Graduate School of Development Studies

A Research Paper presented by:

Sebatware Rutekereza(Rwanda)

in partial fulfillment of the requirements for obtaining the degree of

MASTERS OF ARTS IN DEVELOPMENT STUDIES

Specialization:[Economics of Development]

(ECD)

Members of the examining committee:

Dr Robert Sparrow [Supervisor] Prof. Dr Michael Grimm [Reader]

Economic effects of Health Insurance in Rwanda:

Case of Community Based Health Insurance (CBHI)

The Hague, The NetherlandsNovember, 2011

ii

Disclaimer:This document represents part of the author’s study programme while at the Institute of Social Studies. The views stated therein are those of the author and not necessarily those of the Institute.Research papers are not made available for circulation outside of the Institute.

Inquiries:

Postal address: Institute of Social StudiesP.O. Box 297762502 LT The HagueThe Netherlands

Location: Kortenaerkade 122518 AX The HagueThe Netherlands

Telephone: +31 70 426 0460

Fax: +31 70 426 0799

iii

AcknowledgmentsI owe my deepest appreciation to my supervisor Dr Robert Sparrow, whose valuable guidance, support and encouragement from the initial to the final stage helped me to develop an understanding of this paper. Your advice with originality has played an outstanding role in shaping this paper. I am very grateful to Professor Michael Grimm for his inestimable suggestions, amendments, and support to several draft chapters of this paper. Your comments and observations were vital inputs which enabled me to improve this paper.

It is a pleasure to express my gratitude to my colleagues Nalishiwa and Binyam for giving such a pleasant time when sharing experiences and discussing courses. Many thanks go to Renate for tireless attitudes towards my many questions addressed to her.

My special thanks to my parents, brothers and sisters for your priceless support. Your contribution and sacrifice in all aspects of my life encouraged me to pursue my studies.

Words fail me to express my gratitude to my wife Esperance and my son Ivan for encouragement, support and prayers. I really appreciate my wife for taking headship of the family and taking care of our son Ivan whom I left when he was 5 months.

May God bless all of you!

iv

Contents

List of Tables viList of Figures viList of Acronyms viiAbstract viii

Chapter 1 Introduction1

Chapter 2 Literature Review and analytical Frame work 52.1 Introduction 52.2 Effects of illness on household consumption and its

implications for poverty 52.3 Welfare effects of community based health

insurance in developing countries 10

Chapter 3 An Overview of health System in Rwanda15

3.1 Introduction 153.2 Health system organisation 15

3.2.1 Public sector 153.2.2 Private sector 153.2.3 Traditional medicine 16

3.3 Health insurance system 163.4 Community based health insurance in Rwanda 17

3.4.1 Background of CBHI in Rwanda 173.4.2 Participation in community based health

insurance 183.4.3 Community based health insurance

organisation in Rwanda 183.4.4 Benefits package composition under CBHI 19

3.5 Health financing in Rwanda 20

Chapter 4 Data analysis and empirical strategy21

4.1 Introduction 214.2 Data 21

4.2.1 Quantitative data 214.2.2 Qualitative data 22

4.3 Descriptive Analysis 22

v

4.3.1 Summary statistics of variables included in the analysis 22

4.3.2 Socio-economic differences between participants and non-participants 23

4.3.3 Poverty Incidence 244.3.4 Head of household health insurance

membership 254.3.5 Age group 254.3.6 Health insurance Membership 264.3.7 Health problems incidence 26

4.4 Qualitative sample 274.5 Model specification and econometrics concerns 28

4.5.1 Model specification: PSM model 284.5.2 CBHI effects on household consumption and

impoverishment: Constructing the Counterfactual 29

4.5.3 Community based health insurance participation model 31

4.5.4 Choice of variables for PSM 314.5.5 Household Consumption response to health

problems 32(OLS Model) 32

Chapter 5 Results Discussion34

5.1 Introduction 345.2 Participation Characterizing Model 345.3 Balancing property 355.4 Effects of CBHI on household consumption and

poverty (NNM-PSM) 365.5 Robustness check and sensitivity analysis 38

5.5.1 Effects of CBHI on household consumption (OLS) 38

5.5.2 CBHI effects on household consumption and poverty (NNM compared to Kernel, Radius, Probit and OLS estimates) 40

5.5.3 The Effects of CBHI on sub-groups of population 42

5.6 Qualitative analysis 44

Chapter 6 Conclusion47

vi

References 49

Appendices53

vii

List of TablesTable 3.1: Community based health insurance

membership trend in (%) 18Table 3.2: Descriptive statistics EICV2, Insured and

insured members 23Table 3.3: Social-economic differences between insured

and non-insured members 24Table 3.4: Poverty incidence and health insurance

membership 25Table 3.5: CBHI distribution by sex of head of household 25Table 3.6: Percentage of prevalence of health insurance 26Table 3.7: Percentage of people reporting illness 27Table 5.1: Probit Model of determinants for Participation

in CBHI (Marginal effects) 35Table 5.2: Community based health insurance Effects

(PSM) 36Table 5.3: Community based health insurance Effects

(OLS) 39Table 5.4 : Community based health insurance Effects

(People who reported sick) 41Table 5.5: CBHI effects on outcome variables for

different population groups (full sample) 43Table A.1 Descriptive statistics 53Table A.2: Balancing properties of the matched samples

(Reported sick) 54Table A.3 Balancing properties of the matched samples

(Full sample) 55Table A.4: Community based health insurance Effects

(OLS) 56Table A.5: Community based health insurance Effects

(Probit Model) 59Table A.6: Robustness check Full sample including

Illness 60

List of FiguresFigure 2.1: Simplified Flow-Chart of Key Issues Relating

to the Economic Consequences of Illness and the role of health insurance 9

viii

Figure 2.2: Simplified Flow-Chart Relating health insurance to income generation 12

Figure 4.1: Head of household age distribution 26Figure A .1: Propensity Score Graph (Reported sick) 57Figure A .2: Propensity Score Graph (Full sample) 58

ix

List of Acronyms

$ US dollarCBHI Community Based Health insuranceCFH Community Health FundsCIA Conditional Independence Assumption EICV Enquête Integral sur les Conditions des Vies des ménagesGoR Government of RwandaHh HouseholdIRST Institut de Recherche, Sciences et TechnologieMDG Millennium Development GoalsMMI Military Medical Insurance MoH Ministry of Health NHA National Health AccountsNGO Non-Governmental OrganisationNISR National Institute of Statistics of RwandaNNM Nearest Neighbour Matching OLS Ordinary Least SquaresOOP Out Of PocketPSM Propensity Score MatchingRAMA La Rwandaise d’Assurance MaladiesRWF Rwandan Francs (Rwandan currency)S.E Standard errorsVUP Vision 2020 Umurenge Program WHO World Health Organisation

x

Abstract

This study investigates the effects of community based health insurance in preventing households from consumption expenditures disruption and impoverishment effects resulting from health problems and associated costs. Many theoretical models argued that health insurance protects households from health problems and associated costs. Using integrated household living condition survey (EICV 2) data and qualitative data, we explore whether food and non-food, health and education expenditures of insured people are not disrupted in the wake of health problems, and whether they are prevented from impoverishment effects of health expenditure payments. In addition, we tested whether insured households were protected from dropping their children out of school as adjustment mechanisms when they face severe illness. The results suggested that community based health insurance has prevented consumptions disruption for insured people which would result from health problems. Non-food consumption was found to be positively related to community based health insurance, but was not statistically significant. The findings are consistent with the impact of health insurance on poverty reduction and school dropping out whereby insured households were prevented from falling into poverty and dropping their children out of school following episodes of illness. These findings indicate that community based health insurance plays a crucial role in addressing issues preventing poor people to lead decent life.

Relevance to Development StudiesCommunity based health insurance schemes are increasingly considered as important tools in protecting poor people from health problems and its associated costs in developing countries. This has therefore drawn the attention of many scholars to assess its impacts. Previous studies have revealed potential impact of community based health insurance on demand for modern health care, mitigation of out-of-pocket catastrophic health payments. However, little is known on its role on preventing the disruption of household consumption expenditures,

xi

household impoverishment and school dropout. Identifying its impacts on these outcomes will help to design appropriate policy, aiming at improving the standard of living in developing countries.

KeywordsCommunity based health insurance, Food consumption, Education expenditures, Health expenditures, Non-food consumption, School dropout, Illness and Poverty.

xii

Chapter 1Introduction

Providing health care for poor people who work in informal sector, or live in rural areas is considered as one of the most difficult challenges that many developing countries are facing (Preker and Carrin 2004). Despite remarkable efforts in controlling these challenges by many development agents and states, they remain as severe barrier to economic growth (Sachs and World Health Organization (WHO) 2001) since illness does not only affect the welfare but also increases risks of impoverishment. This is because of high cost associated with health problems, especially in the absence of any form of health insurance. Subsequently, households may decide to leave illness untreated or opt for use of poor quality health care or even self-administration medication (Ataguba et al. 2008). It is argued that more than 150 million people face catastrophic health expenditures each year and most of them fall into poverty worldwide because of out of pocket health payments (Kawabata et al as cited in Saksena et al 2011). This is an indication that health problems and associated costs are main causes that drive people into poverty, especially in developing countries where the health care payment is still made out of pocket. The World Bank reports 1993 and 1995 (as cited in WHO 2002) reveal that illness, death, and injuries stand as the main causes that have led people into poverty. From this analysis it is evident that health problems can hold back any effort made by poor people to improve their standards of living, reason why poverty reduction policies should incorporate health facilities improvement, since health problems and poverty are much related. Poverty is also argued to be among root causes of many health problems, such that poor people can neither afford modern medical care nor decent living conditions.

To better address the problem, community based health insurance schemes (CBHI) are therefore considered to be potential instruments mitigating the impoverishment effects associated with health expenditures, especially in developing countries. The effectiveness of community based health insurance resides in the facts that it can reach a big number of poor people, who would not have been able to insure themselves against health problems and associated costs (Dror and Acquire as cited in Jütting 2004). By pooling illness risks, unpredictable medical expenditures

1

are therefore reassigned to premiums. This will result in increasing access to health care that may mitigate the adverse impact related to health problems on poor households and improve the access to quality health care. Consequently, good health status resulting from access to health will improve productivity, which in turn will increase income leading to good living conditions for insured households (Asfaw and Jütting: 2007).

As a developing country, Rwanda is not spared from health care provision challenges. The National Institute of Statistics of Rwanda (NISR) and Macro International (2008) report reveals that, after introducing the direct payment1 in 1996, after 1994 genocide, health care utilisation decreased considerably because more households were no longer able to pay medical care services. In an attempt to increase health care access, the government of Rwanda has witnessed an increase in health expenditures such that health expenditures per capita increased from US $ 12.68 in 1998 (Schneider et al 1998) to US $ 17 in 2003 (MoH 2006). Therefore, reconciling health care provision and internal resources mobilisation in order to increase financial viability of health care services became a major challenge in Rwandan health system (NIS and Macro International 2008). In addressing such issues, the government of Rwanda initiated community based health insurance as a strategy of improving financial accessibility to health care services for poor people from rural and urban settings (Schneider and Diop 2001). The introduction of this health-risk pooling system intended to help most poor people to be fully integrated in health insurance system and reduce the health care associated costs which push many people into poverty (MoH, 2010). Community based health insurance was established with three specific objectives: to improve financial access to health care, to improve financial situation of health facility and improve the overall health status of population2. Furthermore, Rwanda has integrated the development of community based health insurance in its priorities, basing on the fact that human capital investment

1 Direct payment refers to the situation where sick person pays medical care on his own without any external financing source such as health insurance, employer or other institutions.

2 See the CBHI objectives on this link: http://www.cbhirwanda.org.rw/index.php?option=com_content&view=article&id=49&Itemid=34 (accessed 15 April 2011).

2

is one of poverty reduction strategy pillars that guide the country towards attainment of its vision 2020 objectives as well as international development objectives such as the MDGs (MoH, 2010). Evidences from qualitative analysis reveals that CBHI have prevented insured people from substantial health expenditures they used to incur before joining the schemes. This has enabled insured people to allocate such money to other household expenses and savings, such that the saved money was used for schooling, food and non-food consumption and for further investment. This was emphasized by an experience of an insured person with 6 children who, expressing his satisfaction related to CBHI said:

With community based health insurance, the health expenditures for my family has substantially reduced. I was admitted at National Military Hospital at Kanombe for surgery, and was hospitalized for long period. The bill for health expenditures was around 170 000 RWF (262 US$) but being community based health insurance member, I paid only around 17 000 RWF (26 US$). Whenever any household member falls sick, I am not afraid of taking him to the health centre, because of health insurance membership, he added (Sekavumba 2011, personal Interview).

This story is an indication that, Community based health insurance help reducing health expenditures for poor households in Rwanda. It is also evident that household consumption expenditures are not disrupted due to excessive health problems and associated costs.

Recently, community based health insurance has drawn attention of scholars in the study of development economies. It is in this regard that many research works on community based health insurance in Rwanda were conducted to assess its impact on demand for modern health care, mitigation of out-of-pocket catastrophic health expenditures. Schneider and Diop (2001) using household survey data from three pilot health districts in Rwanda, found that community based health insurance has substantially reduced the out of pocket health payments following illness episodes and improved the equity in health service delivery for members. Likewise, Saksena et al. (2011) findings suggest that community based health insurance has considerably increased the health service utilisation and has had high level of financial risk protection by reducing the catastrophic health expenditures. This study used data collected from 2005/2006 integrated

3

household living conditions survey to estimate the program effects. Using the same data, Shimeles (2010) found that community based health insurance in Rwanda have effectively reduced catastrophic health expenditures.

However, little is known on how community based health insurance prevents households from impoverishment and disruption of household consumption expenditures through prevention of catastrophic health expenditures emanating from health problems. This study, therefore analyses how Community based health insurance has mitigated the negative impact of health problems on food and non-food consumptions, education expenditures, school dropout, out of pocket health payment and prevented impoverishment among insured people. Considering that poor people, particularly in developing countries rely essentially on their labour productivity and on their assets in generating revenues (Jütting 2004).

This study uses both qualitative and quantitative data to achieve its objectives and get a deep understanding of community based health insurance effects in Rwanda. Combining qualitative and quantitative information is essential for the analysis of the effects of the program, considering that health insurance programmes involve some social economic issues. The quantitative data used was from the integrated household living condition survey, conducted in Rwanda covering the period of 2005/2006. Qualitative information was collected through interviews with individuals and groups of people from insured and non-insured households’ members and government officials in charge of community based health insurance from the Ministry of Health. This type of information was used to complement quantitative data by understanding the behaviour of people with and without community based health insurance, mostly in the wake of health problems. This paper uses Propensity Score Matching (PSM) to estimate the effects of community based health insurance on household consumption expenditures and preventing people from impoverishments effects resulting from ill -health and associated costs.

The remainder of the paper is structured as follows: Chapter two reviews the literature related to the effects of illness on household consumptions and its implication for poverty and the effects of community based health insurance in preventing these risks. Chapter 3 gives an overview of health system and community based health insurance in Rwanda background. In chapter 4 the paper

4

analyses the data and lay empirical strategy. A discussion of the results is in chapter 5 while the last chapter draws a conclusion.

5

Chapter 2Literature Review and analytical Frame work

2.1 IntroductionThere is a growing literature on community based health insurance in developing countries and its impacts on households living conditions (Jütting 2004, Msuya et al 2004, Ranson and John 2001 Chankova et al. 2008). Community based health insurance schemes are deemed as “local initiative which is built on traditional coping mechanisms to provide small scale health insurance products specially designed to meet the needs of low-income households ’’ (Carrin et al as cited in Mugisha and Mugumya 2010: 181). These schemes increase health care services of poor people by offering both preventive and curative health care (Hamid et al. 2011). It is further argued that community based health insurance help insured people to recover fast as they are not delayed in seeking health care (Jutting 2004). Given that better health status increases productivity and labour supply, which boost household income level (Hamid et al. 2011), community based health insurance is therefore considered as potential tool in improving standards of living of poor people.

This chapter presents the effects of illness on household consumptions and its relationship with poverty in section two, and the last section points out the effects of community based health insurance in mitigating illness effects based on empirical evidences and highlights empirical evidence gap that will be investigated in this paper.

2.2 Effects of illness on household consumption and its implications for poverty

The effects of illness on household consumption have received a great deal of attention from many scholars for the last decades (Flores et al 2008, Gertler and Gruber 2002, Wagstaff 2007). It is argued that poor people in developing countries pay for their healthcare expenses mainly out of pocket. In case of serious illness, households are forced to reduce their consumption expenditures so as

6

to cope with health expenses incurred. Health expenditures reduce household expenditures such as food, education, farming expenses and others (Wang et al 2006). Similarly, Gertler and Gruber (2002) revealed using a large set of the panel data that, Indonesian households are not able to protect consumption against costs associated with illness, particularly when the degree of illness is severe. In case of small health shocks, their findings reveal partial changes in household consumptions. Similarly, in his empirical analysis, Wagstaff (2007) suggests that households are not able to smooth their food consumption in the face of illness associated with long time hospitalization.

Moreover, it is argued that, when a household member falls sick; household may respond by reducing consumption due to the reduction of income generated from labour supply of the ill person or health expenditures incurred because of health care services, in the absence of health insurance (World Bank 2001). This is because; excessive health expenditures substantially affect household ability to maintain the same level of consumption and force people to make use of income threatening coping mechanisms to deal with medical care expenses. In their study carried out in Burkina Faso, Sauerborn et al. (1996) found that medical expenditures are paid mostly from savings, livestock and other assets sales, contracting loans and labour substitution. Coping mechanisms which are used to maintain the same level of consumption disrupted by health problems and associated costs are argued also to have long term economic effects and undermine the future income earnings capacity (Rajeswari et al., Sen, Krishna van Damme, and Krishna et al., as cited in Flores et al 2008). When a household member faces severe illness, he is compelled to spend a big share of household budget on medical expenditures, which have implications on household consumption of other goods and services such as food, cloth, education and shelter. Health problems, therefore constitute major hindrance to households’ welfare not only on short term, but also in the long run, as assets and savings depletion due to health problems may serve as buffer stock for future household consumptions and investments. Following illness episodes, health expenditures amplify household expenditures, while the income level remains constant or even reduces. In this

7

case, household income level is no longer enough to cover total expenditures.

Moreover, illness does not only involve household consumption disruption but also pushes many household into poverty as the latter is often consumption based. It is argued that poverty is one of the most vexing challenges that developing countries are faced with, with illness being among the main factors that push households to slip into deep poverty. This has raised concern among many development agencies and government authorities from different countries. It is further argued that ‘‘concern about link between ill-health and impoverishment has placed health at the centre of development agencies’ poverty reduction targets and strategies’’ (World Bank and WHO as cited in Russell 2004:147). This concern is based on the fact that health problems undermine household’s income generating capacity by limiting household labour participation on one hand, and increase financial hardship emanating from catastrophic health expenditures on the other hand (World Bank 1997; Barnett et al. 2001). It is therefore clear that health problems and associated costs have severe implications on households’ impoverishment especially in developing countries where health services payment is made out of pocket. According to WHO 2003 data (as cited in Scheil-Adlung et al. 2006), the share of out of pocket health payment stands at 1/3 of total health care expenditures in 2/3 of all developing countries. Similarly, Su et al (2006) added that out of pocket health payment is generally high in developing countries such that poor households are not able to cover them from the existing income, and households are forced to use income threatening coping mechanisms. It is therefore apparent that this draws back any effort made by poor people to lead a decent life and undermines household’s future ability to improve living standards. According to Ataguba et al. (2008), catastrophic health expenditures are argued to increase impoverishment risks and reduce people’s welfare. Likewise, Scheil-Adlung et al. (2006) argued that catastrophic health expenditures lead households into poverty as they are forced to sell their productive assets and reduce their consumption so as to deal with such kind of expenditures. Catastrophic health expenditures therefore constitute one of key factors aggravating poverty by affecting negatively household consumption expenditures patterns (WHO 2000). It is argued that when a household is no longer able to meet basic needs such as food, clothing and shelter, due to excessive medical care costs, it is

8

vulnerable and is forced to persist into poverty. Poverty therefore appears as linked with health problems so much that, in the absence of adequate financial markets poor people are considerably compelled to remain in poverty.

Furthermore, in addition to medical care expenses, illness does entail indirect costs also that drive into poverty. It is argued that following an occurrence of serious illness, household labour supply will drop not only because the ill person is not able to work but also because some household members are assigned to take care of sick person who also is unable to perform tasks (Asenso-Okyere and Dzator 1997). Household may also hire workers to deal with the loss of worker or contract loan to pay treatment and substitute the lost income in order to sustain household financial conditions threatened by illness and related costs (Russell 2004). From this point of view, health problems seem to be a driving force of household impoverishment that needs to be addressed in an effort to improving household living conditions in low-income countries. It sounds that the level of households’ impoverishment will therefore depend on the means that are used to deal with health problems. That is why major illness is considered for developing countries as one of most challenging issue to economic development in the absence of health insurance for developing countries (Gertler 1997).

Additionally, illness affects also labour productivity and productive assets depletion in case household uses those coping mechanisms to deal with health problems and associated costs. Schultz and Tansel (1997) empirically assessed the effects of adult sickness on labour productivity using lost days and wages as measures. Their findings indicate that health problems affected productivity and the wages were lowered for each disabled day. Health problems constitute therefore an impediment to economic growth in generally as it affects individual’s income earning capacity (Cole and Neumayer 2006) and Barro RJ (1996). Similarly, Claeson et al. (2001) stressed that illness is one of the main factors that influences households to slip into poverty as it affects the income earnings capacity and the ability to cope financially. In their study assessing indirect costs associated with illness in Ghana, Asenso-Okyere and Dzator (1997) found that, people spent more time on seeking malaria care and taking care of the sick persons. This involves also opportunity costs for lost days taking care of sick person or looking for treatments. Likewise (Babu et al. 2002) found that illness affect working days

9

and increases absenteeism among the sick persons, which pose burden not only to the family but also to the community.

However, it is worth noting that, despite many findings on economic consequences of illness and health care, most of them have devoted low attention to indirect costs of illness. Much is discussed on the effects of illness on out-of pocket health payments, or on household income, but the channels through which illness leads to poverty are less discussed. This paper will attempt to fill that gap based on Rwandan experience.

The figure below reveals the economic consequences of illness and different coping mechanism and the role of health insurance.

10

Indirect costs

Intra and inter-house labor substitution

Reducing consumption

No

Use of savings made

Family members time costs

Sale of Assets

Hire labor and other strategies

Medical Expenditures

Perceived Illness

Other strategies to cope with financial costs

Other Financial costs

Income of ill person

Borrowings

Seek care if so, type of costs

Impoverishment

Yes

Direct costs

Health insurance schemes

Protection against catastrophic health expenditures

Figure 2.1: Simplified Flow-Chart of Key Issues Relating to the Economic Consequences of Illness and the role of health insurance

Illness Experience Treatment seeking behaviour Economic Consequences Coping strategies and social resources

Source: Modified from McIntyre et al. (2006)

9

The figure above indicates that the strategies that households adopt to finance health care in the absence of health insurance involve financial risks. Due to high health expenditures, households are forced to reduce their basic needs’ spending or sell their assets in order to cope with financial losses incurred (Flores et al. 2008). Besides direct costs associated with illness, it also affects household productivity and labour supply (Asfaw and Jütting 2007). Under such circumstances, households facing health problems are not even able to hire out labour for casual work during illness period which eventually affects households’ income level.

However, community based health insurance schemes play important role in curbing negative effects of illness on poor people in developing countries. This was underlined by Scheil-Adlung et al (2006) stressing that health insurance play a crucial role in reducing household impoverishment through reducing the shortfall in generating income resulting from illness, and protect households from risky and wealth-threatening health coping strategies in order to cover medical care expenses.

2.3 Welfare effects of community based health insurance in developing countries

The role of community based health insurance in preventing negative effects associated with health problems in developing countries is increasingly being noticed and esteemed (Jütting 2005, Schneider and Diop 2001). While health problems constitute one of major barriers to economic growth in these countries, community based health insurance schemes are featuring among prominent poverty reduction strategies. Tabor (2005:8) argued that “improving access to affordable health care is central to boosting growth and help to break the vicious cycle of poverty and ill health”. This is an indication that population health insurance play a crucial role in a country’s economic development, as they are interrelated. In addition, by investing in human capital and particularly in health, developing countries will get through persistent poverty status that has characterised them for long. Therefore, community based health insurance seems to play an important role in alleviating health problems impoverishment effects for poor people in developing countries.

10

Moreover, improvement of health status is believed to lower workdays lost and increase productivity (Hamid et al. 2011), while illness undermines all effort made by poor people. According to Scheil-Adlung et al. (2006), good health determines economic growth through improvement of labour productivity and education (good health status enables people to gain more from education). Jack and Lewis (2009) underlined that, better health is believed to stimulate economic growth that development agents and, nongovernmental agencies could not have achieved. That is why, millennium development goals are mostly related to health problems3. In addition, Hamid et al. (2011) suggest that community based health insurance improves health status of insured members which increases productivity and labour supply. Households are therefore prevented from using income threatening mechanisms to deal with health problems and are more likely to increase their income and consumption level (Ibid). Community based health insurance comes out therefore as an important instrument in improving living conditions of large number of people from health poverty trap in developing countries because it reaches a big number of them. The Sachs and WHO (2001:i) report states that: ‘‘Extending of crucial health services including a relatively small number of specific interventions, to the world’s poor could save millions of lives each year, reduce poverty, spur economic development, and promote global security”. It is therefore imperative that suitable policies are devised for countries to extend health insurance coverage in an effort to curb negative effects associated with health problems. The below flow-chart indicates how health insurance can lead to high level of income by protecting people from illness and improve health status:

3 For details see http://www.un.org/millenniumgoals/ (accessed on 14/11/2011)

11

Health insurance schemes

Reduced uncertainty of health expenditure

Increased utilization formal health care

High income level

Increased investment in all form of capital

Improvement in health status

Income above subsistence level

Higher investment in physical capital

Higher investment in Human capital

Enhanced Income via Higher productivityHigher labor supply

Low level of health Expenditure and lost Working days

Reduction of vulnerability

Figure 2.2: Simplified Flow-Chart Relating health insurance to income generation

Source: Modified from (Hamid et al. 2011)

12

Figure 2 presents the effects of community based health insurance in improving household welfare. By making their pre-payment premiums, households share risks and are prevented from catastrophic health expenditures which may lead to household impoverishment. Health insurance schemes help insured members to take advantage of economic opportunities through reducing the uncertainty resulting from health expenditures (Hamid et al. 2011). As poor people rely mainly on labour productivity in generating income, access to health care resulting from community based health insurance will improve health status and decrease the lost working days due to illness and consequently household production output will increase. In addition, as household are no longer exposed to catastrophic health expenditures, their consumption patterns will increase. Increasing household consumption and improving health status of household members will considerably increase income. Community based health insurance schemes are therefore considered as potential instruments which reduce uncertainty of health expenditures, improve health care utilisation and enhance income through high productivity and labour supply (Ibid).

Based on these views, community based health insurance has drawn attention of many scholars such that many empirical studies were conducted to investigate its impacts in developing countries. According to Jütting (2004) the Senegalese community based health insurance schemes have increased the access to health care for insured members compared to non-member, and health expenditures have substantially reduced for beneficiaries’ members. Similarly, Shimeles (2010) using matching estimator technique and traditional regression found that community based health insurance in Rwanda has successfully increased the utilisation of modern health care and significantly reduced catastrophic health expenditures. This study used data collected from 2005/2006 Integrated Households Living Conditions Survey, which data cover 6 900 and with about 35 000 individual information. Community based health insurance does not only increase access to health care but also reduces uncertainty that may result from health problems and associated costs. .Moreover, health insurance was found protecting households from catastrophic out-of- pocket health payment and its impoverishment effects (Asfaw and Jütting 2007). It is therefore suggested that by preventing households from catastrophic health expenditures, CHBI allow insured people to allocate resources to other household needs.

13

Likewise, Ranson and John (2001) found that CBHI in rural Gujarat plays a crucial role in improving access to modern health care, and preventing indebtedness and impoverishment for insured poor people. Furthermore, using Probit model Msuya et al (2004) found that CHF have improved the access to health care and prevented insured households from relying on risk coping mechanisms such as selling assets and children school dropout following episodes of sickness. Similarly, Chankova et al. (2008) in their study on Mali, Ghana and Senegal found community based health insurance to be an effective tool in protecting households from catastrophic health expenditures, particularly in the area where health care is mostly made out-of-pocket payment. Additionally, findings from Uganda revealed that insured people were protected from selling their assets and financial risks that lead to impoverishment in the wake of illness (Dekker and Wilms 2010). It is therefore clear from all the above findings that, community based health insurance schemes, play a substantial role in improving access to health care which leads to improved health care status in particular and stimulates economic development in general.

However, despite many empirical studies, there is still little empirical evidence on the impact of community based health insurance in countries like Rwanda on household’s food and non-food consumption, education expenditures, school dropout and impoverishment effects following health problems episodes. As mentioned previously, most studies focused on directs effects of illness on households income and the impact of community based health insurance in mitigating these negative consequences. But little attention was devoted on mechanisms through which illness affects households and the role of health insurance in preventing households. In the presence of health problems, poor households cope by adjusting their expenditures such that, some households may reduce education expenditures by dropping children out of school; others may reduce their food and non-food consumption while other may contract credit so as to deal with health expenditures resulting from illness. This has implications on poor households not only in the short run through disrupting households’ consumption expenditures, but also in the long run through impeding future income generating capacity. Nevertheless, community based health insurance is one of the mechanisms of reducing negative effects associated with health problems. Referring to Rwandan CBHI, this paper tries to shed light on health insurance role in preventing

14

poor people from resorting to welfare threatening coping mechanisms. Given that insured households will be able to sustain their level of income in the presence of health problems; this will prevent them from selling productive assets and contracting loans so as to meet health needs (Hamid et al. 2011). As a result, income uncertainty will reduce and allow households to increase their human and physical capital investment.

15

Chapter 3An Overview of health System in Rwanda

3.1 IntroductionThe Rwandan health sector has undergone major changes over the last decades. For the period before colonisation, Rwandan health system was characterised by traditional healing which continued even after the introduction of modern health care but less popular. In addition to the introduction of modern medicine, the health system witnessed a strong centralised health system, whereby religious institutions played an important role in the system (NISR and Macro International 2008). Following the decentralisation policy process that Rwanda has embarked on, health care services reinforces its decentralisation process starting on provincial level reaching district level. Moreover, despite the consequences of macabre 1994 genocide that destroyed the big part of health infrastructure; Rwandan health system has witnessed a remarkable recovery. Health indicators provide evidence of the progress made over the last year whereby, under-five mortality rate dropped by nearly 50% between 2000 and 2007 (NISR 2008). This chapter describes the health system in Rwanda starting by operating authorities in first section; the second section represents the health insurance while the last section gives detail on community based health insurance, its background, management, and benefits packages.

3.2 Health system organisationThe Rwandan health system is made of public and private sector, traditional medicine, and community health. However, the private sector and traditional medicine systems are not well developed as health care provision still depend much on subsidies.

3.2.1 Public sectorThe public sector has three levels that coordinate the health system in order to improve health care provision. The central level develops health policy and technical framework under which health services have to be provided, while the decentralised level facilitates the

16

implementation of different health policies designed by the central level at grass root level. In addition, health centre level provide primary health care services while other health facilities offers a package of activities reduced from that offered by health centre (NISR and Macro International 2008).

3.2.2 Private sectorThe private sector health services provision currently cover small group of people who live mainly in urban areas. This is basically due to low purchasing power of most Rwandan living particularly in rural areas. However, the Ministry of health in partnership with private sector is aimed at improving the participation of this sector in providing health services to a large number the population and upholding its participation in increasing modern health care accessibility (GoR 2005). It is expected that this partnership will boost the access to health care services not only to the urban people but also to the remote areas of the country.

3.2.3 Traditional medicineAccording to NISR and Macro International (2008), a number of the Rwandan population still make use of traditional medical services when they face health problems. This has resulted in the sustenance of traditional medicine despite the introduction of modern medicine. Seeking traditional medicine however, depends mostly on the nature of the health problems faced and the costs of modern medicine. It is believed that there are some diseases that are only treated through the use of traditional medicine. The Ministry of Health in conjunction with the Institute of Scientific and Technological Research (IRST) embarked on developing traditional medicine in order to improve the quality of the services they provide (Ibid).

3.3 Health insurance system The health insurance in Rwanda is made of four types of health insurance schemes: civil servant health insurance (RAMA), Military health insurance (MMI), Private health insurance schemes and Community based health insurance. The Civil servant health insurance is a mandatory public health insurance which started in 2001 and currently covers civil servants and other government agents and some private workers (MoH 2009). The premium under this scheme is shared by both the employer and the employee.

17

The Military medical insurance was initiated in 2005 and covers servicemen and their dependants. The private health insurance serves as optional health insurance and is run mostly by private insurance companies and covers mainly self-employed and private companies’ employees. Community based health insurance serve as a supplement to existing health insurance schemes. However, it is worth noting that this health insurance scheme covers a big number of populations, because around 89.6%4 of them live as farmers and are not able to pay for private health insurance. As the Community based health insurance constitutes the main focus of this paper, the next section will give further detail on this health insurance scheme by looking at its background, organisation and participation process and benefits packages that it provides.

3.4 Community based health insurance in Rwanda

Rwandan Community based health insurance schemes are health insurance organizations which are based on the partnership of the community and health care providers with support from the Ministry of health. These schemes are autonomous such that members can adopt their own internal rules, regulations structure and program (NISR and Macro International 2008).

3.4.1 Background of CBHI in RwandaIn an effort to improve the access to modern health care and protect poor people from financial risks associated with health problems, the government of Rwanda through its Ministry of health in partnership with local population initi-ated community based health insurance program in January 1999, where 54 schemes started in three districts (Schneider and Diop 2001). These prepayment schemes were assigned to mobilize internal resources so as to in-crease financial sustainability of health services in Kabu-tare, Kabgayi and Byumba pilot districts. The selection of these pilot districts by the Ministry of health was based on districts health infrastructure, political will to participate in health insurance introduction and frequent demand for technical assistance from the people in those districts in im-plementing and developing Community based health insur-

4 See the detail on : http://statistics.gov.rw/index.php?option=com_content&task=view&id=252&Itemid=307 (accessed on 22nd September 2011)

18

ance (Ibid). The discussions that were held between local population and health care representatives for CBHI imple-mentation came out with a formation of an insurance feder-ation at district and fixed the annual premium fees for en-rolment to RWF 2 500 (US $ 4.5) per family up to seven persons paid to the schemes affiliated with preferred health centre (Ibid). In case of sickness people visit the nearest centre mostly public or church owned health centres for treatment. In order to avoid excessive costs that may result from high demand for health care, the Ministry of health with medical services providers and community based health insurance managers decided to use capitation pro-vider payment to the health centre (Ibid). In addition, dis-trict hospitals were compensated by insurance federation for services provided to insurances members using free-for-service payments.

After realizing that the pilot CBHI witnessed success in improving the access to health services and preventing financial risks, they have become very popular such that, community and political authorities tried to scale them up at national level (Kagubare 2006). In 2007, the annual subscription was raised to RWF 1000 (around US $ 1.8) 5 per person per household. This increase was made so as to raise internal resource mobilisation for sustainability of community based health insurance and to improve health services provision and expanding basic package of curative services (MoH 2009).

3.4.2 Participation in community based health insurance

Community based health insurance covers all members who pay their required contribution fees. However, given that poor people cannot afford paying their premiums, the government in partnership with non-government organisations and donors subsidize the fees for this category of people. In addition for any household to be entitled to get benefits, all household members have to fully pay their premiums. The subscription fees payments are made on annual basis and have to be made three months before so as to avoid self-selection problems, especially for people sick person. Although, community based health insurance is voluntary, the current law on community based health insurance specifies that every person who resides in Rwanda who is not insured under any other health

5 The exchange rate (RWF) in 2007 was 1 $ US = RWF 555,50

19

insurance schemes should join community based health insurance schemes6. This is due to the government’s willingness to increase the health care access and prevent financial risks that may result from catastrophic health expenditures for insured households. The trend of coverage rate since 2003 when the schemes were expanded across the country is depicted in table 3.1.

Table 3.1: Community based health insurance membership trend in (%)

Year 2003 2004 2005 2006 2007 2008 2009 2010

Coverage rate 7 27 44 73 75 85 86 91

Source: Rwandan Ministry of health

Table 3.1 reveals that community based health insurance has experienced continuous growth whereby the number of beneficiaries has increased from 7% in 2003 to 91% in 2010. However, sustainability and attracting those remaining people to join the schemes is still a challenge.

3.4.3 Community based health insurance organisation in Rwanda

Community based health insurance schemes in Rwanda are autonomous establishments that are managed by their members (NISR and Macro International 2008). The regulations and rules governing the schemes program and functioning are adopted by insured members. Moreover, beneficiaries at grass-root level in all districts, elect management committee members and define their roles and responsibilities7. In order to maintain the proximity, community based health insurance has an implementation unit at each sector8 level which is in charge of collecting contributions, sensitizing the population and pool risks at grass root level. This is because each sector has at least one health centre that provides medical services to the beneficiaries. The unit also coordinates and monitors planned activities at (cell and village level) alongside grass-root level committees in all districts. These committees monitor the turn up and identify the poorest people among

6 See Law N° 62/2007 of 30/12/2007, related to the establishment, organization, functioning and management of community based health insurance. Available at http://www.cbhirwanda.org.rw/documents/Mutuelle%20Law.pdf (accessed on 27 September 2011).

7 Ibid.8 A Sector: is A third level administrative subdivision

made of many villages and is under District20

the village to be subsidized9. Community based health insurance hosts funds titled “District risk pool” at each district level serving many health centres at sector level and district hospital.

However, in order to avoid risks of adverse selection, the entire family is required to enrol so that household member may benefit from the schemes. As previously mentioned, the premium payment system considers the low purchasing power of poor people by providing subsidies from government and development partners. It should be noted that, whenever an enrolled member obtains health services, he pays 10% of medical care costs in an effort to avoid moral hazards that may take place because of overusing health services. In addition, in order to ensure CBHI schemes sustainability, national risks pooling funds was established under the Ministry of Health to assist health insurance schemes experiencing financial problems and allocate grants to CBHI funds at district level10. Despite the setting up of the Fund, some district schemes were becoming more unsustainable with heavy unpaid debts. Therefore, the government decide to increase the premiums starting by end of year 2011, whereby households were classified in three categories based on their level of income whereby rich people will be paying premium up to RWF 7 000 ($US 12) per person per year, and the lowest will be paying RWF 2000 ($US 4) per person per year (MoH 2010). However, the government will still provide subsidies to the poorest people, who are not able to pay the premiums. The subsidies allocation will be based on poverty incidence information collected from VUP (Vision 2020 Umurenge Program) in collaboration with grass-root CBHI committee’s members. Besides ensuring financial sustainability, the premiums increase will extend members medical service access to all hospitals including private hospital and pharmacies and enlarge the package for Universal coverage (Ibid).

9 See Law N° 62/2007 of 30/12/2007, related to the establishment, organization, functioning and management of community based health insurance. Available at http://www.cbhirwanda.org.rw/documents/Mutuelle%20Law.pdf (accessed on 27 September 2011).

10 Ibid21

3.4.4 Benefits package composition under CBHI

The medical services package that are provided by community based health insurance schemes at health centre includes vaccination, consultation with doctors, maternity care, nursing care, medication, physiotherapy, dental care minor surgery, laboratory analyses, radiology and scanning and transportation to hospital11. However, for more complex hospital treatment, members can receive a package of additional services and pay a contribution of 10 per cent of the total cost of the service12. Community based health insurance covers all drugs in all health centres and hospitals. In case the drugs are not available in hospital, the patient will be required to buy from private pharmacies.

3.5 Health financing in RwandaReconciling access to health care and resource mobilisation to cover the expenses is a challenge in Rwanda (NISR and Macro International 2008). It is reported that, in 2008 over five million new outpatient consultations were registered, whereby the principal causes of outpatient visits were malaria, pulmonary infections and other related diseases (MoH 2009). In an effort to improve practices in health financing system, Rwanda has developed over the last few years an inclusive financing framework comprising two main channels for health financing from supply and demand side (Ibid). The Rwandan health sector is mainly financed by government support from the budget, assistance from donors and out of pocket payments made by people paying for health care services (NISR and Macro International 2008). There has been a considerable increase in health expenditures for the last years. According to NHA 2006, the shares of health expenditures as percentage of GDP have increased from 2.5% in 1998 to about 6.8% in 2006, but the country is still facing low level of public domestic expenditures challenges, as it is still dependent much on external support to finance health sector (MoH 2006, 2009). The health sector financing dependence on external support raises concerns on financial sustainability of the sector, as it is reported that the shares of external assistance in health expenditures have increased to 53% in 2006 compared to 42% in 2003 (MoH 2009). In an effort to address this challenge, Rwanda embarked on strengthening

11 Ibid12 Ibid.

22

internal resources mobilisation, by increasing government allocation to health and increasing prepayments into health insurance schemes based on individuals ability to pay (Ibid).

23

Chapter 4Data analysis and empirical strategy

4.1 IntroductionThis chapter presents the description and analysis of data that was used in this research, in order to estimate the economic effects of community based health insurance. It starts with the description of both quantitative and qualitative data and it ends with model specification and econometric concerns.

4.2 DataThis study used quantitative data from the Integrated Household Living Condition Survey, conducted in Rwanda in 2005-2006 (EICV2)13 to analyse the economic effects of Community based health insurance. In addition to quantitative data, qualitative data were used to explore in depth the Community based health insurance in Rwanda and get richer understanding of the full range of opinions and experiences on this topic.

4.2.1 Quantitative dataThe quantitative data used in this study were from EICV2, conducted from November 2005 to December 2006 by National institute of statistics of Rwanda. It was a national representative survey, whereby the sample selection was based on a stratified two-stage sample design using the 2002 Rwanda Census frame14. Rwanda’s capital city Kigali was allocated a large sample compared to other area because of its diversity in terms of socioeconomic characteristics. This survey gathered data from around 6

13 EICV2 : Enquête Intégrale sur les Conditions de vie des ménages 2.14 For detail see Megill David J. (2004) Recommandations on

Sample Design and Estimation Methodology for the Rwanda Enquête Intégrale sur les Conditions de Vie des Ménages 2005 OPM available athttp://imisrwanda.gov.rw/redatam/RpHelp/EICV/

survey0/data/docs/sampling/Samples%20Megill%20Mission%201.pdf (accessed on 20 October 2011).

24

900 households and 34 000 individuals. The collected information was in relation to households’ consumption expenditures on education, food and non-food items and out-of-pocket health expenditures which include hospitalization, consultation, medical tests, health exams, laboratory tests; and medication and supplies costs, transportation related to health care and source of income such as farm land and livestock. The individual collected information includes socio-economic indicators and health insurance status, self-reported health need and utilization of services. It is worth noting that, the questions related to community based health insurance membership as well as other health insurance schemes were captured in the survey, whereby all respondents provided their information on nature of their health insurance membership. In addition, in order to improve the reliability of health expenditures, two different recall periods were used, 12 months for inpatient health expenditures and 2 weeks for outpatient health expenditures (Saksena et al. 2011).

4.2.2 Qualitative dataThe use of qualitative data in assessing the effects of community based health insurance is crucial, considering that health insurance programmes involve social-economic issues that need to be explored in deep. Qualitative data helped to get richer understanding of the full range of opinions and experiences on community based health insurance in Rwanda. It is in this regard that this paper collected qualitative information to supplement the existing quantitative data. The interviews focused mostly on the following issues: (i) Community based health insurance role in reducing the financial cost of illness and its sustainability (ii) Out of pocket health payment effects on household consumption (iii) untreated sickness consequences on earnings and CBHI financing issues

4.3 Descriptive Analysis

4.3.1 Summary statistics of variables included in the analysis

Table 3.2 highlights the mean and standard deviation of outcome and control variables used. The table reveals that the family size is 5.9 and 6.2 respectively non-insured and insured members, while the level of education appears to be the same between two groups. The mean value of non-

25

insured male head of household is higher than those who are insured. The table reveals also that the rate of insured individuals who have livestock and farm land is higher than those who are not insured. School dropout appears to be high among non-insured member compared to those who are insured, while education expenditures are higher among insured people compared to those who are not insured. Surprisingly, non-insured members appear to spend high on non-food consumptions compared to insured members. However, in order to see whether the differences between key variables are statistically significant we performed T-tests and their mean differences are detailed in table 3.3. In addition the summary statistics is on the appendices table A.1, where the means and standards of all variables are shown.

Table 3.2: Descriptive statistics EICV2, Insured and insured members

VariablesNoninsured members Insured members

Number of Observa-tion

Mean Std. Dev. Number of observa-tion

Mean Std. Dev.

Food consumption 19 783 239348.5 356733.8 12 671 238901.4 356816.8

Non-food consumption 19 783 310313.9 1352263 12 671 280167.6 669466.7

Health Expenditures 19 783 6665.34 38876.64 12 671 4223.723 24629.69

Education expenditures 19 783 35369.7 123195.7 12671 37101.58 105505.4

School dropout 17 971 0.144288 0.35190651 11 576 0.122515 0.3390651

Sex 19 783 0.4739423 0.4993332 12 671 0.4743114 0.4993594

Age 19 783 21.09159 17.54438 12 671 21.73404 18.23159

Illness 19 783 0.2049338 0.4036634 12 671 0.1799384 0.3841513

Farm Land 19 783 0.8900066 0.3128895 12 671 0.9358377 0.2450514

Live stock 19 783 0.7008543 0.4578954 12 671 0.8113803 0.3912216

head of household age 19 783 44.94197 14.26739 12 671 45.33675 13.54134

Primary 19 783 0.558993 0.4965202 12 670 0.5667719 0.495541

Vocational 19 783 0.0152664 0.1226136 12 670 0.0213891 0.1446833

Secondary 19 783 0.0457992 0.2090547 12 670 0.0575375 0.2328759

26

University 19 783 0.0046001 0.0676699 12 670 0.0048145 0.0692223

Household size 19 783 5.943285 2.407025 12 671 6.275985 2.387383

Male head of household 19783 0.8117749 0.3909072 12 671 o.733913 0.4419215

Urban 19 783 0.250417 0.4332642 12 671 0.1575251 0.3643095

Source: Author’s own computation from the Household living condition survey EICV2 2005-2006.

4.3.2 Socio-economic differences between participants and non-participants

Table 3.3 presents the means differences of socio economic characteristics of community based health insurance beneficiaries and non-beneficiaries members for 2005/2006 EICV. The T-test results reveal that, on average, food expenditures between the two group was almost equal, while non-food consumption was slightly different, where non-insured members spent more on non-food compared to insured member. The mean health expenditures were high among non-insured members compared to insured members. The table also reveals that, among the people who reported sick, the mean of non-insured members is higher than that of insured people. This may be due the fact that health insurance increases access to health care, which improves subsequently health status. On average insured households have large family size compared to non-insured. The table also reveals that education expenditures of insured people were higher than that of non-insured people; this may be due to the fact that non-insured members may pull their children out of school as coping mechanism when they incur catastrophic health expenditures. Given that there are differences in mean values between insured and insured members on key variables, the use of PSM seems to lead to efficient estimates compared to OLS. However, the latter was used for robustness check purpose.

Table 3.3: Social-economic differences between insured and non-insured members

Variables Treatment Control Difference(t-test)

Farm land 0.9358377 0.8900066 0.0458312***Livestock 0.8113803 0.7008543 0.110526***

Gender 1.525689 1.526058 0.0003691Age 21.73404 21.09159 0.6424446***Education expenditures 37101.58 35369.7 1731.885*Health expenditures 4223.723 6665.34 - 2441.618***Food expenditures 238901.4 239348.5 -447.1058Non-food expenditures 280167.6 310313.9 -30146.27**School dropout 0.1325152 0.144288 -0.0117725***

Poorest 0.1478179 0.2197341 -0.0719163***

27

Poor 0.1752821 0.209644 -0.0343423***Illness 0.1799384 0.2049338 -0.0249953***

Household size 6.275985 5.943285 0.3326999***Male head of household 0.8117749 0.733913 0.077862***

Head of household age 45.33675 44.94197 0.3947828No education 0.349487 0.3753412 - 0.0258542

Primary 0.5667719 0.558993 0.0077789Vocational 0.0213891 0.0152664 0.0061227***secondary 0.0575375 0.0457992 0.0117383***University 0.0048145 0.0046001 0.0002144Source: Author’s own computation from the Household living condition survey EICV2 2005/2006***, **, * indicate statistically significance at 1%, 5% and 10% levels respectively.

4.3.3 Poverty Incidence Table 3.4 gives the distribution of insured and non-insured people based on poverty status. The poverty was measured based on threshold level of consumption such that people below that were considered as poor. Based on 2006 prices of consumption goods, the absolute poverty line was set at RFW 90 000 per adult per year and extreme poverty line at RWF 63,50015. The table shows that among the non-insured people 41% were relatively rich while 59% were poor. For insured people the poverty rate was 51%, while 49% percent were poor. This indicates that being poor could not be the main reason of not joining community based health insurance schemes given that the table reveals some rich people among the non-insured people and some poor people among insured members.

Table 3.4: Poverty incidence and health insurance membership

Rich Poor Total

Population % population % Population

CBHI members 6,265 49% 6,406 51% 12,671

Non- insured 8,140 41% 11,643 59% 19,783 Source: Author’s own computation from the Household living condition survey EICV2 2005-2006.

15 For details see, Methods Used for Poverty Analysis in Rwanda Poverty

Update Note available at : http://196.44.242.24/eicv/survey0/data/docs/studies/Master%20Report.pdf : (accessed at 20 September 2011)

28

4.3.4 Head of household health insurance membership

Table 3.5 illustrates that most households surveyed were headed by male and the coverage rate among males’ head of household was 81%, while that of household headed by female was 19%. However, the table shows that among males head of household who were surveyed, only 41% were CBHI members, whereas the rate of coverage among the female head of household was 31%.

Table 3.5: CBHI distribution by sex of head of household

Between sex Within sex Total

Non-insured CBHI member Non-insured CBHI mem-ber

female 27% 19% 69% 31% 100%

Male 73% 81% 59% 41% 100%

Total 100% 100%

Source: Author’s own computation from the Household living condition survey EICV2 2005-2006.

4.3.5 Age groupThe graph below shows the distribution of age of head of household, whereby the majority of people surveyed was ranged between 25 and 54 years old. This indicates that most people are economically active, such that any health shock that affects them, as breadwinner, it will affect household consumptions expenditures patterns.

Figure 4.1: Head of household age distribution

29

15-24 25-34 35-44 45-54 55-64 65 -980

1000

2000

3000

4000

5000

6000

CBHI membersCBHI Non-memberS

Age

Freq

uence

Source: Author’s calculation from the EICV2 2005-2006 dataset

4.3.6 Health insurance Membership Health insurance information, are captured on individual basis because enrolment in the schemes is done by person. The actual membership status for the period when the survey was conducted is shown in table 3.6. It reports that out of 34 661 people, 36.6% were insured under CBHI, 57, 08% were not insured, while 6, 36 % of people are insured with the remaining insurance schemes. However, for the purpose of this paper, population covered by other health insurance schemes, are not concerned with this study since, the main focus is on community based health insurance. People insured with CBHI serve as treatment group while people without insurance are considered as control group.

Table 3.6: Percentage of prevalence of health insurance

Beneficiary of Health insurance Insured members %

Rwandan National Insurance (RAMA) 944 2.7%

Community based health insurance (CBHI) 12,671 36.6%

Employer 126 0.4%

Other insurance 1,137 3.3%

Without any insurance 19,783 57.1%

Total 34,661 100%Source: Author’s own computation from the Household living condition survey EICV2 2005-2006.

4.3.7 Health problems incidence Table 3.7 reveals that one-third of illnesses that immobi-lized households were attributed to malaria. This may be due to the great incidence of malaria in the country. Over

30

all people reported sick, 72% were seriously ill and 28 % cases were reported less severe.

Table 3.7: Percentage of people reporting illness

Illness Serious illness Not severe

Malaria 0.30 0.04Intestinal parasites 0.14 0.07

Respiratory infection 0.11 0.08

Skin disease 0.02 0.02

Accident or injury 0.02 0.01

Diarrhea 0.01 0.00

Dental problem 0.01 0.01

Gynecological problem 0.01 0.00

Other 0.10 0.05

Total 0.72 0.28

Source: Author’s own computation from the Household living condition survey EICV2 2005-2006.

4.4 Qualitative sampleQualitative data were collected through individual interview, and focused group discussions from two different districts, using household as sample unit. Households from the two districts were selected for capturing disparities related to the household location and behaviour. The survey was conducted in Gasabo district and Bugesera16, whereby a purposive sampling technique was used, to select 15 households for semi-structured interviews. The interviews focused on households that faced health problems for the last 3 months. In order to get such information we consulted hospitals and health centres from two districts where ill persons were treated, and used a list of patients and their addresses so as to get their location easily and selected people located in the area of our interest. The selection of people to participate in interview was not random, as we selected people who can easily be reached, and whom we assumed have good understanding of the program. Along with people from the two districts, managers from the Ministry of Health in charge of CBHI were interviewed so as to provide general information about the organization and management, the background and other relevant information related to the sustainability of the schemes as they have sufficient information. The

16 Gasabo District is located in capital city while Bugesera is situated in Eastern province which is a rural area setting.

31

semi-structured approach was used to capture necessary information. In order, to get more information about health problems, the role of CBHI and its sustainability based on a group opinion and contrasting experiences from different interviewees, we conducted 2 focused group discussions, made of seven to ten people. People, who participated in the discussion, were from households that experienced health problems, and were made of people with and without community based health insurance. In order to capture the viewpoints of people who did not seek health care in the wake of sickness, we selected some households based on information provided by people in charge of health problems awareness at grass-root level.

4.5 Model specification and econometrics concerns

4.5.1 Model specification: PSM model This research paper used quasi-experimental impact evaluation techniques for estimating the impact of community based health insurance on household consumption expenditures and poverty reduction. However, this analysis is not based on a fully randomised experiment, as the data used were produced from a field survey, implying that community based health insurance membership is not likely to be completely random. There is a possibility of self-selection into community based health insurance such that establishing its effects on outcome variables may produce biased results. Based on the nature of subscription fees, flat premiums, rich people may more easily join community based health insurance than poor people. Moreover, people with an acute illness or with pre-existing conditions may have a higher incentive to enrol in CBHI than healthy people, raising adverse selection issue. People from regions where malaria prevalence is high may be more likely to join community based health insurance compared to those from regions where malaria is not prevalent. Households with more children may also be more inclined to join health insurance schemes, because the prevalence of diseases among children is relatively high compared to adults. There are also other unobserved factors that may influence people to join health insurance schemes. One of the unobserved factors that may increase CBHI membership in case of Rwanda is the pressure from local authorities who have signed performance contracts for increasing membership overtime.

32

In addition, community based health insurance targets generally poor people who work in informal sector and rural famers who do not have regular income. Any health shock that affects these people in absence of health insurance may have a large impact on their living conditions. It is therefore likely that these factors may bias the results, while setting up the effects of community based health insurance on outcome variables. However, it is worth noting that, in order to avoid adverse selection and moral hazard issues, community based health insurance membership is made at household level such that , for any household member to be entitled to get medical care service the entire family have to fully pay their premiums. This will reduce the chances for households to select only vulnerable individuals or people with pre-existing conditions to join the schemes. In addition, for any subscriber to benefit from insurance schemes, the premiums have to be paid three months prior. This also reduces chances for ill persons to join community based health insurance after realizing they are experiencing an illness problem. Given that all households are not able to pay for themselves the subscription fees, the Government in partnership with donors subsidize the poorest segment of people. This makes therefore, community based health insurance to serve not only rich people who can easily pay premiums, but also poor people have access to medical care services.

This paper uses Propensity Score Matching (PSM) to estimate the effects of community based health insurance on household’s consumption expenditures and poverty reduction considering the estimation issues highlighted in the previous sections. It is worth noting that, though matching estimator cannot deal with bias through unobservables, it has the potential to deal with selection biases originating from observed covariates (Shimeles 2010). However, based on the facts that ill people unlikely select themselves into CBHI, and that household cannot select any household member like children, women or other vulnerable person to join schemes; it is therefore realistic that the selection criteria that may introduce bias can be observed and controlled for. PSM allows estimating CBHI effects by pairing insured members with a group of non-insured members that have similar characteristics in the probability of participating in health insurance program. It is further argued that bias in PSM program estimate can be also minimised, especially when the data

33

used to estimate the effects are from the same source and when the sample of non-participants who are eligible is big (Heckman, Ichimura, and Todd as cited in Khandker et al. 2009). This study uses data from the same source which have a big number of eligible nonparticipants, therefore we expect bias from our PSM estimates to be minimised. However, it is argued that when the number of characteristics to be used in the match increases, the chances of finding matches reduces (Bryson et al. 2002).

4.5.2 CBHI effects on household consumption and impoverishment: Constructing the Counterfactual

For community based health insurance effects to be credible, treatment group (insured members) need to be accurately compared to control group (non-insured members). In this study, the effects are estimated by the difference between observed outcome of insured members and the outcome of the constructed counterfactual. The t-statistics reveal that insured members and non-insured present differences, which is not only related health insurance membership status but also on set of covariates. Therefore, based on observable variables, insured households are matched with non-insured households such that insurance membership variable will be the only difference between the two groups. However, given that we can only observe E(Y 1|M=1 , X i ) and not E(Y 0|M=1 , X i) outcomes, the counterfactual is estimated by using control group conditional on a set of observable covariates X not affected by CBHI. Underlining this, Heckman et al (1998:264) stated that: “in the absence of data from an ideal social experiment, the outcome of self-selected nonparticipants E (Y0 | D =0, X) is often used to approximate E (Y0|D =1, X)”17. However, the difference between E(Y 0|M=1 , X i)-E(Y 0|M=0 , X i )

is not necessarily equal to zero; which may raise selection bias problem for our estimation. To deal with this issue we called upon conditional independence assumption (CIA)

17 In Heckman et al (1998) paper , D stands for program participation, while in this paper, the CBHI membership is represented by M, where M= 1 when household member is insured and 0 otherwise. Therefore, referring to Heckman et al (1998), M was used as compared to D.

34

(unconfoundedness) developed by Rosenbaum and Rubin in1983 (Guo et al. 2006, Heckman et al. 1997), expressing

that [Y 0 , Y 1 ]⊥ M|X i , where a set of observables variables X were matched on such that the distributions of X are identical among treated and untreated. These variables are not affected by community based health insurance. However, because of big number of conditioning variables, it is practically difficult to match estimator (Diaz and Handa 2004). Taking into account such a problem, Rosenbaum and Rubin studies (as cited in Dehejia and Wahba 2002), advise using propensity score, which is in this case, the probability of participating into community based health insurance conditional on X covariates. We assume that p(xi) is the probability of participating in community based health insurance, which can be expressed p(xi) : Pr (M=1|X i )= E(M|X i). Then, if the outcome variables are orthogonal to

CBHI conditional on X[Y 0 , Y 1 ]⊥ M|X i , this will apply also forp( X i )); [Y 0 , Y 1 ]⊥ M|p( X i ) .

In estimating the probit model, we obtain Propensity Score

(PS) as predicted probability of participating in community

based health insurance. Then, matching insured members with no-insured members is performed on basis of that PS. Moreover, in matching insured and non-insured households, common support condition is needed for ensuring that there is sufficient overlap in the characteristics of insured

and non-insured members in order to obtain adequate matches. When estimating the effects of community based health insurance using Propensity score matching, we con-sider the mean difference in outcome over the common sup-port. Here, insured members have to be similar to non-in-sured members in terms of probability of participating in CBHI; hence, some units from non-insured people will be dropped to ensure comparability (Khandker et al.2009). Given that the data used for comparison between in-sured and non-insured households derive from the same source, this guarantee that measures that were applied for both outcome and control variables are the same. In addi-tion, based on variables available in our dataset used for matching, we consider that non-insured households have the same outcomes that insured households would have had in the absence of community based health insurance. The effects of the program will be the mean difference in out-come between insured and non-insured members:

35

ATT=E(Y 1−Y 0|M=1 , X i )=E(Y 1|M=1, X i )−E (Y 0|M=1 , X1 ). .. .. . .. .. . .. ... (1)

WhereATT : is Average Treatment for treated group Y 1 is the outcome of CBHI insured households, Y 0 is the outcome of individual who are not insured with CBHI, M=1 if household is member of CHBI and Xi: is socioeconomic characteristics. The use of PSM in estimating CBHI effects on outcome variables involves different steps. The probit model is firstly estimated to give propensity score as predicted probability of participating in community based health insurance whereby the matching between insured and non-insured is performed on basis of this PS. We test the balancing properties in order to see whether the mean values of observations are not different after matching and introduce identifying assumptions. The next section analyses, the probit model of participating into CBH.

4.5.3 Community based health insurance participation model

In order to consider the determinants of participating in community based health insurance, we estimated a probit model which is adjusted following equation developed by (Jütting 2005).

mi=f ( wi , X i , H i , E i ,u i)…………………………………… (2)Where, the probability of household i to participate in community based health insurance (mi ) depends on household wealth (w i ), socio-demographic characteristics of households (X i ) such as age, household size, gender, head of household sex, age of head of household, education level, a health status term H i which takes 1 if household member was sick and 0 otherwise, a community characteristics (Ei ) that considers districts of residence and the error term (ui

) reflecting unobserved characteristics that affect mi . The estimation of probability of participating in community based health insurance was performed using the following probit model: mi¿=βw i+φX i+δH +γE i+u............................................... (3)

36

Where a household i is mi= 1 if mi¿ > 0, implying that household i is community based health insurance member, while mi= 0 otherwise.

4.5.4 Choice of variables for PSM Matching strategy is built on conditional independence assumption such that it includes only variables that are not affected by community based health insurance (Heinrich et al. 2010). Based on this, our analysis takes exogenous covariates which cannot be affected by community based health insurance. In selecting variables to be included in the model, this paper considered exogenous variables that are likely to affect community based health insurance.

In our model, we controlled socio-economic characteristics such as gender, age, age of head of household, sex of head of household, household size and educational level. These variables are likely to affect the community based health insurance uptake. In most cases females may seek medical care than males, especially for maternal and related care, that is why they more likely to join CBHI, the age also may affect insurance in the sense that, households with children may decide to enrol in insurance because, the morbidity level is high among children than among adult people, which will increase the health problems related costs. Education is also considered essential in health insurance registration in the sense that, the level of awareness among educated people is higher than that of uneducated ones. Considering the fact that most people in Rwanda live in rural area and are subsistence farmers, we selected variables farm land and livestock as affecting the program, since health insurance involves premiums. These variables are considered as wealth or assets that can allow poor people to get insured under the schemes. The location may also influence people’s joining health insurance schemes such that people living in district where the level of morbidity is high or where health facilities are good, are more likely to uptake health insurance than people with poor health facilities. This was the reason why district dummies were included in our model specification.

However, it is worth noting that there are also other unobserved factors that may influence people to join health insurance schemes. As mentioned in the previous section, there is a possibility that people may select themselves to join CBHI, which may bias the results. Membership may for

37

instance be driven by sickness, where an ill person or people having pre-existing conditions may easily join the schemes avoiding the risks associated with medical care cost. This will raise self-selection issues. In Rwandan context, people from areas where malaria prevalence is high may join highly CHBI than those from other regions. As discussed earlier, for any household member to be entitled community to get medical care services the entire households needs to fully pay their premiums, which reduce the chance for households to select only vulnerable or sick people to join the schemes. In addition, to be entitled to medical services, a subscriber needs to pay premiums three months before. This implies that the selection criteria that may introduce bias can be observed and controlled for.

4.5.5 Household Consumption response to health problems

(OLS Model) The OLS model is used in a bid to compare its results with those from PSM model for robustness checks. From the literature review, it is argued that health problems disrupt households’ consumption in the absence health insurance. The following model is performed in order to analyse CBHI the effects on household consumption resulting from health problem and associated costs.

C i=α+λM i+βw i+φX i+δH i+γEi+εt......................... (4)

This regression includes dependent variable C i which refers to consumption of the household i, M i which is a binary variable that takes value 1 if any member of household i is insured with community based health insurance and 0 oth-erwise, a wealth dummy w i , taking into account if house-hold possess land and livestock or not and X i the household characteristics including, age and gender, household size, age of head household, head of household sex, a dummy re-lated to highest level of education attained, a health status term H i which takes 1 if household member was sick and 0 otherwise, a dummy Ei which considers the districts of residence of household member i and the error term ε t .

38

The estimation will be performed using separately food con-sumption from nonfood consumption education expendi-tures and OOP health expenditures model. The model tries to control variables that can influence households’ con-sumptions and is performed for robustness check purpose.

39

Chapter 5Results Discussion

5.1 IntroductionAs discussed in the literature and analytical framework, health problems and associated costs disrupt household consumptions and push people into poverty in the absence of insurance. This chapter discusses the findings of different estimations of the Community based health insurance’s effects on household’s food and non-food consumption, education expenditures, school dropout and out of pocket health expenditures. Section one presents CBHI determinants using Probit estimation, section two reports the empirical results using PSM while the last section discusses the robust check and sensitivity analysis using different estimation techniques.

5.2 Participation Characterizing Model

As mentioned in the previous section, identification of variables which determine community based health insurance intake is vital. However, determining a comprehensive list of relevant variable is not possible in most cases. In an attempt to provide an unbiased effect, we selected a set of variables that may influence participation in community based health insurance and estimated the probability of being covered.

Table 5.1 presents the marginal effects of probit model of determinants for participating in community based health insurance in Rwanda. The results show that people who own farm lands are 3% more likely to be insured, while people with livestock are 10% more likely to be insured than households without it. People with high level of education are more likely to join community based health insurance than people with no education, while household size increases the probability of joining insurance schemes by 1%. Households headed by male are 9% more likely to be insured compared to household headed by female. Geographical location influences the probability of getting health insurance. In this case, people from Rutsiro district are 27 % more likely to be insured compared to people from Nyarugenge district (reference district). However, illness variable is found to reduce the probability of joining CBHI.

40

This is mainly due to CBHI subscription rules, whereby subscription fees are to be paid three months before getting benefits which reduces chances for ill persons to join community based health insurance after realizing they are experiencing an illness problem. This applies mainly in case of illness that is not chronic as it is the case for our illness variable.

Table 5.1: Probit Model of determinants for Participation in CBHI (Marginal effects)

variables  Coeffi-cients

Std. Err. P>z Variables  Coeffi-cients

Std. Err. P>z

Age 0.0005*** (0.0002) 0.001 Kamonyi 0.0445* (0.0250) 0.075

Gender 0.0108* (0.0056) 0.051 Karongi 0.1379*** (0.0207) 0.000

Householdsize 0.0072*** (0.0013) 0.000 Rutsiro 0.2741*** (0.0219) 0.000

Livestock 0.1052*** (0.0068) 0.000 Rubavu 0.1482*** (0.0219) 0.000

Farm Land 0.0345*** (0.0118) 0.004 Nyabihu 0.2301*** (0.0229) 0.000

illness -0.0207*** 90.0070)

0.003 Ngororero 0.1443*** (0.0224) 0.000

Male head of Household -0.0462*** (0.0036) 0.000 Rusizi 0.0696*** (0.0206) 0.001

Head of household age -0.0001* (0.0002) 0.507 Nyamasheke 0.1568*** (0.0197) 0.000

No education: reference

Primary 0.0185*** (0.0060) 0.002 Rulindo 0.2584*** (0.0216) 0.000

Convetional 0.1313*** (0.0221) 0.000 Gakenke 0.2124*** (0.0224) 0.000

Secondary 0.1166*** (0.0140) 0.000 Musanze 0.0115 (0.0222) 0.604

University 0.1384*** (0.0430) 0.001 Burera 0.1565*** (0.0229) 0.000

Nyaruenge: Reference

Gasabo 0.0013 (0.0184) 0.944 Gicumbi 0.1493*** (0.0203) 0.000

Kicukiro -0.0022 (0.0203) 0.914 Rwamagana 0.1541*** (0.0235) 0.000

Nyanza -0.0365 (0.0223) 0.103 Nyagatare 0.1295*** (0.0199) 0.000

Gisagara 0.0176 (0.0234) 0.452 Gatsibo 0.1904*** (0.0211) 0.000

Nyaruguru -0.0531* (0.0218) 0.015 Kayonza -0.0179 (0.0225) 0.425

Huye -0.0163 (0.0203) 0.423 Kirehe 0.0104 (0.0243) 0.668

Nyamagabe 0.0092 (0.0204) 0.652 Ngoma 0.0868*** (0.0219) 0.000

Ruhango 0.0082 (0.0231) 0.722 Bugesera 0.0790*** (0.0238) 0.001

Muhanga 0.1081*** (0.0220) 0.000

Observations:Pseudo R-square

32 4510.433

32 4510.433

Source: Author’s own computation from the Household living condition survey EICV2 2005-2006. Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

41

5.3 Balancing property This test was performed to check whether after matching, the mean values of covariates are equal between CBHI members (treatment group) and non-insured (control group) people for matched group. Tables A.2 and A.3 presents results on balancing property tests, and they show that after matching the differences are no longer statistically significant; which implies that the matching have reduced the bias associated with observable covariates18 because the balancing property test is fulfilled.

5.4 Effects of CBHI on household consumption and poverty (NNM-PSM)

This paper used NNM-PSM to estimate the effects of community based health insurance between treated and control group. As specified, the dependent variables are food and Non-food, health expenditures, education expenditures, school dropout and poverty incidence. The determination of poverty incidence was based on household consumption expenditures which were expressed in terms of Rwandan francs per year. In addition, the findings from variables of interest are presented in table 5.2 which highlights the CBHI effects for two groups (people who reported sick and the entire population surveyed). Table 5.2 summarises, the main findings from NNM-PSM model:

Table 5.2: Community based health insurance Effects (PSM)

People who reported sick Full sample

FIVE Nearest Neighbor FIVE Nearest Neighbor

ATT ATT

Outcome Variables Treated Controls Difference T-stat Obs. Treated Controls Difference T-stat Obs.

(1) (2) (1) - (2) (1) (2) (1)-(2)

Food consump-tion

226472.7 199624.5 26848 3.05 6 334 238714.8 221654 17061 3.74 32 451

(8811) (4565)

Non-food-con-sumption

272193.6 260014.4 12179 0.33 6 334 279671.3 301215.9 -21545 -1.49 32 451

(36909) (14420)

Poorest 0.1304 0.2034 -0.073 -6.43 6 334 0.148 0.21 -0.062 -12.4 32451

(0.0113) (0.005)

18 For detail see Heinrich, C., A. Maffioli and G. Vázquez (2010) 'A Primer for Applying Propensity-Score Matching', SPD Working Papers.

42

Poor 0.1612 0.2097 -0.0486 -4.15 6 334 0.1754 0.2161 -0.0407 -7.96 32 451

(0.0117) (0.0051)

Education Expen-ditures

29915.3 22973.1 6942 2.65 6 334 36705.8 34743.6 1962 1.36 32 451

(2621) 6 334 (1439)

Health expendi-tures

6953.7 10515.3 -3562 -3.03 6 334 4225.3 5736 -1511 -3.69 32 451

(1177) (409.8)

School dropout 0.12985 0.15254 -0.0227 -2.08 5 628 0.1325 0.1426 -0.0101 -2.14 29 545

(0.0109) (0.0047)

Source: Author’s own computation from the Household living condition survey EICV2 2005-2006, standard errors in parentheses.

Table 5.2 Indicates the effect of the community based health insurance on household food and nonfood consumption, education expenditures, out of pocket health payment, school dropout and impoverishment effects. The results estimated using the PSM five nearest neighbor matching (NNM) suggest that food consumption for insured households was not disrupted compared with non-insured people. The estimated results show that CBHI members spent around RWF 26 848 (2 237 per month) and RWF 17 061 (1422 per month) on food, respectively insured member who reported sick and insured member from full sample, compared to non-insured members. The results are statistically significant and the difference among both insured and non-insured members is attributed to community based health insurance since households are expected to have similar characteristics after covariates are matched. The mean difference is higher for the sub group of people who reported sick compared to mean difference from the full sample. This may be due to the direct effects of health problems and associated costs that disrupt household consumption in the absence of health insurance, when reported sick. Non-food consumption presents no difference among two groups as the differences are not statistically significant. This may be to the fact that it is made of items which are not used on daily basis; whereby any health problem that affects households for short period cannot affect substantially non-food consumption.

Moreover, the findings confirm that community based health insurance has prevented insured people from incurring excessive health expenditures compared to non-insured people. Insured people who reported sick paid on average, RWF 3 562 less than non-insured people for medical care, while the results from the full sample show that insured people paid RWF 1 511 less than non-insured. Out of pocket health payments constitute one of channels

43

through which health problems affects poor people. Therefore, extending health insurance to this category of people will prevent them from running financial risks resulting from paying catastrophic health expenditures and will enable them to smooth their consumption. This is consistent with, Dekker and Wilms (2010) Chankova et al. (2008) Asfaw and Jütting (2007) findings, stressing the effects of community based health insurance in preventing catastrophic health expenditures for insured member in developing countries.

Furthermore, the results from matching estimator sug-gest that insured households spent much on education than non-insured members for people who reported sick. How-ever, for the full sample results show no significant differ-ence among treated and control group. This adjustment mechanism involves also school dropout, whereby poor households may cope with health problems and associated costs by dropping their children out of school for reducing education expenditures or for taking care of sick person or for child labor purpose. The findings reveal that, the school dropout was 2% higher among the non-insured compared to insured members for people who reported sick, while the results from full sample shows a 1% high among non-in-sured members compared to insured members. Community based health insurance therefore appears as potential tool in improving human capital development for poor people. This is consistent with qualitative results, where some peo-ple reported that with community based health insurance the health status improved, and secured income for alterna-tive uses, including investment in human capital. Scheil-Adlung et al. (2006) revealed also that good health determ-ine economic growth through improvement of labour pro-ductivity and education.

Community based health insurance constitutes an im-portant tool of preventing people from impoverishment re-sulting from health problems associated costs. In Rwandan case, the matching estimator reveals the differences among insured and non-insured members in terms of poverty inci-dence. For the poorest segment of people, the results re-veal that the rate of poverty among non-insured people was 7% higher compared to insured people for households that reported sick and 6% higher for the full sample than in-sured people. This large effects among the poorest is not only due to health spending made out of pocket, but also to loss of income due illness. Given that most people under this category depend on irregular income and are casual workers, any health problem that affects them will have

44

severe consequence on their consumption patterns. On the other side, for middle-poor people, the matching estimator reveal that the mean of non-insured members was 4% higher compared to that of insured people for both full sam-ple and people who were reported sick. This implies that community based health insurance prevented people from incurring catastrophic health expenditures, which are con-sidered as one of the channels through which people are deeply pushed into poverty. Therefore, the results suggest that, if community based health insurance in Rwanda was extended to non- insured households, poverty should have dropped by 4 % among non-insured middle-poor people, 7% for the poorest people who reported sick and 6% for the full sample. This is consistent with McIntyre et al. (2006) who stressed that out of pocket health care payment constitutes a major financial problem for poor household in developing countries such that, combined with the costs of not being able to perform normal tasks due to poor health lead to household impoverishment.

Generally, matching estimator indicates strong evidences of insuring household consumption and preventing impover-ishment because of community based health insurance in Rwanda. However, it is worth noting that there might be target bias whereby some of poorest people did not receive the subsidies from government, or reporting bias from the survey where some households may fail to recall their con-sumption expenditures. This may lead to overestimate the effects on poverty.

5.5 Robustness check and sensitivity analysis

The robustness check and sensitivity analysis were performed in order to confirm the results and give more credibility to the identification strategy. We estimated the results first, using OLS, Kernel and Radius Matching techniques and Probit to be compared with NNM-PSM results. Under NNM- PSM, unit from control group is chosen as partner for nearest unit in treated group on basis of propensity score. In addition to these different estimation techniques, we estimated the CBHI effects by including illness variable to check whether the results are sensitive to its inclusion. Finally, we re-estimated the effects for different population groups, more particularly; by rural and urban areas and by male household headed and female household headed to see impacts heterogeneity among these groups.

45

5.5.1 Effects of CBHI on household consumption (OLS)

The results of Community based health insurance effects on household food and non-food consumption, education expenditures and out of pocket payment health expenditures are reported in table 5.3 using OLS. It shows that insured people spent RWF 19473 per year (around RWF 1623 per month) higher on food consumption compared to non-insured people. In addition, insured people have spent more on education compared to non-insured members while out of pocket health expenditures was higher among non-insured people compared to insurance. This implies that expenditures on schooling, food and non-food consumption were not disrupted by health problems for insured people. The table indicates the out of pocket health payment for insured members were RWF 1542 lower compared to non-insured members, while education expenditures for CBHI beneficiaries were RWF 2250 higher compared to non-beneficiaries.

However, the results suggest that, community based health insurance effects on non-food consumption are positive but not statistically significant. Similarly food consumption is negatively related with illness, but not statistically significant. On the other hand, education level, income sources are positively associated with household consumption expenditures.Table 5.3: Community based health insurance Effects (OLS)

(1) (2) (3) (4)

VARIABLES Foodconsumption

NonfoodConsumption

EducationExpenditures

OOP Health Expenditures

Insurance 19,473*** -10,598 2,550** -1,542***(3,246) (12,627) (1,081) (344.6)

Illness -5,343 19,821 -1,947* 5,457***(3,462) (17,374) (1,132) (560.8)

Age 299.9*** 882.8*** 130.8*** -15.18(83.57) (296.0) (28.54) (9.607)

Head of Household age -738.9*** -1,218*** 259.2*** 27.21**(111.3) (307.2) (37.68) (13.01)

Male head of Householde -50,607*** -72,144*** 8,606*** 239.9(3,229) (7,520) (1,228) (500.1)

Gender 4,933 18,778 1,611 -72.62(3,179) (12,152) (1,119) (368.0)

Household size 39,513*** 80,490*** 13,307*** 788.1***(900.2) (4,020) (380.4) (96.25)

Farm Land 138,721*** 275,719*** 33,799*** 2,881**(10,672) (37,822) (4,379) (1,412)

Livestock 47,975*** 73,798*** 5,708*** 1,765***(4,026) (12,044) (1,408) (653.1)

No education: referencePrimary 26,418*** 64,916*** 7,941*** 681.0**

(2,857) (11,314) (934.7) (336.7)

46

Vocational 79,216*** 121,172*** 27,563*** 2,167(18,799) (32,842) (5,346) (1,478)

Secondary 149,576*** 362,448*** 112,786*** 3,189***(12,221) (32,525) (4,562) (1,194)

University 507,466*** 1.503e+06*** 390,800*** 52,937***(50,064) (174,565) (33,221) (15,352)

Constant 227,532*** -159,168** -79,466*** 2,167(22,234) (71,280) (8,408) (3,818)

Districts dummies: Yes Yes Yes Yes

Observations 32,451 32,451 32,451 32,451R-squared 0.383 0.120 0.284 0.046

Source: Author’s own computation from the Household living condition survey EICV2 2005-2006, standard errors in parentheses. Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

5.5.2 CBHI effects on household consumption and poverty (NNM compared to Kernel, Radius, Probit and OLS estimates)

While NNM may face bad matches’ problems, especially when the nearest neighbor is far away, the radius may involve tolerance level on the maximum distance of propensity score and Kernel method uses weighted averages of all non-participants from the control group to create the counterfactual outcome (Caliendo and Kopeinig 2008). The Ordinary Least Squares model applies functional form in estimating the impact of program.

Table 5.4 presents a summary of the results for the ef-fects of community based health insurance on different out-comes of our interest for people who reported sick. Though the magnitudes of differences differ depending on tech-niques used, the results reveal similarity for different tech-niques used in checking the robustness of the findings. All different techniques used indicate that food consumption and education expenditures among insured household were not disrupted even in the presence of health problems. In addition, the results are consistent on CBHI effects in pre-venting excessive out of pocket health payments and school dropout among insured people. The findings reveal also similarities of CBHI effects on poverty reduction for differ-ent techniques used. Insured households were protected from impoverishment effects resulting from out of pocket health expenditures payment. Given that OLS could not capture the effects on poverty and school dropout because as they include binary response, Probit model was per-formed and estimates are consistent with PSM model. Moreover, all techniques used suggest a slight difference in non-food consumption between insured household and non-insured household, even though the difference was not sta-tistically significant. In addition, table A.4 shows that the results obtained after introducing illness variable still have

47

not significantly changed with respect to food consumption, health expenditures, school dropout and poverty. This con-firms the conclusion drawn in section 5.2. Education expen-ditures were also found not significantly different between insured and non-insured members.

48

Table 5.4 : Community based health insurance Effects (People who reported sick)

FIVE Nearest Neighbor Radius Kernel OLS Probit

ATT ATT ATT

Outcome Variables Treated Controls Difference T-stat Treated Controls Difference T-stat Treated Controls Difference T-stat Coeff. Coeff.

(1) (2) (1)-(2) (1) ( 2) (1)-(2) (1) (2) (1)-(2)

Food consumption 226472.7 199624.5 26848.2 3.05 226473 192143 34330 5.38 226473 197513 28960 3.6 19473***

(8810.9) (6377) (8039) (3246)

Non-Food consumption 272193.6 260014.4 12179.2 0.33 272193.6 248192.8 24000.8 1.8 272194 271321 872.8 0.03 -10598

(36909) (13368.1) (29608) (12627)

Poorest 0.1304 0.2034 -0.0730 -6.4 0.1304 0.2234 -0.0930 -13.1 0.1303 0.2005 -0.0701 -6.8 -0.06***

(0.0113) (0.0071) (0.0102) (0.004)

Poor 0.1612 0.2097 -0.0486 -4.15 0.1612 0.2195 0.0584 -7.5 0.1612 0.2154 -0.0543 -5.1 0.04***

0.0117 (0.077) (0.011) (0.004)

Education Expenditures 29915.3 22973.1 6942.3 2.65 29915.3 21537.3 8378.1 4.4 29915.3 22367.6 7548 3.1 2550**

(2621.1) (1923.7) (2416) (1081)

OOP Health expenditures 6953.7 10515.3 -3561.7 -3.03 6953.7 10611.1 -3657.4 -6.6 6953.7 9934.3 -2981 -2.96 - 1542***

(1176.8) (551.7) (1008) (344.6)

School dropping out 0.1298 0.1525 -0.02269 -2.08 0.1298 0.1434 -0.0136 -1.81 0.12985 0.1499 -0.0201 -2.01 -0.011**

(0.01090) (0.00753) (0.0100) (0.0042)

Source: Source: Author’s own computation from the Household living condition survey EICV2 2005-2006, standard errors in parentheses *** Statistically significant at 1%, ** at 5% and * at 10% level of confidence.

41

5.5.3 The Effects of CBHI on sub-groups of population

The effects of community based health insurance on outcome variables for different groups of population are reported in table 5.5. The table reveals the CBHI effects in protecting insured households from impoverishment and consumption expenditures disruption. The effects of community based health insurance on outcome variables are highly observed among insured members living in rural areas compared to urban areas. This makes sense because diseases are predominant in rural areas than urban areas in most developing countries. In addition, people in rural areas rely mostly on physical demanding labour as source of income whereas, urban areas’ sources of income are much developed compared to rural areas. Generally, the results are consistent with conclusion drawn from our previous estimates in sections 5.4 and from other estimation strategies.

However, few remarkable observations have been made after estimating the effects. While the findings reveal that poverty rate from middle poor was 5% high among non-insured from rural areas, the matching estimator indicates no difference in urban areas. In addition, the table reveal no significant difference among insured and non-insured households headed by female for CBHI effects on OOP health expenditures and school dropout. Insured female headed households spent higher on non-food consumption compared to non-insured female headed households and the difference is statistically significant.

42

Table 5.5: CBHI effects on outcome variables for different population groups (full sample)

5 NNM- PSMMale head of household Female head of household Urban areas Rural AreasATT ATT ATT ATT

Outcome Variables Treated Controls Difference T-stat Treated Controls Difference T-stat Treated Controls Difference T-stat Treated Controls Difference T-stat

(1) (2) (1)-(2) (1) (2) (1) - (2) (1) (2) (1)-(2) (1) (2) (1)-(2)

Food consumption 25117 230199 20918 3.77 185610 158586 27024 4.61 593260 554889 38372 2.21 171076 147555 23521 8.29

(5547) (5861) (17393) (2837)

Non-food consump-tion

297773 325226 -27454 -1.47 202320 175002 27318 2.84 851206 862000 -10794 -0.23 169284 310192 -21118 -1.5

(18628) (9612) (46012) (14174)

Poorest 0.1460 0.1966 -0.0535 -5.4 0.1342 0.2461 -0.1119 -9.6 0.0247 0.0765 -0.0518 8.3 0.1711 0.2358 -0.0647 -10.8

(0.0055) (0.0110) (0.0062) (0.0132)

Poor 0.1730 0.2207 -0.0476 -8.24 0.1858 0.1976 -0.0117 -1.08 0.1003 0.2161 -0.0096 -1.05 0.1895 0.2403 -0.0508 -8.51

(0.0058) (0.0109) (0.0091) (0.0060)

Education Expendi-tures

35198 34115 1083 0.63 441991 31638 12553 4.6 107409 108751 -1342 -0.21 23249 19086 4163 4.8

(1725) (2753) (6481) (873)

OOP Health expendi-tures

3884 5970 -2086 -4.52 5702 5377 -324 -0.3 11688 15102 -3413 -1.84 28278 3689 -862 - 3.0

(1304.4) (2907.6) (1853) (286)

School dropping 0.1314 0.1433 -0.0119 -2.22 0.13774 0.13774 0.0012 0.12 0.1475 0.1609 -0.0134 -1.18 0.12940 0.13932 -0.0099 -1.91

(0.0053) (0.0012) (0.0113) (0.0052)

Source: Author’s own computation from the Household living condition survey EICV2 2005-2006, standard errors in parentheses.

43

5.6 Qualitative analysisThe findings from qualitative data reveal that, insured households consumption expenditures were not disrupted following illness episodes. The main reason given was that community based health insurance considerably prevented them from incurring catastrophic medical spending. They further argued that access to modern health care has significantly improved because of CBHI. The following story from a 30 years old man, illustrates the point:

I experienced a chronic disease since 1999 when I was 18 years old such that I had to see the doctor at least four times per year, and could spend at least RWF 90 000 per year (145 US $). Sometimes I could not afford to seek medical care because of excessive health expenditures I was incurring as compared to my earnings. As my income was running low, this disrupted my consumption expenditures such that I was always under food insecurity and could not dare to make any investment. However, with the introduction of community based health insurance, my health expenditures have drastically reduced such that the entire household can only spend RWF 7000 (12US $) per year, as a result, this allowed me to construct a house and acquire a farm for agriculture purpose. In addition to this, my health status has improved a lot as it is easy for me to meet any specialist doctor for treatment without fearing medical costs, which was not the case before (Muragijimana 2011, personal Interview).

This narrative indicates that community based health insurance has not only protected insured household from financial risk but also improved the likelihood of desired health care. By improving the access to quality health care, CBHI has played an important role in bettering health status of poor people who could not benefit from such services before. This is consistent with Saksena et al (2011) who revealed that community based health insurance schemes in Rwanda have increased the access to modern health care and have prevented insured households from incurring catastrophic health expenditures. A contrasting experience is from one respondent, a 40 years old female head of household who is not insured under community based health insurance who declared:

When I was sick, I went to see traditional healers for treatment as I could not afford modern medical care; unfortunately, I could not recover and used the savings to get basic drugs for appeasing. As a casual worker, the

44

situation was worsened by the fact that I was not able to go to work and get money to buy food for my three children who were starving; on an empty stomach, my eldest child who is 10 years old could not even attend school (Mukamrenzi 2011, personal interview).

This story indicates that non-insured household are likely to face catastrophic health expenditures and fall into poverty because of medical care payments made out of pocket or may face human capital depreciation when they are not able to pay for health care (Hodgson and Meiners 1982).

A large number among insured people who were interviewed argued that community based health insurance schemes have improved their living standards by preventing them from excessive health expenditures. Given that the majority of them are subsistence farmers and informal sector workers, they are mostly exposed to health shocks than rich people. However, with the introduction of CBHI, health status was argued to have improved as they can easily access health care. They further stressed that with CBHI, the health expenditures have substantially reduced and resources that were previously spent on health care in the absence of health insurance are now being used for other expenditures. These resources are now being allocated to children schooling, hiring labour for farming, and making savings in microfinance institutions. It is therefore noticeable from this point of view that community based health insurance plays an important role in protecting insured member from impoverishment effects resulting from health problems and associated costs, and improves the income generating capacity because of good health status (Scheil-Adlung et al. 2006).

On the other side, the respondents from non-insured members revealed that it is hard to access modern health care because of associated high costs. However, in the presence of severe illness, respondents revealed that the main coping mechanisms used are selling farm lands in most cases, goats and sheep, stored food and other assets so as to meet medical needs. In addition, households that do not own these assets are forced to drop their children out of school for work and adjustment of education expenditures. As result, they testified experiencing impoverishment effects following an episode of severe health problems. This was supported by the story of a mother of 6 children, who said:

45

My elder son experienced health problems, as we were not insured with mutual health, at each visit to health centre; we could spend a considerable amount of money for medical care expenses. The situation got worse when we were no longer able to go to a health centre after selling all assets at our disposal, my son’s health status deteriorated to the extent that he was no longer able to work. As he was the breadwinner for the household, we are no longer able to pay rent for housing and other consumption expenditures (Mukandayisenga 2011, personal interview).

The above story clearly shows that in the absence of health insurance, poor households are exposed to different risks. This may consequently lead not only to depreciation of human capital but also into poverty trap, as they are no longer able to perform any income generating activity.

On the reasons why some people are not insured despite the benefit associated with community based health insurance, some people said that it is too expensive to get the premiums for all household members at once, so as to be entitled to get benefits. Their suggestions were to make deferred payments which will enable each household member to access medical care service even before full payment. They claimed that this will also allow them to extend the payment schedule throughout the year given their income level. However, most people said that being cash constrained was not the main reasons for turn up of not joining the schemes at all. They said that unwillingness to pay premiums was key factor, because the poorest households are in most cases subsidised by government and other development partners to pay their insurance contributions. They added that the time allocated to payment of premium is adequate to get payment since people are advised five months before the starting of the year.

However, community based health insurance financial sustainability constitutes a major challenge for the schemes in Rwanda, given that they are financed by both subscription fees and subsidies from government and development partners. This concern is based on the fact that external funds may be cut abruptly and the subscription fees may not be enough to cover medical care services provided. To ensure sustainable health insurance schemes, the gap between members’ contributions and medical care services costs need to be filled. On one hand, people in charge of CBHI said that solidarity and equity of

46

contributions fees among community based health insurance members is one of the key factors for the sustainability of the schemes. Categorising people into different socioeconomic categories for contribution will not only establish equity and solidarity among members but also will improve the coverage rate which results in increasing contributions. Furthermore, the government is considering increasing taxes on consumption of tobacco and alcohol as they are considered to undermine people’s health. (MoH 2010). On the other side, by electing mobilisation committees at village level, people admitted that the turn up rate will decrease because of intensive sensitisation. By increasing also the awareness on community based health insurance benefits; many people will join the schemes, which will increase subscription fees. Given the increase of contributions resulting from enlarging the coverage rate, people believe that the health insurance schemes will be able to meet the needs of beneficiaries in a sustainable way.

In general, despite some contradicting views especially on the reasons of joining the schemes, all interviewed people agreed on the role of community based health insurance in improving the access to modern health care and preventing insured member from incurring excessive health expenditures and from using income threatening coping mechanisms. The analysis suggests that, community based health insurance has improved the household welfare by preventing them from cutting back on consumption expenditures or using other coping mechanisms so as to deal with health problems and associated costs. In addition, with CBHI, future households’ income generating capacity is not threatened, as insured poor people’s access to health care increases and they are prevented from borrowings, saving depletion, sale of assets to finance health care. Furthermore, the analysis reveals that non-insured members are exposed to different risks and are likely to remain in poverty, as any effort made by them to improve their living conditions is drawn back by health problems and associated costs which contribute to perpetual poverty.

47

48

Chapter 6Conclusion

There is a growing evidence of health problems effects on poor households particularly in developing countries. In most cases a sizeable part of health spending is made out of pocket, which not only affects household consumption expenditures but also may push them into poverty. Despite remarkable efforts in controlling health problems and associated costs by the government and development agents, this remains a major challenge in Rwanda. Community based health insurance schemes constitute potential instrument to mitigate negative effects associated with health problems.

This paper investigates the effects of community based health insurance in Rwanda on households’ food and non-food consumptions, education expenditures, school dropout and preventing households from impoverishment effects resulting from substantial medical expenditures. Based on data from integrated household living conditions survey, this paper estimates the effects of CBHI program using NNM- PSM model whereby non-insured people were used as control group and insured members as treatment group for comparison. To see whether the results from our estimation are robust, Kernel, Radius and OLS techniques were used for comparing estimates. In addition, the effects were re-estimated for different population groups to see the impact heterogeneity. Qualitative data from interviews was also applied to supplement quantitative findings.

The findings suggest that community based health insurance schemes in Rwanda have protected insured people from households food consumption and education expenditures disruption. In addition, the out of pocket health payments and school dropout were found high among non-insured people compared to insured people. Implying that insured households are cushioned from substantial medical expenses and dropping their children out of school, as their consumption level is not disrupted by health problems and associated costs. From the qualitative analysis, it was found that community based health insurance improved the access to health care for insured members and they are no longer constrained by medical expenses. This implies that insured people’s health status is likely to improve compared to non-insured as they can

49

access easily modern health care. The results also reveal that community based health insurance schemes have protected insured households from impoverishment effects resulting from substantial medical expenses. Comparing insured and non-insured members between two categories of poor people, the matching estimator reveals that the rate of poverty was higher among non-insured people as compared to insured members. Health problems and associated costs in the absence of health insurance are therefore considered as one of the main factors that drive households into poverty particularly in developing countries. From these results it is therefore useful to extend community based health insurance to all people in an effort to reducing the impoverishment effects resulting from health problems and associated costs.

Generally, both qualitative and quantitative data stressed the role of community based health insurance in improving households’ living conditions of insured members by protecting them from incurring substantial health expenditures and enabling them to maintain their consumption expenditures level even in the wake of severe illness. It is therefore suggested that any policy aiming at improving households, living conditions, especially in developing countries should incorporate health insurance and particularly community based health insurance in the poverty reduction program, particularly because the majority of people in these countries are poor and cannot afford other form of health insurance.

Nevertheless, this paper presents some limitations to be pointed out which may shed shadow on the findings. The baseline on household characteristics before community based health insurance program was implemented is not available. Therefore, by expecting strict exogeneity of covariates, this paper considers the difference between insured and non-insured households as due to community based health insurance. In addition, because of lack of information on labour supply and depletion of assets, this paper did not consider these coping mechanisms in estimating CBHI effects. It is worth noting also that this study used data from a cross sectional survey, which may present a limitation of not taking into account dynamic of health problems and other change along the time.

50

51

ReferencesAsenso-Okyere, W. and J.A. Dzator (1997) 'Household Cost of

Seeking Malaria Care. A Retrospective Study of Two Districts in Ghana', Social science & medicine 45(5): 659-667.

Asfaw, A. and J.P. Jutting (2007) 'The Role of Health Insurance in Poverty Reduction: Empirical Evidence from Senegal', International Journal of Public Administration, 30 8(9): 835-858.

Ataguba, J.E., H.E. Ichoku and W.M. Fonta (2008) 'Estimating the Willingness to Pay for Community Healthcare Insurance in Rural Nigeria', Cahiers de recherche PMMA.

Babu, B., A. Nayak, K. Dhal, A. Acharya, P. Jangid and G. Mallick (2002) 'The Economic Loss due to Treatment Costs and Work Loss to Individuals with Chronic Lymphatic Filariasis in Rural Communities of Orissa, India', Acta Tropica 82(1): 31-38.

Barnett, T., A. Whiteside and C. Desmond (2001) 'The Social and Economic Impact of HIV/AIDS in Poor Countries: A Review of Studies and Lessons', Progress in Development Studies 1(2): 151.

Barro, R.J. (1996) 'Health, Human Capital and Economic Growth', Washington, DC: Pan American Health Organization.

Bryson, A., R. Dorsett and S. Purdon (2002) 'The use of Propensity Score Matching in the Evaluation of Active Labour Market Policies'.

Caliendo, M. and S. Kopeinig (2008) 'Some Practical Guidance for the Implementation of Propensity Score Matching', Journal of economic surveys 22(1): 31-72.

Chankova, S., S. Sulzbach and F. Diop (2008) 'Impact of Mutual Health Organizations: Evidence from West Africa', Health policy and planning 23(4): 264.

Claeson, M., C. Griffin, M. Johnston, A. McLachlan, A. Soucat, A. Wagstaff et al. (2001) 'Poverty Reduction and the Health Sector', Poverty Reduction Strategy Sourcebook.

Cole, M.A. and E. Neumayer (2006) 'The Impact of Poor Health on Total Factor Productivity', Journal of Development Studies 42(6): 918-938.

Dehejia, R.H. and S. Wahba (2002) 'Propensity Score-Matching Methods for Nonexperimental Causal Studies', Review of Economics and statistics 84(1): 151-161.

Dekker, M. and A. Wilms (2010) 'Health Insurance and Other Risk-Coping Strategies in Uganda: The Case of Microcare Insurance Ltd.', World Development 38(3): 369-378.

Diaz, J.J. and S. Handa (2004) 'Estimating the Selection Bias of Matching Estimators using Experimental Data from PROGRESA', Department of Economics, University of

52

Maryland, and Department of Public Policy, University of North Caroline at Chapel Hill .

Flores, G., J. Krishnakumar, O. O'Donnell and E. Van Doorslaer (2008) 'Coping with health‐care Costs: Implications for the Measurement of Catastrophic Expenditures and Poverty', Health economics 17(12): 1393-1412.

Gertler, P. and J. Gruber (2002) ‘Insuring Consumption against Illness’, American Economic Review, Vol. 92 No.1, pp. 51-70.

Gertler, P. and J. Gruber (1997) “Insuring Consumption Against Illness.” National Bureau of Economic Research (Cambridge, MA) Working Paper No. 6035.

Government of Rwanda (2005) Rwanda National Health Sector Policy.

http://www.eac.int/health/index.php?option=com_docman&Itemid=47 (accessed on 22 October 2011).

Guo, S., R.P. Barth and C. Gibbons (2006) 'Propensity Score Matching Strategies for Evaluating Substance Abuse Services for Child Welfare Clients', Children and Youth Services Review 28(4): 357-383.

Hamid, S.A., J. Roberts and P. Mosley (2011) 'Can Micro Health Insurance Reduce Poverty? Evidence from Bangladesh', Journal of Risk and Insurance.

Heckman, J.J., H. Ichimura and P. Todd (1998) 'Matching as an Econometric Evaluation Estimator', The Review of Economic Studies 65(2): 261.

Heckman, J.J., H. Ichimura and P.E. Todd (1997) 'Matching as an Econometric Evaluation Estimator: Evidence from Evaluating a Job Training Programme', The Review of Economic Studies 64(4): 605.

Heinrich, C., A. Maffioli and G. Vázquez (2010) 'A Primer for Applying Propensity-Score Matching', SPD Working Papers .

Hodgson, T.A. and M.R. Meiners (1982) 'Cost-of-Illness Methodology: A Guide to Current Practices and Procedures', The Milbank Memorial Fund Quarterly.Health and Society : 429-462.

Jack, W. and M. Lewis (2009) Health Investments and Economic Growth: Macroeconomic Evidence and Microeconomic Foundations. World Bank.

Jütting, J.P. (2004) 'Do Community-Based Health Insurance Schemes Improve Poor People's Access to Health Care? Evidence from Rural Senegal', World Development 32(2): 273-288.

Jütting, J.P. (2005) Health Insurance for the Poor in Developing Countries: Johannes P. Jütting. Aldershot [etc.]: Ashgate.

Kagubare M., (2006) The impact of community based health insurance on health care utilisation and financial

53

sustainability: the example of Rwanda, The Johns Hopkins University 2006. 3213725.

http://search.proquest.com/docview/305317354 (accessed on 10 October 2011).

Khandker, S.R., G.B. Koolwal and H.A. Samad (2009) Handbook on Impact Evaluation: Quantitative Methods and Practices. World Bank Publications.

McIntyre, D., M. Thiede, G. Dahlgren and M. Whitehead (2006) 'What are the Economic Consequences for Households of Illness and of Paying for Health Care in Low-and Middle-Income Country Contexts?', Social science & medicine 62(4): 858-865.

Ministry of Health, Government of Rwanda (2006) National Health Accounts With HIV/AIDS, Malaria and Reproductive health subaccounts, Rwanda.

http://statistics.gov.rw/images/PDF/NHA%20Rwanda_Fin.pdf (accessed on 20 September 2011).

Ministry of health, Government of Rwanda (2009), Rwanda Health Financing Policy, Rwanda.

Ministry of Health, Government of Rwanda (2010) Rwanda community based health insurance policy, Rwanda.

http://www.ilo.org/gimi/gess/RessShowRessource.do?ressourceId=23210 (accessed on 27 September 2011).

Msuya, J. M., J. P. Jütting and A. Asfaw (2004) ‘Impacts of Community Health Insuance Schemes on Health Care Provision in Rural Tanzania'. ZEF. Germany.

Mugisha M. M. and F. Mugumya (2010) ‘Rethinking the role of the State in Health Care Financing: Can Community-Based Health Insurance Show the Way?' in Awortwi, N. and A. Okwany (eds) Issues in Social Development and Local Governance in Uganda, pp. 178-211. Maastricht: Shaker Publishing BV.

National Institute of Statistics of Rwanda and Macro International Inc (2008) Rwanda Service Provision Assessment Survey. Calverton, Maryland, U.S.A.

National Institute of Statistics of Rwanda (2008) Preliminary results of interim demographic and health survey 2007-2008, Kigali.

Preker, A.S. and G. Carrin (2004) Health Financing for Poor People: Resource Mobilization and Risk Sharing. Vol. 434. World Bank Publications.

Ranson, M.K. and K. John (2001) 'Quality of Hysterectomy Care in Rural Gujarat: The Role of Community-Based Health Insurance', Health policy and planning 16(4): 395.

Russell, S. (2004) 'The Economic Burden of Illness for Households in Developing Countries: A Review of Studies Focusing on Malaria, Tuberculosis, and Human Immunodeficiency virus/acquired Immunodeficiency

54

Syndrome', The American Journal of Tropical Medicine and Hygiene 71(2 suppl): 147.

Sachs, J.D. and World Health Organization (WHO) (2001) Macroeconomics and Health: Investing in Health for Economic Development: Report / Commission on Macroeconomics and Health, Chaired by Jeffrey D. Sachs. Geneva: WHO.

Saksena, P., A.F. Antunes, K. Xu, L. Musango and G. Carrin (2011) 'Mutual Health Insurance in Rwanda: Evidence on Access to Care and Financial Risk

Sauerborn, R., A. Adams and M. Hien (1996) 'Household Strategies to Cope with the Economic Costs of Illness', Social science & medicine 43(3): 291-301.

Scheil-Adlung, X., G. Carrin, J. Juetting, K. Xu and International Labour Office. Social Security Policy and Development Branch (2006) What is the Impact of Social Health Protection on Access to Health Care, Health Expenditure and Impoverishment?: A Comparative Analysis of Three African Countries. ILO.

Schneider, P., A K. Ndakumar, M. Bhawalkar, D. Butera and G. Barnett (1998) ' Rwanda National Health Account', Howard University International Affairs center. University Research Co., LLC.

http://www.who.int/nha/country/rwanda_1998.pdf (accessed on 9 October 2011).

Schneider, P. and F. Diop (2001) 'Synopsis of Results on the Impact of Community-Based Health Insurance on Financial Accessibility to Health Care in Rwanda', Health, Nutrition and Population (HNP) Discussion Paper.Washington, DC: The World Bank.

Schultz, T.P. and A. Tansel (1997) 'Wage and Labor Supply Effects of Illness in Côte d'Ivoire and Ghana: Instrumental Variable Estimates for Days Disabled* 1', Journal of Development Economics 53(2): 251-286.

Shimeles, A. (2010) 'Community Based Health Insurance Schemes in Africa: The Case of Rwanda', rapport nr. Working Papers in Economics 463.

Su, T.T., B. Kouyaté and S. Flessa (2006) 'Catastrophic Household Expenditure for Health Care in a Low-Income Society: A Study from Nouna District, Burkina Faso', Bulletin of the World Health Organization 84(1): 21-27.

Tabor, S. (2005) 'Community-Based Health Insurance and Social Protection Policy', World Bank.

Wagstaff, A. (2007) 'The Economic Consequences of Health Shocks: Evidence from Vietnam', Journal of health economics 26(1): 82-100.

55

Wang, H., L. Zhang and W. Hsiao (2006) 'Ill Health and its Potential Influence on Household Consumptions in Rural China', Health Policy 78(2-3): 167-177.

World Bank (1997) Confronting AIDS: Public Priorities in a Global Epidemic. Oxford University Press.

World Bank (2001) World Development Report 2000/2001. Attacking poverty. Oxford University Press.

World Health Organisation (2000) The World Health Report 2000 Health Systems: Improving performance, Geneva.

World Health Organisation (2002) Health, Economic Growth, and Poverty Reduction. The Report of Working Group I of the Commission on Macroeconomics and Health, Geneva.

56

Appendices

Table A.1 Descriptive statistics

VariableS Obs Mean Std. Dev. Variable Obs Mean Std. Dev.

Food 32578 239466 357434 Huye 32578 0.039 0.195

Nonfood 32578 299292 1135821 Nyamagabe 32578 0.042 0.201

Health expenditures 32578 5711 33996 Ruhango 32578 0.025 0.157

Education expenditures 32578 36239 116833 Muhanga 32578 0.032 0.177

school dropout 29636 0.140 0.347 Kamonyi 32578 0.021 0.142

Insurance 32454 0.390 0.488 Karongi 32578 0.043 0.202

Illness 32465 0.195 0.396 Rutsiro 32578 0.026 0.159

Age 32578 21.337 17.798 Rubavu 32578 0.030 0.172

Head of household age 32578 45.102 13.991 Nyabihu 32578 0.023 0.151

Male head of household 32578 0.764 0.424 Ngororero 32578 0.030 0.171

Gender 32578 1.526 0.499 Rusizi 32578 0.042 0.200

Household size 32578 6.074 2.410 Nyamasheke 32578 0.051 0.220

Farm Land 32578 1.092 0.289 Rulindo 32578 0.029 0.169

Livestock 32578 1.256 0.436 Gakenke 32578 0.027 0.162

Primary 32577 0.560 0.496 Musanze 32578 0.030 0.169

Vocational 32577 0.018 0.132 Burera 32578 0.028 0.164

Secondary 32577 0.050 0.218 Gicumbi 32578 0.044 0.205

University 32577 0.005 0.068 Rwamagana 32578 0.024 0.153

Nyarugenge 32578 0.046 0.209 Nyagatare 32578 0.050 0.217

Gasabo 32578 0.055 0.228 Gatsibo 32578 0.036 0.185

Kicukiro 32578 0.038 0.191 Kayonza 32578 0.028 0.165

Nyanza 32578 0.027 0.162 Kirehe 32578 0.023 0.149

Gisagara 32578 0.024 0.155 Ngoma 32578 0.034 0.180

Nyaruguru 32578 0.029 0.168 Bugesera 32578 0.025 0.155

Source: Source: Author’s own computation from the Household living condition survey EICV2 2005-2006.

57

Table A.2: Balancing properties of the matched samples (Reported sick)

Unmatched Matched

Variables Treated Controls Treated Control Difference T-stat

Age 25.4645 23.6838 25.4396 25.2563 0.1833 0.73

Head of hh age

45.5048 45.2107 45.4949 45.3528 0.1421 0.33

Male head of hh

0.8039 0.7114 0.8037 0.8100 -0.0063 -0.5

Gender 1.5561 1.5634 -0.0038 -0.0038 -0.0038 -0.25

Household size 5.8882 5.5424 5.8858 5.8848 0.0011 0.01

Farm land 0.9342 0.9038 0.9341 0.9260 0.0082 1.02

Livestock 0.7842 0.6855 0.7839 0.7809 0.0031 0.24

Primary 0.4978 0.4914 0.4985 0.4937 0.0047 0.32

Vocational 0.0250 0.0136 0.0246 0.0235 0.0011 0.27

Secondary 0.0364 0.0281 0.0356 0.0371 -0.0015 -0.28

University 0.0009 0.0017 0.0009 0.0010 -0.0001 -0.09

Gasabo 0.0404 0.0543 0.0404 0.0461 -0.0057 -0.9

Kicukiro 0.0268 0.0397 0.0268 0.0253 0.0015 0.29

Nyanza 0.0162 0.0390 0.0162 0.0155 0.0008 0.18

Gisagara 0.0175 0.0385 0.0176 0.0158 0.0018 0.39

Nyaruguru 0.0123 0.0437 0.0123 0.0127 -0.0004 -0.1

Huye 0.0355 0.0557 0.0356 0.0352 0.0004 0.06

Nyamagabe 0.0399 0.0557 0.0400 0.0394 0.0006 0.1

Ruhango 0.0272 0.0338 0.0272 0.0282 -0.0010 -0.19

Muhanga 0.0351 0.0266 0.0351 0.0351 0.0000 0.005

Kamonyi 0.0202 0.0192 0.0202 0.0205 -0.0003 -0.06

Karongi 0.0452 0.0382 0.0452 0.0475 -0.0023 -0.37

Rutsiro 0.0285 0.0094 0.0277 0.0278 -0.0001 -0.02

Rubavu 0.0342 0.0326 0.0343 0.0361 -0.0018 -0.33

Nyabihu 0.0254 0.0138 0.0255 0.0247 0.0008 0.19

Ngororero 0.0263 0.0242 0.0264 0.0285 -0.0021 -0.44

Rusizi 0.0452 0.0451 0.0452 0.0425 0.0027 0.43

Nyamagabe 0.0711 0.0493 0.0711 0.0704 0.0008 0.11

Rulindo 0.0338 0.0170 0.0334 0.0300 0.0034 0.72

Gakenke 0.0272 0.0165 0.0272 0.0281 -0.0009 -0.2

Musanze 0.0224 0.0264 0.0224 0.0243 -0.0019 -0.41

Burera 0.0175 0.0150 0.0176 0.0178 -0.0003 -0.07

Gicumbi 0.0557 0.0427 0.0558 0.0563 -0.0005 -0.08

Rwamagana 0.0254 0.0237 0.0255 0.0229 0.0025 0.54

Nyagatare 0.0671 0.0409 0.0672 0.0731 -0.0059 -0.85

Gatsibo 0.0491 0.0289 0.0492 0.0501 -0.0009 -0.15

Kayonza 0.0250 0.0333 0.0250 0.0217 0.0033 0.68

Kirehe 0.0232 0.0306 0.0233 0.0233 0.0000 0.00

Ngoma 0.0482 0.0402 0.0483 0.0442 0.0041 0.66

Bugesera 0.0268 0.0313 0.0268 0.0271 -0.0003 -0.05

Source: Author’s calculation from EICV2 dataset

58

Table A.3 Balancing properties of the matched samples (Full sample)

Unmatched Matched

Variable Treated Control Treated Control Differece T-Stat

Age 21.735 21.09 21.729 21.68 0.049 0.22

Head of hh age 45.337 44.941 45.335 45.323 0.012 0.07

Male head of HH 1.1882 1.2661 1.1884 1.1888 -0.0004 -0.07

Gender 1.5257 1.5261 1.5259 1.5258 1E-04 0.01

Household size 6.2762 5.9431 6.2698 6.2802 -0.0104 -0.34

Farm Land 1.0642 1.11 1.0642 1.0701 -0.0059 -1.86

Livestock 1.1886 1.2991 1.1888 1.1952 -0.0064 -1.28

Primary 0.56677 0.55897 0.56735 0.56434 0.00301 0.48

Vocational 0.02139 0.01527 0.02133 0.02158 -0.00025 -0.14

Secondary 0.05754 0.0458 0.05681 0.05753 -0.00072 -0.25

University 0.00481 0.0046 0.00466 0.00525 -0.00059 -0.66

Gasabo 0.04159 0.0639 0.04164 0.04184 -0.0002 -0.08

Kicukiro 0.02612 0.04565 0.02615 0.02617 -2E-05 -0.01

Nyanza 0.01902 0.03235 0.01904 0.01989 -0.00085 -0.49

Gisagara 0.02052 0.02705 0.02054 0.01787 0.00267 1.55

Nyaruguru 0.01926 0.03559 0.01928 0.02032 -0.00104 -0.6

Huye 0.02976 0.0458 0.02979 0.02938 0.00041 0.19

Nyamagabe 0.0352 0.04656 0.03524 0.03435 0.00089 0.38

Ruhango 0.02044 0.02826 0.02046 0.01842 0.00204 1.17

Muhanga 0.03552 0.03048 0.03555 0.03535 0.0002 0.09

Kamonyi 0.01886 0.02159 0.01888 0.0177 0.00118 0.7

Karongi 0.04878 0.03867 0.04883 0.04982 -0.00099 -0.37

Rutsiro 0.03875 0.01764 0.0384 0.03919 -0.00079 -0.33

Rubavu 0.0337 0.02841 0.03374 0.03565 -0.00191 -0.83

Nyabihu 0.03212 0.01795 0.03168 0.03088 0.0008 0.37

Ngororero 0.03512 0.0272 0.03516 0.03415 0.00101 0.44

Rusizi 0.04065 0.04241 0.04069 0.03955 0.00114 0.46

Nyamasheke 0.05967 0.0456 0.05973 0.06115 -0.00142 -0.47

Rulindo 0.04246 0.02078 0.04235 0.04417 -0.00182 -0.71

Gakenke 0.03662 0.02083 0.03666 0.03506 0.0016 0.68

Musanze 0.02384 0.03311 0.02386 0.02275 0.00111 0.58

Burera 0.0322 0.02447 0.03224 0.03345 -0.00121 -0.54

Gicumbi 0.05059 0.03913 0.05064 0.05091 -0.00027 -0.1

Rwamagana 0.0277 0.02159 0.02773 0.02712 0.00061 0.3

Nyagatare 0.05651 0.0453 0.05657 0.05847 -0.0019 -0.65

Gatsibo 0.04522 0.02947 0.04527 0.04575 -0.00048 -0.18

Kayonza 0.02084 0.03261 0.02086 0.02024 0.00062 0.35

Kirehe 0.01839 0.02523 0.01841 0.01778 0.00063 0.38

Ngoma 0.03394 0.03301 0.03397 0.03449 -0.00052 -0.23

Bugesera 0.02478 0.02487 0.02481 0.02343 0.00138 0.71

Illness 0.17995 0.20494 0.18006 0.17879 0.00127 0.26

Source: Author’s calculation from EICV2 dataset.

59

Table A.4: Community based health insurance Effects (OLS)

(1) (2) (3) (4)

VARIABLES Foodconsumption

NonfoodConsumption

EducationExpenditures

OOP Health Expenditures

insurance 19,473*** -10,598 2,550** -1,542***

(3,246) (12,627) (1,081) (344.6)

Illness -5,343 19,821 -1,947* 5,457***

(3,462) (17,374) (1,132) (560.8)

Age 299.9*** 882.8*** 130.8*** -15.18

(83.57) (296.0) (28.54) (9.607)

Head of HH age -738.9*** -1,218*** 259.2*** 27.21**

(111.3) (307.2) (37.68) (13.01)

Male head of Hhold -50,607*** -72,144*** 8,606*** 239.9

(3,229) (7,520) (1,228) (500.1)

Gender 4,933 18,778 1,611 -72.62

(3,179) (12,152) (1,119) (368.0)

Household size 39,513*** 80,490*** 13,307*** 788.1***

(900.2) (4,020) (380.4) (96.25)

S8BQ1 138,721*** 275,719*** 33,799*** 2,881**

(10,672) (37,822) (4,379) (1,412)

S8SA 47,975*** 73,798*** 5,708*** 1,765***

(4,026) (12,044) (1,408) (653.1)

No education: reference

Primary 26,418*** 64,916*** 7,941*** 681.0**

(2,857) (11,314) (934.7) (336.7)

Vocational 79,216*** 121,172*** 27,563*** 2,167

(18,799) (32,842) (5,346) (1,478)

Secondary 149,576*** 362,448*** 112,786*** 3,189***

(12,221) (32,525) (4,562) (1,194)

University 507,466*** 1.503e+06*** 390,800*** 52,937***

(50,064) (174,565) (33,221) (15,352)

Nyaruenge district: ReferenceGasabo -96,549*** 49,583 651.5 3,662

(17,618) (46,580) (8,091) (2,904)

Kicukiro -33,312 296,305*** 9,550 -3,085

(22,858) (65,104) (8,512) (2,678)

Nyanza -434,503*** -454,968*** -54,319*** -11,970***

(14,495) (37,310) (6,320) (2,274)

60

Gisagara -476,487*** -519,911*** -51,110*** -12,629***

(14,136) (36,666) (6,323) (2,258)

Nyaaruguru -486,469*** -551,856*** -56,017*** -10,408***

(14,062) (36,413) (6,126) (2,370)

Huye -429,632*** -484,400*** -59,976*** -11,513***

(14,673) (36,414) (6,233) (2,269)

Nyamagabe -442,686*** -522,041*** -62,875*** -12,201***

(14,566) (37,493) (6,149) (2,262)

Ruhango -439,280*** -459,827*** -58,227*** -10,011***

(14,997) (36,894) (6,106) (2,289)

Muhanga -417,094*** -512,621*** -55,485*** -10,198***

(14,530) (35,794) (6,326) (2,310)

Kamonyi -432,339*** -465,954*** -54,772*** -11,446***

(14,526) (37,250) (6,292) (2,290)

Karongi -433,048*** -141,217 -61,225*** -11,140***

(14,952) (116,305) (5,995) (2,257)

Rutsiro -449,491*** -515,546*** -65,005*** -11,568***

(14,251) (37,674) (6,134) (2,287)

Rubavu -407,409*** -521,397*** -66,298*** -4,508

(14,999) (35,626) (6,195) (2,951)

Nyabihu -449,808*** -530,899*** -46,125*** -9,312***

(14,534) (36,464) (6,936) (2,258)

Ngororero -499,219*** -553,643*** -65,160*** -12,436***

(14,115) (36,680) (6,034) (2,261)

Rusizi -401,166*** -492,879*** -59,031*** -10,393***

(19,667) (36,265) (6,256) (2,221)

Nyamasheke -447,106*** -527,424*** -58,781*** -10,565***

(14,315) (36,569) (6,203) (2,246)

Rulindo -463,539*** -499,795*** -64,975*** -11,471***

(14,463) (37,194) (6,044) (2,258)

Gakenke -503,970*** -517,042*** -56,428*** -10,614***

(14,472) (37,280) (6,684) (2,260)

Musanze -435,153*** -446,138*** -61,270*** -4,930*

(14,201) (46,715) (6,088) (2,694)

Burera -494,528*** -577,199*** -68,362*** -12,117***

(14,248) (36,680) (6,064) (2,227)

Gicumbi -472,031*** -546,141*** -67,321*** -10,896***

(14,134) (36,005) (6,138) (2,245)

Rwamagana -371,148*** -378,870*** -48,373*** -9,761***

61

(16,076) (37,806) (6,393) (2,268)

Nyagatare -452,298*** -350,533*** -60,678*** -9,946***

(14,221) (51,274) (6,087) (2,240)

Gatsibo -454,329*** -444,111*** -54,197*** -11,740***

(14,361) (37,552) (6,134) (2,243)

Kayonza -414,653*** -456,839*** -54,740*** -11,443***

(14,824) (36,623) (6,158) (2,280)

Kiehe -464,259*** -485,981*** -65,067*** -6,641***

(15,059) (37,337) (6,184) (2,522)

Ngoma -448,139*** -444,204*** -46,781*** -11,447***

(14,361) (37,011) (6,640) (2,286)

Bugesera -441,724*** -467,792*** -44,593*** -11,782***

(14,112) (36,883) (6,710) (2,258)

Constant 227,532*** -159,168** -79,466*** 2,167

(22,234) (71,280) (8,408) (3,818)

Observations 32,451 32,451 32,451 32,451

R-squared 0.383 0.120 0.284 0.046

Source: Author’s own computation from the Household living condition survey EICV2 2005-2006Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Figure A .1: Propensity Score Graph (Reported sick)

0

.125 .25 .375 .5 .625 .75 .875Propensity Score

Untreated Treated: On supportTreated: Off support

Source: Author’s analysis based on Household living condition survey EICV2 2005-2006

62

Figure A .2: Propensity Score Graph (Full sample)

0 .2 .4 .6 .8Propensity Score

Untreated Treated: On supportTreated: Off support

Source: Author’s analysis based on Household living condition survey EICV2 2005-2006

58

Table A.5: Community based health insurance Effects (Probit Model)

School dropout Poorest PoorVariables Coeff. Std. Err. P>z Coeff. Std. Err. P>z Coeff. Std. Err. P>z

Insurance -0.0106 ** 0.004 0.012 -0.0607*** 0.0041 0.000 -0.0405**** 0.00445 0.000Age -0.0001 0.000 0.334 -0.0002* 0.0001 0.098 -8.7E-05 0.00013 0.505Age of head of household 0.0001 0.000 0.741 0.0001 0.0002 0.892 0.000556**

*0.00017 0.001

Male head of Household -0.0049 0.005 0.343 0.0665*** 0.0051 0.000 0.008876 0.00551 0.107Gender -0.0011 0.004 0.794 0.0016 0.0041 0.697 -0.00057 0.00437 0.897Householdsize -0.0004 0.001 0.664 0.0181*** 0.0009 0.000 0.007854**

*0.00097 0.000

Land 0.0057 0.008 0.484 -0.0616*** 0.0108 0.000 -0.09155*** 0.0116 0.000Livestock -0.0057 0.005 0.273 0.1226*** 0.0050 0.000 -0.00373 0.00562 0.506Illness 0.0003 0.005 0.958 -0.0151 0.0050 0.52 -0.00226 0.00548 0.68No education (Reference)Primary -0.0035 0.004 0.421 -0.0287*** 0.0043 0.000 -0.01517*** 0.00463 0.001Convetional 0.0127 0.016 0.428 -0.1092*** 0.0091 0.000 -0.07151*** 0.01412 0.000Secondary 0.0021 0.010 0.828 -0.1377*** 0.0046 0.000 -0.11021*** 0.00781 0.000University 0.0127 0.029 0.665 -0.1450*** 0.0086 0.000Nyarugenge (reference)Gasabo 0.0067 0.012 0.582 0.0658** 0.0281 0.019 0.153343**

*0.02822 0.000

Kicukiro 0.0158 0.014 0.245 0.0140 0.0278 0.615 0.158463***

0.03087 0.000

Nyanza -0.0319** 0.013 0.016 0.4562*** 0.0365 0.000 0.346836***

0.03299 0.000

Gisagara -0.0088 0.015 0.559 0.6778*** 0.0250 0.000 0.325728***

0.03388 0.000

Nyaruguru -0.0255* 0.013 0.057 0.6934*** 0.0234 0.000 0.392901***

0.03197 0.000

Huye -0.0270** 0.012 0.027 0.5524*** 0.0314 0.000 0.371707***

0.03091 0.000

Nyamagabe -0.0362*** 0.012 0.002 0.5971*** 0.0293 0.000 0.379949***

0.03072 0.000

Ruhango 0.0123 0.016 0.444 0.3910*** 0.0388 0.000 0.359026***

0.0334 0.000

Bugesera -0.0073 0.014 0.597 0.3541*** 0.0379 0.000 0.270273***

0.03247 0.000

Muhanga -0.0113 0.016 0.471 0.3069*** 0.0417 0.000 0.367342***

0.0344 0.000

Kamonyi -0.0023 0.013 0.866 0.4766*** 0.0343 0.000 0.373348***

0.03087 0.000

Karongi -0.0066 0.015 0.663 0.4750*** 0.0364 0.000 0.401262***

0.03255 0.000

Rutsiro 0.0038 0.014 0.79 0.4753*** 0.0346 0.000 0.277749***

0.03267 0.000

Rubavu 0.0015 0.016 0.926 0.4020*** 0.0386 0.000 0.224179***

0.03448 0.000

Nyabihu -0.0177 0.014 0.196 0.5455*** 0.0327 0.000 0.375742***

0.03232 0.000

Ngororero -0.0539*** 0.010 0.000 0.5068*** 0.0330 0.000 0.292211***

0.03109 0.000

Rusizi -0.0085 0.012 0.495 0.4847*** 0.0329 0.000 0.323604***

0.03038 0.000

Nyamasheke -0.0256 0.013 0.056 0.4361*** 0.0368 0.000 0.279206***

0.03316 0.000

Rulindo -0.0439*** 0.013 0.000 0.4893*** 0.0356 0.000 0.253978***

0.0335 0.000

Gakenke 0.0047 0.015 0.753 0.4150*** 0.0371 0.000 0.288904***

0.03288 0.000

Musanze -0.0066 0.015 0.652 0.5855*** 0.0308 0.000 0.350147***

0.03294 0.000

Burera -0.0150 0.013 0.233 0.4549*** 0.0343 0.000 0.375946***

0.03045 0.000

Gicumbi -0.0091 0.018 0.622 0.1797*** 0.0382 0.000 0.179607***

0.03361 0.000

Rwamagana -0.0345*** 0.012 0.003 0.2884*** 0.0354 0.000 0.229414***

0.03023 0.000

59

Nyagatare -0.0459*** 0.011 0.000 0.4291*** 0.0361 0.000 0.264798***

0.03206 0.000

Gatsibo 0.0103 0.019 0.589 0.1427*** 0.0372 0.000 0.316131***

0.03297 0.000

Kayonza 0.4521*** 0.0377 0.000 0.305*** 0.03435 0.000Kirehe 0.3642*** 0.0376 0.000 0.241912**

*0.03229 0.000

Ngoma -0.0553*** 0.012 0.000 0.5257*** 0.0343 0.000 0.203153***

0.03376 0.000

ObservationPseudo R-square

295450.0055

324510.1327

322990.044

Source: Author’s own computation from the Household living condition survey EICV2 2005-2006. Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

60

Table A.6: Robustness check Full sample including Illness

FIVE Nearest Neighbor Radius Kernel OLS

ATT ATT ATT

Outcome Variables Treated Controls Difference T-stat Treated Controls Difference T-stat

Treated Controls Difference T-stat  Coeff.

(1) (2) (1)-(2) (1) (2) (1) - (2) (1) (2) (1)-(2)

Food consumption 238711.3 219287.5 19423.8 4.26 279674 310201 -30528 -3.66 238711 220182 18529.6 4.35 19473***

(4563) (8335) (4258) (3246)

Non-food consumption 279673.6 297654.3 -17981 -1.26 238711.3 239286.5 -575.2 -0.16 279674 297038 -17363.9 -1.41 -10598

(14323) (3526) (12304) (12627)

Poorest 0.1480 0.2078 -0.0598 -12.0 0.1480 0.2198 -0.0718 -19.8 0.1480 0.2109 -0.0630 -13.79

(0.0050) (0.0036) (0.0046)

Poor 0.1754 0.2220 -0.0466 -9.12 0.1754 0.2096 -0.0342 -8.98 0.1754 0.2181 -0.0427 -9.12

(0.0051) (0.0038) (0.0047)

Education Expenditures 36708.0 34142.4 2565.7 1.75 36708.0 35372.2 1335.8 1.3 36708.0 34297.2 2410.9 1.79  2550**

(1462) (1069) 91350)  (1081)

OOP Health expenditures 4225.0 5316.2 -1091 -3.03 4225.0 6662.1 -2437.1 -8.8 4225.0 5760.8 -1535.9 -4.05  -1542***

(1177) (276) (379)   (344.6)

School dropping 0.1325 0.1436 -0.0111

(0.0047)

-2.36 0.13251 0.14430 -0.0118

(0.0035)

-3.4 36705.85 0.1433 -0.0108

(0.0043)

-2.49

Source: Author’s own computation from the Household living condition survey EICV2 2005-2006, standard errors in parentheses

60

*** Statistically significant at 1%, ** at 5% and * at 10% level of confidence

61