marital Status: Not married(Reference) -0.007(- -0.007(-3.84) -0.007(- Married 3.95)*** 3.79)*** 0.017(6.00)* 0.017(5.91) 0.017(5.90)* Divorced/Separated/Widowed ** ** Religion: Traditionalist/Atheist/Others(Reference) Catholic 0.005(1.43) 0.005(1.47) 0.005(1.41) 0.009(2.76)* 0.009(2.74)* 0.009(2.63)* Protestant ** ** ** -0.011(- -0.011(- -0.011(- Muslim 2.84)*** 2.75)*** 2.73)*** 0.0004(1.84) 0.0004(1.74) 0.0004(1.66) Household size * * * Education Level: No education (Reference Primary Education 0.003(1.71)* 0.004(1.71)* 0.003(1.27)

198

-0.012(- -0.012(- -0.013(- Secondary 5.34)*** 4.74)*** 5.13)*** -0.028(- -0.027(- -0.028(- College/University education 9.09)*** 7.74)*** 8.02)*** Employment Status: No (Reference) -0.016(- -0.016(- -0.016(- Yes 10.30)*** 10.19)*** 10.29)*** Access to piped water: No (Reference) -0.038(- -0.039(- -0.039(- Yes 12.36)*** 11.15)*** 11.26)*** Area of residence: Rural (Reference) 0.006(4.26)* 0.006(3.37)* 0.007(3.77)* Urban ** ** ** Poverty residuals 0.001(0.19) 0.004(1.04) Interaction of wealth index and poverty -0.018(- residuals 6.59)*** Number of observations 80742 80450 80450 Note: ***, ** and * indicates statistical significance at 1 per cent, 5 per cent and 10 per cent levels of significance, respectively. (.)=Robust Z Statistics, (.)a=P- value Source: Author computation, Study Data, 2013.

Table A9: Average Marginal Effects of Probability of Reporting Own Health as Good Self Rated Health Status=Good Baseline 2SRI CFA Variable Model(1) Model(2) Model(3) -0.006(- -0.007(- -0.006(- Wealth Index 11.83)*** 5.91)*** 5.39)*** 0.0003(4.49)** 0.0003(4.45) 0.0003(4.57) Age * *** *** 6.94e- 6.93e- 6.81e- Age Squared 06(8.12)*** 06(8.12)*** 06(7.98)*** Sex: Male(Reference) 0.006(12.81)** 0.006(12.86) 0.006(12.79) Female * *** *** Marital Status: Not married(Reference) -0.002(- -0.002(- -0.002(- Married 3.93)*** 3.83)*** 3.78)*** 0.006(5.78)* 0.006(5.77)* Divorced/Separated/Widowed 0.006(5.87)*** ** ** Religion: Traditionalist/Atheist/Others(Reference) Catholic 0.002(1.43) 0.002(1.47) 0.002(1.40) Protestant 0.003(2.75)*** 0.003(2.73)* 0.003(2.62)*

199

** ** -0.004(- -0.004(- -0.004(- Muslim 2.82)*** 2.73)*** 2.71)*** 0.0002(1.74) 0.0001(1.66) Household size 0.0002(1.84)* * * Education Level: No education (Reference Primary Education 0.001(1.71)* 0.001(1.71)* 0.001(1.27) -0.004(- -0.004(- -0.005(-5.07) Secondary 5.28)*** 4.70)*** -0.010(- -0.009(- -0.010(- College/University education 8.82)*** 7.57)*** 7.83)*** Employment Status: No (Reference) -0.005(- -0.005(- -0.005(- Yes 9.82)*** 9.73)*** 9.81)*** Access to piped water: No (Reference) -0.013(- -0.013(- -0.014(- Yes 11.65)*** 10.63)*** 10.72)*** Area of residence: Rural (Reference) 0.002(3.35)* 0.002(3.74)* Urban 0.002(4.22)*** ** ** Poverty residuals 0.0002(0.19) 0.001(1.04) Interaction of wealth index and poverty -0.006(- residuals 6.46)*** Number of observations 80742 80450 80450 Note: ***, ** and * indicates statistical significance at 1 per cent, 5 per cent and 10 per cent levels of significance, respectively. (.)=Robust Z Statistics, (.)a=P- value Source: Author computation, Study Data, 2013.

200