Graduate School of Development Studies
Economic effects of Health Insurance in Rwanda :
Case of Community Based Health Insurance (CBHI)
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]
The Hague, The Netherlands
November, 2011
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 Studies
P.O. Box 29776
2502 LT The Hague
The Netherlands
Location:
Kortenaerkade 12
2518 AX The Hague
The Netherlands
Telephone:
+31 70 426 0460
Fax:
+31 70 426 0799
ii
Acknowledgments
I 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!
iii
Contents
List of Tables
List of Figures
List of Acronyms
Abstract
Chapter 1 Introduction
vi
vi
vii
viii
1
Chapter 2 Literature Review and analytical Frame work
5
2.1 Introduction
5
2.2 Effects of illness on household consumption and its implications for
poverty
5
2.3 Welfare effects of community based health insurance in developing
countries
10
Chapter 3 An Overview of health System in Rwanda
3.1 Introduction
3.2 Health system organisation
3.2.1 Public sector
3.2.2 Private sector
3.2.3 Traditional medicine
3.3 Health insurance system
3.4 Community based health insurance in Rwanda
3.4.1 Background of CBHI in Rwanda
3.4.2 Participation in community based health insurance
3.4.3 Community based health insurance organisation in Rwanda
3.4.4 Benefits package composition under CBHI
3.5 Health financing in Rwanda
15
15
15
15
15
16
16
17
17
18
18
19
20
Chapter 4 Data analysis and empirical strategy
4.1 Introduction
4.2 Data
4.2.1 Quantitative data
4.2.2 Qualitative data
4.3 Descriptive Analysis
4.3.1 Summary statistics of variables included in the analysis
4.3.2 Socio-economic differences between participants and nonparticipants
4.3.3 Poverty Incidence
4.3.4 Head of household health insurance membership
4.3.5 Age group
4.3.6 Health insurance Membership
iv
21
21
21
21
22
22
22
23
24
25
25
26
4.4
4.5
4.3.7 Health problems incidence
Qualitative sample
Model specification and econometrics concerns
4.5.1 Model specification: PSM model
4.5.2 CBHI effects on household consumption and impoverishment:
Constructing the Counterfactual
4.5.3 Community based health insurance participation model
4.5.4 Choice of variables for PSM
4.5.5 Household Consumption response to health problems
(OLS Model) 32
26
27
28
28
29
31
31
32
Chapter 5 Results Discussion
5.1 Introduction
5.2 Participation Characterizing Model
5.3 Balancing property
5.4 Effects of CBHI on household consumption and poverty (NNMPSM)
5.5 Robustness check and sensitivity analysis
5.5.1 Effects of CBHI on household consumption (OLS)
5.5.2 CBHI effects on household consumption and poverty (NNM
compared to Kernel, Radius, Probit and OLS estimates)
5.5.3 The Effects of CBHI on sub-groups of population
5.6 Qualitative analysis
34
34
34
35
Chapter 6 Conclusion
References
47
49
Appendices
53
v
36
38
38
40
42
44
List of Tables
Table 3.1: Community based health insurance membership trend in (%)
Table 3.2: Descriptive statistics EICV2, Insured and insured members
Table 3.3: Social-economic differences between insured and non-insured
members
Table 3.4: Poverty incidence and health insurance membership
Table 3.5: CBHI distribution by sex of head of household
Table 3.6: Percentage of prevalence of health insurance
Table 3.7: Percentage of people reporting illness
Table 5.1: Probit Model of determinants for Participation in CBHI
(Marginal effects)
Table 5.2: Community based health insurance Effects (PSM)
Table 5.3: Community based health insurance Effects (OLS)
Table 5.4 : Community based health insurance Effects (People who
reported sick)
Table 5.5: CBHI effects on outcome variables for different population
groups (full sample)
Table A.1 Descriptive statistics
Table A.2: Balancing properties of the matched samples (Reported sick)
Table A.3 Balancing properties of the matched samples (Full sample)
Table A.4: Community based health insurance Effects (OLS)
Table A.5: Community based health insurance Effects (Probit Model)
Table A.6: Robustness check Full sample including Illness
18
23
24
25
25
26
27
35
36
39
41
43
53
54
55
56
59
60
List of Figures
Figure 2.1: Simplified Flow-Chart of Key Issues Relating to the Economic
Consequences of Illness and the role of health insurance
Figure 2.2: Simplified Flow-Chart Relating health insurance to income
generation
Figure 4.1: Head of household age distribution
Figure A .1: Propensity Score Graph (Reported sick)
Figure A .2: Propensity Score Graph (Full sample)
vi
9
12
26
57
58
List of Acronyms
$
CBHI
CFH
CIA
EICV
GoR
Hh
IRST
MDG
MMI
MoH
NHA
NGO
NISR
NNM
OLS
OOP
PSM
RAMA
RWF
S.E
VUP
WHO
US dollar
Community Based Health insurance
Community Health Funds
Conditional Independence Assumption
Enquête Integral sur les Conditions des Vies des ménages
Government of Rwanda
Household
Institut de Recherche, Sciences et Technologie
Millennium Development Goals
Military Medical Insurance
Ministry of Health
National Health Accounts
Non-Governmental Organisation
National Institute of Statistics of Rwanda
Nearest Neighbour Matching
Ordinary Least Squares
Out Of Pocket
Propensity Score Matching
La Rwandaise d’Assurance Maladies
Rwandan Francs (Rwandan currency)
Standard errors
Vision 2020 Umurenge Program
World Health Organisation
vii
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 Studies
Community 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, 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.
Keywords
Community based health insurance, Food consumption, Education
expenditures, Health expenditures, Non-food consumption, School dropout,
Illness and Poverty.
viii
Chapter 1
Introduction
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 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
1
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 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).
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=4
9&Itemid=34 (accessed 15 April 2011).
1
2
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 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.
3
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 analyses the data and lay empirical strategy. A discussion of
the results is in chapter 5 while the last chapter draws a conclusion.
4
Chapter 2
Literature Review and analytical Frame work
2.1 Introduction
There 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 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.
5
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 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
6
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 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 and increases absenteeism among
the sick persons, which pose burden not only to the family but also to the
community.
7
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.
8
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
Income of ill
person
No
Indirect
costs
Family
members
time costs
Intra and inter-house
labor substitution
Hire labor and other
strategies
Reducing consumption
Impoverishment
Perceived
Illness
Seek care if
so, type of
costs
Use of savings made
Other Financial
costs
Direct
costs
Medical
Expenditures
Yes
Sale of Assets
Borrowings
Other strategies to cope
with financial costs
Health insurance
schemes
Source: Modified from McIntyre et al. (2006)
9
Protection against
catastrophic health
expenditures
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.
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.
3
For details see http://www.un.org/millenniumgoals/ (accessed on 14/11/2011)
10
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:
11
Figure 2.2: Simplified Flow-Chart Relating health insurance to income generation
Increased utilization
formal health care
Improvement in
health status
Low level of health
Expenditure and lost
Working days
Higher investment in
Human capital
Enhanced Income via
Higher productivity
Higher labor supply
Health insurance
schemes
Income above subsistence level
High income level
Higher investment in
physical capital
Reduced uncertainty
of health expenditure
Increased investment
in all form of capital
Source: Modified from (Hamid et al. 2011)
12
Reduction
vulnerability
of
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. 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
13
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 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.
14
Chapter 3
An Overview of health System in Rwanda
3.1 Introduction
The 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 organisation
The 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 sector
The 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 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 sector
The 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
15
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 medicine
According 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. 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.
See the detail on :
http://statistics.gov.rw/index.php?option=com_content&task=view&id=252&Itemi
d=307 (accessed on 22nd September 2011)
4
16
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 Rwanda
In 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
initiated 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 increase
financial sustainability of health services in Kabutare, 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 implementing and developing Community based
health insurance (Ibid). The discussions that were held between local population and health care representatives for CBHI implementation came out with a
formation of an insurance federation at district and fixed the annual premium
fees for enrolment 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 provider payment to the health centre (Ibid). In addition, district 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).
5
The exchange rate (RWF) in 2007 was 1 $ US = RWF 555,50
17
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 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
Coverage rate
2003
7
2004
2005
2006
2007
2008
2009
2010
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
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 District
6
18
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 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).
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
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 Ibid
11 Ibid
9
19
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 Rwanda
Reconciling 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 internal resources mobilisation,
by increasing government allocation to health and increasing prepayments into
health insurance schemes based on individuals ability to pay (Ibid).
12
Ibid.
20
Chapter 4
Data analysis and empirical strategy
4.1 Introduction
This 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 Data
This 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 data
The 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 6 900 households and
34 000 individuals. The collected information was in relation to households’
consumption expenditures on education, food and non-food items and out-ofpocket 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
EICV2 : Enquête Intégrale sur les Conditions de vie des ménages 2.
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 at
http://imisrwanda.gov.rw/redatam/RpHelp/EICV/survey0/data/docs/sampling/Sa
mples%20Megill%20Mission%201.pdf (accessed on 20 October 2011).
13
14
21
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 data
The 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-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 noninsured 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.
22
Table 3.2: Descriptive statistics EICV2, Insured and insured members
Noninsured members
Variables
Number of
Observation
Food consumption
19 783
239348.5
356733.8
Non-food consumption
19 783
310313.9
1352263
Health Expenditures
19 783
6665.34
Education expenditures
19 783
School dropout
17 971
Sex
Age
Mean
Std. Dev.
Insured members
Number of
observation
Mean
Std. Dev.
12 671
238901.4
356816.8
12 671
280167.6
669466.7
38876.64
12 671
4223.723
24629.69
35369.7
123195.7
12671
37101.58
105505.4
0.144288
0.35190651
11 576
0.122515
0.3390651
19 783
0.4739423
0.4993332
12 671
0.4743114
0.4993594
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
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 nonparticipants
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 noninsured 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
23
efficient estimates compared to OLS.
robustness check purpose.
However, the latter was used for
Table 3.3: Social-economic differences between insured and non-insured members
Variables
Treatment
Control
Difference(t-test)
Farm land
Livestock
0.9358377
0.8113803
0.8900066
0.7008543
0.0458312***
0.110526***
Gender
Age
Education expenditures
Health expenditures
Food expenditures
Non-food expenditures
1.525689
21.73404
37101.58
4223.723
238901.4
280167.6
1.526058
21.09159
35369.7
6665.34
239348.5
310313.9
0.0003691
0.6424446***
1731.885*
- 2441.618***
-447.1058
-30146.27**
School dropout
0.1325152
0.144288
-0.0117725***
Poorest
Poor
Illness
0.1478179
0.1752821
0.1799384
0.2197341
0.209644
0.2049338
-0.0719163***
-0.0343423***
-0.0249953***
Household size
6.275985
5.943285
0.3326999***
Male head of household
0.8117749
0.733913
0.077862***
Head of household age
No education
45.33675
0.349487
44.94197
0.3753412
0.3947828
- 0.0258542
Primary
Vocational
secondary
0.5667719
0.0213891
0.0575375
0.558993
0.0152664
0.0457992
0.0077789
0.0061227***
0.0117383***
University
0.0048145
0.0046001
0.0002144
Source: 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.
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)
15
24
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.
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
Non-insured
Total
Non-insured
CBHI member
CBHI member
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 group
The 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.
25
Figure 4.1: Head of household age distribution
6000
Frequence
5000
4000
3000
CBHI members
2000
CBHI Non-memberS
1000
0
15-24 25-34 35-44 45-54 55-64 65 -98
Age
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 immobilized households were
attributed to malaria. This may be due to the great incidence of malaria in the
26
country. Over 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
0.04
Intestinal 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 sample
Qualitative 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 semistructured 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
Gasabo District is located in capital city while Bugesera is situated in Eastern
province which is a rural area setting.
16
27
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.
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
28
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 used to
estimate the effects are from the same source and when the sample of nonparticipants 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 (noninsured 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 (Y1 | M  1, X i ) and not E (Y0 | 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.
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
17
29
However, the difference between E (Y0 | M  1, X i ) - E (Y0 | 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) (unconfoundedness) developed by Rosenbaum and Rubin
in1983 (Guo et al. 2006, Heckman et al. 1997), expressing that
[Y0 , Y1 ]  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 [Y0 , Y1 ]  M | X i , this will apply also for p( X i ) );
[Y0 , Y1 ]  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 consider the mean difference in outcome over the common support. Here, insured members have to be similar to
non-insured 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 insured 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 addition,
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 outcome between insured
and non-insured members:
ATT  E (Y1  Y0 | M  1, X i )  E (Y1 | M  1, X i )  E (Y0 | M  1, X 1 )............. ... (1)
member is insured and 0 otherwise. Therefore, referring to Heckman et al (1998), M
was used as compared to D.
30
Where ATT : is Average Treatment for treated group Y1 is the outcome of
CBHI insured households, Y0 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 , Ei , ui ) …………………………………… (2)
Where, the probability of household i to participate in community based health
insurance ( mi ) depends on household wealth ( wi ), 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 ( u i ) 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  wi  X i  H  Ei  u ............................................... (3)
*
*
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
31
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 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.
Ci    M i  wi  X i  H i  Ei   t ......................... (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 otherwise, a wealth dummy wi , taking into account if household possess
land and livestock or not and X i the household characteristics including, age
32
and gender, household size, age of head household, head of household sex, a
dummy related 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 E i
which considers the districts of residence of household member i and the error term  t . The estimation will be performed using separately food consumption from nonfood consumption education expenditures and OOP
health expenditures model. The model tries to control variables that can influence households’ consumptions and is performed for robustness check purpose.
33
Chapter 5
Results Discussion
5.1 Introduction
As 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. 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.
34
Table 5.1: Probit Model of determinants for Participation in CBHI (Marginal effects)
variables
Coefficients
Std. Err.
P>z
Variables
Coefficients
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
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
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
32 451
0.433
No education: reference
Nyaruenge: Reference
32 451
0.433
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
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 noninsured (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.
For detail see Heinrich, C., A. Maffioli and G. Vázquez (2010) 'A Primer for
Applying Propensity-Score Matching', SPD Working Papers.
18
35
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
Outcome Variables
Food consumption
Treated
Controls
(1)
(2)
226472.7
199624.5
ATT
Difference
T-stat
Obs.
(1) - (2)
26848
Treated
(1)
3.05
6 334
238714.8
Controls
(2)
221654
(8811)
Non-foodconsumption
272193.6
260014.4
12179
0.1304
0.2034
-0.073
0.33
6 334
279671.3
301215.9
0.1612
0.2097
-0.0486
-6.43
6 334
0.148
0.21
29915.3
22973.1
Health expenditures
6953.7
10515.3
6942
-4.15
6 334
0.1754
0.2161
2.65
6 334
36705.8
34743.6
4225.3
5736
6 334
-3.03
6 334
0.12985
0.15254
-0.0227
3.74
32 451
-21545
-1.49
32 451
-0.062
-12.4
32451
-0.0407
-7.96
32 451
1962
1.36
32 451
-3.69
32 451
-2.14
29 545
(1439)
(1177)
School dropout
17061
(0.0051)
(2621)
-3562
(1)-(2)
(0.005)
(0.0117)
Education Expenditures
Obs.
(14420)
(0.0113)
Poor
T-stat
(4565)
(36909)
Poorest
Difference
-1511
(409.8)
-2.08
5 628
(0.0109)
0.1325
0.1426
-0.0101
(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
36
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
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 suggest that insured
households spent much on education than non-insured members for people
who reported sick. However, for the full sample results show no significant
difference 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-insured
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 people
reported that with community based health insurance the health status improved, and secured income for alternative uses, including investment in human capital. Scheil-Adlung et al. (2006) revealed also that good health determine economic growth through improvement of labour productivity and education.
Community based health insurance constitutes an important tool of
preventing people from impoverishment resulting from health problems associated costs. In Rwandan case, the matching estimator reveals the differences
37
among insured and non-insured members in terms of poverty incidence. For
the poorest segment of people, the results reveal 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 insured
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 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 sample and people who were
reported sick. This implies that community based health insurance prevented
people from incurring catastrophic health expenditures, which are considered
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 impoverishment 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 consumption 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.
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
38
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 noninsured 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
Food
consumption
Nonfood
Consumption
Education
Expenditures
OOP Health
Expenditures
Insurance
19,473***
(3,246)
-5,343
(3,462)
299.9***
(83.57)
-738.9***
(111.3)
-50,607***
(3,229)
4,933
(3,179)
39,513***
(900.2)
138,721***
(10,672)
47,975***
(4,026)
-10,598
(12,627)
19,821
(17,374)
882.8***
(296.0)
-1,218***
(307.2)
-72,144***
(7,520)
18,778
(12,152)
80,490***
(4,020)
275,719***
(37,822)
73,798***
(12,044)
2,550**
(1,081)
-1,947*
(1,132)
130.8***
(28.54)
259.2***
(37.68)
8,606***
(1,228)
1,611
(1,119)
13,307***
(380.4)
33,799***
(4,379)
5,708***
(1,408)
-1,542***
(344.6)
5,457***
(560.8)
-15.18
(9.607)
27.21**
(13.01)
239.9
(500.1)
-72.62
(368.0)
788.1***
(96.25)
2,881**
(1,412)
1,765***
(653.1)
26,418***
(2,857)
79,216***
(18,799)
149,576***
(12,221)
507,466***
(50,064)
227,532***
(22,234)
Yes
64,916***
(11,314)
121,172***
(32,842)
362,448***
(32,525)
1.503e+06***
(174,565)
-159,168**
(71,280)
Yes
7,941***
(934.7)
27,563***
(5,346)
112,786***
(4,562)
390,800***
(33,221)
-79,466***
(8,408)
Yes
681.0**
(336.7)
2,167
(1,478)
3,189***
(1,194)
52,937***
(15,352)
2,167
(3,818)
Yes
Illness
Age
Head of Household age
Male head of Householde
Gender
Household size
Farm Land
Livestock
No education: reference
Primary
Vocational
Secondary
University
Constant
Districts dummies:
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 2005-2006,
standard errors in parentheses. Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
39
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 effects of community
based health insurance on different outcomes of our interest for people who
reported sick. Though the magnitudes of differences differ depending on techniques used, the results reveal similarity for different techniques 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 preventing excessive out of pocket health
payments and school dropout among insured people. The findings reveal also
similarities of CBHI effects on poverty reduction for different 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 performed and estimates are consistent with
PSM model. Moreover, all techniques used suggest a slight difference in nonfood consumption between insured household and non-insured household,
even though the difference was not statistically significant. In addition, table
A.4 shows that the results obtained after introducing illness variable still have
not significantly changed with respect to food consumption, health expenditures, school dropout and poverty. This confirms the conclusion drawn in section 5.2. Education expenditures were also found not significantly different
between insured and non-insured members.
40
Table 5.4 : Community based health insurance Effects (People who reported sick)
FIVE Nearest Neighbor
Radius
ATT
Outcome Variables
Treated
(1)
Food consumption
226472.7
Kernel
ATT
Controls
(2)
199624.5
Difference
T-stat
(1)-(2)
26848.2
Treated
(1)
3.05
226473
272193.6
260014.4
12179.2
Controls
Difference
( 2)
(1)-(2)
192143
34330
272193.6
248192.8
(36909)
Poorest
0.1304
0.2034
-0.0730
0.1612
0.2097
-0.0486
-6.4
0.1304
0.2234
29915.3
22973.1
OOP Health expenditures
6953.7
10515.3
6942.3
-4.15
0.1612
0.2195
0.0584
0.1298
0.1525
-0.02269
5.38
226473
(2)
197513
2.65
29915.3
21537.3
-3.03
6953.7
10611.1
8378.1
1.8
272194
271321
-13.1
0.1303
0.2005
0.1298
Coeff.
Coeff.
0.1434
-0.0136
(0.01090)
-7.5
0.1612
0.2154
4.4
29915.3
22367.6
-6.6
6953.7
9934.3
3.6
872.8
-0.0701
19473***
(3246)
0.03
-10598
(12627)
-6.8
-0.06***
-0.0543
(0.004)
-5.1
0.04***
7548
(0.004)
3.1
(2416)
-2981
-1.81
0.12985
0.1499
-0.0201
(0.0100)
2550**
(1081)
-2.96
(1008)
Source: Source: Author’s own computation from the Household living condition survey EICV2 2005-2006, standard errors in parentheses
41
28960
(0.011)
(0.00753)
*** Statistically significant at 1%, ** at 5% and * at 10% level of confidence.
(1)-(2)
(0.0102)
(551.7)
-2.08
T-stat
(29608)
(1923.7)
-3657.4
Difference
(8039)
(0.077)
(1176.8)
School dropping out
(1)
Controls
(0.0071)
(2621.1)
-3561.7
24000.8
-0.0930
0.0117
Education Expenditures
Treated
(13368.1)
(0.0113)
Poor
T-stat
(6377)
0.33
Probit
ATT
(8810.9)
Non-Food consumption
OLS
- 1542***
(344.6)
-2.01
-0.011**
(0.0042)
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 noninsured 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- PSM
Outcome Variables
Male head of household
Female head of household
Urban areas
Rural Areas
ATT
ATT
ATT
ATT
Treated
Controls
(1)
(2)
Food consumption
25117
230199
Non-food consumption
297773
325226
Difference
T-stat
Treated
Controls
(1)
(2)
3.77
185610
158586
-1.47
202320
175002
(1)-(2)
20918
(5547)
-27454
0.1460
0.1966
-0.0535
0.1730
0.2207
-0.0476
-5.4
0.1342
0.2461
35198
34115
1083
-8.24
0.1858
0.1976
3884
5970
School dropping
0.1314
0.1433
-2086
0.63
441991
31638
(0.0053)
(2)
27024
4.61
593260
554889
2.84
851206
862000
-4.52
5702
5377
-2.22
0.13774
0.13774
Difference
T-stat
Treated
Controls
(1)
(2)
2.21
171076
147555
-0.23
169284
310192
(1)-(2)
38372
(17393)
27318
-10794
-0.1119
-9.6
0.0247
0.0765
-0.0518
-1.08
0.1003
0.2161
-0.0096
8.3
0.1711
0.2358
4.6
107409
108751
-1342
-1.05
0.1895
0.2403
-0.3
11688
15102
0.12
0.1475
0.1609
(2907.6)
-3413
-0.21
23249
19086
(0.0012)
-0.0134
(0.0113)
Source: Author’s own computation from the Household living condition survey EICV2 2005-2006, standard errors in parentheses.
43
8.29
-21118
-1.5
-0.0647
-10.8
-0.0508
-8.51
4163
4.8
(873)
-1.84
28278
3689
-1.18
0.12940
0.13932
(1853)
0.0012
23521
(0.0060)
(6481)
-324
(1)-(2)
(0.0132)
(0.0091)
12553
T-stat
(14174)
(0.0062)
-0.0117
Difference
(2837)
(46012)
(2753)
(1304.4)
-0.0119
(1)
(1) - (2)
(0.0109)
(1725)
OOP Health expenditures
Controls
(0.0110)
(0.0058)
Education Expenditures
Treated
(9612)
(0.0055)
Poor
T-stat
(5861)
(18628)
Poorest
Difference
-862
- 3.0
(286)
-0.0099
(0.0052)
-1.91
5.6 Qualitative analysis
The 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 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
44
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:
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
45
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 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.
46
Chapter 6
Conclusion
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 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.
47
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.
48
References
Asenso-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): 3172.
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): 369378.
Diaz, J.J. and S. Handa (2004) 'Estimating the Selection Bias of Matching Estimators
using Experimental Data from PROGRESA', Department of Economics, University of
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.
49
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 PropensityScore 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 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.
50
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 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 CommunityBased 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.
51
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.
52
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 20052006.
53
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
Male head of
hh
Gender
45.5048
45.2107
45.4949
45.3528
0.1421
0.33
0.8039
0.7114
0.8037
0.8100
-0.0063
-0.5
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
54
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.
55
Table A.4: Community based health insurance Effects (OLS)
(1)
(2)
(3)
(4)
VARIABLES
Food
consumption
Nonfood
Consumption
Education
Expenditures
OOP Health
Expenditures
insurance
19,473***
(3,246)
-5,343
(3,462)
299.9***
(83.57)
-738.9***
(111.3)
-50,607***
(3,229)
4,933
(3,179)
39,513***
(900.2)
138,721***
(10,672)
47,975***
(4,026)
-10,598
(12,627)
19,821
(17,374)
882.8***
(296.0)
-1,218***
(307.2)
-72,144***
(7,520)
18,778
(12,152)
80,490***
(4,020)
275,719***
(37,822)
73,798***
(12,044)
2,550**
(1,081)
-1,947*
(1,132)
130.8***
(28.54)
259.2***
(37.68)
8,606***
(1,228)
1,611
(1,119)
13,307***
(380.4)
33,799***
(4,379)
5,708***
(1,408)
-1,542***
(344.6)
5,457***
(560.8)
-15.18
(9.607)
27.21**
(13.01)
239.9
(500.1)
-72.62
(368.0)
788.1***
(96.25)
2,881**
(1,412)
1,765***
(653.1)
64,916***
(11,314)
121,172***
(32,842)
362,448***
(32,525)
1.503e+06***
(174,565)
7,941***
(934.7)
27,563***
(5,346)
112,786***
(4,562)
390,800***
(33,221)
681.0**
(336.7)
2,167
(1,478)
3,189***
(1,194)
52,937***
(15,352)
49,583
(46,580)
296,305***
(65,104)
-454,968***
(37,310)
-519,911***
(36,666)
-551,856***
(36,413)
-484,400***
(36,414)
-522,041***
(37,493)
-459,827***
(36,894)
-512,621***
(35,794)
-465,954***
(37,250)
-141,217
(116,305)
-515,546***
(37,674)
-521,397***
(35,626)
-530,899***
(36,464)
-553,643***
(36,680)
-492,879***
(36,265)
-527,424***
(36,569)
-499,795***
(37,194)
-517,042***
651.5
(8,091)
9,550
(8,512)
-54,319***
(6,320)
-51,110***
(6,323)
-56,017***
(6,126)
-59,976***
(6,233)
-62,875***
(6,149)
-58,227***
(6,106)
-55,485***
(6,326)
-54,772***
(6,292)
-61,225***
(5,995)
-65,005***
(6,134)
-66,298***
(6,195)
-46,125***
(6,936)
-65,160***
(6,034)
-59,031***
(6,256)
-58,781***
(6,203)
-64,975***
(6,044)
-56,428***
3,662
(2,904)
-3,085
(2,678)
-11,970***
(2,274)
-12,629***
(2,258)
-10,408***
(2,370)
-11,513***
(2,269)
-12,201***
(2,262)
-10,011***
(2,289)
-10,198***
(2,310)
-11,446***
(2,290)
-11,140***
(2,257)
-11,568***
(2,287)
-4,508
(2,951)
-9,312***
(2,258)
-12,436***
(2,261)
-10,393***
(2,221)
-10,565***
(2,246)
-11,471***
(2,258)
-10,614***
Illness
Age
Head of HH age
Male head of Hhold
Gender
Household size
S8BQ1
S8SA
No education: reference
Primary
26,418***
(2,857)
Vocational
79,216***
(18,799)
Secondary
149,576***
(12,221)
University
507,466***
(50,064)
Nyaruenge district: Reference
Gasabo
-96,549***
(17,618)
Kicukiro
-33,312
(22,858)
Nyanza
-434,503***
(14,495)
Gisagara
-476,487***
(14,136)
Nyaaruguru
-486,469***
(14,062)
Huye
-429,632***
(14,673)
Nyamagabe
-442,686***
(14,566)
Ruhango
-439,280***
(14,997)
Muhanga
-417,094***
(14,530)
Kamonyi
-432,339***
(14,526)
Karongi
-433,048***
(14,952)
Rutsiro
-449,491***
(14,251)
Rubavu
-407,409***
(14,999)
Nyabihu
-449,808***
(14,534)
Ngororero
-499,219***
(14,115)
Rusizi
-401,166***
(19,667)
Nyamasheke
-447,106***
(14,315)
Rulindo
-463,539***
(14,463)
Gakenke
-503,970***
56
Musanze
Burera
Gicumbi
Rwamagana
Nyagatare
Gatsibo
Kayonza
Kiehe
Ngoma
Bugesera
Constant
Observations
R-squared
(14,472)
-435,153***
(14,201)
-494,528***
(14,248)
-472,031***
(14,134)
-371,148***
(16,076)
-452,298***
(14,221)
-454,329***
(14,361)
-414,653***
(14,824)
-464,259***
(15,059)
-448,139***
(14,361)
-441,724***
(14,112)
227,532***
(22,234)
(37,280)
-446,138***
(46,715)
-577,199***
(36,680)
-546,141***
(36,005)
-378,870***
(37,806)
-350,533***
(51,274)
-444,111***
(37,552)
-456,839***
(36,623)
-485,981***
(37,337)
-444,204***
(37,011)
-467,792***
(36,883)
-159,168**
(71,280)
(6,684)
-61,270***
(6,088)
-68,362***
(6,064)
-67,321***
(6,138)
-48,373***
(6,393)
-60,678***
(6,087)
-54,197***
(6,134)
-54,740***
(6,158)
-65,067***
(6,184)
-46,781***
(6,640)
-44,593***
(6,710)
-79,466***
(8,408)
(2,260)
-4,930*
(2,694)
-12,117***
(2,227)
-10,896***
(2,245)
-9,761***
(2,268)
-9,946***
(2,240)
-11,740***
(2,243)
-11,443***
(2,280)
-6,641***
(2,522)
-11,447***
(2,286)
-11,782***
(2,258)
2,167
(3,818)
32,451
0.383
32,451
0.120
32,451
0.284
32,451
0.046
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
0
Figure A .1: Propensity Score Graph (Reported sick)
.125
.25
.375
.5
Propensity Score
Untreated
Treated: Off support
.625
.75
Treated: On support
Source: Author’s analysis based on Household living condition survey EICV2 2005-2006
57
.875
Figure A .2: Propensity Score Graph (Full sample)
0
.2
.4
Propensity Score
Untreated
Treated: Off support
.6
Treated: On support
Source: Author’s analysis based on Household living condition survey EICV2 2005-2006
58
.8
Table A.5: Community based health insurance Effects (Probit Model)
School dropout
Poorest
Poor
Variables
Coeff.
Std. Err.
P>z
Coeff.
Std. Err.
P>z
Coeff.
Std. Err.
P>z
Insurance
Age
-0.0106 **
-0.0001
0.004
0.000
0.012
0.334
-0.0607***
-0.0002*
0.0041
0.0001
0.000
0.098
-0.0405****
-8.7E-05
0.00445
0.00013
0.000
0.505
Age of head of household
Male head of Household
Gender
Householdsize
Land
Livestock
Illness
No education (Reference)
Primary
Convetional
Secondary
University
Nyarugenge (reference)
Gasabo
Kicukiro
Nyanza
Gisagara
Nyaruguru
Huye
Nyamagabe
Ruhango
Bugesera
Muhanga
Kamonyi
Karongi
Rutsiro
Rubavu
Nyabihu
Ngororero
Rusizi
Nyamasheke
Rulindo
Gakenke
Musanze
Burera
0.0001
-0.0049
-0.0011
-0.0004
0.0057
-0.0057
0.0003
0.000
0.005
0.004
0.001
0.008
0.005
0.005
0.741
0.343
0.794
0.664
0.484
0.273
0.958
0.0001
0.0665***
0.0016
0.0181***
-0.0616***
0.1226***
-0.0151
0.0002
0.0051
0.0041
0.0009
0.0108
0.0050
0.0050
0.892
0.000
0.697
0.000
0.000
0.000
0.52
0.000556***
0.008876
-0.00057
0.007854***
-0.09155***
-0.00373
-0.00226
0.00017
0.00551
0.00437
0.00097
0.0116
0.00562
0.00548
0.001
0.107
0.897
0.000
0.000
0.506
0.68
-0.0035
0.0127
0.0021
0.0127
0.004
0.016
0.010
0.029
0.421
0.428
0.828
0.665
-0.0287***
-0.1092***
-0.1377***
-0.1450***
0.0043
0.0091
0.0046
0.0086
0.000
0.000
0.000
0.000
-0.01517***
-0.07151***
-0.11021***
0.00463
0.01412
0.00781
0.001
0.000
0.000
0.0067
0.0158
-0.0319**
-0.0088
-0.0255*
-0.0270**
-0.0362***
0.0123
-0.0073
-0.0113
-0.0023
-0.0066
0.0038
0.0015
-0.0177
-0.0539***
-0.0085
-0.0256
-0.0439***
0.0047
-0.0066
-0.0150
0.012
0.014
0.013
0.015
0.013
0.012
0.012
0.016
0.014
0.016
0.013
0.015
0.014
0.016
0.014
0.010
0.012
0.013
0.013
0.015
0.015
0.013
0.582
0.245
0.016
0.559
0.057
0.027
0.002
0.444
0.597
0.471
0.866
0.663
0.79
0.926
0.196
0.000
0.495
0.056
0.000
0.753
0.652
0.233
0.0658**
0.0140
0.4562***
0.6778***
0.6934***
0.5524***
0.5971***
0.3910***
0.3541***
0.3069***
0.4766***
0.4750***
0.4753***
0.4020***
0.5455***
0.5068***
0.4847***
0.4361***
0.4893***
0.4150***
0.5855***
0.4549***
0.0281
0.0278
0.0365
0.0250
0.0234
0.0314
0.0293
0.0388
0.0379
0.0417
0.0343
0.0364
0.0346
0.0386
0.0327
0.0330
0.0329
0.0368
0.0356
0.0371
0.0308
0.0343
0.019
0.615
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.153343***
0.158463***
0.346836***
0.325728***
0.392901***
0.371707***
0.379949***
0.359026***
0.270273***
0.367342***
0.373348***
0.401262***
0.277749***
0.224179***
0.375742***
0.292211***
0.323604***
0.279206***
0.253978***
0.288904***
0.350147***
0.375946***
0.02822
0.03087
0.03299
0.03388
0.03197
0.03091
0.03072
0.0334
0.03247
0.0344
0.03087
0.03255
0.03267
0.03448
0.03232
0.03109
0.03038
0.03316
0.0335
0.03288
0.03294
0.03045
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
Gicumbi
Rwamagana
Nyagatare
Gatsibo
-0.0091
-0.0345***
-0.0459***
0.0103
0.018
0.012
0.011
0.019
0.622
0.003
0.000
0.589
0.1797***
0.2884***
0.4291***
0.1427***
0.0382
0.0354
0.0361
0.0372
0.000
0.000
0.000
0.000
0.179607***
0.229414***
0.264798***
0.316131***
0.03361
0.03023
0.03206
0.03297
0.000
0.000
0.000
0.000
0.000
0.4521***
0.3642***
0.5257***
32451
0.1327
0.0377
0.0376
0.0343
0.000
0.000
0.000
0.305***
0.241912***
0.203153***
32299
0.044
0.03435
0.03229
0.03376
0.000
0.000
0.000
Kayonza
Kirehe
Ngoma
Observation
Pseudo R-square
-0.0553***
29545
0.0055
0.012
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
59
Table A.6: Robustness check Full sample including Illness
FIVE Nearest Neighbor
ATT
Outcome Variables
Food consumption
Treated
Controls
(1)
(2)
238711.3
219287.5
Difference
T-stat
(1)-(2)
19423.8
4.26
Radius
Kernel
ATT
ATT
Treated
Controls
Difference
(1)
(2)
(1) - (2)
279674
310201
(4563)
Non-food consumption
279673.6
297654.3
Poorest
0.1480
0.2078
-17981
-1.26
238711.3
239286.5
-12.0
0.1480
0.2198
0.1754
0.2220
-0.0466
36708.0
34142.4
2565.7
-9.12
0.1754
0.2096
4225.0
5316.2
School dropping
0.1325
0.1436
-1091
1.75
36708.0
35372.2
238711
220182
-575.2
-0.0718
-0.0342
1335.8
-3.03
4225.0
6662.1
-2.36
0.13251
0.14430
-2437.1
-0.16
279674
297038
-19.8
0.1480
0.2109
(0.0047)
-8.98
(1)-(2)
18529.6
4.35
-17363.9
-0.0630
19473***
(3246)
-1.41
-10598
(12627)
-13.79
0.1754
0.2181
-0.0427
-9.12
(0.0047)
1.3
36708.0
34297.2
2410.9
1.79
91350)
-8.8
4225.0
5760.8
-3.4
36705.85
0.1433
-1535.9
-0.0108
(0.0043)
2550**
(1081)
-4.05
(379)
Source: Author’s own computation from the Household living condition survey EICV2 2005-2006, standard errors in parentheses
60
Coeff.
(0.0046)
(0.0035)
*** Statistically significant at 1%, ** at 5% and * at 10% level of confidence
T-stat
(12304)
(276)
-0.0118
Difference
(4258)
(1069)
(1177)
-0.0111
(2)
(0.0038)
(1462)
OOP Health expenditures
(1)
(0.0036)
(0.0051)
Education Expenditures
Controls
(3526)
(0.0050)
Poor
-3.66
Treated
(8335)
(14323)
-0.0598
-30528
Tstat
OLS
-1542***
(344.6)
-2.49